Conference Agenda

Glacier velocity fields are typically derived through offset tracking techniques applied to optical and/or SAR remote sensing images. This is mainly because offset tracking is highly effective at detecting small changes in image features caused by glacier motion, which often results in strong decorrelation. Correlation algorithms extract the pixel-level offset, which can then be refined to a sub-pixel level using various interpolation techniques. However, the accuracy of these interpolation algorithms incorporated in different offset tracking software has rarely been evaluated or compared. In addition, the lack of in-situ observations to confirm the sub-pixel precision of derived offset can cause uncertainties. For above reason, a digital image processing method was used to evaluate the precision of various software and algorithms. The study aimed to assess the sub-pixel precision of derived offset and suggested an algorithm to correct possible offset tracking bias. This will ultimately help improve the accuracy of glacier velocity fields, which is crucial for climate change research and hazard assessment.

This study focused on the two largest glaciers in Greenland, Petermann Glacier and Kangerlussuaq Glacier, which account for roughly 4% each of the entire ice sheet's glacier mass loss and flow in northwestern and southeastern directions, respectively. To evaluate the precision of different algorithms, six pairs of Sentinel-2 images were used.The study combined the offset tracking results obtained from different algorithms, including COSI-Corr, autoRIFT, and ImGRAFT (CCF-O and NCC), and treated them as pre-set offset fields. Using the Sinc interpolation, which is an optimal interpolation method according to the sampling theory, simulated offset images are generated using the pre-set offset fields and pre-event images. The mentioned software and algorithms were then used to obtain offset tracking results based on the pre-event images and simulated offset images. Precision was assessed and possible bias inspected at the sub-pixel level only, as all algorithms first established a dependable offset value at the pixel level and then interpolated to the sub-pixel level. The displacement results and the pre-set offset fields were wrapped to a range of [-0.5, 0.5], designated as y and x, respectively. A cubic function, y=ax+4(1-a)x^3 (where a is the correction parameter), was chosen for the regression. The precision was exhibited by the fitting's RMSE, while parameter a indicated the presence of bias. if a equals 1, then no bias exists, but if not, there is a bias. Finally, the inverse function of the fitting can rectify potential systematic errors at the sub-pixel level.

The regression results showed that the sub-pixel systematic error of COSI-Corr is negligible and could be disregarded, whereas autoRIFT and ImGRAFT (CCF-O or NCC) displayed a certain degree of systematic errors in their offset results. Specifically, the values of a were 1.008, 0.778, 0.915, and 0.886 for COSI-Corr, autoRIFT, ImGRAFT (CCF-O), and ImGRAFT (NCC), respectively. In COSI-Corr, the Sinc function was used to interpolate the correlation coefficient matrix, while ImGRAFT applied bicubic interpolation regardless of the correlation algorithm being CCF-O or NCC. autoRIFT utilized a rapid Gaussian pyramid upsampling algorithm for estimating the sub-pixel displacement with a precision of 1/64 pixel. The results suggest that the use of COSI-Corr may be more reliable regarding the interpolation technique for obtaining sub-pixel precision in offset tracking.

Sub-pixel systematic error correction yielded the most significant improvement in the autoRIFT algorithm, reducing RMSE by an average of 0.0054 pixels in a single direction and increasing precision by 11%. This demonstrates the significance of performing this type of correction. ImGRAFT's RMSE also decreased slightly: ImGRAFT (CCF-O) and ImGRAFT (NCC) decreased by an average of 0.0014 and 0.0012 pixels in a single direction, respectively. However, whether to apply this correction to ImGRAFT depends on the desired level of precision as it only resulted in a 1.5% increase. Furthermore, as no noticeable systematic sub-pixel errors were detected in COSI-Corr, this correction is unnecessary. The similar regression results across different study sites and deformation directions indicate that sub-pixel systematic error is dependent on interpolation algorithm. After systematic correction, all algorithms showed reliable results. For instance, COSI-Corr and autoRIFT showed higher precision than ImGRAFT, with RMSEs of 0.04~0.14 pixels at Kangerlussuaq. In contrast, ImGRAFT had slightly lower precision with RMSEs of 0.08-0.10 pixels at Petermann, and 0.09~0.13 pixels at Kangerlussuaq. ImGRAFT (CCF-O) shows slightly better precision than ImGRAFT (NCC). Finally, it is worth noting that autoRIFT has much higher computational efficiency than the other algorithms, this study recommends combining it with a post-correction step for systematic error.

Synthetic Aperture Radar (SAR) images are commonly used to monitor glacier flow velocity at Greenland Ice Sheet (GrIS). However, in summer, offset-tracking with SAR imagery in summer usually show poor quality because the rapid ice surface freezing-melting cycles contaminating the surface backscattering characteristic. Optical images are less sensitive to this phenomenon. In this study, we combine Sentinel-1 and -2 images to create the glacier velocity time series for the Petermann glacier, located in the northern GrIS. Firstly, the offset-tracking technique is employed to acquire the initial deformation fields with SAR and optical sensors separately, each SAR and/or optical acquisition is tracked with its closest next three acquisitions. Next, after removing the outliers the least squares method based on connected components is employed to calculate the time series of glacier velocity for Sentinel-1 and -2, separately. Finally, these two kinds of derived time series are integrated with a weighted least squares method, where weights are evaluated according to the estimated RMSEs in the last step. Error propagation analysis suggests RMSEs of the single pair of Sentinel-1 and -2 images offset-tracking are ~0.22 m and ~2.5 m for Petermann glaciers. Standard deviation of the difference between Sentinel-1 and Sentinel-2 measured velocity are ~0.25 m/day. Compared with 6-day velocity fields product, NSIDC (National Snow and Ice Data Center) -0766, which is only derived with Sentinel-1observations, our results show good agreement and less defects in summer. The differences are ~0.20 m/day in non-melting seasons and ~0.34 m/day in summer. Longitudinal velocity differences growing in 2019 and 2020 at ~20 Km up to the terminus are consistency with the crevasse expansion, indicating another calving event is approaching. This research finds that the fusion of Sentinel-1 and -2 offset-tracking results improves the completeness of the ice movement time series for polar glaciers.

The validation of satellite soil moisture products has been serving as an active research topic for the application of the products and improvement of the retrieval algorithms, attracting extensive attention. Nevertheless, most existing validation activities focus on the validation of coarse-resolution soil moisture products at regional or global scales, seldom on the validation of high-resolution SM products at the fine scale. To this end, the State Administration of Science, Technology, and Industry for the National Defense of China initiated a research project entitled "Key Technology Research and Standard Specifications for the Validation of High-Resolution Remote Sensing Products" in 2020. Under the framework of the project, a soil moisture experiment was conducted in the middle stream of the Heihe River Basin in northwestern China in the summer of 2021, aiming to validate high-resolution satellite remote sensing products of soil moisture. The paper introduces the design, composite, and preliminary results of the experiment. A ground soil moisture observation network was established, and several synchronized campaigns were conducted. Simultaneously, several satellite remote sensing observations and soil moisture products were collected and validated against the ground measurements. A preliminary analysis shows that the experimental datasets can support the validation of satellite soil moisture products, as well as the monitoring of irritation at the agricultural field scale. Overall, the experiment provides fruitful methodologies and datasets for the validation of high-resolution remote sensing products, benefiting the development and improvement of soil moisture retrieval algorithms and products to support irrigation scheduling and management at a precision agricultural scale in the future.

Monitoring the evolution of of glacier is essential to understanding glacier reaction to climate change. To better track the long-term changes in glacier mass balance, some archived historical images has been widely used, but a knowledge gap of how glaciers evolve across the whole High Asia Mountain remains to be addressed. To this end, we reprocessed all KH-9 stereo images covering glacierized areas of the HMA, and combined them with NASADEM and Copernicus DEM to estimate glacier mass changes over two periods (i.e., 1975-2000 and 2000-2015). The results show that the eastern part of the HMA experienced a sustained acceleration of glacier mass loss, with relatively significant acceleration in the Nyainqentanglha and Hengduan mountains. In contrast, the glacier mass loss rate slowed down in the western part of the HMA, especially Pamir Alay and Eastern Pamir. In addition, there is no significant change in the rate of mass loss in the Gangdise, Karakoram and Hindu Kush mountains.

The terrestrial water cycle is affected by climate change, through changing evaporation and precipitation patterns. Reanalysis products play an important role in monitoring the changing climate, where the past weather and climatological conditions are estimated based on assimilating historical observational data into numerical models. Reanalysis products in the past largely focused on the estimation of atmospheric variables. While some reanalysis products included the usage of land surface models, the hydrological component in these models is often rudimentary or lacking. Furthermore, reanalysis of land surface variables is often done in an offline approach, where the atmospheric forcing is prescribed using already existing datasets: feedback from the land to the atmosphere is not taken into account.

To overcome these limitations and gain a deeper understanding of the terrestrial water cycle, we introduce a novel weakly coupled reanalysis framework. This framework addresses the shortcomings by incorporating a comprehensive representation of land surface processes and a three-dimensional model for subsurface and surface flow. Our study focuses on Europe from 2000 to 2020, utilizing a horizontal spatial resolution of approximately 11km.

The Community Land Model (CLM3.5) is employed to capture crucial land surface processes such as evaporation, transpiration, and infiltration, accounting for land cover and vegetation types across Europe. We used CLM3.5 coupled with ParFlow, a hydrological model that simulates subsurface and surface flow. Initially, we use ERA5 data for atmospheric forcing, but later replace it with the Icosahedral Nonhydrostatic (ICON) model to achieve a fully coupled terrestrial framework. All components of our model are integrated using the Parallel Data Assimilation Framework (PDAF).

To estimate both uncertain state variables (e.g., soil moisture) and uncertain parameters (e.g., hydraulic conductivities) for the land surface and subsurface, we explore the application of Ensemble Kalman Filters and iterative Ensemble Kalman Smoothers. We present preliminary results where Soil Moisture Passive Active (SMAP) data have been assimilated.

These results are part of ongoing work, exploring the added benefit of a fully coupled reanalysis framework. In the weakly coupled reanalysis framework presented here, only state variables and parameters related the the land surface and subsurface are updated in the data assimilation cycle. In the fully coupled reanalysis framework this will be done for all three model components simultaneously.

Soil moisture plays a critical role in governing water and energy exchanges in the land-atmosphere continuum. Accurate knowledge of soil moisture is essential for water resources management, agricultural production, and weather prediction. The assimilation of remotely sensed soil moisture data into land surface models (LSMs) has demonstrated potential in improving land surface states and fluxes. However, the relative value of assimilating microwave soil moisture observations acquired at different frequencies remains uncertain. Limited studies have examined the impact of applying different merging algorithms to generate a merged soil moisture product prior to data assimilation (DA). This study focuses on assimilating soil moisture data obtained from L-band (Soil Moisture Active Passive Mission- SMAP and Soil Moisture and Ocean Salinity Mission - SMOS), C-band (Advanced SCATterometer - ASCAT), and X-band (Advanced Microwave Scanning Radiometer 2 - AMSR2) into the land surface model (CLM, Community Land Model) using the Ensemble Kalman Filter (EnKF) approach. This is done for the North-Rhine-Westphalia region in Germany, for the years 2017 and 2018. Initially, each remotely sensed soil moisture product is assimilated individually. Subsequently, both a conventional linear combination method and a novel Long Short-Term Memory (LSTM) approach are employed to calculate weights for the different remotely sensed soil moisture products. These weights are determined with in situ soil moisture measurements acquired through Cosmic Ray Neutron Sensors (CRNS). The two merged products are then assimilated into the Community Land Model (CLM), a land surface model. The simulated soil moisture time series are evaluated against independent point measurements. The study shows that joint assimilation of merged retrievals can offer improved characterization of soil moisture compared to assimilating each remote sensing product individually. In addition, by analyzing multiple data assimilation results, we are able to assess the variations and similarities in assimilating retrievals from different microwave bands. This analysis allows us to evaluate the impact on data assimilation performance, particularly in situations involving mission or sensor transitions or discontinuities. Furthermore, given that the behavior of different retrieval schemes is influenced by surface characteristics and spatial heterogeneity, this study also examines the spatial patterns of soil moisture and explores the potential for capturing and propagating spatial information of remotely sensed soil moisture in the land surface model.

Snow density varies spatially, temporally, and vertically within the snowpack and is the key to converting snow depth to snow water equivalent. While previous studies have demonstrated the feasibility of retrieving snow density using a multiple-angle L-band radiometer in theory and in ground-based radiometer experiments, this technique has not yet been applied to satellites. In this study, the snow density was retrieved using the Soil Moisture Ocean Salinity (SMOS) satellite radiometer observations at 43 stations in Quebec, Canada. We used a one-layer snow radiative transfer model and added a vegetation model over the snow to consider the forest influence. We developed an objective method to estimate the forest parameters (tau, omega) and soil roughness (SD) from SMOS measurements during the snow-free period and applied them to estimate snow density. Prior knowledge of soil permittivity was used in the entire process, which was calculated from the Global Land Data Assimilation System (GLDAS) soil simulations using a frozen soil dielectric model. Results showed that the retrieved snow density had an overall root-mean-squared error (RMSE) of 83 kg/m³ for all stations, with a mean bias of 9.4 kg/m³. The RMSE can be further reduced if an artificial tuning of three predetermined parameters (tau, omega, and SD) is allowed to reduce systematic biases at some stations. The remote sensing retrieved snow density outperforms the reanalysis snow density from GLDAS in terms of bias and temporal variation characteristics.

The two vegetation transfer parameters of tau (Vegetation Optical Depth, VOD) and Omega (Single Scattering Albedo) could vary significantly across microwave channels in terms of frequencies, polarizations, and incidence angles, and their characteristics of channel dependence have not yet been fully investigated. In this study, we investigate the channel dependence of vegetation effects on microwave emissions from soils using a higher-order vegetation radiative transfer model. Corn was chosen as the research object, and a corn growth model was developed using the multifrequency and multiangle ground-based microwave radiation experiment from the Soil Moisture Experiment in the Luan River (SMELR). After establishing the corresponding database of corn radiation characteristics, the effective scattering albedo under various channels was calculated using the higher-order radiation transfer model. The channel dependence analysis of the vegetation optical depth and effective scattering albedo in the database was performed. The results show that the channel dependence of vegetation optical depth can be described as the polarization dependence parameter (C_P ) and the frequency dependence parameter (C_f ). According to these two parameters, the vegetation optical depth can be calculated at any channel under three adjacent frequencies (L band, C band and X band). The effective scattering albedo has no obvious dependence on the angle, so the effective scattering albedo based on the higher-order radiation transfer model under three adjacent frequencies with different polarizations is obtained. This study is helpful for understanding the differences in vegetation radiation characteristics in different channels, thereby promoting the development of large-scale soil moisture retrieval accuracies in vegetated areas.

Surface soil moisture (SSM) is an important variable in drought monitoring, floods predicting, weather forecasting, etc. and plays a critical role in water and heat exchanges between land and atmosphere. SSM products from L-band observations, such as the Soil Moisture and Ocean Salinity (SMOS) mission and the Soil Moisture Active Passive (SMAP) mission, have proven to be optimal global estimations. Although X-band has a lower sensitivity to soil moisture than that of L-band, Chinese FengYun-3 series satellites (FY-3A/B/C/D) have provided sustainable and daily multiple SSM products from X-band since 2008. This research developed a new global SSM product (NNsm-FY) from FY-3B MWRI from 2010 to 2019, transferred high accuracy of SMAP L-band to FY-3B X-band. The NNsm-FY shows good agreement with in-situ observations and SMAP product and has a higher accuracy than that of official FY-3B product. With this new dataset, Chinese FY-3 satellites may play a larger role and provide opportunities of sustainable and longer-term soil moisture data record for hydrological study.