Conference Agenda

Microwave, optical (including multispectral and hyperspectral) and laser altimetry data, including GF-3, Sentinel-1, Sentinel-2, EnMAP, PRISMA, SWOT, and ICESat-2, were utilized to investigate salt crust classification, regional mapping, salt field identification, sediment monitoring, and water depth inversion in the Salt Lake region. The application of microwave remote sensing, optical remote sensing, and laser altimetry to salt lake resource investigation, development management, and dynamic monitoring was further expanded.

In terms of SAR-based research, studies were mainly focused on the identification of salt crust types in the Qarhan Salt Lake region. Different salt crust types exhibit highly similar polarimetric scattering characteristics, while high-resolution SAR imagery is inevitably affected by speckle noise, which increases the difficulty of fine classification. To address this issue, a neighborhood-informed autoencoder-based method for feature optimization and reconstruction was developed using fully polarimetric GF-3 data and dual-polarization Sentinel-1 data. Experimental results showed that the proposed method reduced feature redundancy, improved feature representation, and achieved satisfactory extraction of the spatial distribution of different salt crust types, providing a basis for subsequent surface parameter inversion and change monitoring.

In terms of optical data, while Sentinel-2 provides continuous seasonal monitoring though robust spectral indices (NDWI, NDSI, SI), its 13-band resolution limits fine mineralogical discrimination. Hyperspectral sensors – PRISMA (400-2500nm) and EnMap (200+ bands), fill this gap decisively. Both instruments detect diagnostic absorption features of gypsum (CaSO₄·2H₂O) at 1900 and 2200nm, enabling the identification of evaporitic crusts well before they become visible in broadband data. At Qarhan, EnMap additionally reveals signatures consistent with sodium carbonates (natron) around 2300-2350nm, and PRISMA detects a characteristics 1300-1500nm peak potentially attributable to hydrated carnallite, mineral level identifications impossible with conventional multispectral imagery.

Across all sites, principal component analysis of hyperspectral data consistently identifies water content as the dominant structuring factor (PC1 > 70-80%), followed by salt crystal properties, halophilic biomass, and sediment composition. Halophilic algae (Dunaliella salina), detectable through a narrow spectral peak at 600-700nm, were identified at Qarhan during spring evaporation cycles. The complementarity between NDSI-1 and NDSI-2 the former saturating rapidly in aquatic conditions, the latter providing finer moisture gradient discrimination, is fully exploitable only through hyperspectral resolution. Taken together, the result demonstrates that the combination of Sentinel-2 for temporal coverage and PRISMA/EnMap for mineralogical depth represents the most robust methodological framework currently available for the operational and scientific monitoring of saline environments from space.

In terms of the integration of SAR and optical remote sensing, the research mainly included fine land-cover mapping in the Qarhan Salt Lake region and salt field identification at the national scale. For regional mapping, Sentinel-1 SAR data and Sentinel-2 optical imagery were jointly used to address the difficulty of simultaneously distinguishing salt crusts and multiple water body types in a complex surface environment. An anisotropic water response index, AWRI, was proposed by combining the anisotropy parameter A and the normalized difference water index (NDWI). This feature improved the separability of different surface types and supported large-scale fine mapping of the study area.

At a broader spatial scale, research on nationwide salt field distribution mapping led to the development of a multimodal self-supervised learning framework, SaltMoMAE, based on Sentinel-1 and Sentinel-2 data. The pretrained framework was fine-tuned for salt field segmentation, and a national salt field distribution map of China at 10m spatial resolution was generated for the period from 2020 to 2022. The results showed that the proposed method achieved high identification accuracy at the national scale, particularly in distinguishing salt fields from spectrally similar water-related land-cover types.

In terms of the integration of optical remote sensing and laser altimetry, emphasis was placed on the dynamic monitoring of sediment in salt ponds. ICESat-2 data and Sentinel-2 imagery were used to conduct multi-year observations of sediment surface elevation and thickness changes from 2020 to 2025 in artificial sodium salt ponds of the Qarhan Salt Lake region. Through photon denoising, classification, and refraction correction, combined with water surface elevation information extracted from optical imagery, pond-scale inversion of sediment surface elevation was achieved. Preliminary results indicated that, under favorable conditions, ICESat-2 can stably reflect the annual trend of sediment accumulation, demonstrating the feasibility of laser altimetry for monitoring shallow artificial salt ponds.

For Dabusun Lake, a technical framework was proposed for lake-bottom topography reconstruction and spectral water depth inversion by combining Sentinel-2 and SWOT data. Multi-temporal Sentinel-2 imagery was used to extract shoreline information, while SWOT-derived water surface elevation data were used to establish shoreline elevation constraints and reconstruct the lake-bottom elevation model. Based on the differences between water surface elevation and lake-bottom elevation, training samples were generated and combined with optical features to build a water depth inversion model. The results showed that the proposed method achieved good accuracy in verifiable areas and provided a reliable solution for salt lake water depth inversion where large-scale in situ bathymetric measurements are lacking.

Overall, the current research has further strengthened the application of multi-source remote sensing data to land-cover classification, parameter inversion, and dynamic monitoring in salt lake regions on the basis of last year’s work. Future studies will focus on salt crust roughness and water content inversion based on polarimetric SAR, as well as water volume and regional salinity estimation based on optical remote sensing and LiDAR, in order to further improve the quantitative monitoring capability of salt lake resource parameters.

This work is also partly supported by the project 62331026 funded by National Natural Science Foundation of China, and the project PID2023-149659OB-C22 funded by MCIN/AEI/10.13039/501100011033/ FEDER, UE.

This project focus on producing high-precision, high-resolution forest type classification maps, monitoring forest growth dynamics based on multi-temporal LiDAR data, retrieving forest vertical structure parameters based on LiDAR data, and developing a multi-scale forest quality assessment method. Up to now, main work of the project including:

(1) Integrating the Vegetation Height Model (VHM) with the Alpha shape method improves LiDAR-based forest gap detection. VHM, derived from CHM and DTM, identifies potential gaps by detecting height reductions. However, VHM thresholding alone fails to accurately delineate irregular gap boundaries in fragmented or disturbed forests. The Alpha shape method reconstructs point set boundaries by tuning the Alpha parameter, preserving local geometry and capturing irregular gap shapes from natural disturbances. Results show that this combined approach enhances detection accuracy and boundary delineation. After applying a height threshold to VHM to separate canopy gaps from low vegetation, the Alpha shape method removes noise and precisely fits gap edges while retaining local geometric features.

(2) To address the challenge of indistinct height boundaries between understory saplings and overstory trees, which complicates mixed canopy segmentation, we propose a method for extracting understory saplings using high-density UAV LiDAR data, combining local region growing and connectivity analysis. By integrating this method with previous airborne LiDAR-based segmentation results, we generated a distribution map of sapling growth status. Detection accuracy ranged from 0.68 to 0.90, decreasing with increasing canopy closure. Sapling heights derived from UAV LiDAR effectively captured field-measured variations (R² = 0.78, rRMSE = 9.07%). Moreover, the integrated growth status showed high consistency with field measurements. This study provides a foundation for enhancing forest carbon sink capacity by regulating gap dynamics and optimizing light resource utilization.

(3) To address the limited generalizability of spaceborne photon-counting LiDAR in complex forest scenes, this study uses denoised ICESat-2 ATL03 point clouds. The along-track data are segmented into 100 m estimation units and quality-controlled. Three types of structural features are extracted: height distribution (HD), canopy height and heterogeneity (HH), and vertical structure (VS). An optimal feature combination is determined using group-based pre-screening combined with regularized all-subsets regression. Model accuracy is evaluated via ten-fold cross-validation and compared with traditional height/density models. Results show that the optimal volume estimation model for mixed coniferous and broadleaved forests in Aargau, Switzerland, incorporates Top Canopy Height (TCH), 65% density quantile (Dp65), Leaf Area Weighted Canopy Volume (LAVH), and mean Vertical Foliage Profile (VFPmean). Ten-fold cross-validation yields R² = 0.78, RMSE = 92.48 m³·hm⁻², and rRMSE = 0.24. Compared to traditional models, R² improves by ~10% on average, and rRMSE decreases from 0.28 to 0.24, demonstrating more stable performance in stands with high canopy heterogeneity. This provides a reference for large-scale forest volume/carbon monitoring and quality assessment using ICESat-2 data.

(4) SAR backscatter is sensitive to forest structure, supporting large-scale monitoring via constellations like Sentinel-1 and NISAR. However, the impact of structural variations on backscatter remains unclear. We addressed this by linking time-series ULS point clouds to S1 C-band and SAOCOM L-band SAR data over a full vegetation period (13 time steps, two 0.25 ha plots in a mixed temperate forest). Forest structure proxies (e.g., PAI) were correlated with ɣ0 backscatter (VV, VH) at plot and pixel levels. Preliminary results shows that moderate PAI–C-band correlation, with seasonal VH changes in the broadleaf-dominated plot (likely phenology-driven), while the conifer-dominated plot showed higher VV correlation and no seasonal trend. This confirms that spaceborne SAR captures temporal forest structural changes, advancing landscape-scale understanding of temperate forest dynamics.