Conference Agenda

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. We focus on different regions across High Mountain Asia (HMA) where glaciers have different average velocities (slow, median, and fast). HMA has the largest glacier coverage outside the polar regions and is considered the water tower of Asia. Previous studies have found that the glaciers in HMA are in tendency slowing down concomitant to losing mass at an accelerating rate. We use both optical and SAR remote sensing data including Landsat-8, Sentinel-1 and -2, Planet and Pléiades images to present multiple remotely sensed calculated glacial velocities for the different regions of HMA over the last decade. We calculate the velocity variations using different tracking methods. By analysing the accuracy of the velocity variations through validation with the higher spatial resolution Pleiades velocity dataset and field data as well as using statistical techniques such as the GLAcier Feature Tracking testkit (Zheng et al., 2023), we provide insights into the accuracy of the different remote sensing data and tracking methods. Finally, we explore possible internal and external drivers of the observed glacial velocity variations, with a focus on mass balance and short-term climate variability.

The AMSR-E/NASA (Advanced Microwave Scanning Radiometer-Earth Observing System/National Aeronautics and Space Administration) daily global-scale soil moisture (SM) product, spanning from 2002 to 2011 with a spatial resolution of 25 km, was provided by the NASA National Snow and Ice Data Center Distributed Active Archive Center (NSIDC DAAC). However, it was noted that the AMSR-E/NASA SM product exhibited limited sensitivity in capturing intra- and inter-annual variability of SM across many regions. Investigation revealed that inaccurate parameter values (A₀ and A₁) in the AMSR-E/NASA SM retrieval algorithm were pivotal in causing this issue. Building upon previous work by Xie et al. (2019) that utilized in-situ SM measurements in the Tibetan Plateau, this study sought to enhance the global-scale AMSR-E/NASA SM product. Parameter values (A₀ and A₁) were recalibrated using a global network of 13 in-situ observation sites (totaling 192 sites), and their relationships with fractional vegetation cover (FVC) across four land cover types (sparse vegetation, grassland, cropland, and forest) were analyzed. Inversion models for A₀ and A₁ parameters tailored to each land cover type were constructed, utilizing a global FVC dataset. Subsequently, the improved AMSR-E/NASA SM product was generated employing the refined algorithm proposed by Xie et al. (2019). Evaluation of the product against SM measurements from 6 in-situ observation networks indicated strong agreement, particularly evident in sparse vegetation areas where a linear relationship between A₀ (or A₁) parameter values and FVC was observed (e.g., A1=-0.61×FVC+1.21 and A0=-0.20×FVC+0.012). Conversely, non-linear relationships were prominent in grassland/cropland/forest areas (e.g., A1=69.04×(FVC)2-28.49×FVC+5.67 for grass). Furthermore, the improved global-scale AMSR-E/NASA SM product demonstrated superior performance compared to AMSR-E/JAXA (Japan Aerospace Exploration Agency) and AMSR-E/LPRM (Land Parameter Retrieval Model) SM products, exhibiting lower Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) values (i.e., 0.026 cm³/cm³ and 0.032 cm³/cm³, respectively).

GaoFen-1 (GF-1) is a satellite launched by China, which has the advantages of high spatial and temporal resolution, wide swath, and high accuracy. Since 2013, it has provided a 10-year time series of multi-spectral satellite imagery, which provides strong data support for glacier change monitoring. However, GF-1 optical image has only four bands, i.e., blue, green, red and near infrared, and the lack of spectral information limits its application in glacier cover extraction.

We developed a machine learning approach to explore the ability to extracting glacier cover using GF-1 satellite imagery and other ancillary data, minimizing the effects of cloud cover and seasonal snow cover. Based on surface reflectance bands, spectral indices, topographic information and textural features, the algorithm established the rules of glacier information extraction by decision tree classification method, and finally obtained the glacier coverage area. A series of comparative experiments showed that introducing topographic information and texture features to the machine learning model could improve the classification results, especially the identification of debris-covered glaciers. To reduce the noise and deal with the problem of poor spatial continuity, morphological analysis was used to post-process the classification results. Firstly, the morphological closing method was used to fill small gaps and make the classification results more continuous, then the morphological opening method was used to remove speckle noise, and finally the adaptive median filtering method was used to remove salt-and-pepper noise. This machine learning algorithm was applied to multi-temporal images to obtain multiple classification results, which were further overlaid to obtain the annual glacier map, effectively reducing the influence of snow cover and cloud cover. The case study in Parlung Zangbo glacier in the southeastern Tibetan Plateau showed that the annual glacier cover result in 2020 had a high coincidence with the glacier boundary in Randolph Glacier Inventory 6.0. The classification performance of the algorithm was evaluated by using the selected testing samples and the overall classification accuracy was above 96%.

In general, Gaofen-1 optical image with four bands has the ability to extract glacier cover with acceptable accuracy.

High Mountain Asia (HMA) hosts the largest mass of ice outside the Polar Regions and provides water to large downstream communities. Glacier mass change has been highly heterogeneous across the region over the last decades, with glaciers in the Pamirs experiencing near-neutral mass balance while fast rates of mass loss are observed in the Southeastern Tibetan Plateau (STP). In a previous modeling study in the STP, we found that precipitation phase changes associated with climate warming were a major accelerator of glacier losses, but this mechanism of mass loss acceleration has yet to be explored across the rest of HMA. Additionally, snow sublimation is a process known to influence glacier mass supply, but its relevance has not been established across HMA. Very few direct observations exist at high elevation, hindering the quantification of glacier mass inputs which is essential to estimate the long-term sensitivity of glaciers to warming.

In this study, we combine in-situ hydro-meteorological observations with remote sensing data to constrain a land-surface model and understand snow accumulation dynamics over the past five decades at three glacierized catchments with contrasting climatic conditions in HMA. The three sites are Kyzylsu in the Northern Pamirs, Parlung No.4 in the STP and Mugagangqiong in the central part of the Tibetan Plateau. We use MODIS, Landsat-8 and Sentinel-2 satellite images to derive snow cover dynamics at high spatial and temporal resolutions, using a combination of NDSI and albedo thresholding to distinguish between bare-ice and snow or firn at very high elevations.The land-surface model is forced with statistically downscaled and bias-corrected reanalysis data (ERA5-Land) at 100m spatial and hourly temporal resolution, from 1970 to 2023. Glacier elevation changes over this 50-year period are assessed using declassified satellite images acquired in the 1970s and used to evaluate our model performance. To tackle the challenges posed by the computational resources needed to run our model at such high spatial and temporal resolution, we use a clustering approach to run the model on a sample of points representing the most important components of the land surface heterogeneity. This approach is tested over the 2015-2020 period and shown to approximate accurate fully distributed simulations.

We find that the fastest glacier mass loss among the three sites occurred at Parlung No.4 in the STP, where it accelerated since the 2000s, and was associated with a decrease in the annual solid to total precipitation ratio (72% to 60%). A slight decrease in catchment sublimation amounts was also found, as a consequence of the reduced snow. On the other hand, Kyzylsu and Mugagangqiong glaciers experienced slightly negative mass balances, without clear trends and mostly related to the decadal variability in total precipitation amount. At these two sites, sublimation amounts, snow cover fraction and annual snowfall ratio have been relatively stable from 1970 to 2023, with the latter potentially explaining the absence of glacier mass loss acceleration. Sublimation is, in absolute amounts (, the lowest at the colder and most arid catchment (central TP) but there it also constitutes to the largest fraction of annual snowfall and snowpack ablation with glacier areas (30% on average and up to 50% on individual years), highlighting that it is an important process shaping the snow and glacier mass balances. This work paves the way towards a better understanding on how snow accumulation processes are affected by climate change and what are the implications for glacier evolution and high-elevation catchment hydrology.

Glaciers are a sensitive indicator of climate change and play a crucial role in the terrestrial water cycle, ecosystem stability and food security in cold and high elevation regions. The lack of high spatial resolution meteorological data limits our understanding of glacier change and its response to climate change. However, new, high-resolution, spatially distributed remote sensing observations create an opportunity to use glaciers to refine high-mountain meteorological reanalysis through inverse modeling.

In this study, we use Tethys-Chloris, a physically based, coupled mass- and energy- balance model, and remote-sensing-derived glacier surface mass balance and Sentinel-2/Landsat glacier albedo, to resolve air temperature, precipitation and incoming shortwave radiations biases through 1000 Monte Carlo simulations. These biases are determined by the simulation with the best agreement with remote sensing result. We test this approach at Abramov Glacier and Parlung No.4 Glacier with long-term and high-quality in-situ mass balance and meteorological observations during 2015-2019, and validate it during 2000-2014.

The results show that model inversion based on remote sensing data can accurately quantify the systematic multiannual bias in reanalysis data, and the meteorological data derived in this way show a good agreement with in-situ observations collected at our two test glaciers. This method paves a way to correct biases in climate reanalysis data to provide high spatial resolution meteorological forcing for individual glaciers, allowing us to better understand recent and future climate change in the mountain cryosphere.

The Pamir Mountains are unique worldwide: they constitute a crucial mountain water tower that is highly vulnerable to future climatic, environmental, and social change; and their heterogeneous cryosphere includes glaciers with an approximately balanced mass budget despite climatic warming. Due to geopolitical factors, the in-situ records in the region have been interrupted, and current assessments of glacier volume and mass change in this region show high uncertainty. One of the most significant contributions of remote sensing to cryospheric research is the continuous provision of surface Digital Elevation Models (DEMs) derived from high-resolution (<5m) optical stereo images, showing promise to compensate for the scarcity of in-situ snow and glacier observation.

In this study, we analyze high-resolution Pléiades stereo satellite images systematically collected twice per year since 2022, in May and in September, over seven mountain alpine catchments in the Pamir mountains. We adapted a stereo image analysis workflow from the Ames Stereo Pipeline to systematically process these data, including coregistration and bias correction, and to remove erroneous artifacts such as jitter-induced undulations. We leverage independent high-resolution UAV-derived DEMs to validate the Pléiades DEMs elevation bias and their estimated uncertainties based on the patch approach, and to determine the effectiveness of the jitter undulation correction. We then used the DEMs to quantify the seasonal snow volumes and annual glacier mass balance across our study domains. We provide a methodological summary of the processing workflow and the uncertainty quantification and propagation to glacier-wide geodetic mass balance elevation change and distributed snow volume assessments, as well as our preliminary results.

Our results demonstrated the potential of very high-resolution satellite imagery for snow and glacier monitoring, despite the challenge of converting these results to mass change. In addition, the results also highlight the need for a consistent measure of DEM accuracy to ensure the reliability of future land elevation assessments.

Remote sensing sea ice monitoring is a current research hotspot. By translating sea ice SAR images into optical images through the Conditional Generative Adversarial Network (CGAN), all-weather and intuitive monitoring data can be obtained. However, the translation results obtained by this method have problems such as blurring of contours, weakening of textures, and color distortion. This article designs an improved CGAN network to address the above issues. Taking into account the current improvement methods, the new model adds a hollow space pyramid module to the network structure and designs a skip connection with a cross feature fusion module. The loss function uses a joint loss function of structural similarity and L1 norm. This article selects 5 Sentinel-1 images and 7 Sentinel-2 images from the East Beaufort Sea area for experiments. The experimental results show that the improved CGAN translated images have better visual effects, with peak signal-to-noise ratio (PSNR) increased by 3.4, structural similarity (SSIM) increased by 0.11, root mean square error (RMSE) decreased by 13%, and the accuracy of the processed images for sea ice classification improved by 7.33% compared to SAR images.

Surface soil moisture (SSM) and vegetation optical depth (VOD) are essential variables in the terrestrial ecosystem. The microwave remote sensing has been providing a unique way to obtain the global SSM and VOD from space with various spaceborne sensors, including the Advanced Microwave Scanning Radiometer for EOS (AMSR-E), Soil Moisture and Ocean Salinity (SMOS), AMSR2, and Soil Moisture Active Passive (SMAP) etc. Those microwave sensors operating at different frequencies possess differentiated vegetation penetration capabilities and might provide significant information of the Soil-Plant-Atmosphere-Continuum (SPAC) system.

In this study, we applied the Multi-Channel Collaborative Algorithm (MCCA) to the AMSR-E/2 and SMAP observations for a physically consistent SSM and VOD products. The main idea of MCCA is applying the relationship between soil and vegetation property at the core channel to the collaborative channels, with three features: (1) unlike other retrieval algorithms using proxy or iterative procedure to retrieve VOD, an analytical solution was proposed to derive vegetation transmittivity, and then VOD; (2) a general function was proposed to describe relationships of VODs between different channels (frequency, polarization and incidence angle); and (3) brightness temperatures (Tb) at the collaborative channels are derived by the Tb at core channel based on the two-component version of τ-ω model, without any assumptions.

The SMAP MCCA retrievals are inter-compared with other SSM and VOD products (MT-DCA version 5, and DCA, SCA-H, SCA-V from SMAP Level-3 products version 8, and SMAP-IB), showing an analogous spatial pattern. The MCCA derived SSM had the lowest unbiased root mean square error ubRMSE of 0.055 m3 /m3 followed by SMAP-IB and DCA (0.061 m3 /m3), and an overall Pearson’s correlation coefficient of 0.744 (SMAP-IB performed best with R=0.764) when evaluated against in situ observations from the International Soil Moisture Network (ISMN). Comparable accuracy also found in widely used validation spare network SCAN. The MCCA generates VOD at both vertical and horizontal polarization. While the magnitude of the polarized VODs is lower than other products. MCCA polarized VODs were found to have a good linearity with live biomass and canopy height, though partial saturation exists in the relationship with live biomass of tropical forests but not canopy height. The polarization difference of L-band VODs is mainly located at densely vegetated and arid areas.

The AMSR-E/2 MCCA retrievals are inter-compared with other SSM products (AMSR-ANN, CCI-passive v07.1, LPRM[1]C/X, JAXA) at ISMN soil moisture networks. Although the R-value of MCCA (0.709) was slightly lower than that of LPRM-X (0.735), MCCA achieved the best scores in terms of RMSE=0.074 m3 /m3, ubRMSE=0.073 m3 /m3 and bias=0.007 m3 /m3. For the indirect evaluation of VOD with aboveground biomass (AGB) and MODIS NDVI, the MCCA product showed the performance comparable to other products (LPRM-C/X, VODCA-C/X/Ku). MCCA-derived VODs exhibited smooth non-linear density distribution with AGB and high temporal correlations with MODIS NDVI over most regions, especially for the H-polarized VOD. MCCA-derived VODs can physically present reasonable variations across the microwave spectrum (values of VOD increase with microwave frequency), which is superior to the LPRM and DCA.

In addition, considering that existing soil moisture products were derived using from different algorithms and from passive microwave observations at different frequencies, it is challenging to fairly compare the sensing capabilities and depths of different microwave frequency sensors. This study addresses this challenge by comparing SMAP MCCA and AMSR2 MCCA to resolve the above issues. The assessment involves a comparison of the satellite soil moisture with 40 globally distributed soil moisture observation networks at regional (dense network) or grid scales. The findings first indicate that both MCCA SMAP and AMSR2 soil moisture products demonstrate superior performance at the regional scale compared to the grid scale. Secondly, for the first time, we have provided evidence that SMAP outperforms AMSR2, with both sensing capabilities decreasing as vegetation cover increases. Furthermore, the analysis of global soil moisture dry-down patterns reveals that SMAP exhibits a lower soil moisture loss rate, an extended soil moisture retention duration, and an effective higher wilting point than AMSR2, which suggests that the L-band contribution depth surpasses that of C-, X-, and Ku-bands at the satellite scale.

Overall, MCCA products developed in this study showed good performance on both SSM and VOD. It is crucial for studies that consider the effects of paired SSM and VOD simultaneously, such as water fluxes in the SPAC system. In addition, the retrieval is implemented on snapshot observations, and MCCA can provide continuous daily data once the daily Tb is updated. It is expected that the MCCA algorithm can be extended to the observations of the upcoming Copernicus Imaging Microwave Radiometer (CMIR) mission.

The launch of the Soil Moisture Active Passive (SMAP) satellite in 2015 has resulted in significant achievements in global soil moisture mapping. Nonetheless, spatiotemporal discontinuities in the soil moisture products have arisen due to the limitations of its orbit scanning gap and retrieval algorithms. To address these issues, this paper presents a physics-constrained gap-filling method, PhyFill for short. The PhyFill method employs a partial convolutional neural network technique to explore the spatial domain features of the original SMAP soil moisture data. Then, it incorporates variations in soil moisture induced by precipitation events and dry-down events as penalty terms in the loss function, thereby accounting for monotonicity and boundary constraints in the physical processes governing the dynamic fluctuations of soil moisture. The PhyFill model was applied to SMAP soil moisture data, resulting in continuous spatially daily soil moisture data on a global scale. Three validation strategies are employed: visual inspection through a global pattern, simulated missing-region validation, and soil moisture validation with in situ measurements. The results indicated that the reconstructed soil moisture achieved a higher spatial coverage with satisfactory spatial continuity with neighbouring pixels. The simulated validation of the missing regions revealed that the averaged unbiased root mean square difference (ubRMSD) and correlation coefficient (R) were 0.0102 m³/m³ and 0.9919, respectively, versus the gap-filled SMAP product. The core validation sites demonstrated that the reconstructed soil moisture data has a consistent ubRMSD compared with the original SMAP soil moisture data (0.041 m³/m³ vs. 0.040 m³/m³). The PhyFill method can generate globally continuous, highly accurate soil moisture estimates, providing remarkable support for advanced hydrological applications, e.g., global soil moisture dry-down events and patterns.