Conference Agenda

Marine plastic monitoring is a challenging task especially by satellite Synthetic Aperture Radar (SAR) sensor. In this paper, a study presented that to an experimental campaign conducted in July 2024 in a lake in Southern Italy. The experimental goal was meant at detecting aggregations of floating macro-plastic using X-band SAR satellite imagery and optical data acquired by drones.

The main results show that floating plastic produces a distinctive spectral signature in optical images, making it recognizable compared to natural debris and SAR data, acquired from the Italian Cosmo-SkyMed Second Generation (CSG) constellation, also show detectable signals in both co-polarized and cross-polarized channels, with significant contrast between plastic and the water surface. In particular, the co-polarized channel proved more effective for discrimination, while the cross-polarized channel provided additional information.

In summary, the paper highlights the importance of a multi-level approach, combining optical and radar data to improve the monitoring of marine plastic litter. The proposed techniques is meant to support marine plastic spatial coverage observations and therefore enhancing the effectiveness of operational strategies for mitigating marine pollution.

The study is part of the ECOMARE project, funded by the Italian Ministry of Research, and contributes to international research on remote sensing of plastic waste in the oceans.

Green macroalgae blooms, typically occurring in coastal regions such as Europe, North America, and Asia, are primarily driven by eutrophication. These blooms pose threats to marine ecosystems, coastal economies, and human livelihoods. Among them, the recurrent “green tide” in the Yellow Sea—dominated by Ulva prolifera—is particularly notable for its large spatial extent and persistent impacts. Given the rapid temporal evolution and vast coverage of green tide events, remote sensing has become a key monitoring tool. Optical sensors are most commonly used due to their interpretability easiness and open access, with medium-resolution sensors as the primary tools, often supplemented by high-resolution sensors especially for the initial and dissipation periods. However, their effectiveness is severely limited under the cloudy and foggy conditions prevalent during summer in the Yellow Sea.

Synthetic aperture radar (SAR) offers an alternative under such conditions, as it is insensitive to clouds and solar illumination. Operational SAR systems usually perform at L-, C-, or X-band frequencies. Previous studies using near-synchronous C-band SAR and optical observations have reached a consensus that green tide patches exhibit a higher normalized radar cross section (NRCS) than surrounding seawater, appearing as bright slicks in C-band images [1][2][3]. However, the signatures of green tide patches in L-band SAR remain largely unexplored. Understanding the multi-frequency SAR imaging characteristics of green tide patches is essential for multi-platform monitoring especially with respect of the improved temporal coverage.

This study systematically investigates the imaging characteristics of green tide in L- and C-band SAR. Four regions of interest (ROIs) were selected, each covered by near-synchronous JAXA ALOS-2 L-band SAR (HH+HV), ESA Sentinel-1 C-band SAR (VV+VH), and optical data (ESA Sentinel-2 or USGS Landsat). After standard preprocessing, SAR backscatter from green tide patches and adjacent seawater was analyzed, with optical imagery used to verify green tide presence.

Experimental results show that, for C-band SAR, green tide increased co-polarized NRCS by around 10 dB relative to seawater, while cross-polarized NRCS showed little or no increase. Consequently, green tide patches are bright on C-band imagery, with brighter features on co-polarized channels than on cross-polarized channels. The backscattering mechanism of green tide patches in C-band is dominated by surface scattering, with the higher roughness of green tide patches compared to the calm seawater surface causing an apparent increase in co-polarized NRCS.

However, L-band SAR exhibited more variable behaviors, i.e., green tide patches can be either bright or dark patches. In one case, green tide patches increased co-polarized NRCS by around 6 dB while no impact on cross-polarized one. All values remain below the system noise floor, indicating a very calm sea state and a roughness-dominated scattering like that of C-band. In other cases, green tide patches reduced co-polarized NRCS by around 8 dB, with sea surface NRCS values mostly above or close to the noise floor. Given the longer wavelength of L-band, for the normal ocean state, the thin green tide patches above sea surface appearing relatively smooth behaving like oil spill with a suppression on Bragg waves of sea surface [5]. Therefore, a suppression-dominant surface scattering process is suggested for the cases with dark green tide patches.

References:

[1] Y. Guo, L. Gao, X. Li, “A deep learning model for green algae detection on SAR images,” IEEE Transactions on Geoscience and Remote Sensing, vol. 60, p. 4210914, 2022.

[2] L. Gao, X. Li, F. Kong, R. Yu, Y. Guo, Y. Ren, “AlgaeNet: A Deep-Learning Framework to Detect Floating Green Algae From Optical and SAR Imagery,” IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens., vol. 15, pp.2782–2796, Mar. 2022.

[3] L. Gao, Y. Guo, and X. Li, “Weekly green tide mapping in the Yellow Sea with deep learning: Integrating optical and SAR ocean imagery,” Earth System Science Data Discussions, vol. 2024, pp. 1–34, 2024.

[4] X. Pan, M. Cao, L. Zheng, Y. Xiao, L. Qi, et. al., “Remote Sensing of Ulva Prolifera Green Tide in the Yellow Sea Using Multisource Satellite Data: Progress and prospects,” IEEE Geoscience and Remote Sensing Magazine, vol. 12, no. 4, pp. 110-131, 2024.

[5] F. Nunziata, A. Buono, M. Migliaccio, “COSMOSkyMed Synthetic Aperture Radar data to observe the deepwater horizon oil spill,” Sustainability, vol.10, n.10, pp.3599-3624, 2018.

This study assesses and compares the accuracy of the widely used ocean wind vector products from two European C-band spaceborne radars: MetOp-A/B Advanced Scatterometer (ASCAT) and Sentinel-1A/B synthetic aperture radar (SAR) in Interferometric Wide (IW) mode. Two years (2019 ~ 2020) SAR/ASCAT winds are validated against the ground truth from the offshore/coastal meteorological buoys of National Data Buoy Center. Winds from radars and buoys are considered spatially matched if the buoy is located within the satellite wind cell, with a time difference limited to 30 min. To evaluate SAR/ASCAT winds against buoys in the same framework, we use 43 offshore buoy stations as common reference, and then validate them at comparable spatial resolution (25 km). Intercomparison results show that ASCAT winds exhibit higher accuracy level with root mean square error (RMSE) of 0.92 m/s (21.71°) in terms of the wind speed (direction). Regarding the Sentinel-1 SAR, offshore wind speed (direction) data downsampled to 25 km present an RMSE of 1.03 m/s (27.48°), showing very little change compared to 1 km products originally provided by ESA. Furthermore, wind accuracy degrades for Sentinel-1 SAR at both 1 km and 25 km resolutions when validated against coastal buoys, where ASCAT winds are usually unavailable. In general, our results demonstrate that the Sentinel-1 winds are as accurate as ASCAT in open ocean, while the performance of coastal SAR winds could be further improved. This finding could be helpful for better exploiting synergies in the ocean wind datasets from these two spaceborne radars.

Urban building change detection has a wide range of applications, including urban planning, disaster assessment, and digital map updating. Using remote sensing data (such as radar data and optical remote sensing data) to achieve urban change detection has become a common approach. Different modalities of remote sensing satellite imagery possess distinct characteristics: SAR satellites can acquire imagery under all weather conditions and regardless of illumination, as they are not affected by light source or atmospheric conditions; however, they are prone to speckle noise, and their backscatter intensity can be disturbed by building layout, orientation, and surface material. In contrast, optical satellites can capture abundant building texture and spectral information, although their acquisition may be compromised by cloud and haze interference. By synergistically utilizing multimodal remote sensing imagery, the complementary advantages of each modality can be fully exploited to improve the precision of urban building change detection. This study proposes a new building change detection method based on an adaptive-feature-fusion and dual-cross-attention-guided dual-stream network (ADDNet). In the proposed network, the dual-stream architecture is employed to extract features in a manner tailored to the distinct characteristics of SAR and optical remote sensing data. The cross-attention mechanism effectively fuses multi-scale features from both the encoder and decoder, reinforcing the complementary relationship between low-level details (such as edge textures) and high-level semantic context (such as global structure). Furthermore, an adaptive feature fusion module dynamically generates spatial weight maps to automatically regulate the contribution ratios of SAR and optical features, thereby achieving efficient integration of the complementary information.

This study obtained Sentinel-1 (S-1) and Sentinel-2 (S-2) satellite imagery data from the Shanghai area, comprising image pairs from two temporal phases (pre-change and post-change). The time span for the S-1 data is from 21 November 2021 to 23 December 2024, while the time span for the S-2 optical remote sensing data is from 24 November 2021 to 28 December 2024. Prior to training, the original data was processed following a standard preprocessing procedure. For S-1 Single Look Complex (SLC) data, a sequential process of multi-looking, radiometric correction, and geocoding was performed to derive intensity information. For S-2 multispectral data, atmospheric correction and radiometric calibration were applied. Subsequently, the processed SAR and optical remote sensing data were resampled to the same spatial resolution and geocoded to ensure complete spatial alignment. ADDNet was applied to detect building changes in Shanghai by employing a dual-stream network architecture. In this architecture, each stream is based on an encoder–decoder structure, with the encoder designed with specialized down-sampling convolution modules that are customized according to the data type (SAR or optical remote sensing data) for precise feature extraction. The encoder then performs up-sampling to restore the feature maps to the original resolution. Concurrently, the encoder outputs are refined using a cross-attention mechanism and merged with the decoder outputs via skip connections to better preserve low-level features. Finally, an adaptive feature fusion module is introduced to seamlessly integrate the complementary information from SAR and optical remote sensing data.

The evaluation results of ADDNet indicate that the proposed method is reliable for urban building change detection. Specifically, the precision, recall, and F1 scores all exceed 85%, and the overall accuracy surpasses 98%.

Keywords: building change detection; synthetic aperture radar (SAR); optical remote sensing; deep learning; multimodal remote sensing

Huaibei Plain is one of the main distribution areas of land subsidence in China. In the 1990s, the intensive exploitation of groundwater in the western part of the plain led to the formation of a regional subsidence bowl. In recent years, initial progress has been made in the control of groundwater overexploitation in this area, which has effectively alleviated land subsidence and even caused land uplift in some areas. The causes of ground deformation in the Huaibei Plain are complex, factors such as groundwater overextraction, coal mining, and sediment consolidation interact with each other, forming complex patterns of ground deformation driven by multiple factors. Therefore, identifying and analyzing different deformation patterns not only helps to reveal the cause mechanism of ground deformation in the Huaibei Plain, but also has great significance for preventing land subsidence disaster and enhancing urban public security.

The accuracy of monitoring ground deformation based on Multi-temporal InSAR (MT-InSAR) has reached the millimeter level, which can accurately capture small ground deformation. Compared with traditional measuring methods, which are difficult to achieve large-scale continuous monitoring, MT-InSAR has been developed as an important methods of regional ground deformation monitoring. Independent Component Analysis (ICA) can extract independent components from mixed signals. By analyzing the time series and spatial distribution of different ground deformation signals components and combining with multi-source data, the main influencing factors of different ground deformation patterns can be explored.

This study inverts the time series and rate field of ground deformation in the Huaibei Plain from May 2017 to October 2023, taking 572 Sentinel-1 satellite images as the data source, based on MT-InSAR method, and analyzes the distribution characteristics and spatio-temporal evolution of ground deformation in the Huaibei Plain. Using the Independent Component Analysis method, the elastic and inelastic ground deformation signals are effectively separated from the time-series field of ground deformation in the Huaibei Plain. In order to deeply understand and analyze the spatiotemporal heterogeneity of ground deformation in the Huaibei Plain, this study clusters the time series of ground deformation based on the time series similarity method. Then we observe 5 typical ground deformation patterns in the Huaibei Plain, including approximate uniform speed subsidence, decelerating subsidence, accelerating subsidence, uplift and elastic deformation of post-subsidence uplift. Combined with multi-source geographic big data, including clay content, land use type, precipitation, groundwater, mining area, building load, etc., we discuss the main influencing factors of different ground deformation patterns.

Keywords: Huaibei Plain, MT-InSAR, Ground deformation, ICA, pattern analysis

Satellite-based Synthetic Aperture Radar (SAR) remote sensing has fundamentally transformed the monitoring of Earth environments and geohazards, providing high-resolution data on inundation, land movement, and risk exposure that are essential for disaster response and planning [1], [2], [3]. In such context, several multi-temporal Interferometric synthetic aperture radar (MT-InSAR) techniques have been developed and successfully applied, mostly for the analysis of independently processed single-pol SAR datasets. To mitigate the noise effects in the differential SAR interferograms, several noise-filtering techniques have been proposed, most of them working independently on single interferograms. Subsequently, new efforts have been made to extend these methods to include space-time information [4], [5], [6]. The method in [6] complemented the extended minimum cost flow (EMCF) space-time phase unwrapping operations with an additional processing step that allowed an increase in the signal-to-noise ratio of a set of multi-temporal multi-looked interferograms by computing a set of optimized phases. The growing availability of SAR data, including dual-pol and full-pol datasets from current and future constellations, supports integrating polarimetric InSAR techniques to enhance interferogram coherence by leveraging spatial, temporal, and polarimetric information. Combining co- and cross-pol data further aids in distinguishing information sources. Furthermore, building on this foundation, our work explores the integration of multi-polarization data as a transformative approach to enhancing the detection, analysis, and forecasting of geohazard events. The problem we address is the inherent complexity in accurately characterizing and predicting geohazard phenomena due to challenges inherent to traditional remote sensing approaches. Our methodology involves combining multi-pol SAR data with high-resolution optical imagery. This general approach allows us to extract more detailed information on surface changes and environmental dynamics. By integrating diverse data sources, we overcome limitations of using a single sensor type, achieving a more comprehensive and nuanced understanding of dynamic Earth processes [1], [2], [3], [7].

REFERENCES

[1] R. Tomás e Z. Li, «Earth Observations for Geohazards: Present and Future Challenges», Remote Sens., vol. 9, fasc. 3, Art. fasc. 3, mar. 2017, doi: 10.3390/rs9030194.

[2] D. Xue, Z. He, e D. Hu, «Application of radar remote sensing in landslide geohazard risk
assessment», in International Symposium on Lidar and Radar Mapping 2011: Technologies and
Applications, SPIE, ott. 2011, pp. 579–584. doi: 10.1117/12.912922.

[3] J. Im, H. Park, e W. Takeuchi, «Advances in Remote Sensing-Based Disaster Monitoring and Assessment», Remote Sens., vol. 11, fasc. 18, Art. fasc. 18, gen. 2019, doi: 10.3390/rs11182181.

[4] A. Ferretti, A. Fumagalli, F. Novali, C. Prati, F. Rocca, e A. Rucci, «A New Algorithm for
Processing Interferometric Data-Stacks: SqueeSAR», IEEE Trans. Geosci. Remote Sens., vol. 49, fasc. 9, pp. 3460–3470, set. 2011, doi: 10.1109/TGRS.2011.2124465.

[5] G. Fornaro, S. Verde, D. Reale, e A. Pauciullo, «CAESAR: An Approach Based on Covariance Matrix Decomposition to Improve Multibaseline–Multitemporal Interferometric SAR Processing», IEEE Trans. Geosci. Remote Sens., vol. 53, fasc. 4, pp. 2050–2065, apr. 2015, doi: 10.1109/TGRS.2014.2352853.

[6] A. Pepe, Y . Yang, M. Manzo, e R. Lanari, «Improved EMCF-SBAS Processing Chain Based on Advanced Techniques for the Noise-Filtering and Selection of Small Baseline Multi-Look DInSAR Interferograms», IEEE Trans. Geosci. Remote Sens., vol. 53, fasc. 8, pp. 4394–4417, ago. 2015, doi: 10.1109/TGRS.2015.2396875.

[7] P. Euillades et al., «Recent advancements in multi-temporal methods applied to new generation SAR systems and applications in South America», J. South Am. Earth Sci., vol. 111, p. 103410, nov. 2021, doi: 10.1016/j.jsames.2021.103410.

Under the background of global warming, extreme climate events are becoming more frequent, and the average sea level keeps rising, posing severe challenges to human society's survival and development. Over the past 40 years, sea levels along the Chinese coast have been accelerating, and with the rapid urbanization process, the impact and risk of sea level rise on coastal areas have further increased. Shanghai is one of China's important coastal mega-cities, with a high concentration of population, economy, and building disaster-bearing bodies. Shanghai has long been affected by land subsidence and typhoon storm surges. In the future, under the combined effects of rising sea levels and storm surges, as well as the weakening of coastal protection capabilities due to continuous land subsidence, the risk of coastal flooding disasters and economic losses in coastal areas will further escalate, threatening the city's sustainable development. Conducting comprehensive risk assessments and economic loss assessments for Shanghai under future disaster coupling scenarios is an urgent need to enhance Shanghai's disaster prevention capabilities and promote the construction of resilient cities.

Based on Sentinel-1A satellite remote sensing data, this study obtained the surface deformation rate of Shanghai from 2017 to 2021. Combined with future sea level rise and water increase data of extreme storm surge events, under the scenarios SSP1-1.9, SSP1-2.6, SSP2-4.5, SSP3-7.0, and SSP5-8.5, We use a hydrodynamic model to simulate the extent and depth of coastal flood inundation in Shanghai under coupled disaster scenarios in 2030, 2050 and 2100. Based on simulated coastal flood inundation depth risk indicators and vulnerability and exposure indicators for buildings, a comprehensive risk assessment index system and model for individual building structures in mega-cities were constructed, and a comprehensive risk assessment was conducted for over 1.2 million buildings in Shanghai (excluding Chongming District) under different disaster scenarios in the future. This study also evaluated the direct economic losses that may result from dike overtopping and breaching in 2030, 2050, and 2100 under different disaster scenarios, and used a multi-regional input-output model to assess the associated economic losses of Shanghai's disaster at the provincial scale.

Under future disaster scenarios of sea level rise, land subsidence, and storm surges, there is a severe risk of coastal flood disasters in Shanghai's coastal areas, which may result in significant economic losses. In scenarios with low emissions and controlled socioeconomic development, such as the SSP1-1.9 scenario, the impact range and resulting economic losses of disasters are smaller compared to other scenarios. The comprehensive risk assessment model based on the scale of individual buildings can finely distinguish and locate high-risk buildings and analyze the reasons for high-risk levels based on vulnerability, exposure, and hazard dimensions. The use of multi regional input-output models achieved an assessment of associated economic losses at the provincial level in China, with results consistent and similar to other scales of associated economic loss studies. This study can provide scientific data and references for targeted disaster reduction strategies, disaster prevention and mitigation in mega-city, and the construction of safe and resilient cities.

Keywords: Land subsidence; Sea level rise; Coastal flooding; Comprehensive risk assessment; Economic loss assessment

The estimation of sea surface carbonate system parameters is critical for understanding ocean acidification dynamics and its ecological impacts, particularly in semi-enclosed marginal seas like the South China Sea (SCS), where complex biogeochemical processes and heterogeneous environmental controls pose significant challenges to quantifying spatiotemporal pH variability. Sea surface pH, a direct proxy for ocean acidification and sea carbonate system, reflects the cumulative effects of anthropogenic CO₂ uptake and natural variability, with studies showing temperature increases driving pH decreases. Traditional empirical methods relying on single-predictor relationships, such as temperature-salinity parameterizations, often fail to capture these multidimensional drivers, while sparse in situ observations further limit accurate assessments. However, generating high-resolution, satellite-derived pH time-series products remains challenging due to insufficient data and the intricate coupling between pH and environmental variables like SST, SSS, chlorophyll-α, and pCO₂.

This study addresses the challenges of estimating sea surface pH in the South China Sea (SCS) by developing a novel two-step machine learning framework integrating clustering-regression algorithms. To overcome the limitations of traditional data-driven approaches constrained by insufficient spatiotemporal training data, we established a remote sensing inversion model using Gradient Boosting Machine (GBM) with six predictive variables: longitude (LON), latitude (LAT), sea surface temperature (SST), chlorophyll-a concentration (Chla), mixed layer depth (MLD), sea surface height anomaly (SSHA) and sea surface salinity (SSS). Through twelve systematic training experiments evaluating three machine learning algorithms and four input parameter configurations, the GBM-based model demonstrated optimal capability in capturing spatial heterogeneity and physicochemical-biological interactions governing pH variability.

Sensitivity analysis revealed differential parameter contributions: SST and SSS dominated spatial gradients (explaining 38% and 24% of variance, respectively), while Chla and MLD modulated biological effects (19% and 12%). The model achieved validation errors of <0.04 for pH, with spatial homogeneity metrics (R²=0.89±0.05 for pH) confirming robust performance across oligotrophic basins and coastal fronts.

The derived 0.25°×0.25° monthly pH dataset (2000-2022) exhibits accuracy comparable to CMEMS and OceanSODA products respectively, particularly in resolving Pearl River plume impacts (ΔpH=0.12±0.03 during summer monsoon) and winter upwelling signatures. This framework advances regional carbonate system monitoring by synergizing underway pH observations, satellite biogeochemical parameters, learning techniques, providing critical baselines for quantifying ocean acidification trends in marginal seas under climate change. This pioneering approach establishes a paradigm shift in ocean acidification monitoring by overcoming historical limitations in pH data availability through intelligent fusion of underway measurements, satellite observations, and machine learning. The operational framework supports improved quantification of marine carbon fluxes and ecosystem vulnerability assessments under climate change scenarios.

Within the framework of the ESA-MOST Dragon 6 cooperation, this research addresses the critical need for advanced monitoring tools to support sustainable agriculture and understand environmental change, particularly focusing on soil characterization. The upcoming Copernicus Land Surface Temperature Monitoring (LSTM) mission, an ESA High Priority Candidate Mission scheduled for launch around 2028-2029, promises to significantly advance Earth observation capabilities in the thermal infrared (TIR) domain. LSTM will provide TIR data with a spatial resolution of 50 m and a revisit frequency of 1-3 days, enabling field-scale monitoring of Land Surface Temperature (LST) and derived evapotranspiration (ET). Beyond its primary agricultural applications, LSTM explicitly includes "mapping and monitoring soil composition" as a key objective, highlighting its potential for detailed soil property assessment. This study investigates the potential of LSTM data for improving the retrieval of key soil biophysical parameters, specifically soil texture, Soil Organic Carbon (SOC), and calcium carbonate (CaCO3) content, within the agriculturally significant Jolanda di Savoia site in the Po Valley, Northern Italy (44.87°N, 11.97°E).. On this study site, an HyTES (https://hytes.jpl.nasa.gov/) survey occurred on 22/06/2023. At the time of the survey 29 soil samples were collected and characterized with wet analysis and emissivity retrieval (i.e. Nicolet FTIR).

While Land Surface Temperature (LST) is an Essential Climate Variable, Land Surface Emissivity (LSE) is linked to surface composition. A fundamental challenge in TIR remote sensing is the accurate separation of LST and LSE which is an inherently ill-posed problem requiring N+1 unknowns to be solved from N measurements. Accurate atmospheric correctionis a prerequisite for reliable LST/LSE retrieval. Algorithms like the Temperature Emissivity Separation (TES) rely on empirical relationships, such as the one linking minimum emissivity (εmin) and spectral contrast (Maximum-Minimum Difference, MMD), to constrain the solution. These empirical relations (e.g., εmin=a−b×MMDc with coefficients specific for each sensors) can be sensitive to noise and may not perform optimally for all surface types or sensor configurations.

This research will focus on modifying and evaluating Temperature Emissivity Separation (TES) approaches and methods like In-Scene Atmospheric Correction (ISAC), to derive high-fidelity Land Surface Emissivity (LSE) spectra suitable for quantitative soil analysis.

The effectiveness of methods like ISAC for accurate atmospheric correction relies on subsequent scaling using radiative transfer models, such as MODTRAN, to obtain physically meaningful transmittance and path radiance values. This dependency introduces sensitivity to the accuracy of the atmospheric inputs used in the model simulations, with water vapor column amount being a particularly impactful parameter. To address this, our research aims to refine the atmospheric correction process by investigating a method to estimate the optimal water vapor scaling factor relative to a reference atmospheric profile. This involves exploring the potential use of normalized difference indices calculated from Modtran simulated transmittance and path radiance spectra within water vapor absorption regions, leveraging their correlation with water vapor content to select the most appropriate atmospheric state suitable for the correction.

Beyond atmospheric correction, another key challenge lies in the empirical relationships often used within TES algorithms to constrain the LST/LSE separation, such as the one linking minimum emissivity (ϵmin) and spectral contrast (Maximum-Minimum Difference, MMD). These relationships can exhibit variability depending on surface cover type. Therefore, we have investigate methods to refine this constraint by incorporating ancillary information derived by the LSTM mission's co-registered optical bands. Specifically, we will explore the NDVI index, calculated from LSTM's Visible and Near-Infrared (VNIR) bands, to update the ϵmin-MMD relationship, for instance, by developing vegetation-dependent parameters or stratifying the relationship based on NDVI values.

Furthermore, this research will explore the integration of spectral libraries (e.g., adapting concepts from algorithms like DirecTES) as an alternative or complementary approach to constrain the TES solution. The ultimate goal remains the derivation of accurate LSE spectra optimized for retrieving soil biophysical parameters.

The derived LSE spectra (8-12.5 µm) will be applied to explore the potential of the LSTM LWIR spectral range to characterize agricultural topsoil characteristics. In particular: soil texture (i.e. clay, silt and sand components), Soil Organic Content (SOC) and Calcium Carbonate (CaCo3) will be considered within this work.

The findings will contribute valuable insights into the capabilities of the LSTM mission for quantitative soil assessment, complementing ongoing analyses using existing VNIR/SWIR sensors (PRISMA, EnMAP) and airborne TIR (HYTES) at the same study site.

Crop mapping plays a key role in supporting sustainable land use by guiding policy decisions. It strengthens resilience against climate change and food security challenges by offering detailed information about spatial distribution and potential yields on a given region.

Crop type mapping by remote sensing is mostly divided into two categories: a) multispectral data-based studies, with a high temporal revisit time and medium to high spectral resolution and b) hyperspectral data-based studies, relying on finer spectral resolutions, but coarser temporal resolution from single- or multiple-date selective acquisitions. This poor temporal resolution does not allow the assessment of phenological development of crops throughout the growing season, potentially limiting the ability to capture critical growth stages required for accurate discrimination and monitoring. Nevertheless, the finer spectral resolution of hyperspectral data offers a valuable asset for discriminating between species, as it captures subtle differences in reflectance that are often indistinguishable with multispectral data.

In this context, this study explores the potential of EnMAP hyperspectral satellite data for in season winter cereal species classification in the Walloon Region (Belgium) in 2023 and 2024, benchmarking it against multispectral time series from Sentinel-2. This analysis has a major advantage: it has access to a robust dataset comprising thousands of agricultural parcels distributed across a wide geographical area. This extensive spatial coverage eliminates potential gradient environmental or meteorological effects, ensuring a robust evaluation. By leveraging such a comprehensive dataset, the study assesses the true capabilities of hyperspectral data in discriminating closely related cereal species and provides a rigorous comparison with the established performance of multispectral data.

The classification on multispectral and hyperspectral data are compared using a cross-validation on various classification algorithms, such as Random Forest (RF) and Support Vector Machine (SVM). Additionally, on hyperspectral data, multiple dimensionality reduction techniques, including Principal Component Analysis (PCA) and Partial Least Squares (PLS), are being explored.

Ultimately, this research aims to provide new insights into crop monitoring, setting a foundation for next-generation agricultural decision-making and precision farming, where hyperspectral imagery could redefine species-level mapping and monitoring across large-scale landscapes.

Nitrogen oxides (NO_x=NO+NO₂), as short-lived reactive gases, serve as crucial precursors for ozone (O₃) and secondary particulate matter (PM_2.5). With the continued global decline in anthropogenic NO_x emissions, the contribution from soil microbial processes has become increasingly significant, particularly in intensive agricultural regions where emission peaks may partially offset anthropogenic reduction achievements. However, current quantification of soil NO_x emissions remains subject to substantial uncertainties, and accurate estimation is essential for optimizing agricultural nitrogen management and developing effective air quality improvement strategies. This study employed Tropospheric Monitoring Instrument (TROPOMI) NO₂ observations from Sentinel-5P satellite, combined with the DECSO inversion algorithm and a grid-based approach incorporating land use types and climatic factors to estimate monthly soil NO_x emissions and their spatial distribution patterns in Eastern China in 2019. The results demonstrate that the satellite-derived soil NO_x emissions show consistency in both magnitude and spatiotemporal distribution when compared with bottom-up estimates from the CAMS global emission inventory. Validation confirms that satellite observations provide rapid and accurate quantification of monthly regional emissions of soil NO_x, offering scientific support for dynamic monitoring and precision management.

Keywords: soil NO_x emissions; satellite NO₂ observations; TROPOMI; remote sensing

Strengthening agricultural production management is an important measure to ensure national grain harvest. The Northeast region of China has state-owned farms that are modernized and the cornerstone of national grain production. The real-time information of crop cultivation practices from ploughing, sowing, managing to harvesting is crucial for the decision on agricultural production management. In recent years, real-time online free sharing of high-resolution satellite data has provided timely data support and the opportunity for the method development of crop cultivation practices information based on remote sensing images. Based on domestic high-resolution satellite data, European Sentinel satellite data, and other sources satellite data, this study integrated a Random Forest classification algorithm to construct a method for extracting key crop cultivation practices in the Jiansanjiang farms. From April to November 2024, 8-date remote sensing images were used, and key crop cultivation practices including paddy field preparation, rice transplanting, transplanted or directly sowed rice, distribution of paddy rice and dry land crops, autumn harvest, and land ploughing have been accurately extracted. The results show that field preparation for paddy rice began in early April and completed in early May, with clear progress in each field on different dates; The directly sowed rice field accounts for about 28.5% of the total rice planted area, mainly distributed in eastern farms; In autumn, the harvest starts in mid-September and ends in late October, and the ploughing followed until November 3rd. This study has promoted the deep application of remote sensing technology in the field of agriculture, and these monitoring information provide timely and scientific data support for agricultural production management, helping to discover problems in agricultural production in a timely manner, and assisting in the sustainable development and modernization of agriculture.

Cosmic Ray Neutron Sensing (CRNS) provides non-invasive soil moisture observations with a spatial footprint of 150–250 m radius and sensing depths of up to 50 cm. While CRNS stations provide continuous time series data at a certain location, mobile CRNS could extend the spatial coverage to tens of square kilometers, thereby bridging the scale gap between point measurements and remote sensing products. In Germany we operate both, networks of CRNS stations and CRNS on mobile platforms, such as cars and trains. These observations could serve as high-quality ground truth for calibrating and validating satellite-derived root-zone soil moisture products while also supporting hydrological modeling and data assimilation efforts. Our long-term goal is the development of high-quality spatial soil moisture products across Germany and Europe.

A key challenge in CRNS is the neutron’s sensitivity to hydrogen in vegetation. While there is a promising potential of CRNS to detect intercepted water in vegetated environments, the dry biomass itself could introduce a bias that needs to be corrected for. Especially in forested regions, high above-ground biomass introduces those systematic biases to CRNS detectors on passing trains. This research investigates potential correction approaches based on remote sensing-derived estimates of vegetation water content (VWC). We integrated multi-source datasets—Sentinel-2 and MODIS NDVI/LAI, DLR Forest Structure data for Germany, and CHIRPS precipitation—to model temporal and regional variations in VWC. The research aims to reduce biomass-induced uncertainties and enhance the reliability of CRNS as a reference dataset for Earth observation-based soil moisture products.

Hot temperature extremes are changing in intensity and frequency. Quantifying these changes is key for developing adaptation and mitigation strategies. The conventional approach to study changes in hot extremes is based on air temperatures. However, soil processes are triggered by soil temperature and it remains unclear whether it changes as does air temperature. Here, we demonstrate that hot extremes based on soil temperatures (soil hot extremes) are intensifying and becoming more frequent faster than air hot extremes over central eastern and western Europe. Based on existing model simulations, we also show that the increase in hot soil extremes could amplify or spread future heat waves by releasing sensible heat during hot days. In light of these results, future research on the impact of hot extremes on agriculture should be complemented with the analysis of soil hot extremes, since the fast intensification of soil hot extremes could pose high risk for crop failure.

Exchanges of water and energy between the lower atmosphere and the shallow subsurface are fundamental in governing the evolution of land surface conditions. Notably, energy exchanges have a significant role in the development of extreme events, largely due to the release of latent and sensible heat into the atmosphere. While surface latent and sensible heat fluxes have been extensively analyzed in the literature, our understanding of the magnitude and temporal dynamics of ground heat flux (GHF), representing conductive heat transfer within the ground, remains limited. The majority of existing GHF data are derived from long-term inversions of subsurface temperature profiles, providing estimates of GHF variability on decadal to millennial timescales. Direct, high-frequency measurements of GHF are sparse, and most available records are too short for robust climate analyses.

Here, we present a multi-layer perceptron model capable of predicting GHF at a daily resolution across geographically extensive regions. This model is trained using variables from remote sensing products as predictive features and direct GHF measurements from eddy-covariance sites as observational targets. We employ rigorous cross-validation methodologies to assess the robustness and generalizability of the model given the constraints of the observational database. Our preliminary results indicate the potential for this model to generate a spatially contiguous, gridded GHF product covering most of the global land surface.

The spectral invariants theory (𝑝-theory) has received much attention in the field of quantitative remote sensing over the past few decades and has been adopted for modeling of canopy solar-induced chlorophyll fluorescence (SIF). However, the spectral invariant properties (SIP) in simple analytical formulae have not been applied for modeling canopy fluorescence anisotropy primarily because they are parameterized in terms of leaf total scattering, which precludes the differentiation between forward and backward leaf SIF emissions. In this study, we have developed the canopy-SIP SIF model by combining geometric-optical (GO) theory to account for asymmetric leaf SIF forward and backward emissions at the first-order scattering and by modeling multiple scattering based on the 𝑝-theory, thus avoiding the dependence on radiative transfer models. The applicability of the model simulations especially over 3D heterogeneous canopies was improved by incorporating canopy structure through multi-angular clumping index, and by modeling single scattering from the four components of the scene in view according to the GO approach. The results show good consistency with both the state-of-the-art SIF models and multi-angular field SIF observations over grass and chickpea canopies. The coefficient of determination (R²) between the simulated SIF and field measurements was 0.75 (red) and 0.74 (far-red) for chickpea, and 0.65 (both red and far-red) for grass. The average relative error was approximately 3 % for 1D homogeneous scenes when comparing the canopy-SIP SIF model simulations to the SCOPE model simulations, and around 4 % for the 3D heterogeneous scene when comparing to the LESS model simulations. The results indicate that the proposed approach for separating asymmetric leaf SIF emissions is a robust way to keep a balance between satisfactory simulation accuracy and efficiency. Model simulations suggest that neglecting the leaf SIF asymmetry can lead to an underestimation of canopy red SIF by 6.3 % to 42.6 % for various leaf biochemical and canopy structural parameters. This study presents a simple but efficient analytical approach for canopy fluorescence modeling, with potential for large-scale canopy fluorescence simulations.

Remote sensing observations of variables affecting the hydrological water cycle like Surface Soil Moisture (SSM), land surface temperature (LST) and vegetation indexes (e.g., NDVI, LAI) are available at increasingly higher spatial, temporal and spectral resolutions. Integrating them into a robust hydrological modelling framework via direct input and data assimilation is a promising way to retrieve information about processes that are poorly gauged like irrigation. In this work, satellite LST and SSM retrievals over a heterogeneous agricultural area in Italy will be assimilated into a hydrological model in order to estimate irrigation volumes and evapotranspiration changes.

The employed hydrological model is the FEST-EWB (flash–Flood Event–based Spatially distributed rainfall–runoff Transformation – Energy Water Balance model), which is a distributed model that closes, at every time step, both the surface energy and water balances, computing LST as an internal variable. This allows the model to run on meteorological forcings and vegetation status information alone, and to use satellite observations of SSM and LST for other purposes, like calibration or data assimilation. In this work, irrigation will be derived from the model with three main strategies: (a) a volume equivalent to the Readily Available Water will be budgeted for every vegetated pixel whenever water stress conditions are met, following FAO guidelines and providing a baseline for unstressed irrigation management; (b) LST data from Sentinel-3 at 1 km spatial resolution and daily revisit time will be assimilated into the model, correcting the soil moisture status to match the thermodynamic conditions observed from satellite and budgeting irrigation water volumes for positive model LST biases; (c) SSM data from Sentinel-1 at 1 km spatial resolution and roughly 6 days of revisit time will be assimilated into the model, inducing a direct correction of the modelled soil moisture status and budgeting irrigation volumes for negative model SSM biases.

The final results are contrasted with official irrigation volumes provided by the local consortium, and show good agreement. However, the spatial gap between satellite observations (1 km²) and average size of the local plots (on average, 7 ha, 14 times lower) is a major issue in capturing the field-scale differences, considering that the area is highly heterogeneous, with many neighbouring fields running on different yearly cultivation schedules (both in terms of crop type and water management).

Modelling energy fluxes, and in particular the accurate estimation of evapotranspiration (ET), is essential in the current context of water resource management. Recognized recently as an Essential Climate Variable (ECV), ET plays a central role in understanding land–atmosphere interactions. Several approaches exist for modelling ET, among which the physically based energy balance methods stand out. These methods often utilize Land Surface Temperature (LST) data for model validation, enabling simulations that are independent of satellite-based ET products, in contrast to operational satellite-based ET models that rely directly on such products.

Furthermore, physically based approaches can also integrate LST data directly into the modelling framework, effectively combining the strengths of both physically based and satellite-based approaches. This work presents preliminary results using the physically based FEST-EWB (Energy-Water Balance) model, which was calibrated and validated at various Eddy Covariance stations across an aridity index gradient, at both daily and hourly timescales. LST data from the geostationary MSG satellite were used both as model input and for validation of the energy balance. Specifically, the FEST-EWB model continuously simulates soil moisture and ET over time and space, resolving land surface temperature by ensuring closure of the energy and water balance equations (Corbari et al., 2011).

The results not only highlight the differences between modelling approaches and the benefits of incorporating LST data but also evaluate the impact each method has on final ET estimates—particularly in arid regions, where increases in temperature or vapor pressure deficit (VPD) can have a significant influence on evapotranspiration.

Traditional methods of assessing irrigation efficiency and water resource carrying capacity based on evapotranspiration derived from remote sensing observations often overestimate irrigation efficiency because of the difficulty in effectively separating green water (water stored in soil from precipitation that sustains rainfed ecosystems and agriculture) and blue water (liquid freshwater resources in surface water bodies (rivers, lakes) and groundwater). In these methods, the contribution of precipitation is often included in the irrigation benefits, so that the irrigation efficiency and the carrying capacity of regional water resources cannot be accurately reflected. To overcome this limitation, this study developed a novel irrigation efficiency index based on blue evapotranspiration (BET). By partitioning evapotranspiration into green evapotranspiration and blue evapotranspiration using remote sensing data and the Budyko hypothesis, the actual contribution of irrigation water can be more accurately assessed. Based on this new index, the irrigation performance of arid northwest China was evaluated from 2001 to 2021. The results indicate that Ningxia has sufficient total available water resources to meet irrigation demand; Xinjiang's water supply and demand are in a fragile equilibrium; while the Hexi Corridor faces increasing risks of unsustainable water use and significant irrigation water deficits, and the water situation in the Hetao Irrigation District has shifted from a previous supply-demand balance to a situation where demand significantly exceeds supply, leading to increasing water scarcity. For the entire northwest region, the average irrigation efficiency from 2001 to 2021 is about 0.54, showing a significant upward trend. Notably, the irrigation efficiency in Ningxia, the Hetao Irrigation District, and Xinjiang has increased over the past two decades (2001-2021). The innovative BET index proposed in this study provides an important reference for a more accurate assessment of irrigation efficiency.

Leaf Area Index (LAI), a key parameter for assessing vegetation growth and ecosystem functionality, holds important value in agricultural management and ecological monitoring. This study introduces LAI-TransNet, a two-stage transfer learning framework that leverages field-measured data at the Unmanned Aerial Vehicle (UAV) scale to effectively extend LAI prediction capabilities to the PlanetScope satellite scale, enabling high-precision, large-scale LAI estimation and mapping without requiring field-measured data at the PlanetScope scale. The PROSAIL radiative transfer model was used to generate simulated canopy reflectance data, and to create a mixed dataset by combining these with field-measured LAI data, referred to as UAVSim-Measured data. These datasets were used for the training of the first stage of LAI-TransNet, involving the development and optimization of traditional machine learning models (including Random Forest, XGBoost, and LightGBM) and deep learning models (including Convolutional Neural Network (CNN), Multilayer Perceptron (MLP), and Transformer). The results indicate that traditional machine learning methods underperformed in the presence of data heterogeneity, whereas deep learning models, particularly CNN, excelled in transfer learning using UAVSim-Measured data, achieving superior performance (R² of 0.81, RMSE of 0.64 m²/m², and rRMSE of 11.47%). Building on this foundation, the second stage of LAI-TransNet further utilized simulated canopy reflectance data generated by the PROSAIL model based on the spectral response functions of PlanetScope satellite sensors, referred to as PSSim data, for transfer learning. By integrating cross-domain mapping and cross-scale feature enhancement strategies, the framework effectively bridged the data distribution gap between UAV and PlanetScope scales. The model demonstrated excellent performance on the validation dataset, with an R² of 0.96, an RMSE of 0.11 m²/m², and a cross-scale determination coefficient (R²) of 0.69 between UAV and PlanetScope LAI predictions, highlighting its robust cross-scale generalization capability. By enhancing spatial adaptability and eliminating the need for field-measured data at the PlanetScope scale, LAI-TransNet provides an efficient and scalable solution for large-scale crop LAI monitoring, demonstrating the substantial potential of PlanetScope satellite data for fine-scale agricultural applications.

Recently, solar-induced chlorophyll fluorescence (SIF) from satellites has shown potential for evaluating vegetation status and stress responses. Fluorescence quantum yield (Φ_F) is essentially linked to vegetation stress. However, the complex physiological and structural responses of SIF and Φ_F to drought need further study. This study normalized SIF as SIF_n to account for angular variations and fluctuations in photosynthetically active radiation (PAR), aiming for more accurate drought monitoring. SIF_n anomalies were compared to historical baselines (2019–2021 averages) of vegetation indices (VIs), raw SIF, and Φ_F during a 2019 drought in the North China Plain (NCP). The results show SIF_n provides an effective method for drought monitoring, showing the earliest decline compared to raw SIF, VIs, and Φ_F. In the first two weeks of drought, SIF_n decreased by 8.2%, 7.0%, 12.5%, and 8.2% across the four NCP subdivisions. SIF_n outperformed other indicators, proving sensitive to early drought detection. SIF_n was also examined for tracking drought alleviation by rainfall. The uncertainty under different viewing geometries was quantified. SIF_n anomalies showed a strong correlation with rainfall anomalies (R: 0.45 ~ 0.52) and meteorological factors like PAR (R: 0.80 ~ 0.84) and relative humidity (R:0.52 ~ 0.54). The correlation of near-infrared reflectance (NIRv) and Φ_F anomalies with SIF was weak during drought onset (R: 0.16 ~ 0.32) but strong at the end (R: 0.83 ~ 0.87). These suggest both canopy structure (mainly characterized by NIRv) and vegetation chlorophyll (Φ_F) are impacted by drought and influence SIF at different stages.

As a high-incidence disease, Southern Corn Rust (SCR) has seriously affected corn yield and food security. Conventional field investigations for assessing SCR severity are time-consuming and difficult to monitor at a large scale. Integrating Unmanned Aerial Vehicles (UAVs) and multispectral remote sensing technology provides a practical approach. In this manuscript, we proposed an Enhanced Disease Detection Index (EDDI) sensitive to SCR based on hyperspectral responses at different severity levels. Three machine-learning (ML) models were constructed based on four combinations of features: texture features, traditional vegetation indices (VIs), EDDI, and combined EDDI with texture features to classify severity into 1-5 levels. The results show that EDDI was highly correlated with SCR severity, and the Pearson Correlation Coefficient was -0.90. The EDDI-based ML model demonstrated a substantial improvement in F1 score, achieving an 11.7% to 24.1% enhancement compared to traditional VIs-based ML models. The accuracy of the XGBoost model improved most when combining EDDI and texture features, increasing its F1 score by 9.0% over EDDI alone and 102.8% over texture features alone. The XGBoost model combining EDDI and texture features had the highest accuracy, with an F1-score of 0.72. Based on this optimal model, the spatial distribution of SCR resistance breeding materials was presented, classifying five levels across the study area, with the highest proportion of Level 3 (34.0%), followed by Level 4 (25.2%). The model of EDDI combined with the texture features can efficiently evaluate SCR severity, supporting the development of precision agriculture through large-area field monitoring.

Agricultural irrigation water accounts for more than 50% of the total annual water use in China, and this has led to increasingly prominent issues of water scarcity issues, particularly in the endorsed basins in northwestern China. The Hexi Corridor, characterized by a arid climate and very limited precipitation, relies heavily on irrigation for the agricultural productivity, which has reduced the water available to ecosystem and resulted in serious water-use conflicts among different sectors. Accurate monitoring of actual irrigated cropland area and irrigation water use, to improve the regional water management, is therefore critical for sustainable development in this region. In this study, we have developed a remote sensing-based method to obtain the actual irrigated cropland area based on blue evapotranspiration (BET) and green evapotranspiration (GET) obtained from multiple remote sensing datasets, e.g., ETMonitor evapotranspiration (ET) data and CHIRPS precipitation data. The proposed approach uses the water-energy-coupled model to partition actual evapotranspiration into BET and GET, and then identifies irrigated areas using a thresholding method after analyzing the differences in BET between rainfed and irrigated croplands. Two water-energy-coupled models were tested in this study: the original parameter-free Budyko model (BD74) and the single-parameter Fu’s model (Fu81), and the parameter ω in the Fu81 model was calibrated following Zhou et al. (2025). The irrigated areas extracted based on the two models were compared and validated against statistical data from water resource bulletins, which shows that the irrigated area based on the Fu81’s result has higher accuracy. The actual volume of irrigated water was also obtained by multiplying the irrigated cropland area by the irrigation quota. On this basis, we analyzed the inter-annual characteristics variations of irrigated cropland area and actual irrigation water use across the Hexi Corridor.

Wheat rust disease has always been a global crop disease. In recent years, with the continuous changes in global climate, farming methods, etc., the hotspots of disease occurrence and the severity of the disease have undergone significant changes, posing a severe challenge to global food security. Therefore, conducting research on the global suitable distribution and evolution patterns of wheat rust disease is of great significance for ensuring global food security. This study divides the main wheat-growing areas worldwide into different research regions, with the occurrence and development of wheat rust disease as the core, combined with the distribution of wheat cultivation and the growing season, based on multi-source data such as meteorological data, soil and terrain data, and remote sensing data, using Lasso regression to process factors, and combining and integrating four algorithms including random forest algorithm, artificial neural network, gradient boosting machine, and maximum entropy model to construct ten species distribution models. The optimal model is selected based on the results of TSS and AUC indicators to evaluate the environmental suitability of wheat rust disease growth, and further combines the Hysplit model to simulate the risk diffusion of wheat rust disease.

The mountainous area of eastern Liaoning is the only debris flow disaster-prone zone in Northeast China. Houses in this area are predominantly built downstream in the valley along river sides, which are highly susceptible to debris flow hazards. This is particularly evident in the Wudaogou region, where multiple debris flow disasters have occurred during the past decades. Field investigations have revealed a significant presence of loose solid materials within the basin, posing a high risk of re-triggering debris flows. To identify potential hazard zones and estimate the damage conditions of buildings under debris flow impacts, numerical simulation using MassFlow and finite element modeling are conducted in this study. In the absence of detailed building damage information, we established a response model of buildings subjected to debris flows, achieving a quantitative risk assessment for debris flow disasters in the Wudaogou area. Using finite element modeling, we constructed a numerical model for the single-story brick-wood structures widely present in rural Northeast China. By quantifying the building damage degrees under different debris flow intensity combinations using inter-story drift angles, we developed a building vulnerability function correlating structural damage with debris flow intensity indices. This vulnerability function is then used to estimate the degree and extent of building damage in the study area under potential debris flows. Experimental results show that such buildings exhibit poor disaster resistance. When the debris flow velocity exceeds 6 m/s or the debris flow thickness reaches 1 m, these buildings are likely subjected to moderate or higher degrees of damage. When the velocity reaches 10 m/s or the thickness exceeds 2 m, these buildings face a risk of collapse. Statistical data on the evaluation results reveal that the most severe damage occurs to buildings located on the eastern side of the river and near the gully mouth. In the most extreme debris flow disaster scenario, 57% of the buildings would suffer varying degrees of damage and 14% would be destroyed by the debris flow. This comprehensive approach generated debris flow risk maps and potential building damage degrees effectively in the absence of detailed building damage information, providing valuable insights for debris flow risk assessment in the eastern Liaoning region.

The study area encompasses approximately 65 km² in the north-western sector of Alicante province, south-eastern Spain. Centred on the municipality of Villena and extending to its municipal boundary with Caudete, this region lies at an average elevation of 504 m a.s.l. Recent satellite-based ground motion observations have revealed that this area is affected by a progressive land subsidence process, the causes and consequences of which are the subject of the present study. Ground deformation was identified using the European Ground Motion Service (EGMS), a product of the Copernicus Land Monitoring Service that provides nationwide interferometric radar data based on long temporal series of Sentinel-1 images. The technique employed Persistent Scatterer Interferometry (PSInSAR) allows millimetric precision in measuring surface displacement over time. According to EGMS ascending and descending track data, the affected zone is subsiding at an average rate of approximately 9 mm/year. Analyses were conducted over two-time intervals: 2015–2021 and 2019–2023, both revealing a predominantly linear downward trend, suggesting continued subsidence beyond the last available date. From a geological perspective, the subsidence affects recent unconsolidated Quaternary sediments, composed primarily of conglomerates, sands, and clays. In contrast, the surrounding carbonate ranges (limestones and dolomites) and underlying Triassic gypsum and clay formations show no evidence of deformation. In some instances, slight uplift of Triassic units is observed, potentially indicating the rise of halite-rich materials beneath the gypsiferous layers. To assess the implications of this displacement on urban infrastructure, a comprehensive damage inventory was conducted in the city of Villena. A total of 128 buildings exhibiting structural damage were surveyed. Cracks were classified according to Cooper (2008), with the majority falling into level 3 (moderate damage) and several into level 4 (severe damage). The spatial distribution of the damaged structures was compared with the ground motion patterns derived from EGMS, aiming to determine potential correlations between subsidence intensity and damage severity. The findings underscore the relevance of InSAR derived data for detecting and monitoring ground displacements in urban areas, especially where subsidence may compromise infrastructure integrity. Moreover, they highlight the necessity for further hydrogeological investigations, as groundwater overexploitation is suspected to be a contributing factor in this case

Abstract：Forest parameters inversion is crucial for global ecological balance and carbon cycle monitoring. The Synthetic Aperture Radar (SAR) Tomography (TomoSAR) technique, which provides large-scale, penetrative, high-resolution three-dimensional (3D) observations under all-time and all-weather conditions, is of great significance for forest parameter inversion. Based on 12-track P-band airborne SAR data acquired over the Saihanba Forest in China, we performed TomoSAR 3D imaging of the forest following the procedures of co-registration, phase flattening, and phase calibration. We first compared the performance of different tomographic imaging algorithms, including Fourier transform, Capon, Multiple Signal Classification (MUSIC), and Compressive Sensing, in forested areas, highlighting their respective advantages and limitations. The underlying topography and forest height of the Saihanba Forest are retrieved based on the reconstructed tomograms. The root mean square error (RMSE) between the underlying topography derived from airborne TomoSAR and that from airborne LiDAR data is approximately 4.17 meters. The forest height estimated by airborne TomoSAR is between 0~25m, which is generally consistent with the SAR reference image in Pauli false color.

The main contributions of this study are as follows:

1. Based on 12-track P-band airborne SAR data acquired over the Saihanba Forest in China, we performed TomoSAR 3D imaging of the forest following the procedures of co-registration, phase flattening, and phase calibration.

2. We compared the performance of different tomographic imaging algorithms in forested areas, highlighting their respective advantages and limitations.

3. The underlying topography and forest height of the Saihanba Forest are retrieved based on the reconstructed tomograms.

Acknowledgments: The authors would like to thank Chinese Academy of Forestry for providing airborne P-band TomoSAR data.

The interferometric phase histogram (PH) technique typically requires only single-polarization and single-baseline data (or a limited number of interferograms) to retrieve low-resolution vertical forest structures and estimate forest physical parameters, such as sub-canopy terrain and forest height [1], [2], [3] . However, unstable airborne trajectories can lead to significant variations in the interferometric vertical wavenumber, potentially degrading inversion performance. To mitigate this issue, this study proposes an interferometric baseline optimization strategy. By constraining the phase height of ambiguity range, suitable baselines for PH-based inversion are selected, ensuring continuous vertical backscattering intensity distributions across the test site. Based on forest scattering mechanisms, three-dimensional parameters including forest height, intensity at specific height bin, and the volume-to-ground intensity ratio—can be derived from the backscattering intensity distribution. By incorporating these vertical structure-related parameters and integrating them with regression models, the estimation accuracy of above-ground biomass (AGB) is significantly improved.

This study aims to leverage the baseline-optimized PH technique to retrieve vertical forest structures from airborne SAR data and subsequently estimate key forest physical parameters. The specific objectives are:

1) To reconstruct vertical forest structures from airborne SAR data by applying the PH technique with optimized interferometric baselines.

2) To extract three-dimensional forest structural parameters from the PH-derived backscattering intensities and integrate them with multi-band, multi-dimensional regression models to enhance AGB estimation accuracy.

The proposed method is validated using tomographic SAR data from ESA’s airborne TomoSense campaign, which includes both P-band and L-band acquisitions. The TomoSense experiment flown in 2020 at the Kermeter area in Eifel National Park, North-West Germany. The analyzed dataset includes 28 P-band and 30 L-band overpasses in South-East flight heading[4].

This study first validates the capability of the multi-baseline optimized PH technique to reconstruct vertical profiles of forest scattering. Using the backscattering intensity distribution derived from the PH technique, canopy height, PH scattering intensity at specific height layers, and volume-to-ground intensity ratio parameters are extracted separately for P-band and L-band data. Subsequently, the effectiveness of AGB estimation is assessed by integrating multi-band and multi-dimensional parameters into multiple linear regression and random forest models.

Experimental results demonstrate that, compared to the scattering intensity profiles obtained from single-baseline PH technique, the multi-baseline optimized PH technique effectively mitigates discontinuities in histogram results caused by significant variations in airborne trajectories. The three-dimensional parameters derived from the PH scattering intensity distribution exhibit a strong positive linear correlation with AGB, confirming the validity of the proposed PH-derived three-dimensional parameters for characterizing vertical forest structures. Furthermore, compared to traditional regression-based AGB estimates relying solely on two-dimensional parameters, the integration of multi-band and multi-dimensional parameters into multivariate linear stepwise regression and random forest models improves AGB inversion accuracy by up to approximately 30%.

Reference

[1] G. H. X. Shiroma and M. Lavalle, “Digital Terrain, Surface, and Canopy Height Models from InSAR Backscatter-Height Histograms,” IEEE Trans. Geosci. Remote Sens., vol. 58, no. 6, pp. 3754–3777, Jun. 2020, doi: 10.1109/TGRS.2019.2956989.

[2] R. N. Treuhaft et al., “Vegetation profiles in tropical forests from multibaseline interferometric synthetic aperture radar, field, and lidar measurements,” J. Geophys. Res. Atmos., vol. 114, no. D23, Dec. 2009, doi: 10.1029/2008JD011674.

[3] Y. Lei, R. Treuhaft, and F. Gonçalves, “Automated estimation of forest height and underlying topography over a Brazilian tropical forest with single-baseline single-polarization TanDEM-X SAR interferometry,” Remote Sens. Environ., vol. 252, p. 112132, Jan. 2021, doi: 10.1016/j.rse.2020.112132.

[4] S. Tebaldini et al., “TomoSense: A unique 3D dataset over temperate forest combining multi-frequency mono-and bi-static tomographic SAR with terrestrial, UAV and airborne lidar, and in-situ forest census,” Remote Sens. Environ., vol. 290, p. 113532, May 2023, doi: 10.1016/j.rse.2023.113532.

Phase unwrapping (PU) is a critical step in Synthetic Aperture Radar Interferometry (InSAR) that directly affects the accuracy of deformation measurements. Traditional phase unwrapping methods, such as the Minimum Cost Flow (MCF) algorithm, often perform poorly in regions with large deformation gradients because the basic assumption of phase continuity is no longer valid. To address this issue, this study proposes a Phase Gradient Rate Constrained Minimum Cost Flow (PGR-MCF) method. The method utilizes the Phase Gradient Rate (PGR) extracted from time series differential interferograms as constraint information to optimize the construction of the unwrapping path. In the PGR-MCF method, high-coherence points are first selected based on a predefined coherence threshold, and a Delaunay triangulation network is subsequently constructed. The PGR is then computed by stacking the interferograms to reduce random noise. The resulting PGR is used to guide the construction of the unwrapping path, ensuring that regions with disrupted phase continuity due to high gradients are avoided. Specifically, areas where the PGR exceeds a preset threshold are assigned zero weight, thereby automatically excluding them from the unwrapping path. This strategy directs the unwrapping process from regions of lower deformation gradient into the interior of the deformation field, thus improving the reliability of the unwrapped results. The performance of the PGR-MCF method was verified through simulations and experiments with real data. Simulation results indicate that the PGR-MCF method achieves 99.5% phase unwrapping accuracy in high deformation gradient regions, effectively addressing problems that traditional methods struggle with. Results from the Guobu landslide case show that the proposed method reduces the root mean square error (RMSE) between the InSAR-derived deformation time series and Global Navigation Satellite System (GNSS) measurements by more than 70% compared to the conventional MCF method, demonstrating a clear advantage in complex environments. One significant benefit of the PGR-MCF method over traditional approaches is its ability to effectively relax the time baseline constraints typically required by MCF. In areas with large deformation gradients, this relaxation enables more effective handling of complex deformation fields, particularly when using SAR images with longer temporal baselines. Moreover, the PGR-MCF method does not depend on external data or prior models but rather relies solely on high-coherence, low-gradient regions to guide the unwrapping path, exhibiting strong adaptability and versatility across various SAR data types and environmental conditions. However, the PGR-MCF method does have certain limitations. For example, when the entire boundary of an area exhibits high PGR, the method finds it challenging to recover absolute phase values. Nevertheless, in regions with large deformation gradients, the method still demonstrates significant improvements in both unwrapping accuracy and deformation monitoring capabilities. Future research could further extend the applicability of this method to more complex deformation scenarios, such as nonlinear temporal deformations or cases with missing short time-baseline data. In summary, the PGR-MCF method provides a robust and reliable new approach for phase unwrapping, significantly enhancing the accuracy and reliability of deformation monitoring in real-world applications.

Persistent Scatterer Interferometric Synthetic Aperture Radar (PS-InSAR) technology has been applied to deformation monitoring for transportation infrastructures such as bridges, highways, and railways. However, extracting and interpreting information from the time-series deformation data obtained using PS-InSAR remains a complex and challenging task. In this study, the Lupu Bridge in Shanghai is selected as the research subject, and 18 ascending and 18 descending track Stripmap (SM) images collected by the COSMO-SkyMed satellite between January 13, 2009 and February 7, 2010 were utilized to monitor bridge deformation and quantify the influence of temperature on the observed deformation.

The time-series PS-InSAR technique was employed to preprocess the SAR image data through procedures including data co-registration, differential interferometry, flat-earth phase removal, and filtering, thereby extracting time-series deformation information. Subsequently, representative Persistent Scatterer (PS) points were selected based on the structural segmentation of the bridge for detailed analysis. Initially, the Seasonal-Trend decomposition using Loess (STL) method was applied to partition the time series into trend, seasonal, and residual components, facilitating the estimation of various periodic and trend components within the overall time-series deformation. To capture potential delayed effects, cross-correlation analysis and dynamic lag feature construction were utilized. A multiple linear regression model was then developed to quantify the influence of temperature on the bridge’s seasonal deformation by correlating the different periodic components with temperature variations. This approach transforms the complex effects of temperature on structural deformation into quantifiable engineering indicators.The results are summarized as follows:

1. The spatiotemporal evolution of deformation at the Lupu Bridge was obtained using the PS-InSAR technique. Between January 13, 2009, and February 7, 2010, the deformation rates of the Lupu Bridge ranged from −9.25 mm/year to 20.49 mm/year, exhibiting significant spatial variability. The mid-span region of the bridge experienced the largest variation in deformation rates. The deformation in the mid-span area is the result of multiple combined factors, among which temperature cycles and traffic loads are the primary contributors.The time-series deformation results derived from ascending and descending orbit PS-InSAR data show that the deformation rates of most PS points in the ascending imagery are concentrated between 11.64 mm/year and 20.49 mm/year, while in the descending imagery, most PS points exhibit deformation rates ranging from −9.25 mm/year to 4.63 mm/year.Due to the annual temperature difference in Shanghai reaching up to 40°C (with bridge deck temperatures exceeding 60°C in summer and dropping below 20°C in winter), uneven thermal expansion occurs between the bridge deck and the arch ribs, making them prone to differential deformation.

2. A joint analysis of time-series deformation and temperature variations was conducted to estimate the thermal expansion coefficients of the bridge. Significant spatial differences in these coefficients were observed across different regions of the bridge. Specifically, the thermal expansion coefficients of the Lupu Bridge ranged from –0.53 mm/°C to 1.98 mm/°C, indicating a spatial variation from the side spans to the mid-span. By correlating the seasonal component obtained through Seasonal-Trend decomposition using Loess (STL) with temperature variations, the influence of temperature on the bridge’s seasonal deformation was quantified. Analysis of representative Persistent Scatterer (PS) points on the Lupu Bridge revealed that the same-day temperature had a positive contribution to deformation at a rate of 0.63 mm/°C. Conversely, the temperature recorded one day later exhibited a negative correlation with the current deformation, with a 1°C increase in temperature leading to a decrease in deformation of approximately 0.1430mm.

As a Core Component of Modern Transportation Systems, High-Speed Railways Serve as a Vital Engine for Regional Economic Development and Social Connectivity. The China-Laos Railway, a Flagship Project Under China’s Belt and Road Initiative, Commenced Operation in December 2021, Connecting Kunming (China) and Vientiane (Laos). The Kunmo Section (Kunming to Mohan) Traverses the Geologically Complex Yunnan-Guizhou Plateau, Characterized by Rugged Terrain, Densely Distributed Fault Zones, Abundant Rainfall, Frequent Mountain Floods, and Geological Hazards. The Expanding Operational Scale of the Railway and the Intricate Geological Conditions Necessitate Systematic Monitoring and Evaluation of Surface Deformation and Potential Geological Hazards Along the Route, Which Is Critical for Ensuring Long-Term Operational Safety.

Synthetic Aperture Radar Interferometry (InSAR) Enables Large-Scale Surface Deformation Monitoring, Providing an Effective Approach for Detecting Deformation Anomalies and Identifying Hazards Along Linear Infrastructure. However, Conventional Time-Series InSAR Techniques Face Computational Challenges in Large-Scale Deformation Monitoring, Hindering Early Anomaly Detection. The InSAR Phase Gradient Stacking Technique Addresses This Limitation by Directly Extracting Deformation Signals From the Phase Gradients of Multi-Temporal Interferograms. Price and Sandwell First Applied Phase Gradient Methods for Deformation Monitoring. Lv Fu Et al. Developed Azimuth and Range-Direction Phase Gradient Stacking Algorithms. Ziming Liu and Teng Wang ^[3] map the Shear Strain Rates Along the North Anatolian Fault (NAF) by Stacking Phase Gradients of Sentinel-1 Interferograms, Provides a Valuable Reference for Deformation Monitoring of Similar Large-Scale Linear Infrastructures. Existing Studies Validate the Efficacy of Phase Gradient Stacking for Large-Scale Deformation Monitoring. Since the China-Laos Railway’s Operation in 2021, Systematic Research on Deformation Anomalies and Hazard Identification Along Its Route Remains Lacking. This Study Employs the InSAR Phase Gradient Stacking Technique To Monitor Deformation Along the China Section of the Railway, With the Following Objectives:

1. Construct a Recent Deformation Field (2022–2024) Using Sentinel-1 Descending Data From the European Space Agency (ESA), Generating a Deformation Map of the Entire China Section To Support Anomaly Detection and Hazard Identification.
2. Integrate Deformation Data, Optical Remote Sensing Imagery, and Geological Datasets to Preliminarily Identify Anomalies and Potential Hazards. Investigate the Developmental Characteristics and Triggering Mechanisms of Geological Hazards Along the Route, Analyzing the Spatial Distribution of Disaster-Prone Conditions and Contributing Factors.

Keywords: China-Laos Railway; InSAR; Deformation monitoring; Hazard identification

Author Contributions:

Yangwei Yu (Master’s student, Yunnan University): Research focuses on the application of spaceborne InSAR in geological hazard interpretation. Email: yuyangwei@stu.ynu.edu.cn

Mengshi Yang (Associate Professor, Yunnan University): Research focuses on radar remote sensing methodologies and applications. Email: yangms@stu.ynu.edu.cn

Ramon F. Hanssen (Professor): Research focuses on radar remote sensing, geodesy, geostatistics. Email: R.F.Hanssen@tudelft.nl.

Kang Yin (Master’s student, Yunnan University): Research focuses on the InSAR time series analysis for deformation monitoring and interpretation. Email: yink@stu.ynu.edu.cn.

Weibing Huang (Master’s student, Yunnan University): Research focuses on the application of spaceborne InSAR in urban infrastructure monitoring. email：huangweibin@stu.ynu.edu.cn

Hao Wu (Master's student, Yunnan University): Research focuses on InSAR data processing and polarimetric SAR data processing. Email: wh_ynu@163.com

Siran Chen (Master’s- student, Yunnan University): Research focuses on the application of InSAR in urban road research. Email: 18487172404@163.com.

References:

[1] Price, Evelyn J, and David T Sandwell. n.d. “Small‐scale Deformations Associated with the 1992 Landers, California, Earthquake Mapped by Synthetic Aperture Radar Interferometry Phase Gradients.”

[2] Fu Lv, Qi Zhang, Teng Wang, Weile Li, Qiang Xu, and Daqing Ge. 2022. “Detecting Slow-Moving Landslides Using InSAR Phase-Gradient Stacking and Deep-Learning Network.” Frontiers in Environmental Science 10 (August):963322. https://doi.org/10.3389/fenvs.2022.963322.

[3] Liu, Ziming, and Teng Wang. 2023. “High‐Resolution Interseismic Strain Mapping From InSAR Phase‐Gradient Stacking: Application to the North Anatolian Fault With Implications for the Non‐Uniform Strain Distribution Related to Coseismic Slip Distribution.” Geophysical Research Letters 50 (15): e2023GL104168. https://doi.org/10.1029/2023GL104168.

Point scatterers form the basis of many time series InSAR products. Taking differences between them results in double-difference observations, that can be used to provide estimates of cross-range position differences, relative displacement differences, and thermal differences. Yet, since every point scatterer stems from different situational circumstances, their behavior is far from uniform. In fact, this can vary significantly between points. Additionally, the quality of these scatterers can and will be different as well. We refer to this as the functional model and the stochastic model of a point scatterer, respectively.

In this study, we attempt to maximize information extraction for an arbitrary single point scatterer, assuming that this will eventually lead to an improvement of the information extraction for the ensemble of millions of point scatterers at a later stage. We distinguish several aspects. First, the 3D position of the point scatterer needs to be known in a terrestrila geodetic reference frame, to be able to compare it to objects in the surroundings. This also includes the uncertainty, i.e., stochasticity in this position estimate. We experiment with 3D virtual reality in a gaming environment to assess the scatterers in their 3D context. Second, we exploit the amplitude variability, including abrupt and significant changes, to provide a proxy for the quality of the interferometric phase, and detect potential changes in behavior. For the single scatterer, this temporal information is connected as attributes to the scatterer. Third, we estimate the relevant physical, geometric, or dynamic parameters for the point scatterer relative to another point scatterer. Finally, we link this information to relevant temporal and/or spatial contextual information.

We demonstrate how such an in-depth analysis of a single-point or single-arc InSAR dataset can lead to increased interpretability and better supported actionability. This is applied on test datasets including an Integrated Geodetic Reference Station (IGRS) that provides independent position estimates.

Assessing large-scale soil erosion is essential for designing and implementing spatially targeted mitigation strategies. Interferometric synthetic aperture radar (InSAR) offers significant potential for regional erosion monitoring. However, a comprehensive evaluation of its accuracy in quantifying erosion remains limited. This study proposes a framework for large-scale erosion assessment using multi-source data and systematically examines the capabilities and constraints of multi-temporal InSAR (MT-InSAR) in tracking erosion and deposition dynamics. 21 SAR images (July 2021–June 2022) in the Loess Plateau’s hilly and gully region were processed using MT-InSAR techniques. Topographic and geomorphological characteristics were extracted from high-resolution digital elevation models (DEMs) generated from unmanned aerial vehicle (UAV) laser scanning (ULS) point clouds to assess surface deformation linked to erosion and deposition. The erosion/deposition estimates derived from MT-InSAR were validated against field erosion pin measurements, DEM differencing (DoD) results, and precipitation data. MT-InSAR results aligned well with DoD outputs, though narrow gullies posed detection challenges. Notably, geomorphic changes at gully heads and slopes correlated strongly with precipitation. After excluding sites affected by InSAR decorrelation, MT-InSAR erosion/deposition measurements showed strong agreement with field data. In conclusion, MT-InSAR is viable for long-term erosion monitoring, but its effectiveness is hindered by decorrelation issues (e.g., low spatial resolution, inability to capture rapid/large changes). Future improvements could integrate multi-source data (e.g., L-band SAR, optical remote sensing) to enhance InSAR-based erosion monitoring accuracy.

Interferometric Synthetic Aperture Radar (InSAR) technology extracts land surface information by analyzing phase differences of SAR images captured at different times and locations. The interferometric phase is generated through conjugate multiplication of two complex-valued SAR images, with its principal value wrapped within the interval [-π, π]. Phase unwrapping reconstructs a continuous phase surface by estimating integer multiples of 2π jumps to recover the true phase gradient field. From a mathematical perspective, phase unwrapping is an NP-hard problem in the presence of noise and substantial topographic variations, and its solution is not unique unless further constraints are provided. Existing phase unwrapping algorithms have made certain assumptions about the unwrapping problem but still face two main challenges: noise and the inherent discontinuity of the deformation phase itself.

Given the impressive performance of deep learning in optical image processing, particularly its potential in addressing unwrapping challenges in low-coherence and discontinuous phase regions, we propose a two-stage deep learning approach. The first stage involves preprocessing the noisy interferometric images with a simplified fully convolutional network, OKNet, which effectively reduces noise and maintains the continuity of fringes. The second stage employs an improved SegFormer-based network, U-SegFormer, tailored for the phase unwrapping problem.

The OKNet module used for filtering and preprocessing was initially designed for image restoration, comprising an encoder, a decoder, and the Omni-Kernel Module (OKM) at the bottleneck. The two-channel image containing the real and imaginary parts of the InSAR image is first mapped into an embedding feature space of size C×H×W, and the encoder downsamples and extracts deep features. After processing through the OKM bottleneck layer, the decoder utilizes skip connections to restore the image to high-resolution features, ultimately generating the filtered output. The OKM bottleneck layer is a key component, consisting of three branches: the local branch which captures image details using 1×1 convolutions, the large-scale branch that employs a 63×63 kernel to extract widespread contextual features, and the global branch that captures global dependencies through Dual-Domain Channel Attention Module (DCAM) and Frequency-based Spatial Attention Module (FSAM).

Subsequently, we apply U-SegFormer to perform phase unwrapping on the interferometric phase patches processed by the filtering module. SegFormer is a simple yet efficient image segmentation framework that combines a transformer encoder with a lightweight MLP decoder. The encoder features a hierarchical transformer structure capable of extracting multi-scale features without positional encoding, thereby avoiding interpolation issues during model inference. The MLP decoder aggregates features from different levels to produce a powerful feature representation. However, the lightweight MLP decoder is too simple to capture the detailed texture features of InSAR phase patches, falling short of the accuracy requirements for phase unwrapping. Inspired by U-Net, we incorporated skip connections into the decoder, introduced the Effi-Transformer module for feature extraction, and added a Patch Expanding Module (PEM) for upsampling, modifying SegFormer into a U-shaped architecture for improved results.

We applied the U-SegFormer unwrapping method to both simulated and real datasets, comparing it with Minimum Cost Flow (MCF), Least Squares (LS), and Graph Cuts Based Phase Unwrapping Algorithm (PUMA). U-SegFormer demonstrated good unwrapping accuracy and Root Mean Square Error (RMSE) performance on high noise and discontinuous simulated datasets, although it slightly lagged behind MCF and PUMA in structural similarity. On real datasets, U-SegFormer produced more continuous unwrapped phases compared to traditional methods.

In summary, the SegFormer-based single-stage unwrapping network achieved relatively accurate and continuous unwrapped results. However, its structural similarity index (SSIM) was slightly lower compared to conventional methods, suggesting potential partial degradation of phase fringe consistency in the unwrapped outputs. Future studies may explore hybrid approaches integrating this network with traditional methods (e.g., SNAPHU), aiming to enhance precision through explicit prediction of inter-pixel discontinuities.

Efficient monitoring and prediction of airport reclamation area settlement is crucial for the safe construction and operation of airport facilities. In recent years, synthetic aperture radar interferometry (InSAR) technology has developed rapidly. Due to its ability to obtain surface deformation information over a large range, all-weather, and with high precision, it has become one of the important means of monitoring ground settlement in reclamation areas.

Therefore, based on PS-InSAR technology, this study selected Shanghai Pudong International Airport (SPIA) and Dalian Jinzhou Bay International Airport (DJBIA) for long-term settlement monitoring. The results can be used to analyze the spatiotemporal characteristics of settlement in the airport reclamation area and provide effective support for the operation and maintenance of reclamation airports.

For the SPIA, Sentinel-1 images from 2017 to 2024 and TerraSAR-X images from 2013 to 2016 were processed for a twelve-year time series sedimentation result, as shown in Fig.1.

Based on the distribution and changes of the settlement rate in different time periods of SPIA, three areas with significant settlement were selected, namely Runway 4, Runway 5 and the outer embankment area of the reclamation area, for temporal and spatial characteristics analysis.

First, the deformation level is defined, 10 mm/yr~−10 mm/yr for slight deformation, −10 mm/yr~−30 mm/yr for general subsidence, and −30 mm/yr~−50 mm/yr for severe subsidence. All PS points on Runway 4 have been slightly deformed since 2021, and the deformation in this area has tended to be stable. About 70% of the PS points in the Runway 5 area are stable with slight deformation, and nearly 30% are general subsidence. The number of PS points with general subsidence in the outer embankment area of the reclamation area has increased year by year, and the number of PS points with severe subsidence has decreased year by year. In 2021, the number of PS points with severe subsidence has decreased to 0. The risk of subsidence in the Runway 5 and outer embankment areas of the reclamation area is high and requires continued attention.

For the DJBIA, Sentinel-1 images from 2017 to 2023 were processed for a long- time series sedimentation result, as shown in Fig.2.

Dalian Jinzhou Bay International Airport is located in an artificial island reclamation area. Spatially, the settlement of the embankment increases with the distance from the land due to the change in sediment type in the bay. The large-scale filling area exhibits significant settlement due to soil consolidation, with the main settlement concentrated in its central area. In the study area, the most severe subsidence is located in the middle of the airport reclamation area, where most of the cumulative subsidence exceeds 200 mm. Meanwhile, the maximum uplift in the study area is only about 16 mm. Temporally, the settlement in each part of the area shows a trend of gradually slowing down.

In order to further efficiently and accurately extract the time-series characteristics of airport settlement in reclamation areas and quickly predict the global settlement, this study trained the prediction model based on machine learning methods and zoned settlement time-series data, selected LSTM and CNN networks as the basic time-series prediction models, and carried out settlement time-series prediction research for two airports.

The ESA TomoSense campaign aimed to advance forest parameter retrieval techniques for future spaceborne SAR missions by acquiring multi-baseline, fully polarimetric L-band bistatic SAR data over Germany’s Eifel National Park. These acquisitions enabled three-dimensional forest structure analysis using SAR Tomography (TomoSAR). Key challenges included bistatic clock synchronization errors and trajectory inaccuracies, notably due to the absence of a time-synchronization link.

This paper introduces a four-step calibration methodology that integrates monostatic and bistatic interferometry with natural scatterers. First, clock mismatches are resolved by calibrating bistatic datasets against the corresponding monostatic reference. Second, single-baseline time-varying baseline errors are inverted using partially calibrated single-pass bistatic InSAR data. Third, multi-pass time-varying baseline errors are estimated through multi-baseline monostatic and bistatic InSAR data stacks. Finally, baseline-direction positioning errors are corrected while accounting for target height uncertainties.

The results validate the method’s efficacy, demonstrating enhanced interferometric coherence, improved tomographic reconstruction, and higher vertical resolution in fused monostatic-bistatic TomoSAR imaging.

Accurate detection of forest fire scars from Earth observation data is a critical task for the efficient management of wildfire disasters across wide areas such as the state of California. Post-fire satellite optical image of high-quality is usually difficult to acquire due to unfavorable weather conditions including smoke and cloud. Therefore, it has become an interesting research topic to exploit satellite SAR data as an alternative in forest fire scar detection since its acquisition is independent of weather condition and solar illumination. Existing studies mainly perform multi-temporal change detection upon the magnitude feature of SAR images to identify potential fire scars. However, due to the complex scattering mechanisms and geometric distortions in SAR images, false detection problems may occur in mountainous areas with significant terrain undulation. Meanwhile, the coherence feature derived by InSAR has also been employed for change detection. Nevertheless, coherence exhibits uncertainty due to the non-stationary properties of forest vegetation, and thus false alarms may become a dominant problem in the detection results.

To address above problems, we proposes a semi-supervised deep learning framework based on time-series SAR coherence analysis, which optimizes forest vegetation coherence prediction through a Convolutional-Recurrent Neural Network (CNN-RNN) temporal prediction model. The method utilizes the redundant characteristics of time-series SAR coherence, combines CNN for extracting multi-dimensional (spatial, temporal, and polarimetric) features, and models the dynamic evolution patterns of coherence sequences through RNN, thereby predicting vegetation coherence states under undisturbed conditions. The difference between predicted results and actual observation data is used to accurately extract change areas. Using the 2018 Camp Fire in California as the test area, experimental validation with Sentinel-1 time-series data shows: 1) Compared with traditional bi-temporal methods, the proposed framework significantly reduces errors introduced by uncertain expressions of vegetation coherence; 2) The CNN-RNN model incorporating multi-dimensional features ensures prediction accuracy while maintaining detail preservation capability. This study provides a new technical approach for high-precision fire monitoring in complex vegetation environments.

The goal of this work is to utilize GANs to create synthetic geomagnetic residual datasets that can, in long-term, be used for detecting potential pre-earthquake anomalies in Earth’s geomagnetic field. For the purpose of this study, geomagnetic data, collected in a 3 – year period prior to a major M6.0 earthquake episode in Arzak, China, on 19^th January 2020, was used. Similarly to previously performed smaller scale study, datasets collected by SWARM A and C have been merged for this period, and an 8-day window has been found to provide the optimal spatial coverage. The SWARM satellite dataset, initially corrected for Earth’s main magnetic field, was further processed to remove seasonal and daily geomagnetic variations. The resulting geomagnetic residual was used for the study. Standard GAN architecture was used to generate 8 – day rolling windows of synthetic data and interpolate missing sections resulting from satellite orbits. The approach was successful in generating synthetic geomagnetic residuals and reconstructing data on a 1 – day scale, despite the initial lack of spatial coverage due to missing data. The study of detecting pre-earthquake related anomalies using generated synthetic datasets is currently underway.

Accurate soil moisture estimation is critical for sustainable water management in agriculture, particularly in olive orchards where precise irrigation is essential to maintain yield and quality. With climate change exacerbating water scarcity, there is a growing need for advanced tools to optimize water use. Remote sensing technologies, such as Synthetic Aperture Radar (SAR), offer valuable capabilities for large-scale soil moisture monitoring. When combined with in-situ measurements and data-driven approaches like Artificial Neural Networks (ANNs), these methods can effectively address the complexity of heterogeneous agricultural landscapes.

This study integrates Sentinel-1 SAR imagery with ANN models to estimate soil moisture in olive orchards located in the Vilariça Valley, northeastern Portugal. Soil moisture was measured at a depth of 10 cm every 30 minutes from July 2020 to December 2021. Sentinel-1 SAR data were acquired in dual polarization modes (VV and VH), and additional synthetic bands were generated through arithmetic operations combining polarization and calibration metrics (Beta, Sigma, Gamma, Gamma TF), resulting in 24 input features per image. Two datasets were constructed to evaluate the impact of orbital geometry: (1) D1, consisting of 161 ascending-orbit images, and (2) D2, consisting of 246 images combining ascending and descending orbits.

The ANN regression model, configured with six hidden layers and validated using 20-fold cross-validation, achieved higher performance on the D1 dataset (RMSE: 2.78, R²: 0.69, MAPE: 8.26%) compared to the D2 dataset (RMSE: 3.96, R²: 0.59, MAPE: 12.41%). The decrease in performance for D2 is likely due to the increased variability introduced by mixed orbital geometries, underscoring the importance of dataset homogeneity for SAR-based soil moisture modeling.

This work demonstrates the potential of integrating Sentinel-1 SAR data with ANN models for accurate and scalable soil moisture estimation in olive orchards. Future efforts will aim to address dataset imbalances, incorporate topographic features, and apply advanced data augmentation techniques to enhance model robustness and transferability.

Interferometric Synthetic Aperture Radar (SAR)(InSAR) technology enables the acquisition of wide-area, high-precision surface deformation measurements and is widely applied in geological hazard monitoring ^[1]. However, its performance is limited by decorrelation problem arising from temporal changes in scatterer characteristics. The Phase Linking (PL) method, which recovers systematic phase series through multi-temporal interferometric phase analysis, has become a core technique for addressing the challenge of monitoring coherence-low surfaces in Time-Series InSAR (TSInSAR) ^{[2], [3]}. This report systematically reviews the research motivation, statistical foundation, methodological advancements, result analysis, and future trends of the PL method.

1. Research Motivation. Based on the analysis of InSAR interferometric phase components and decorrelation sources, this section explores the differences in phase consistency between Persistent Scatterers (PS) and Distributed Scatterers (DS), along with their underlying causes. This discussion highlights the necessity of research on the PL theory ^{[4], [5]}.

2. Statistical Foundation. The time-series interferometric scattering vector follows a multivariate complex circular Gaussian distribution, and the multi-look complex covariance matrix conforms to a complex Wishart distribution ^{[3], [6]}. These statistical properties provide a robust theoretical foundation for the parameter estimation techniques in the PL theory.

3. Methodological Advancements. Starting from the modeling of the time-series coherence matrix, this section emphasizes that the differences among various PL algorithms primarily arise from their weighting strategies and solution approaches ^{[3], [7-10]}. It also analyzes the theoretical advantages and limitations of the Maximum Likelihood Estimator (MLE).

4. Result Analysis. Through Monte Carlo simulation and real data experiment covering the Daguangbao landslide, this section evaluates the strengths and weaknesses of several mainstream PL methods in terms of phase optimization, deformation accuracy, and computational efficiency. Experimental results demonstrate that, compared to other methods, the Adaptive Phase Linking (AdpPL) method ^[10] can more closely approach Cramér-Rao Lower Bound, effectively recovering the dominant deformation fringe signal over a long temporal baseline of 744 days, albeit at a higher computational cost.

5. Future Trends. The PL methods have made significant progress in improving phase accuracy and algorithm efficiency, but they still face several challenges, including the low efficiency of regularized solutions and interference from physical factors like soil moisture. As spaceborne SAR satellites advance toward multi-band, multi-polarization, and high spatio-temporal resolution, the application of PL methods in multi-band data fusion, non-Gaussian characteristics of high-resolution imagery, and monitoring in vegetated areas will encounter new technical difficulties.

The data and code related to the experiments in this report have been uploaded to GitHub: https://github.com/shen-pengsar. We invite fellow experts to download and engage in discussions.

Key words: InSAR, interferometric phase optimization, maximum likelihood estimation, distributed scatterers, regularization, decorrelation

[1] Xue F, Lv X, Dou F, et al. A review of time-series interferometric SAR techniques: A tutorial for surface deformation analysis[J]. IEEE Geoscience and Remote Sensing Magazine, 2020, 8(1): 22-42.

[2] Guarnieri A M, Tebaldini S. On the exploitation of target statistics for SAR interferometry applications[J]. IEEE Transactions on Geoscience and Remote Sensing, 2008, 46(11): 3436-3443.

[3] Ferretti A, Fumagalli A, Novali F, et al. A new algorithm for processing interferometric data-stacks: SqueeSAR[J]. IEEE Transactions on Geoscience and Remote Sensing, 2011, 49(9): 3460-3470.

[4] De Zan F, Zonno M, Lopez-Dekker P. Phase inconsistencies and multiple scattering in SAR interferometry[J]. IEEE Transactions on Geoscience and Remote Sensing, 2015, 53(12): 6608-6616.

[5] Zebker H A, Villasenor J. Decorrelation in interferometric radar echoes[J]. IEEE Transactions on Geoscience and Remote Sensing, 1992, 30(5): 950-959.

[6] Goodman N R. Statistical analysis based on a certain multivariate complex Gaussian distribution (an introduction)[J]. The Annals of Mathematical Statistics, 1963, 34(1): 152-177.

[7] Fornaro G, Verde S, Reale D, et al. CAESAR: An approach based on covariance matrix decomposition to improve multibaseline–multitemporal interferometric SAR processing[J]. IEEE Transactions on Geoscience and Remote Sensing, 2014, 53(4): 2050-2065.

[8] Ansari H, De Zan F, Bamler R. Efficient phase estimation for interferogram stacks[J]. IEEE Transactions on Geoscience and Remote Sensing, 2018, 56(7): 4109-4125.

[9] Ansari H, De Zan F, Bamler R. Sequential estimator: Toward efficient InSAR time series analysis[J]. IEEE Transactions on Geoscience and Remote Sensing, 2017, 55(10): 5637-5652.

[10]Shen P, Wang C, An B. AdpPL: An adaptive phase linking-based distributed scatterer interferometry with emphasis on interferometric pair selection optimization and adaptive regularization[J]. Remote Sensing of Environment, 2023, 295: 113687.

Calibration and validation (Cal/Val) of satellite products is required before their use. Comparisons with collocated satellite and suborbital data allow for a comprehensive analysis of instrument performance and the identification of areas that need further calibration and processor improvements. To this end, suborbital collocated measurements play a key role. Research networks, such as ACTRIS (https://www.actris.eu/), EARLINET (European Aerosol Research Lidar NETwork; data.earlinet.org) and POLLYNET (https://polly.tropos.de/), provide ground-based collocated lidar measurements that can be used to evaluate the Aeolus, ATLID, and ACDL lidar and aerosol products.

The Mediterranean basin provides a complex aerosol-cloud environment for the exploitation of EarthCARE's capabilities. The PANhellenic GEophysical observatory of Antikythera, PANGEA, located in the centre of the Eastern Mediterranean basin, at the island of Antikythera (35.86°N Ν, 23.81°E Ε, elevation: 110 m a.s.l.) in Greece, is the main station of the National Observatory of Athens for atmospheric monitoring. PANGEA is a remote site, located in a hotspot region in terms of climate change. In terms of natural sources, the site is influenced by combinations of all distinct aerosol types and exhibits diverse characteristics and variable layering patterns in the atmosphere. PANGEA has a background of sea-salt particles and is a crossroad of air masses, significantly impacted by mineral dust from Africa, smoke from increasing forest fires and agricultural biomass burning, anthropogenic pollution from megacities, and volcanic ash from frequent Etna eruptions. There are very few locations on Earth that experience such complex aerosol and cloud structures, vertical layering, and aerosol mixtures that can dramatically influence cloud evolution, cloud lifetimes, and precipitation processes. In a nutshell, PANGEA is ideal for Cal/Val because: (a) the site is located in a hot-spot region in terms of climate change, (b) the location is representative of a wider region within the Mediterranean, and (c) PANGEA NRI participates in Environmental European RIs (such as ACTRIS, and ICOS) as well as the Greek NRI (i.e. PANACEA).

A complementary station to the PANGEA observatory is the Thessaloniki lidar station. It is situated in the Eastern Mediterranean, in a location where many different types of aerosols coexist. In Thessaloniki station, on top of the background aerosol layers observed in the region (as discussed for PANGEA), a background of continental aerosol is present in the boundary layer, attributed to mixtures of anthropogenic pollution.[1] Both PANGEA and Thessaloniki stations are used for the evaluation of space-based lidar missions. Currently, the stations support the evaluation of the Aeolus and EarthCARE lidar aerosol product. Within Dragon 6,[2] the measurements will be used for the evaluation of DQ-1 aerosol lidar products.

The ESA-JAXA EarthCARE satellite mission, launched in May 2024, delivers vertical profiles of aerosols, clouds, and precipitation properties together with radiative fluxes, utilizing an instrumental suite of a high spectral resolution ATmospheric LIDar (ATLID), a Doppler Cloud Profiling Radar (CPR), a MultiSpectral Imager (MSI), and a BroadBand Radiometer (BBR). The simultaneous measurements will be utilized to improve our understanding of aerosol-cloud interactions (ACI) and their radiative effects and to assess the representation of clouds, precipitation, aerosols, radiative fluxes, and heating rates in weather and climate models.

The Chinese atmospheric environment monitoring satellite DQ-1 was successfully launched on 16 April 2022. The DQ-1 equips five sensors, including an Aerosol and Carbon Detection Lidar (ACDL), a Particulate Observing Scanning Polarimeter (POSP), a Directional Polarization Camera (DPC), an Environmental Trace Gas Monitoring Instrument (EMI) and a Wide Swath Imaging system (WSI). As the primary payload among them, ACDL is a HSRL with two-wavelength polarization detection that can be utilized to derive the aerosol optical properties. The aerosol and cloud optical properties products of the ACDL include total depolarization ratio, backscatter coefficient, extinction coefficient, lidar ratio and color ratio.

The Aeolus satellite carries ALADIN (Atmospheric LAser Doppler INstrument), the first space-based High Spectral Resolution Lidar (HSRL) Doppler ever placed in space. ALADIN emits a linear polarized beam, which, after going through a quarter-wave plate, is transmitted with a circular polarization (at 355 nm) and receives the co-polarized backscatter component from molecules and particles or hydrometeors in two separate channels. The main product of the mission includes profiles of the horizontally projected line-of-sight winds, and spin-off products. As spin-off products, ALADIN provides the particle optical properties, namely the particle backscatter and extinction coefficients from particles and hydrometeors, known as Level 2A (L2A) products, which are separately retrieved through the Rayleigh (Ray) and Middle Bin (Mid Bin) algorithms.

For the validation of the Aeolus, EarthCARE, and DQ-1 products in the Mediterranean, collocated cases with the ground based lidar over PANGEA and Thessaloniki are used towards achieving the following objectives: (i) validate Aeolus and EarthCARE aerosol products using state-of-the-art ground-based, (ii) perform intercomparisons with DQ-1 aerosol products, and (iii) provide information for harmonizing and bridging past and future missions, to deliver Climate Data Records on aerosols. The rationale for this comparison is based on lessons learned from the JATAC campaign in the Atlantic [1, 2], and the comparison will be aligned with the Best Practice Protocol for the validation of Aerosol, Cloud, and Precipitation Profiles (ACPPV) project (https://zenodo.org/records/15025627). The presentation includes examples of the validation of Aeolus aerosol products and ATLID aerosol products using ground-based lidar measurements, as well as an intercomparison with DQ-1 aerosol products.

References

[1] Fehr, T., et al., “The Joint Aeolus Tropical Atlantic Campaign 2021/2022 Overview—Atmospheric Science and Satellite Validation in the Tropics”, In Proceedings of the EGU General Assembly 2023, Vienna, Austria, 24–28 April 2023. https://doi.org/10.5194/egusphere-egu23-7249, 2023.

[2] Marinou, E. Et al., “An Overview of the ASKOS Campaign in Cabo Verde”. Environ. Sci. Proc., 26, 200. https://doi.org/10.3390/environsciproc2023026200, 2023.

Acknowledgments: This research was financially supported by the PANGEA4CalVal project (Grant Agreement 101079201) funded by the European Union, the EarthCARE DISC project funded by ESA, and the CERTAINTY project (Grant Agreement 101137680) funded by the Horizon Europe program.

Spaceborne High Spectral Resolution Lidars (HSRLs) are capable of measuring global aerosol profiles. Before the operational application of spaceborne lidar systems, dedicated and strict calibrations and validations activities have to be conducted. This study will focus on the validation of three spaceborne lidar missions. Aeolus was launched on 22 August 2018 and finished its lifetime on 30 April 2023. The Aeolus satellite’s only payload ALADIN is a HSRL, working at the wavelength of 355 nm. It measured the global wind profiles and aerosol profiles for more than 4 years. The reprocessing of the Aeolus aerosol data products has been accomplished with the new processor version baseline 16 by 2024. The ESA’s EarthCARE satellite has been launched in May 2024. The lidar payload of the spacecraft is the Atmospheric Lidar (ATLID) with HSRL capability, working also at the wavelength of 355 nm. The ATLID provides vertical profiles of aerosols and thin clouds. At present, the commissioning phase of ATLID has been finished and the aerosol products have been published. The Chinese atmospheric environment monitoring satellite DQ-1 has running for about 3 years since 16 April 2022. The lidar equipped by DQ-1 is an Aerosol and Carbon Detection Lidar (ACDL). As the primary payload, ACDL is a HSRL with two-wavelength (532 nm and 1064 nm) and polarization detection. The aerosol and cloud optical properties products of the ACDL include total depolarization ratio, backscatter coefficient, extinction coefficient, lidar ratio and color ratio.

The procedures of the cross-validation of these three spaceborne lidars will be presented. The cross-validation will target some large-scale and representative cases, e.g., smoke over Amazon and Africa, dust over Sahara, high-altitude cirrus and polar stratosphere clouds. Track match (temporally and spatially), aerosol and cloud optical properties transfer from different wavelength (355 nm and 532 nm) data quality control are the key procedures of the cross-validation. The data products involved in the cross-validation will be Level 2A of ALADIN (extinction coefficient, backscatter coefficient, lidar ratio at 355 nm), Level 1 (attenuate backscatter coefficient at 355 nm) and Level 2A (extinction coefficient, backscatter coefficient, lidar ratio, depolarization ratio at 355 nm) of ATLID, Level 1 (attenuate backscatter coefficient at 532 nm) and Level 2A (extinction coefficient, backscatter coefficient, lidar ratio, depolarization ratio at 532 nm) of ACDL. The cross-validation results will be investigated and presented, including both profiles comparison and statistical results. The data assessment of the three spaceborne lidar will be conducted by the cross-validation and the aerosol and clouds detection performance of different wavelength will be explored.

In May 2024, the new Earth Cloud Aerosol and Radiation Explorer (EarthCARE, Wehr et al., 2023) satellite was launched. As a joint project of the European Space Agency (ESA) and the Japan Aerospace Exploration Agency (JAXA), this mission will provide unique new insights in the vertical structure of aerosol, clouds and precipitation properties in combination with measurements of radiative fluxes. The novelty of this mission is the synergy of four different instruments which are co-located onboard of EarthCARE. These are two active instruments – the ATmospheric LIDar (ATLID) and the Cloud Profiling Radar (CPR) – and two passive instruments – the Multi-Spectral Imager (MSI) and the Broad Band Radiometer (BBR). In this presentation, the focus will be on ATLID, which is a high spectral resolution lidar (HSRL) operated at 355 nm. With ATLID, direct measurements of the particle backscatter and extinction coefficients as well as of the lidar ratio, and the particle linear depolarization ratio are possible (Level 2a data, Eisinger et al., 2024).

To ensure an accurate operation of the instruments and a high quality of the retrieved products, intensive calibration and validation (Cal/Val) activities are required. The Leibniz Institute for Tropospheric Research (TROPOS) strongly contributes to the ATLID Cal/Val activities with its global network PollyNET (Baars et al., 2026), including ground-based lidar systems of the type PollyXT (Engelmann et al., 2016). These systems are multiwavelength-Raman-polarization lidars and enable measurements of the particle backscatter and extinction coefficients, the lidar ratio, and the particle linear depolarization ratio at 355, 532, and 1064 nm. Thus, a direct comparison with the ATLID Level 2a products is possible.

For the validation of the ATLID products, special focus is given on PollyXT ground-based observations from three stationary ground sites with diverse atmospheric conditions, which are Mindelo (Cabo Verde), Leipzig (Germany), and Dushanbe (Tajikistan). In the vicinity of Cabo Verde, a validation campaign called Organized Convection and EarthCARE Studies over the Tropical Atlantic (ORCESTRA, orchestra-campaign.org) was conducted in August and September 2024. Numerous reference measurements were collected via aircraft and shipborne observations and at the ground site at Mindelo, which contributed to ORCESTRA in the framework of the CLoud and Aerosol Remote sensing for EarThcare (CLARINET) sub-campaign. Another validation cruise of the research vessel Meteor, equipped with a PollyXT lidar, took place at the beginning of 2025.

We will present validation results based on case studies, using vertical profiles of aerosol optical properties of the ATLID extinction, backscatter and depolarization (A-EBD) product and the ATLID aerosol (A-AER) product (Donovan et al., 2024). In this early mission phase, all products, but especially the single-instrument ones (Level 1 and 2a) at the beginning of the processing chain (Eisinger et al., 2024), are continuously improved. This allows us to validate products which are further down in the EarthCARE processing chain. Therefore, aerosol layer information and mean optical properties from the ATLID aerosol layer descriptor (A-ALD) product (Wandinger et al., 2023) will be examined as well. The comparison of the different baselines will highlight the improvements in the data quality, due to updates in the processing algorithms for ATLID.

References:

Baars, H. et al.: An overview of the first decade of Polly^NET: an emerging network of automated Raman-polarization lidars for continuous aerosol profiling, Atmos. Chem. Phys., 16, 5111–5137, https://doi.org/10.5194/acp-16-5111-2016, 2016.

Donovan, D. P., van Zadelhoff, G.-J., and Wang, P.: The EarthCARE lidar cloud and aerosol profile processor (A-PRO): the A-AER, A-EBD, A-TC, and A-ICE products, Atmos. Meas. Tech., 17, 5301–5340, https://doi.org/10.5194/amt-17-5301-2024, 2024.

Eisinger, M., Marnas, F., Wallace, K., Kubota, T., Tomiyama, N., Ohno, Y., Tanaka, T., Tomita, E., Wehr, T., and Bernaerts, D.: The EarthCARE mission: science data processing chain overview, Atmos. Meas. Tech., 17, 839–862, https://doi.org/10.5194/amt-17-839-2024, 2024.

Engelmann, R., Kanitz, T., Baars, H., Heese, B., Althausen, D., Skupin, A., Wandinger, U., Komppula, M., Stachlewska, I. S., Amiridis, V., Marinou, E., Mattis, I., Linné, H., and Ansmann, A.: The automated multiwavelength Raman polarization and water-vapor lidar Polly^XT: the neXT generation, Atmos. Meas. Tech., 9, 1767–1784, https://doi.org/10.5194/amt-9-1767-2016, 2016.

Wandinger, U., Haarig, M., Baars, H., Donovan, D., and van Zadelhoff, G.-J.: Cloud top heights and aerosol layer properties from EarthCARE lidar observations: the A-CTH and A-ALD products, Atmos. Meas. Tech., 16, 4031–4052, https://doi.org/10.5194/amt-16-4031-2023, 2023.

Wehr, T., Kubota, T., Tzeremes, G., Wallace, K., Nakatsuka, H., Ohno, Y., Koopman, R., Rusli, S., Kikuchi, M., Eisinger, M., Tanaka, T., Taga, M., Deghaye, P., Tomita, E., and Bernaerts, D.: The EarthCARE mission – science and system overview, Atmos. Meas. Tech., 16, 3581–3608, https://doi.org/10.5194/amt-16-3581-2023, 2023.

The distribution of aerosols and clouds is one of the most uncertain factors in the atmosphere through direct or indirect forcing. The study of the macroscopic distribution, optical properties and interaction processes of aerosols and clouds can contribute to the understanding of atmospheric science and to effectively assessing the global climate change situation. Improving the observation capability of aerosols and clouds will reduce the uncertainty in climate model simulation and prediction systems. With the continuous development of satellite technology, spaceborne remote sensing is gradually playing an important role in atmospheric observation, and high spectral resolution lidar (HSRL) is an important instrument for achieving high-precision and high-resolution global comprehensive monitoring of aerosols and clouds.

China successfully launched the atmospheric environment monitoring satellite Daqi-1 (DQ-1) on 16 April 2022. This was done to meet two key objectives: first, to meet the need for remote sensing data in monitoring atmospheric environmental pollution; and secondly, to facilitate scientific research on global climate change. The main payload of DQ-1 is the Aerosol and Carbon Detection Lidar (ACDL), which uses 532nm and 1064nm wavelengths, with 532nm polarization detection capability and separates molecular backscatter signals using iodine molecular filters. Therefore, ACDL is able to achieve independent retrieval of optical parameters such as extinction coefficient, backscattering coefficient, depolarisation ratio, and lidar ratio for aerosols and clouds. The EarthCARE satellite, which has been launched in May 2024, also carries a high spectral resolution lidar (ATLID) as its primary payload. ATLID provides polarization and HSRL detection at 355 nm, with high-resolution profiling of optical and microphysical properties of clouds, as well as aerosol layers. In this paper, the observation capability and data accuracy of ACDL are verified through the joint satellite-ground comparison between ACDL and AERONET in optical thickness and other observation products. The ACDL and ATLID long-time global observation data products are also classified according to the polarization and extinction characteristics of different scenarios, such as dust, cirrus, and polar stratospheric clouds. With the ACDL aerosol classification and cloud phases classification algorithms completed in this research, the classification results of the two spaceborne lidar are compared and global distribution characteristics of different aerosol and cloud types are analysed. Based on the comparison results, the application capability of dual-wavelength HSRL detection in aerosol scene classification and cloud phases identification will be discussed in the subsequent study.

Synthetic Aperture Radar (SAR) and optical remote sensing imagery offer complementary perspectives of the surface, with SAR providing structural and textural information regardless of weather or lighting conditions, while optical imagery offers high-resolution spectral and spatial details. However, integrating these two modalities remains a significant challenge due to their fundamentally different imaging principles. Our experiment investigates the use of Conditional Generative Adversarial Networks (cGANs) for improving image matching between SAR and optical domains. The proposed framework aims to translate SAR imagery into optical-like images to facilitate more accurate image registration, feature alignment, and pixel-level correspondence.

Our methodology combines both deep learning and traditional image processing approaches. Using a cGANs model trained on the QXS-SAROPT dataset, we generate pseudo-optical images from SAR data. These synthetic images are then compared with true optical imagery using a range of evaluation metrics, including Structural Similarity Index (SSIM), pixel-wise absolute difference, shape-based contour matching, and SIFT-RANSAC keypoint matching. For baseline comparison, we also assess the performance of traditional SAR-optical matching using the same metrics.

Results from the SAR-to-optical matching experiments reveal considerable challenges. The SSIM score was low (0.10), and the pixel-wise absolute difference was moderately high (0.67), indicating substantial dissimilarity between the two image types. These findings are expected given the inherent geometric distortions, noise, and speckle in SAR imagery, which make direct pixel-wise or feature-wise comparisons with optical images less effective. Shape-based contour matching provided better results with an average score of 0.94, though this metric favors structural similarity rather than radiometric or textural alignment.

In contrast, the results significantly improved when matching optical images with the pseudo-optical images generated from SAR data using the cGAN model. The SSIM score increased to 0.20, while the pixel-wise absolute difference rose to 0.81, indicating a closer alignment in visual structure and intensity between the generated and real optical images. This improvement demonstrates the potential of cGANs in learning complex cross-domain mappings that preserve critical visual and semantic information necessary for effective image registration.

The shape-based matching score for the optical-to-generated optical image pair remained at 0.94, consistent with the SAR-to-optical case, suggesting that structural information is well-preserved in both scenarios. However, the improvement in keypoint matching and pixel-level similarity strongly indicates that cGANs facilitate a better modality translation, bringing SAR data closer to the optical domain and enabling more effective image fusion workflows.

These findings have important implications for remote sensing applications. By generating high-quality, optical-like imagery from SAR data, cGANs allow for more accurate and robust feature matching and image alignment. This is particularly beneficial in tasks requiring high precision, such as change detection, land-use classification, and environmental monitoring. Traditional SAR-optical fusion approaches often struggle with modality gaps, but our results suggest that cGANs offer a viable solution to this persistent problem.

Despite the promising results, several limitations and challenges remain. The quality of generated images can vary depending on the input SAR data and model architecture. Additionally, while our experiments focus on two-dimensional registration, future work should explore how height information and 3D geometry can be integrated into GAN-based frameworks to support more complex applications. Furthermore, advancements in generative modeling—such as the use of CycleGANs, diffusion models, or transformer-based architectures—could further enhance translation quality and improve alignment metrics.

Our experimental study demonstrates the potential of Conditional GANs in addressing the challenge of SAR-optical image matching. The ability to generate realistic pseudo-optical images from SAR inputs represents a significant step forward in multi-sensor data fusion. Our comparative analysis shows that deep learning approaches, particularly cGANs enhances traditional methods in key performance metrics, making them a valuable tool for the remote sensing community. Future research will focus on improving the generalizability of the model, refining domain adaptation strategies, and integrating semantic and contextual information to further enhance fusion performance.

Single-modal change detection methods based on optical or Synthetic Aperture Radar (SAR) images face challenges such as degradation due to adverse weather or noise interference. In contrast, multimodal change detection struggles with significant domain gaps between different modalities. Inspired by the SAM2 model's temporal memory mechanism for video segmentation, this paper introduces the concept of memory into change detection and proposes a novel approach called Memory-based Multimodal Change Detection. By treating change detection as a temporal problem and modeling remote sensing images as video sequences, the proposed method integrates historical optical images with current SAR images to enhance detection accuracy. Additionally, a difference map enhancement module is introduced to mitigate false changes caused by modality discrepancies.

Treasure hunters are quickly destroying archeological artifacts that scholars have spent years unearthing, making looting one of the biggest risks to cultural heritage in the world. Therefore, it is essential to identify unlawful looting operations in historical locations in order to preserve cultural assets.

In order to evaluate the potential of Synthetic Aperture Radar (SAR) imaging for identifying and evaluating small-scale looting holes, we carried out an experiment in Wuhan, China. We excavated four simulated looting holes of varying sizes and then created photogrammetric models of them. Several imaging parameters were used to gather TerraSAR-X data from the experimental area. To simulate the external factors on looting activities, we created SAR simulations by using 3D models of the looting holes that we created.

We can modify the hole geometry and apply environmental impacts by using SAR simulations, which provides a more adaptable and economical option than field-based research. This method saves time and money by enabling an in-depth review of looting without requiring ongoing fieldwork. We completed our investigation by locating a real looting hole and validating our experimental findings following the fieldwork and computational analysis.

Our findings highlight the potential of SAR imaging and simulations as valuable tools in the detection and protection of archaeological sites, offering a simplified process for assessing looting damage and enhancing conservation efforts. Ultimately, this study contributes to the development of more effective and efficient methods for safeguarding cultural heritage from the growing threat of looting.

In a recent study (Kouremadas et al., 2023), our team introduced a novel methodology for assessing the deterioration and surface soiling of various construction materials by utilizing satellite-derived environmental and air pollution data. This technique, termed Satellite Sensed Data – Dose Response Functions (SSD-DRFs), enables the estimation of material degradation and soiling at virtually any geographic location, overcoming the spatial limitations of traditional Dose-Response Functions (DRFs). Conventional DRFs depend on ground-based environmental monitoring data, which are typically sparse and unevenly distributed, particularly in remote or under-monitored regions. By contrast, SSD-DRFs leverage globally available satellite data, providing a consistent and scalable approach to assessing environmental impacts on materials.

As part of the initial phase of the DRAGON 6 project, we applied the SSD-DRFs to model limestone deterioration and material soiling at several culturally and historically significant sites in China. These include the Fujian Tulou (25°01'23.0"N, 117°41'09.0"E), the Historic Centre of Macao (22°11'28.7"N, 113°32'11.3"E), the Ancient Building Complex in the Wudang Mountains (32°28'00.0"N, 111°00'00.0"E), the Dazu Rock Carvings (29°42'04.0"N, 105°42'18.0"E), and the Circular Mound Altar – Temple of Heaven in Beijing (39°52'56.1"N, 116°24'23.3"E). The study period spans five years (2020–2024), during which annual estimates of both corrosion and soiling have been produced using the appropriate satellite-derived variables.

This preliminary analysis focuses on detecting and understanding the year-to-year variations in the estimated levels of material degradation. The findings support the ongoing development of two diagnostic indices: the Limestone Deterioration Index (LDI) and the Soiling Index (SI). These indices are central to our broader project titled Remote Sensing of Environmental Effects on Materials – Application to the Degradation of Cultural Heritage Monuments (Project ID: 95481), which aims to create robust tools for heritage conservation through remote sensing technologies.

Kouremadas G., Christodoulakis J., Varotsos C., Xue Y., 2023, Satellite Sensed Data-Dose Response Functions: A Totally New Approach for Estimating Materials’ Deterioration from Space. Remote Sensing, 15, 3194. https://doi.org/10.3390/rs15123194

SAR coherence estimation plays a crucial role in surface deformation monitoring and change detection, with its accuracy directly affecting the reliability of subsequent interferometric measurements. However, when dealing with small samples and low coherence, traditional coherence estimation methods often suffer from large biases and variances, and have high computational overhead. In this paper, machine learning methods are used to estimate coherence, and the complex numerical integration and bootstrap sampling process are avoided through automated non-parametric statistical inference. The ratio change statistics that characterize the intensity difference are introduced as training features to avoid the limitation of the image equal variance assumption, further improving the accuracy and computational efficiency of the estimation. Experimental results show that the proposed method significantly reduces the root mean square error of the estimation in small sample sizes and low coherence areas, and improves the contrast in low coherence areas; this method has broad application potential, especially in scenarios where fast and efficient estimation of coherence amplitude is required.

Flood events rank among the most destructive natural hazards, necessitating comprehensive risk management strategies to mitigate their impact on human health, the environment, cultural heritage, and economic activities. Various approaches have been developed to identify flood-prone areas and create flood susceptibility maps; however, there remains a pressing need to enhance their capabilities to account for dynamic changes in landscapes and infrastructure.

This study investigates the potential of the Random Forest (RF) model to map flood-prone areas and assess the flood susceptibility of the Italian territory. The optimal set of predictor factors (SoF) was determined by evaluating a total of 25 variables spanning hydrologic, topographic, and categorical information. Particular attention was given to maximizing the information and avoiding redundancy and high correlations among factors through innovative methods such as the Average Merit of Information (AMI) and Variance Inflation Factor (VIF).

The model Calibration and validation were conducted using Copernicus Emergency Management Service maps, along with regional data from historical flood event records and permanent water areas. The results highlight that AMI effectively identifies the most relevant factors. Furthermore, the RF model trained with Set of Factors - 1 (SoF-1) demonstrated superior generalization capacity compared to other tested sets, as evidenced by a quantitative comparison with official flood hazard maps (AUCROC greater than 0.9). In particular, SoF-1 includes maximum daily precipitation (MDP), the Geomorphic Flood Index (GFI), horizontal distance from the nearest river (HDNR), elevation, lithology, organic matter, NDVI, land use, and silt. The inclusion of GFI significantly improved prediction accuracy in unexplored areas, though challenges persist in extensively flat regions where geomorphic indicators are less distinct. Therefore, the integration of satellite-derived information and complementary datasets facilitates the accurate identification of flood-prone areas, streamlining computational processes and enabling preliminary analyses for decision-makers. These findings underscore the importance of leveraging advanced modelling techniques and continuously updated data to inform effective flood risk management policies and practices.

Aerosol quantitative remote sensing products based on geostationary series satellites provide high-timely data support for atmospheric environmental monitoring, data assimilation, and other fields. Internationally, satellites such as Himawari-9/AHI and GEMS have developed inversion products that include dark pixels and dust-like pixels. In contrast, the FY-4B/AGRI aerosol operational products lack the ability to invert in dust areas. Therefore, this study uses historical FY-4B/AGRI L1 observation data to construct a seasonal hourly surface reflectance library for the solar zenith angle dependence model, and analyzes the spatiotemporal variation characteristics of aerosols by combining the advantages of multi-temporal observations of geostationary stars. This is to achieve product accuracy quality control based on spatiotemporal characteristics constraints, thereby improving the accuracy of inversion products. Subsequently, a sensitivity analysis of inversion errors is conducted from four dimensions: spatiotemporal matching strategy, L1 data, aerosol model, and surface reflectance. The successful implementation of this study will enable the Fengyun geostationary satellite to compare with international similar products, and will fully exert the application benefits of Fengyun geostationary satellite data in hot event monitoring, dust storm warning, and climate change analysis.

Ozone is one of the most important trace gas in the Earth’s atmosphere. Its effect on the biosphere depends on the altitude: stratospheric ozone is essential to protect the biosphere from biologically damaging ultraviolet radiation, while tropospheric ozone has a negative impact on human health, ecosystem services, and agriculture. Moreover, ozone plays an important role in the Earth’s climate, being the third most important anthropogenic greenhouse gas in the middle-upper troposphere. In recent years, there has been a noticeable increase in tropospheric ozone pollution in highly polluted areas worldwide. Tropospheric ozone is now one of the primary pollutants affecting summer air quality in major economically developed regions in China, such as the Beijing–Tianjin–Hebei and the Yangtze River Delta, or in Europe, such as in the Po Valley. As the variation of tropospheric ozone strongly depends on geographical characteristics and climatic conditions, it is crucial to monitor daily global tropospheric ozone to have a better understanding not only of the tropospheric ozone distribution but also of its transport, at a local to global scale.
The TROPOMI ozone profile retrievals provide a good opportunity to study the daily global 3-dimensional ozone distribution at an high spatial resolution of 28x28km2. The retrieved ozone information is distributed over five to six vertical sub-columns of independent information (degrees of freedom, DFS), with a vertical resolution depending on the altitude. As expected from a nadir-viewing instrument, the vertical sensitivity of the retrieval decreases for altitudes <20km in the atmosphere, therefore towards the surface more information comes from the a-priori ozone profile.
In order to enhance the tropospheric ozone detection sensitivity of the ozone profile retrievals, several algorithms refinements can be implemented. In this contribution, we will focus on the following crucial aspects. First, the radiometric calibration, or “soft-calibration”. Because ozone profile retrievals are very sensitive to systematic effects in the measured radiance, an absolute radiometric calibration is essential to obtain accurate tropospheric ozone column. Moreover, the retrieval algorithm currently applies a signal-to-noise ratio (SNR) floor at 150 to reduce the computational time. Increasing the SNR floor, enlarges the number of independent pieces of information (DFS), especially in the troposphere, at the same time increasing the number of un-converged retrieval because of the higher measurement precision requirement. Finally, the choice of the a-priori ozone profile and the assumptions on its errors is another fundamental parameter, as the latter drives the vertical sensibility and stability of the retrieved profile.
In this contribution, we will show the algorithm’s improvements and post-processing techniques (complete data fusion) which are essential to prepare the dataset for the trend and seasonal study of the 0-6km ozone partial column in the selected polluted/clean areas in China.

Atmospheric aerosols strongly influence airquality, climate forcing, and ecosystem health, yet their vertical structure remains one of the largest sources of uncertainty in regional climate assessments. Leveraging the high–spectral resolution oxygen A-band observations collected by NASA’s Orbiting Carbon Observatory2 (OCO-2), we develop and validate a machine learning framework for simultaneously retrieving aerosol optical depth (AOD) and altitude resolved aerosol extinction coefficients over China and adjacent regions (16.9–54.9° N, 72–136° E).

Matched OCO-2 (Level1B, version 11r) and Cloud-Aerosol Lidar with Orthogonal Polarization (CALIPSO, CAL_LID_L2_05 km APro V451) data from 2016–2017 form the core training and validation data sets. OCO-2 radiances in the O₂  A-band, weak CO₂ (1.61 µm) and strong CO₂ (2.06 µm) bands are compressed with principal component analysis (PCA); the first 40 principal components (10, 10, 20 per band) preserve >99.99% of the spectral variance. These components, together with geolocation (latitude, longitude, elevation) and viewing geometry (solar and satellite zenith angles), feed a random forest (RF) regression ensemble. Because RF supports only single output targets, 114 separate trees are trained—one for each CALIPSO layer from the surface to 6.8 km (277–390 in CALIPSO indexing). Model hyper-parameters are tuned by incrementally increasing the number of decision trees to 300, beyond which performance saturates.

Validation against independent 2017 CALIPSO profiles demonstrates robust skill. The AOD retrieval attains a correlation coefficient R = 0.676, root mean square error (RMSE) = 0.168, and bias = 0.01. Extinction profile retrievals yield R = 0.535 and RMSE = 0.107 km⁻¹ (bias = 0.01 km⁻¹) across all layers. Performance is seasonally dependent: the model excels in autumn (R = 0.557) and is weakest in summer (R = 0.442). Height stratified diagnostics reveal declining accuracy with altitud: R decreases from 0.585 below 0.5 km to <0.30 above 5 km while RMSE falls from 0.12 km⁻¹ near the surface to 0.04 km⁻¹ aloft. Retrievals are most reliable when column AOD ≤ 0.3, for which layerwise RMSE drops below 0.06 km⁻¹.

Case studies of individual overpasses show that the RF derived extinction profiles reproduce CALIPSO observed layering, including near surface maxima and elevated aerosol layers. Compared with traditional physical inversion schemes, the data driven approach exploits the global coverage and 16 day revisit of passive spectrometers, bypasses assumptions about aerosol type or lidar ratio, and is readily transferable to future green house gas missions carrying O₂  A-band channels.

The proposed framework furnishes a practical pathway for generating gapfree, altitude resolved aerosol climatologies from passive satellites, thereby strengthening constraints on aerosol–radiation interactions and improving the representation of aerosol processes in chemical transport and climate models. Ongoing work focuses on extending the methodology to multiple aerosol types, incorporating surface reflectance information, and adapting the model for global application.

Greenhouse gases and their contribution to the ongoing global warming is a major
concern in today's world. In this study, we focus on methane which is the most potent
greenhouse gas but the second most contributor after carbon dioxide to the global warming
due to its relatively short lifetime of around 12 years. Majority of methane emissions
originate from anthropogenic sources (human activity) such as farming, factories, power
plants and oil fields. However, natural sources also contribute to the global methane but
quantitative studies of wetland emissions is not yet fully understood.
In this study, we focus on natural methane emissions from a wetland. Emission from
wetlands are too weak to be detected with current Earth observation instruments so we
perform in-situ measurements of wetland methane concentrations using a drone equipped
with AirCore sampling system. The AirCore, which is typically used with weather balloons for
vertical profiles, collects air into a container and produces the ppm portion of methane of the
collected air after analyzing the sample.
Using the collected data and by studying the methane behavior via simulations, we
attempt to both estimate the locations and quantify the emission rates of these methane
sources over the wetland. Ultimately, we hope to be able to estimate overall emission rate
from the wetland, which could then be used for calibration and/or validation for both
ongoing and upcoming satellite missions focused on observing methane.

After two years of operation in orbit with DQ-1, the performance of its LiDAR system has been comprehensively evaluated. The LiDAR measurements were compared with TCCON, and 70 valid synchronous observations were selected (within 1° of latitude/longitude from each TCCON site and within a one-hour observation window). The R² between the XCO₂ values from ACDL and TCCON sites is 0.92, and the root mean squared error (RMSE) is 0.95 ppm. For SNO, comparisons were made with OCO-3. For the DQ-1 satellite, we leveraged its observational capabilities to compare nighttime and daytime CO₂ measurements while excluding diurnal variations unrelated to CO₂ concentration changes (e.g., differences in the diurnal boundary layer, temperature, and humidity). Our results indicate that the diurnal differences in CO₂ observations can effectively quantify the strength of terrestrial vegetation sinks. Due to limitations in the active and passive observation mechanisms, we analyze the differences between these measurements to explore their ability to quantify CO₂ sources and sinks. This approach represents a very promising observation method for the forthcoming DQ-2 satellite, which will be equipped with both active and passive instruments. Consequently, an active–passive collaboration technology has been developed. However, we are currently employing simulation techniques to verify the feasibility of this technology prior to launch, and its viability has been preliminarily confirmed through external SNO observations from DQ-1 and other satellites.

Passive microwave soil moisture products are constrained by their coarse spatial resolution (typically >25 km), struggling to meet the requirements for refined applications at regional scales. While downscaling methods based on optical and thermal infrared data can enhance resolution to some extent, these approaches remain susceptible to cloud interference, further compromising their spatiotemporal continuity. Notably, existing SMAP soil moisture products exhibit distinct complementary characteristics: the L3 9km enhanced product (SPL3SMP_E) demonstrates superior spatial coverage but still suffers from insufficient resolution, whereas the SMAP/Sentinel-1 active-passive product (SPL2SMAP_S) achieves high resolution (3km/1km) at the expense of limited coverage due to Sentinel-1's narrow swath observation. To address these challenges, this study proposes a Generative Adversarial Network (GAN)-based soil moisture downscaling methodology. Leveraging spatiotemporal overlapping regions between the two products during 2015-2024, we constructed large-scale low-resolution to high-resolution (LR-HR) sample pairs to comprehensively integrate the advantages of both datasets through deep learning. Results demonstrate that the downscaled product generated by this method maintains high spatial coverage while significantly enhancing the capacity for detailed characterization of soil moisture spatial patterns, achieving superior accuracy (ubRMSE < 0.05 m³/m³). This advancement provides enhanced data support for regional-scale soil moisture monitoring.

Glacier flow is a sensitive indicator of mass balance and dynamics. Monitoring changes in glacier flow at high temporal resolutions enables understanding of the glacier’s sensitivity to short term climate variability. Previous studies have found that the glaciers in High Mountain Asia (HMA) are in tendency slowing down concomitant to losing mass at an accelerating rate. However, only few investigated seasonal velocity variations and the difference between debris-covered and debris-free glaciers flow dynamics. We focus on eight debris-covered and four debris-free glaciers in HMA, which have different climates, glaciological, topographic and terminating environments. The eight debris-covered glaciers include Xibu, Khumbu, Kangshung, Ngozumpa, Lirung, Satopanth, Dokriani and Kekesayi glaciers. The four debris-free glaciers include Yanong, Abramov, Petrov and Tuyuksu glaciers. Sentinel-1 and -2 images are selected to calculate the glacial velocities on seasonal timescales using the feature tracking method CARST (Cryosphere And Remote Sensing Toolkit). Additionally, we use a novel statistical method to extract the seasonally resolved glacial velocity time series with a 6-day sampling interval and 60 m spatial resolution. We find strong seasonal signals in the velocities for three of the four debris-free glaciers. However, we find a different seasonal velocity signal for the debris-covered glaciers in terms of timing, magnitude and location of the signal. Our study highlights that debris-covered and debris-free glaciers respond differently to external and internal drivers on seasonal timescales. We suggest that changes in the subglacial drainage system is driving the observed seasonal variations in velocities, and this mechanism will likely continue in the future due to increased surface melt rates and changes in precipitation patterns across HMA. We also highlight that icefalls may alter the glaciers response by blocking the development of subglacial drainage channels, and thus seasonal propagation of velocities. Finally, work is underway to compare the seasonal velocities with seasonal mass balance estimates derived from satellite altimetry data. We suggest that winter- and/or summer-accumulation type glaciers exhibit stronger seasonal signals in velocities compared to transitional-accumulation type glaciers. Our novel methodology enables further understanding of seasonal flow dynamics of both debris-covered and debris-free glaciers, which is crucial to capture in order to study the response of glaciers today, and in the future to climate change in HMA.

Accurate identification of precipitation types is crucial for the simulation of water-energy balance, hydrological cycle, and cryospheric processes. To enhance the accuracy of precipitation phase recognition over China, this study explores the performance of five machine learning methods within temperature ranges where traditional approaches struggle to accurately differentiate precipitation phases ([-7℃，10℃]) based on the dataset of daily surface climate variables of China. The results show that the MLP model performs well in identifying precipitation types across China, achieving an overall accuracy of 88.1% on the test set, a 4.4% improvement over traditional methods. In the Tibetan Plateau, the accuracy reached 91.4%, an increase of 12.8% compared to traditional methods. The MLP model notably improves the identification of rain and sleet, especially in areas with frequent rainfall and sleet in eastern Tibetan Plateau and southern China. The performance of MLP model is influenced by sample distribution, achieving lower identification errors for rainfall in South China and snowfall in North China. Analysis shows that wet-bulb temperature, which incorporates information on temperature, humidity and pressure, has the greatest impact on MLP model’s performance. Precipitation amount also significantly affects the probability of sleet classification. Additionally, the MLP model successfully reproduces the observed dependence of precipitation types on elevation and relative humidity. Specifically, it captures the increase in the temperature threshold for precipitation phase discrimination with elevation, as well as the higher probability of sleet occurrence with increasing humidity, demonstrating its ability to represent key physical characteristics with a certain degree of interpretability. Overall, MLP-based machine learning method significantly improves the accuracy of precipitation type identification across China, facilitating imporved understanding and modeling of hydrological and cryospheric processes.

Glaciers are sensitive indicators of climate variability, and their movement patterns are changing in response to climate warming. Monitoring glacier surface velocity is crucial for understanding the relationship between glacier motion and mass balance, as well as climate variability, and provides a scientific basis for studying glacier dynamics and mitigating glacier-related hazards. The southeastern Tibetan Plateau is rich in maritime glaciers with clear dynamic features, including the response to seasonal atmospheric forcing. The extensive datasets provided by satellite observations allow large-scale and long-term monitoring of glacier velocity.

In this study, we utilized high spatial and temporal resolution Sentinel-2 MSI (MultiSpectral Imager) satellite imagery and applied an improved method to estimate glacier velocity, and then analyzed the dynamics of the Yanong Glacier in southeastern Tibetan Plateau. First, satellite images with minimal cloud and snow cover interference were carefully selected. Subsequently, the Co-registration of Optically Sensed Images and Correlation (COSI-Corr) method was used to estimate glacier surface flow velocities over different time periods. It was observed that both too short and too long time intervals between image pairs can affect the accuracy of velocity estimate. In addition, due to the monsoon climate, the study area was frequently covered by clouds during the summer (early May to late September), resulting in limited availability of usable satellite images. To address these challenges, we implemented an optimized image pair selection scheme and a redundant measurement method to improve the accuracy of velocity retrievals. Furthermore, the integration of multi-source satellite remote sensing data was considered to increase the number of usable images. To minimize missing and invalid values in the velocity results, post-processing techniques such as bilateral filtering were applied to remove outliers and noise. Considering the seasonal variability of glacier velocity, seasonal and interannual velocities were synthesized using a time-interval weighted averaging approach.

Finally, using Sentinel-2 MSI remote sensing images from 2020 to 2024 and the improved glacier velocity retrieval method, we successfully obtained reliable estimates of seasonal and interannual velocities of the Yanong Glacier.

In High Mountain Asia (HMA), declines in water stored in glaciers and seasonal snowpacks have been widespread in recent decades. Changes are however highly heterogeneous, with glaciers in the Pamirs experiencing near-neutral mass balance while fast rates of mass loss are observed in the Southeastern Tibetan Plateau. While precipitation phase shifts from snowfall to rain were shown to contribute to accelerated glacier mass loss in the latter region, the impact of this process in other regions of HMA and its role in shaping future snow and glacier mass changes remains largely unknown. Quantifying changes in precipitation phase over several decades at glacier elevations requires model simulations at a spatial and temporal resolution not achievable by observations or reanalysis products. The absolute changes in solid and liquid precipitation amounts require knowledge of all-phase amounts, which is challenging due to uncertainties in reanalysis products and rarely available precipitation measurements. Differences in accumulation regimes and precipitation decadal variability complicate the assessment of precipitation phase change and its role in glacier and snow mass changes under warming conditions.

In this study, we leverage in-situ hydro-meteorological observations, climate reanalysis and CMIP6 model outputs to run a mechanistic, highly resolved land-surface model and reconstruct snow and glacier mass changes from 1970 to 2100 at three benchmark glacierized catchments with contrasting climatic conditions in HMA. The catchments cover areas between 100 and 200 km2, span elevations ranging from 2000 to 6000 m a.s.l., and are located in the Northwestern Pamir (Kyzylsu), Nepalese Himalayas (Trambau-Trakarding) and Southeastern Tibetan Plateau (Parlung No.4). The land-surface model is run at hourly and 100 meters resolution, and its performance is evaluated using in-situ snow depth, surface albedo, remotely sensed snow cover and multi-decadal geodetic glacier mass balance. At all sites, we find declining trends in snowfall, snow depth and glacier mass balance between 1970 and 2023. A decline in the snowfall to total precipitation ratio was found at all sites (-0.005, -0.005 and -0.03 decade-1 at Kyzylsu, Trambau-Trakarding and Parlung No.4 respectively), but was only pronounced at the Southeastern Tibetan site. The decadal variability in precipitation amount, rather than phase, controlled most of the snowfall and glacier mass changes, although the shift in precipitation type from snowfall to rainfall had a substantial contribution to the recent snowfall decline at Parlung No.4 (30% of the snowfall decrease between 1970-1999 and 2000-2023), where we simulate the most rapid glacier mass loss, in agreement with regional assessments of geodetic mass balances. Glacier mass loss has only been marked at Kyzylsu since 2018, following a near-neutral mass balance period characteristic of the Pamir-Karakoram Anomaly. Despite a possible increase in total precipitation, future projections indicate further snowfall decline, which along with warmer and longer ablation seasons, will lead to substantial mass loss. While the studied catchments have different climates, elevation ranges and glacier coverage, all will experience a decline in snowmelt, an increase in evapotranspiration and a decline in sublimation, and peaking icemelt due to glacier retreat.

The depth to the bottom of magnetic sources (DBMS) is commonly regarded as equivalent to the Curie depth, which can be estimated by inverting magnetic anomaly data in specific regions. Spectral analysis based on a fractal (self-similar) model primarily extracts the long-wavelength components of magnetic data, with satellite-derived long-wavelength magnetic anomalies and aeromagnetic-compiled medium-to-short-wavelength components demonstrating higher reliability. This study employs the Revised Spherical Cap Harmonic Analysis 2D (R-SCHA2D) method (Thébault, 2008; Zhang et al., 2022;) to replace the low-frequency (degrees 16–90) components of aeromagnetic data with corresponding LCS-1 data (Olsen et al., 2017), thereby improving the accuracy of Curie depth inversion.

Wavelet spectral analysis is adopted for power spectrum calculation, offering significant advantages over conventional Fourier-based methods. Traditional approaches require sufficiently large sliding windows for power spectrum estimation, suffering from spectral leakage issues. Moreover, the optimal window size typically needs to be 6–10 times the depth of the magnetic source bottom (Ravat et al., 2007), which compromises the spatial resolution of Curie depth estimates. Wavelet transform effectively overcomes this limitation by eliminating the need for grid segmentation and enabling point-by-point power spectral density (PSD) function calculation across the input grid (Bouligand et al., 2009; Audet and Gosselin, 2019; Mather and Fullea, 2019; Gaudreau et al., 2019). Additionally, Bayesian inversion is implemented instead of conventional linear or nonlinear fitting methods. The advantages of the Bayesian approach include: (1) flexible incorporation of prior information on inversion parameters; (2) simultaneous provision of mean solutions, maximum a posteriori (MAP) solutions, and error analysis.

To enhance inversion precision, the latest sedimentary thickness dataset (ECM1) is used as the magnetic layer top boundary, and Curie depth is inverted for fixed fractal indices (β = 2.5, 3.0, 3.5). Key results include: (1)β-dependence: Curie depth decreases (shallows) with increasing β. (2)Tibetan Plateau dynamics: The most pronounced Curie depth variations occur within the triple-suture zone of the Tibetan Plateau, indicating intense intra-crustal thermal activity. (3)Spatial distribution: Regionally, Curie depth exhibits an approximately E-W-trending banded distribution, correlating with magnetic anomaly patterns but not their amplitudes—consistent with Shao et al. (2010). (4)Tectonic correlation: Lateral Curie depth variations align well with plate boundaries and fault zones (e.g., the Altyn Tagh Fault marking the Tibet-Tarim boundary), showing a distinct W-to-E transition from shallow to deep depths. (5)Data replacement effects: Long-wavelength component replacement exerts stronger influences at smaller β. Uncertainties decrease with increasing β, and post-replacement uncertainties are systematically lower for the same β, validating the necessity of component replacement.(6)Uncertainty trends: Inversion uncertainties show no linear or other deterministic relationship with β selection but correlate positively with Curie depth itself (larger depths correspond to higher uncertainties), reflecting the inherent challenge in retrieving long-wavelength information.

Understanding and modeling the Earth's magnetic field, especially its lithospheric component, is fundamental to both geoscientific research and practical applications such as geological interpretation, resource exploration, and magnetic navigation. Among the various contributors to the geomagnetic field, the lithosphere produces relatively high-amplitude, small-scale magnetic anomalies—often reaching tens of thousands of nanoteslas in regions with strong magnetization—which require high-resolution models for accurate characterization.While satellite-based models have advanced our understanding of the global magnetic field, their spatial resolution is often limited. Purely satellite-derived models typically achieve resolutions of around 100 km. In contrast, when near-surface data are incorporated, the spatial resolution can be improved to 20–25 km. However, existing global models—such as EMM2017, WDMAMv2, and SH1050—rely on near-surface datasets that lack sufficient coverage and resolution in East Asia, particularly in western China. This data scarcity results in missing small- to medium-scale lithospheric features in regions like southern Xinjiang and Tibet. To address these limitations, we introduce a new version of the China University of Geosciences Lithospheric Magnetic Field Model 3D (CUG_CLMFM3Dv2), developed using revised spherical cap harmonic analysis (R-SCHA) and a combination of well-processed near-surface and satellite magnetic data. Our model integrates the high-resolution airborne magnetic anomaly map of continental China and TMI and vector measurements from satellite missions such as MSS-1. The revised model improves upon the previous version (CUG_CLMFM3Dv1) by refining satellite data preprocessing (including leveling and along-track filtering) and incorporating updated and denser near-surface data. Compared with earlier regional models such as CLAS and the Xinjiang-Tibet 3DSS model, the CUG_CLMFM3Dv2 achieves a higher spatial resolution (down to 5.7 km) and offers enhanced fidelity in representing the lithospheric field. These improvements allow the model to resolve complex magnetic structures in data-sparse regions and to maintain consistency with global satellite-derived magnetic field models at higher altitudes. This study highlights the importance of integrating multi-source geomagnetic data with physically robust modeling techniques for developing regional magnetic field models. The methodology and results presented provide a valuable foundation for further investigations into crustal structure, tectonic evolution, and magnetic anomaly interpretation across China and adjacent areas. Our model sets the stage for next-generation regional magnetic field modeling and serves as a reference for similar efforts in other complex geological settings.

In practical scenarios, remote sensing semantic segmentation plays a crucial role in applications such as natural resource management, and environmental monitoring. Most of these methods depend on supervised learning, necessitating a substantial number labeled data for training deep neural networks. However, acquiring a well-annotated training dataset is expensive due to high collection and labeling costs. Hence, developing remote sensing semantic segmentation models under limited labeled data remains critical.

Remote sensing pre-training methods based on self-supervised learning have shown feasibility to train deep learning models with insufficient or low-quality annotated training data. Contrastive learning and masked image modeling (MIM) represent two prevalent image-based self-supervised methods. Contrastive learning learns representations by distinguishing between positive and negative image pairs. At present, some effective contrastive learning methods have been proposed, such as MOCO^[2], SimCLR^[1], BYOL^[3], and Barlow Twins^[4]. Masked image modeling (MIM) learns representations by reconstructing masked regions. Many large-scale remote sensing pre-trained models have been constructed based on MIM, including RingMo^[5], AST^[6], and S2FL^[7].

However, the pretext tasks of the contrastive learning and MIM are not designed for semantic segmentation, which makes it difficult to qualitatively or quantitatively evaluate the pre-training process without fine-tuning, and may also affect the transfer to the semantic segmentation task. Motivated by Wang et al.^[8], we think that if pixels from the same class are clustered together in the feature space during pre-training, it will become easier to transfer to the semantic segmentation task. Hence, we propose a clustering-based self-supervised pretraining method for remote sensing semantic segmentation. This method extends pixel-level self-distillation frameworks while integrating feature clustering modules.

The method contains five parts: data augmentation, student and teacher networks, cluster assignment module, self-distillation training, and clustering constraint module. Data augmentation adopts a multi-crop augmentation similar to VicregL^[9] and LeoPart^[10], which generates n augmented views from an input image. Then, two views are selected and fed into the student network and teacher network respectively to extract the feature maps. Subsequently, a cluster assignment module is designed to assign the feature vectors to a set of K learnable prototypes. Finally, a self-distillation training process is constructed to train the student network to learn features from the teacher network. In addition, clustering constraint module, including Sinkhorn-Knopp algorithm^[11], and group-level contrastive loss^[12], are introduced into the self-distillation framework to avoid the situation where feature vectors from different types of objects are assigned to one prototype while preserving the semantic information of clustering results. After pre-training, a segmentation model can be obtained with few labeled data through fine-tuning.

To assess the application capabilities of the proposed method in real-world scenarios, we conduct a few-shot segmentation experiment for building extraction along waterways in Shaanxi Province. The application utilizes Sentinel-2 imagery and is divided into three steps: 1) The original imagery is cropped into numerous 512×512 image patches, which are then used for pre-training with the proposed method to obtain a Shaanxi-specific pre-trained model. 2) Some representative image patches are selected and manually annotated. In this application, only 200 512×512 image patches are annotated, which is significantly fewer than typical datasets. 3) Fine-tuning is performed based on the pre-trained model and the labeled data to obtain the building extraction model. Experiment result illustrates the extraction results of province-wide building extraction based on Sentinel-2 imagery, which demonstrates that the proposed method can achieve the extraction of buildings with different distribution types across the entire province using only 200 labeled data with size of 512×512.

References

[1] Chen, T., Kornblith, S., Norouzi, M. and Hinton, G., 2020, November. A simple framework for contrastive learning of visual representations. In International conference on machine learning (pp. 1597-1607).

[2] He, K., Fan, H., Wu, Y., Xie, S. and Girshick, R., 2020. Momentum contrast for unsupervised visual representation learning. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition (pp. 9729-9738).

[3] Grill, J.B., Strub, F., Altché, F., Tallec, C., Richemond, P., Buchatskaya, E., Doersch, C., Avila Pires, B., Guo, Z., Gheshlaghi Azar, M. and Piot, B., 2020. Bootstrap your own latent-a new approach to self-supervised learning. Advances in neural information processing systems, 33, pp.21271-21284.

[4] Zbontar, J., Jing, L., Misra, I., LeCun, Y. and Deny, S., 2021, July. Barlow twins: Self-supervised learning via redundancy reduction. In International conference on machine learning (pp. 12310-12320).

[5] Sun, X., Wang, P., Lu, W., Zhu, Z., Lu, X., He, Q., Li, J., Rong, X., Yang, Z., Chang, H. and He, Q., 2022. RingMo: A remote sensing foundation model with masked image modeling. IEEE Transactions on Geoscience and Remote Sensing, 61, pp.1-22.

[6] He, Q., Sun, X., Yan, Z., Wang, B., Zhu, Z., Diao, W. and Yang, M.Y., 2023. AST: Adaptive Self-supervised Transformer for optical remote sensing representation. ISPRS Journal of Photogrammetry and Remote Sensing, 200, pp.41-54.

[7] Xue, Z., Yu, X., Yu, A., Liu, B., Zhang, P. and Wu, S., 2022. Self-supervised feature learning for multimodal remote sensing image land cover classification. IEEE Transactions on Geoscience and Remote Sensing, 60, pp.1-15.

[8] Wang, T. and Isola, P., 2020, November. Understanding contrastive representation learning through alignment and uniformity on the hypersphere. In International conference on machine learning (pp. 9929-9939).

[9] Bardes, A., Ponce, J. and LeCun, Y., 2022. Vicregl: Self-supervised learning of local visual features. Advances in Neural Information Processing Systems, 35, pp.8799-8810.

[10] Ziegler, A. and Asano, Y.M., 2022. Self-supervised learning of object parts for semantic segmentation. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition (pp. 14502-14511).

[11] Caron, M., Misra, I., Mairal, J., Goyal, P., Bojanowski, P. and Joulin, A., 2020. Unsupervised learning of visual features by contrasting cluster assignments. Advances in neural information processing systems, 33, pp.9912-9924.

[12] Wen, X., Zhao, B., Zheng, A., Zhang, X. and Qi, X., 2022. Self-supervised visual representation learning with semantic grouping. Advances in neural information processing systems, 35, pp.16423-16438.

Solar-induced chlorophyll fluorescence (SIF) is a physiological byproduct of photosynthesis and has received increasing attention in the field of vegetation remote sensing monitoring in recent years. Compared to conventional optical remote sensing vegetation indices, SIF provides a more direct proxy for photosynthetic activity and plant physiological status, making it a critical tool for monitoring vegetation dynamics, assessing ecosystem functioning, and improving the accuracy of terrestrial carbon uptake estimates in the context of global change.

Forest ecosystems, as one of the most critical vegetation systems on land, characterized by diverse vegetation types, massive biomass, and extensive global distribution. SIF observations derived from satellite offer a promising capability for monitoring photosynthetic dynamics in forests. However, due to the structural complexity of forest canopy, significant uncertainties remain in satellite SIF-based forest monitoring, and the SIF-GPP relationship in forest remains insufficiently characterized.

The TROPOspheric Monitoring Instrument (TROPOMI) aboard ESA's Sentinel-5P satellite possesses the capability to detect SIF. With its exceptional spatiotemporal continuity enabling near-daily global coverage, the observations of TROPOM provide new opportunities for SIF research at regional to global scales. Therefore, this study employs time-series TROPOMI SIF observations to monitor diverse forest types and investigate the dynamic patterns of SIF signals and their relationship with GPP. Furthermore, we utilize a 3D radiative transfer model to simulate satellite-forest observation processes, enabling quantitative analysis of how variations in vegetation canopy structural parameters influence SIF observations. On the other hand, the complex three-dimensional structure of forest canopies leads to significant view-angle effects in SIF observations. We therefore quantified the impacts of observation geometry on both canopy SIF and its relationship with GPP, enabling systematic evaluation of the effectiveness of satellite SIF for monitoring different forest types.

Overall, this study systematically assesses how both forest canopy characteristics and observation geometry influence the capacity of SIF to monitor forests. The findings provide scientific references for applying SIF in forest ecosystem monitoring and assessment.

This study uses radiance closure to validate simulations by the Reference Forward Model (RFM) through comparison with spectral measurements taken by the Infrared Atmospheric Sounding Interferometer (IASI) on 13th January 2022. First, internal parameters of the RFM were adjusted to determine which were essential for accurate atmospheric simulation. Inclusion of CO2 line-mixing and chi-factor modification of the Voigt line shape to account for sub-Lorentzian dependence led to a decrease in residuals in the CO2 nu_2 and nu_3 bands, although increasing overall computation time. When compared with IASI observations, inclusion of non-local-thermodynamic equilibrium (LTE) states decreased overall residuals for daytime observations by accounting for solar scattering but made no significant difference for nighttime observations. Next, RFM inputs were iteratively adjusted using retrievals of temperature and gas concentrations. There was good final agreement between IASI and the RFM with RMS residuals minimise to within IASI instrument noise. However, this could not be achieved for the CO2 nu_2 Q-branch at 700 cm-1 and the CO2 nu_3 band centred at 2300 cm-1. This was most likely to be caused by both errors in specifiying tropospheric CO2 concentration as well as limitations of the RFM in modelling scattering effects, with implications for future development.

Driven by global climate change, the frequency of extreme weather and climate events has increased significantly in recent years. Climate change, extreme events, and human activities have significantly influenced regional dual water cycle processes, leading to a series of geological hazards such as karst collapses and land subsidence. Therefore, accurate simulation and prediction of groundwater levels are urgently needed. To address this issue, this study selects the Linyi monocline hydrogeological unit as a case area and introduces a Physics-Informed Neural Network (PINN) model for groundwater level simulation, using the two-dimensional unsteady groundwater flow equation as a physical constraint. Compared to traditional numerical models, the PINN model reduces the reliance on detailed regional geological structures and hydrogeological conditions. Unlike conventional neural networks, the PINN incorporates physical equations into the loss function, enabling the coupling of linear physical constraints and nonlinear neural network modeling, thereby enhancing the physical interpretability of the model. By embedding physical equations, the PINN model achieves physical consistency and transferability, allowing it to be applied beyond the current study area to various groundwater systems. The results show that the PINN model achieves high accuracy in groundwater level simulation, with an R² of 0.87 and RMSE of 1.98 m. By solving physical equations during training, the importance of hydrological parameters to the PINN model is significantly enhanced, with hydraulic conductivity making the greatest contribution to the model. Moreover, the PINN is capable of simultaneously achieving accurate groundwater level simulation and dynamic inversion of hydrogeological parameters even when some system parameters are missing. This study provides a novel and transferable approach to groundwater level modeling and offers scientific guidance for the sustainable management and protection of groundwater resources.

Land Surface Temperature (LST) data within urban environments provides insights into phenomena such as the Urban Heat Island (UHI) effect. Effective urban planning, environmental monitoring, and climate adaptation strategies increasingly depend on access to LST information with both high spatial resolution and high temporal frequency. Currently available satellite thermal sensors offer a trade-off between these two characteristics. Instruments like the Visible Infrared Imaging Radiometer Suite (VIIRS) provide daily global LST observations (Hulley et al., 2018), but their spatial resolutions (750 meters) is too coarse to capture the detailed thermal patterns within heterogeneous urban landscapes. Conversely, Thermal Infrared Spectrometer (TIS) onboard the Sustainable Development Goals Science Satellite 1 (SDGSat-1) offer much finer spatial resolution (30 meters), which is well-suited for detailed urban analysis (Teng et al., 2024). However, SDGSat-1 has a longer revisit period (around 11 days), and like all optical/thermal sensors, its observations can be hindered by cloud cover, making regular updates on LST mapping difficult. This creates a significant gap: the lack of consistent, frequent LST data at a finer spatial resolution for meaningful urban applications. Bridging this gap requires techniques to fuse multi-sensor data, combining the temporal frequency of VIIRS with the spatial detail of SDGSat-1.

This objective of this study is to evaluate a deep learning framework for generating daily, high-resolution (30-meter) LST maps specifically for urban areas using SDGSat-1 and VIIRS Data and a conditional Denoising Diffusion Probabilistic Model (DDPM). Diffusion models are a class of generative models that learn to reverse a process of gradually adding noise to data. In this application, the model learns to remove noise from a randomly perturbed high-resolution LST map to reconstruct the original, clean LST map. This process is conditioned on lower-resolution input data, guiding the generation towards a realistic high-resolution output consistent with the provided context.

The network architecture will be U-Net-based. Modifications incorporating efficient Transformer blocks, leveraging concepts such as those presented in U-DiT which downsample tokens within attention blocks (Tian et al., 2024), will be explored for enhanced feature representation and computational efficiency, though a standard convolutional U-Net will serve as a baseline. The target data for training the diffusion model is SDGSat-1 TIS LST data, generated following methodologies developed and validated by Teng et al. (2024). This 30-meter resolution LST data represents the high-resolution ground truth. The model is trained on spatially and temporally aligned pairs of the low-resolution conditional inputs and the high-resolution SDGSat-1 LST target. The training objective is minimizing the difference (i.e., Mean Squared Error) between the noise predicted by the model and the actual noise added during the forward diffusion process.

Evaluation will involve quantitative and qualitative assessments on a held-out test dataset of urban scenes. Quantitative metrics will include Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Peak Signal-to-Noise Ratio (PSNR), and Structural Similarity Index (SSIM) comparing the generated 30m LST maps against the corresponding derived SDGSat-1 LST data. Qualitative analysis will involve visual inspection of the generated maps, focusing on the model's ability to reproduce fine-scale thermal variations within the urban environment, such as temperature differences between roads and parks, or distinct heat signatures of industrial zones versus residential areas. Performance will be compared against baseline super-resolution techniques like bicubic interpolation.

References

[1] G. C. Hulley, N. K. Malakar, T. Islam, and R. J. Freepartner, "NASA's MODIS and VIIRS Land Surface Temperature and Emissivity Products: A Long-Term and Consistent Earth System Data Record," IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 11, no. 2, pp. 522-535, Feb. 2018, doi: 10.1109/JSTARS.2017.2779330.

[2] Y. Teng, H. Ren, Y. Hu, and C. Dou, "Land surface temperature retrieval from SDGSAT-1 thermal infrared spectrometer images: Algorithm and validation," Remote Sensing of Environment, vol. 315, p. 114412, 2024, doi: 10.1016/j.rse.2024.114412.

[3] Y. Tian, Z. Tu, H. Chen, J. Hu, C. Xu, and Y. Wang, "U-DiTs: Downsample Tokens in U-Shaped Diffusion Transformers," arXiv preprint arXiv:2405.02730, 2024. [Online]. Available: https://arxiv.org/abs/2405.02730

Optical remote sensing demonstrates outstanding capabilities in water body monitoring. However, its application in monitoring salt lakes rich in resources such as potassium (K) and lithium (Li) is still in deficient, especially in key resource-producing salt ponds, such as carnallite ponds. Due to the importance of carnallite pools in salt lake resources, monitoring their sedimentation status not only helps monitor the resource's distribution but also provides a key basis for the timing of collection.Therefore, the fine classification of sediment thickness in carnallite ponds by optical remote sensing is of practical engineering guidance.

In this paper, the multispectral data of Sentinel-2 was selected for a classification study, and four types of sediment thicknesses with different degrees were labeled with the actual mining situation. In order to fully utilize the working mode of linear mining, a line target recognition method combining attention mechanism is proposed. Based on this method, a semi-empirical and semi-structured classification model is developed, which fully combines empirical linear characteristics with structured target information to realize the classification of sediment thickness in carnallite ponds. Different degrees of classification of carnallite sediment thicknesses were successfully classified, and the method's feasibility was preliminarily validated through field investigation.

As a unique water resource, changes in the water depth and volume of the salt lake will not only directly affect the design of the subsequent salt precipitation process in the salt field, brine concentration, and the quality and yield of the final salt products, but also change the water depth of the salt lake due to the different precipitation in different seasons, so the accurate measurement of the water depth of the salt lake has an important practical value and economic significance. This study not only provides a feasible method for remote sensing inversion of salt lake water depth, but also its results have a good promotion prospect. By continuously optimizing the model and exploring the data fusion strategy, it is expected that the method will be extended to the prediction of salt water depth and the intelligent management of salt field processes in the future, so as to provide finer and real-time technological guidance for salt mining operations and further enhance the production and operational efficiency of salt fields.

In this study, the multilayer perceptron (MLP) model is utilized to realize the inversion of salt lake water depth. The multispectral data, which can completely cover the salt lake area due to its wide spatial coverage, is used as the model input. LiDAR data provides high-precision water depth information in point form, which is used as labels for training and fitting. Specifically, we use seven spectral bands of Landsat8 L2 data from 2019 to 2024 as inputs and ICESat2 ATL13 data as labels for training and fitting. As a result, the detection of temporal changes in the water depth of the salt lake is realized based on the data from different periods. By combining multispectral data with LiDAR data, it is expected to achieve more than 85% accuracy.

paceborne and airborne laser scanning (SLS, ALS) provide essential three-dimensional remote sensing data for the accurate estimation of forest biomass at regional scales. The Area-Based Approach (ABA) typically constructs statistical relationships between field-measured biomass and metrics derived from SLS or ALS , enabling the generation of forest biomass distribution maps at spatial resolutions of 20–30 m. However, conventional ABA modeling heavily relies on field-plot data obtained via stratified sampling, which is limited by forest accessibility and high fieldwork costs, consequently restricting overall model accuracy. . In contrast, Individual Tree Segmentation (ITS) methods exhibit high precision and strong transferability but necessitate high-quality point cloud data, thus restricting their scalability for large-area applications. To address these limitations, we propose a novel methodology termed Progressively Synergizing ITS and ABA (ITS-ABA). Specifically, this method leverages high-density airborne LiDAR data in combination with a robust individual-tree structural metric, the LiDAR Biomass Index (LBI), to generate augmented biomass samples. By extending these augmented samples into scenarios, including both low-density airborne LiDAR and photon-counting point clouds from sensors such as ALS and ICESat-2, the method enables accurate plot-scale biomass estimation across large areas. Results indicate that individual-tree LBI models derived from high-density point clouds consistently achieve high accuracy (R² > 0.85, rRMSE < 0.2). For ALS data, biomass estimations based on augmented samples exhibit strong agreement with field-measured biomass (R² > 0.81, rRMSE < 2.00%) and achieve accuracy comparable to traditional ABA models built from actual field plots (|ΔR²| < 0.05, |ΔrRMSE| < 2.00%). For ICESat-2 data, high precision can also be achieved (R2=0.63, rRMSE=26.51%).Consequently, the proposed method effectively utilizes airborne LiDAR data as a substitute for traditional field-plot surveys, enabling accurate large-scale forest biomass estimation with minimal or no field data requirements.

Effectively identifying the spatiotemporal distributions and structure characteristics of understory saplings and monitoring their growth is beneficial to exploring the internal mechanisms of plant regeneration and providing technical assistances for continues cover forest management and tending practices. LiDAR signals can penetrate the forest canopy to obtain three-dimensional structural information from the top to the bottom of the forest, which provided a good data foundation for the extraction and segmentation of regenerated saplings under the large trees.

Our existing research has achieved the segmentation of overstory large trees and lower saplings based on airborne LiDAR data, but this method can only be applied to double layered forests with clear vertical height difference [1]. As the lower saplings grow, they have reached a high position under the first live branches of the overstory large trees, making it difficult to separate them using a uniform threshold. Therefore, this study proposed an automatic and improved method to detect the regenerated understory saplings in an experimental forest tree farm in North China based on the three-dimensional (3-D) structural information generated by LiDAR data. By delineating the individual tree crowns above 10 m using the improved watershed algorithm, we first obtain the local mixed point cloud clusters of overstory large trees and lower layer saplings. Furthermore, each local point cloud cluster is vertically divided into three layers based on a unified threshold (the threshold is generally set through empirical and combine with the field measurement data), and the point cloud in the middle layer are used as seed points. The tree trunk point cloud of the large trees are detected combined with the downward 45 degree region growing algorithm and the seed points. Although the above methods can remove crown points above the threshold and trunk points of the overstory large tree, there are still some crown points of large tree below the threshold mixed with regenerated understory saplings, and these large tree crown points have significant discontinuity with saplings. Therefore, the saplings are further detected through the points composing and connectivity method (as shown in Figure 1). Then the individual understory saplings were segmented using an adaptive bandwidth mean shift-based clustering algorithm. In order to analyze the growth of samplings, we applied this method to the unmanned aerial vehicle (UAV) LiDAR data obtained in 2023. And the extraction results of sapling parameters are compared with the results extracted using ALS in 2018.

Our results showed that the detection rates of understory saplings ranged from 94.41% to 152.78%, and the matching rates increased from 62.59% to 95.65% as canopy closure went down. The ULS-based sapling heights well captured the variations of field measurements (R2 = 0.71, N=3241, RMSE = 0.26 m, p < 0.01) and terrestrial laser scanning (TLS)-based measurements (R2 = 0.78, N=443, RMSE = 0.23 m, p < 0.01) using the linear regression statistical models. Overall, the growth of saplings has a strong correlation with measured data, and are greatly affected by the gaps formed through the overstory of large trees. This study provides a foundation for increasing the capacity of forest carbon sequestration through regulating the dynamics of forest gaps to better utilizing light resources.

Grassland shrub encroachment represents a widespread ecological challenge in arid and semi-arid grassland ecosystems. Traditional remote sensing monitoring faces limitations due to spectral and scattering characteristic similarities between shrubs and grasses. This study developed an optical-SAR fusion approach for large-scale shrub encroachment monitoring in Xilingol grassland, utilizing Sentinel-1 SAR and Sentinel-2 multispectral data. Shrub coverage samples were obtained through field surveys and high-resolution image segmentation. Multi-temporal features (growth season phases: early, peak, and late) including spectral reflectance, spectral angle, optical vegetation indices, backscatter coefficients, polarimetric characteristics, and radar vegetation indices were extracted. Sensitivity and correlation analyses identified optimal monitoring periods and shrub coverage-sensitive features. The results demonstrated: (1) Early growth season as the optimal monitoring window; (2) Existence of distinct optical features and SAR features sensitive to shrub coverage; (3) A novel fused index developed through weighted integration of selected features effectively amplified shrub signals in heterogeneous shrub-herb-soil landscapes. The proposed methodology enhanced shrub detection capability in mixed vegetation scenarios and showed better scalability for regional-scale monitoring compared to conventional approaches. This integrated optical-SAR framework provides technical support for large-scale grassland shrub encroachment assessment and ecological management.

Salt-lake information is important in ecological protection, water resource management, economic development, and tourism culture. This above information is closely related to the ecology of salt-lake area, and the salt production of different kinds (for industry, for alimentation, and also in terms of rare earth elements’ exploitation).

Observing salt lakes from space has great advantages in efficiency and continuity, which is mainly reflected in the following aspects: Firstly, salt lakes are often located in remote areas, which are difficult to reach and monitor directly. Remote sensing satellites can cover a wide surface area and provide high-resolution data, making it easier and more comprehensive to monitor and study salt lakes. Secondly, remote sensing satellites can obtain data over a long period of time, which can help us understand the hydrological characteristics, water volume changes, water quality status and other information of salt lakes, and provide a scientific basis for water resources management and ecological protection

In this context, the aim of this DRAGON 95458 YS work is to develop approaches to characterize salt lakes and its surrounding environments, using optical and techniques, which could output the segmentation of salt-lake areas, the classification of the salt crystals states in different salted water fields, and establish the dynamic maps of salt lakes across years. Indeed optical remote sensing has significant advantages in salt-lake monitoring as water change in terms of geochemistry components, associated with modification of alga and bacteria’s populations involving change in water colours. First, optical remote sensing sensors can provide high-resolution image data, which makes it possible to clearly observe the detailed features and boundaries of the salt-lake.

In a first step, a bibliographical search has been done, with mixed results, as if laboratory’s works on salted crust are available, only few papers on mapping have been found allowing however some methodological approaches based on spectral indices, such as the NDSI (Normalized Difference Salinity Index).

Then, on the Salins d’Aigues-Mortes, covering 10 000 ha in South of France, more geared to the production of food salt (450 000 t/year) a database containing the geometry of the salted fields parcels have been done, differencing the supply pond, the settling ones, and finally the production eyelets.

On a second step, a set Sentinel 2 (L2A) has been selected in order to carry an analysis, on the temporal signature of the different salt system elements. This allowed to characterize the spectral answers of the targets in regards to the cycle of the salt production. Several indices can be tested including NDWI[1], SI[2] and NDSI[3] to characterize water presence, salinity, and salt crust formation. Preliminary NDSI results suggest a strong correlation between the reflectance of salt crystals and salinity levels, supporting the relevance of these indices for Salt Lake monitoring.

Next steps will correspond to the comparison with in situ da over this Aiges-Mortes site and to transpose the approach to Elbro Delta site as well to the remote Qarhan salt lake (China). In addition, the potential of hyperspectral domain will be investigated.

[1] Normalized Difference Water index. Formula: (NIR-SWIR)/(NIR+SWIR)

[2] Salinity Index. The most common formula: (Red+SWIR)/2

[3] Normalized Difference Salinity Index. Formula: (SWIR-NIR)/(SWIR+NIR)