Wet and Dry Snow Detection Using Sentinel-1 SAR Data for Mountainous Areas with a Machine Learning Technique

Artificial intelligence for geoscience: Progress, challenges, and perspectives

This paper explores the evolution of geoscientific inquiry, tracing the progression from traditional physics-based models to modern data-driven approaches facilitated by significant advancements in artificial intelligence (AI) and data collection techniques. Traditional models, which are grounded in physical and numerical frameworks, provide robust explanations by explicitly reconstructing underlying physical processes. However, their limitations in comprehensively capturing Earth's complexities and uncertainties pose challenges in optimization and real-world applicability. In contrast, contemporary data-driven models, particularly those utilizing machine learning (ML) and deep learning (DL), leverage extensive geoscience data to glean insights without requiring exhaustive theoretical knowledge. ML techniques have shown promise in addressing Earth science-related questions. Nevertheless, challenges such as data scarcity, computational demands, data privacy concerns, and the "black-box" nature of AI models hinder their seamless integration into geoscience. The integration of physics-based and data-driven methodologies into hybrid models presents an alternative paradigm. These models, which incorporate domain knowledge to guide AI methodologies, demonstrate enhanced efficiency and performance with reduced training data requirements. This review provides a comprehensive overview of geoscientific research paradigms, emphasizing untapped opportunities at the intersection of advanced AI techniques and geoscience. It examines major methodologies, showcases advances in large-scale models, and discusses the challenges and prospects that will shape the future landscape of AI in geoscience. The paper outlines a dynamic field ripe with possibilities, poised to unlock new understandings of Earth's complexities and further advance geoscience exploration.

Monitoring 23-year of shoreline changes of the Zengwun Estuary in Southern Taiwan using time-series Landsat data and edge detection techniques

More than 80% of sandy beaches in Taiwan have been experiencing a severe recession, although the sediment discharge of rivers in Taiwan is significantly higher than the world average producing almost 2% of global fluvial sediment discharge. This contradiction is primarily due to the widespread constructions of reservoirs and intensive anthropogenic activities in coastal regions. In addition, coasts are particularly vulnerable to hazards due to climate change, such as sea-level rise, as they are located at the transition zone of terrestrial and marine environments. Along with the fact that Taiwan is an island and is one of the most climate-vulnerable regions globally, coastal management and sustainability are nationally critical topics, especially considering the ongoing reformation and legislation of Taiwan's coastal conservation laws. As stated in the Sustainable Development Goals (SDGs) goal 14, accurate and continuous shoreline positions information is essential for coastal conservation. However, by reviewing previous global studies and projects commissioned by the Taiwanese government aiming at monitoring shoreline changes, they usually exhibit several limitations, such as limited band selections or conservative band ratio-derived water indices, relying on either manual digitization or simple thresholding methods, focusing on either artificial or smoothly shaped coasts, and using images acquired at considerably different tidal height levels. Therefore, in the present study, a subpixel shoreline extraction approach based on a sustainable cross-generation dataset and a robust edge detection algorithm is proposed. This approach is exemplified by the Zengwun River Estuary located in southwestern Taiwan-Taiwan's most critical coastal preservation region. By quantitatively analyzing the resultant time-series shoreline positions from 1999 to 2021, several hotspots of shoreline recession have been identified: an extreme erosional rate up to -69.4 m year^-1 is revealed in the northern sand bank; while the offshore sand bar demonstrates an overall landward retreat rate of -35.4 ± 1.24 m year^-1.

High-Resolution Snow-Covered Area Mapping in Forested Mountain Ecosystems Using PlanetScope Imagery

Improving high-resolution (meter-scale) mapping of snow-covered areas in complex and forested terrains is critical to understanding the responses of species and water systems to climate change. Commercial high-resolution imagery from Planet Labs, Inc. (Planet, San Francisco, CA, USA) can be used in environmental science, as it has both high spatial (0.7–3.0 m) and temporal (1–2 day) resolution. Deriving snow-covered areas from Planet imagery using traditional radiometric techniques have limitations due to the lack of a shortwave infrared band that is needed to fully exploit the difference in reflectance to discriminate between snow and clouds. However, recent work demonstrated that snow cover area (SCA) can be successfully mapped using only the PlanetScope 4-band (Red, Green, Blue and NIR) reflectance products and a machine learning (ML) approach based on convolutional neural networks (CNN). To evaluate how additional features improve the existing model performance, we: (1) build on previous work to augment a CNN model with additional input data including vegetation metrics (Normalized Difference Vegetation Index) and DEM-derived metrics (elevation, slope and aspect) to improve SCA mapping in forested and open terrain, (2) evaluate the model performance at two geographically diverse sites (Gunnison, Colorado, USA and Engadin, Switzerland), and (3) evaluate the model performance over different land-cover types. The best augmented model used the Normalized Difference Vegetation Index (NDVI) along with visible (red, green, and blue) and NIR bands, with an F-score of 0.89 (Gunnison) and 0.93 (Engadin) and was found to be 4% and 2% better than when using canopy height- and terrain-derived measures at Gunnison, respectively. The NDVI-based model improves not only upon the original band-only model’s ability to detect snow in forests, but also across other various land-cover types (gaps and canopy edges). We examined the model’s performance in forested areas using three forest canopy quantification metrics and found that augmented models can better identify snow in canopy edges and open areas but still underpredict snow cover under forest canopies. While the new features improve model performance over band-only options, the models still have challenges identifying the snow under trees in dense forests, with performance varying as a function of the geographic area. The improved high-resolution snow maps in forested environments can support studies involving climate change effects on mountain ecosystems and evaluations of hydrological impacts in snow-dominated river basins.

A Machine Learning Snowfall Retrieval Algorithm for ATMS

This article describes the development of a machine learning (ML)-based algorithm for snowfall retrieval (Snow retrievaL ALgorithm fOr gpM–Cross Track, SLALOM-CT), exploiting ATMS radiometer measurements and using the CloudSat CPR snowfall products as references. During a preliminary analysis, different ML techniques (tree-based algorithms, shallow and convolutional neural networks—NNs) were intercompared. A large dataset (three years) of coincident observations from CPR and ATMS was used for training and testing the different techniques. The SLALOM-CT algorithm is based on four independent modules for the detection of snowfall and supercooled droplets, and for the estimation of snow water path and snowfall rate. Each module was designed by choosing the best-performing ML approach through model selection and optimization. While a convolutional NN was the most accurate for the snowfall detection module, a shallow NN was selected for all other modules. SLALOM-CT showed a high degree of consistency with CPR. Moreover, the results were almost independent of the background surface categorization and the observation angle. The reliability of the SLALOM-CT estimates was also highlighted by the good results obtained from a direct comparison with a reference algorithm (GPROF).

Impact of Image-Processing Routines on Mapping Glacier Surface Facies from Svalbard and the Himalayas Using Pixel-Based Methods

Glacier surface facies are valuable indicators of changes experienced by a glacial system. The interplay of accumulation and ablation facies, followed by intermixing with dust and debris, as well as the local climate, all induce observable and mappable changes on the supraglacial terrain. In the absence or lag of continuous field monitoring, remote sensing observations become vital for maintaining a constant supply of measurable data. However, remote satellite observations suffer from atmospheric effects, resolution disparity, and use of a multitude of mapping methods. Efficient image-processing routines are, hence, necessary to prepare and test the derivable data for mapping applications. The existing literature provides an application-centric view for selection of image processing schemes. This can create confusion, as it is not clear which method of atmospheric correction would be ideal for retrieving facies spectral reflectance, nor are the effects of pansharpening examined on facies. Moreover, with a variety of supervised classifiers and target detection methods now available, it is prudent to test the impact of variations in processing schemes on the resultant thematic classifications. In this context, the current study set its experimental goals. Using very-high-resolution (VHR) WorldView-2 data, we aimed to test the effects of three common atmospheric correction methods, viz. Dark Object Subtraction (DOS), Quick Atmospheric Correction (QUAC), and Fast Line-of-Sight Atmospheric Analysis of Hypercubes (FLAASH); and two pansharpening methods, viz. Gram–Schmidt (GS) and Hyperspherical Color Sharpening (HCS), on thematic classification of facies using 12 supervised classifiers. The conventional classifiers included: Mahalanobis Distance (MHD), Maximum Likelihood (MXL), Minimum Distance to Mean (MD), Spectral Angle Mapper (SAM), and Winner Takes All (WTA). The advanced/target detection classifiers consisted of: Adaptive Coherence Estimator (ACE), Constrained Energy Minimization (CEM), Matched Filtering (MF), Mixture-Tuned Matched Filtering (MTMF), Mixture-Tuned Target-Constrained Interference-Minimized Filter (MTTCIMF), Orthogonal Space Projection (OSP), and Target-Constrained Interference-Minimized Filter (TCIMF). This experiment was performed on glaciers at two test sites, Ny-Ålesund, Svalbard, Norway; and Chandra–Bhaga basin, Himalaya, India. The overall performance suggested that the FLAASH correction delivered realistic reflectance spectra, while DOS delivered the least realistic. Spectra derived from HCS sharpened subsets seemed to match the average reflectance trends, whereas GS reduced the overall reflectance. WTA classification of the DOS subsets achieved the highest overall accuracy (0.81). MTTCIMF classification of the FLAASH subsets yielded the lowest overall accuracy of 0.01. However, FLAASH consistently provided better performance (less variable and generally accurate) than DOS and QUAC, making it the more reliable and hence recommended algorithm. While HCS-pansharpened classification achieved a lower error rate (0.71) in comparison to GS pansharpening (0.76), neither significantly improved accuracy nor efficiency. The Ny-Ålesund glacier facies were best classified using MXL (error rate = 0.49) and WTA classifiers (error rate = 0.53), whereas the Himalayan glacier facies were best classified using MD (error rate = 0.61) and WTA (error rate = 0.45). The final comparative analysis of classifiers based on the total error rate across all atmospheric corrections and pansharpening methods yielded the following reliability order: MXL > WTA > MHD > ACE > MD > CEM = MF > SAM > MTMF = TCIMF > OSP > MTTCIMF. The findings of the current study suggested that for VHR visible near-infrared (VNIR) mapping of facies, FLAASH was the best atmospheric correction, while MXL may deliver reliable thematic classification. Moreover, an extensive account of the varying exertions of each processing scheme is discussed, and could be transferable when compared against other VHR VNIR mapping methods.

An Automated Snow Mapper Powered by Machine Learning

Snow preserves fresh water and impacts regional climate and the environment. Enabled by modern satellite Earth observations, fast and accurate automated snow mapping is now possible. In this study, we developed the Automated Snow Mapper Powered by Machine Learning (AutoSMILE), which is the first machine learning-based open-source system for snow mapping. It is built in a Python environment based on object-based analysis. AutoSMILE was first applied in a mountainous area of 1002 km2 in Bome County, eastern Tibetan Plateau. A multispectral image from Sentinel-2B, a digital elevation model, and machine learning algorithms such as random forest and convolutional neural network, were utilized. Taking only 5% of the study area as the training zone, AutoSMILE yielded an extraordinarily satisfactory result over the rest of the study area: the producer’s accuracy, user’s accuracy, intersection over union and overall accuracy reached 99.42%, 98.78%, 98.21% and 98.76%, respectively, at object level, corresponding to 98.84%, 98.35%, 97.23% and 98.07%, respectively, at pixel level. The model trained in Bome County was subsequently used to map snow at the Qimantag Mountain region in the northern Tibetan Plateau, and a high overall accuracy of 97.22% was achieved. AutoSMILE outperformed threshold-based methods at both sites and exhibited superior performance especially in handling complex land covers. The outstanding performance and robustness of AutoSMILE in the case studies suggest that AutoSMILE is a fast and reliable tool for large-scale high-accuracy snow mapping and monitoring.

Prediction of Active Microwave Backscatter Over Snow-Covered Terrain Across Western Colorado Using a Land Surface Model and Support Vector Machine Regression

The main objective of this article is to develop a physically constrained support vector machine (SVM) to predict C-band backscatter over snow-covered terrain as a function of geophysical inputs that reasonably represent the relevant characteristics of the snowpack. Sentinel-1 observations, in conjunction with geophysical variables from the Noah-MP land surface model, were used as training targets and input datasets, respectively. Robustness of the SVM prediction was analyzed in terms of training targets, training windows, and physical constraints related to snow liquid water content. The results showed that a combination of ascending and descending overpasses yielded the highest coverage of prediction (15.2%) while root mean square error (RMSE) ranged from 2.06 to 2.54 dB and unbiased RMSE ranged from 1.54 to 2.08 dB, but that the combined overpasses were degraded compared with ascending-only and descending-only training target sets due to the mixture of distinctive microwave signals during different times of the day (i.e., 6 A.M. versus 6 P.M. local time). Elongation of the training window length also increased the spatial coverage of prediction (given the sparsity of the training sets), but resulted in introducing more random errors. Finally, delineation of dry versus wet snow pixels for SVM training resulted in improving the accuracy of predicted backscatter relative to training on a mixture of dry and wet snow conditions. The overall results suggest that the prediction accuracy of the SVM was strongly linked with the first-order physics of the electromagnetic response of different snow conditions.

Monitoring Wet Snow Over an Alpine Region Using Sentinel-1 Observations

The main objective of this study was to monitor wet snow conditions from Sentinel-1 over a season, to examine its variation over time by cross-checking wet snow with independent snow and weather estimates, and to study its distribution taking into account terrain characteristics such as elevation, orientation, and slope. One of our motivations was to derive useful representations of daily or seasonal snow changes that would help to easily identify wet snow elevations and determine melt-out days in an area of interest. In this work, a well-known approach in the literature is used to estimate the extent of wet snow cover continuously over a season and an analysis of the influence of complex mountain topography on snow distribution is proposed taking into account altitude, slope, and aspect of the terrain. The Sentinel-1 wet snow extent product was compared with Sentinel-2 snow products for cloud free scenes. We show that while there are good agreements between the two satellite products, differences exist, especially in areas of forests and glaciers where snow is underestimated. This underestimation must be considered alongside the areas of geometric distortion that were excluded from our study. We analysed retrievals at the scale of our study area by examining wet snow Altitude–Orientation diagrams for different classes of slopes and also wet snow Altitude–Time diagrams for different classes of orientations. We have shown that this type of representation is very useful to get an overview of the snow distribution as it allows to identify very easily wet snow lines for different orientations. For an orientation of interest, the Altitude–Time diagrams can be used to track the evolution of snow to locate altitudes and dates of snow loss. We also show that ascending/descending Sentinel-1 image time series are complementary to monitor wet snow over the French alpine areas to highlight wet snow altitude ranges and identify melt-out days. Links have also been made between Sentinel-1 responses (wet snow) and snow/meteorological events carefully listed over the entire 2017–2018 season.

Spatial and Temporal Changes in Vegetation in the Ruoergai Region, China

Timely monitoring of the changes in coverage and growth conditions of vegetation (forest, grass) is very important for preserving the regional and global ecological environment. Vegetation information is mainly reflected by its spectral characteristics, namely, differences and changes in green plant leaves and vegetation canopies in remote sensing domains. The normalized difference vegetation index (NDVI) is commonly used to describe the dynamic changes in vegetation, but the NDVI sequence is not long enough to support the exploration of dynamic changes due to many reasons, such as changes in remote sensing sensors. Thus, the NDVI from different sensors should be scientifically combined using logical methods. In this study, the Global Inventory Modeling and Mapping Studies (GIMMS) NDVI from the Advanced Very High Resolution Radiometer (AVHRR) and Moderate-resolution Imaging Spectroradiometer (MODIS) NDVI are combined using the Savitzky–Golay (SG) method and then utilized to investigate the temporal and spatial changes in the vegetation of the Ruoergai wetland area (RWA). The dynamic spatial and temporal changes and trends of the NDVI sequence in the RWA are analyzed to evaluate and monitor the growth conditions of vegetation in this region. In regard to annual changes, the average annual NDVI shows an overall increasing trend in this region during the past three decades, with a linear trend coefficient of 0.013/10a, indicating that the vegetation coverage has been continuously improving. In regard to seasonal changes, the linear trend coefficients of NDVI are 0.020, 0.021, 0.004, and 0.004/10a for spring, summer, autumn, and winter, respectively. The linear regression coefficient between the gross domestic product (GDP) and NDVI is also calculated, and the coefficients are 0.0024, 0.0015, and 0.0020, with coefficients of determination (R2) of 0.453, 0.463, and 0.444 for Aba, Ruoergai, and Hongyuan, respectively. Thus, the positive correlation coefficients between the GDP and the growth of NDVI may indicate that increased societal development promotes vegetation in some respects by resulting in the planting of more trees or the promotion of tree protection activities. Through the analysis of the temporal and spatial NDVI, it can be assessed that the vegetation coverage is relatively large and the growth condition of vegetation in this region is good overall.

Monitoring Large-Scale Inland Water Dynamics by Fusing Sentinel-1 SAR and Sentinel-3 Altimetry Data and by Analyzing Causal Effects of Snowmelt

The warming climate is threatening to alter inland water resources on a global scale. Within all waterbody types, lake and river systems are vital not only for natural ecosystems but, also, for human society. Snowmelt phenology is also altered by global warming, and snowmelt is the primary water supply source for many river and lake systems around the globe. Hence, (1) monitoring snowmelt conditions, (2) tracking the dynamics of snowmelt-influenced river and lake systems, and (3) quantifying the causal effect of snowmelt conditions on these waterbodies are critical to understand the cryo-hydrosphere interactions under climate change. Previous studies utilized in-situ or multispectral sensors to track either the surface areas or water levels of waterbodies, which are constrained to small-scale regions and limited by cloud cover, respectively. On the contrary, in the present study, we employed the latest Sentinel-1 synthetic aperture radar (SAR) and Sentinel-3 altimetry data to grant a high-resolution, cloud-free, and illumination-independent comprehensive inland water dynamics monitoring strategy. Moreover, in contrast to previous studies utilizing in-house algorithms, we employed freely available cloud-based services to ensure a broad applicability with high efficiency. Based on altimetry and SAR data, the water level and the water-covered extent (WCE) (surface area of lakes and the flooded area of rivers) can be successfully measured. Furthermore, by fusing the water level and surface area information, for Lake Urmia, we can estimate the hypsometry and derive the water volume change. Additionally, for the Brahmaputra River, the variations of both the water level and the flooded area can be tracked. Last, but not least, together with the wet snow cover extent (WSCE) mapped with SAR imagery, we can analyze the influence of snowmelt conditions on water resource variations. The distributed lag model (DLM) initially developed in the econometrics discipline was employed, and the lagged causal effect of snowmelt conditions on inland water resources was eventually assessed.

Classification of Sea Ice Types in Sentinel-1 SAR Data Using Convolutional Neural Networks

A new algorithm for classification of sea ice types on Sentinel-1 Synthetic Aperture Radar (SAR) data using a convolutional neural network (CNN) is presented. The CNN is trained on reference ice charts produced by human experts and compared with an existing machine learning algorithm based on texture features and random forest classifier. The CNN is trained on two datasets in 2018 and 2020 for retrieval of four classes: ice free, young ice, first-year ice and old ice. The accuracy of our classification is 90.5% for the 2018-dataset and 91.6% for the 2020-dataset. The uncertainty is a bit higher for young ice (85%/76% accuracy in 2018/2020) and first-year ice (86%/84% accuracy in 2018/2020). Our algorithm outperforms the existing random forest product for each ice type. It has also proved to be more efficient in computing time and less sensitive to the noise in SAR data. The code is publicly available.

Automated Mapping of Antarctic Supraglacial Lakes Using a Machine Learning Approach

Supraglacial lakes can have considerable impact on ice sheet mass balance and global sea-level-rise through ice shelf fracturing and subsequent glacier speedup. In Antarctica, the distribution and temporal development of supraglacial lakes as well as their potential contribution to increased ice mass loss remains largely unknown, requiring a detailed mapping of the Antarctic surface hydrological network. In this study, we employ a Machine Learning algorithm trained on Sentinel-2 and auxiliary TanDEM-X topographic data for automated mapping of Antarctic supraglacial lakes. To ensure the spatio-temporal transferability of our method, a Random Forest was trained on 14 training regions and applied over eight spatially independent test regions distributed across the whole Antarctic continent. In addition, we employed our workflow for large-scale application over Amery Ice Shelf where we calculated interannual supraglacial lake dynamics between 2017 and 2020 at full ice shelf coverage. To validate our supraglacial lake detection algorithm, we randomly created point samples over our classification results and compared them to Sentinel-2 imagery. The point comparisons were evaluated using a confusion matrix for calculation of selected accuracy metrics. Our analysis revealed wide-spread supraglacial lake occurrence in all three Antarctic regions. For the first time, we identified supraglacial meltwater features on Abbott, Hull and Cosgrove Ice Shelves in West Antarctica as well as for the entire Amery Ice Shelf for years 2017–2020. Over Amery Ice Shelf, maximum lake extent varied strongly between the years with the 2019 melt season characterized by the largest areal coverage of supraglacial lakes (~763 km2). The accuracy assessment over the test regions revealed an average Kappa coefficient of 0.86 where the largest value of Kappa reached 0.98 over George VI Ice Shelf. Future developments will involve the generation of circum-Antarctic supraglacial lake mapping products as well as their use for further methodological developments using Sentinel-1 SAR data in order to characterize intraannual supraglacial meltwater dynamics also during polar night and independent of meteorological conditions. In summary, the implementation of the Random Forest classifier enabled the development of the first automated mapping method applied to Sentinel-2 data distributed across all three Antarctic regions.

A Combination of PROBA-V/MODIS-based Products with Sentinel-1 SAR Data for Detecting Wet and Dry Snow Cover in Mountainous Areas

In the present study, we explore the value of employing both vegetation indexes as well as land surface temperature derived from Project for On-Board Autonomy – Vegetation (PROBA-V) and Moderate Resolution Imaging Spectroradiometer (MODIS) sensors, respectively, to support the detection of total (wet + dry) snow cover extent (SCE) based on a simple tuning machine learning approach and provide reliability maps for further analysis. We utilize Sentinel-1-based synthetic aperture radar (SAR) observations, including backscatter coefficient, interferometric coherence, and polarimetric parameters, and four topographical factors as well as vegetation and temperature information to detect the total SCE with a land cover-dependent random forest-based approach. Our results show that the overall accuracy and F-measure are over 90% with an ’Area Under the receiver operating characteristic Curve (ROC)’ (AUC) score of approximately 80% over five study areas located in different mountain ranges, continents, and hemispheres. These accuracies are also confirmed by a comprehensive validation approach with different data sources, attesting the robustness and global transferability. Additionally, based on the reliability maps, we find an inversely proportional relationship between classification reliability and vegetation density. In conclusion, comparing to a previous study only utilizing SAR-based observations, the method proposed in the present study provides a complementary approach to achieve a higher total SCE mapping accuracy while maintaining global applicability with reliable accuracy and corresponding uncertainty information.

Remote Sensing of Snow Cover Using Spaceborne SAR: A Review

The importance of snow cover extent (SCE) has been proven to strongly link with various natural phenomenon and human activities; consequently, monitoring snow cover is one the most critical topics in studying and understanding the cryosphere. As snow cover can vary significantly within short time spans and often extends over vast areas, spaceborne remote sensing constitutes an efficient observation technique to track it continuously. However, as optical imagery is limited by cloud cover and polar darkness, synthetic aperture radar (SAR) attracted more attention for its ability to sense day-and-night under any cloud and weather condition. In addition to widely applied backscattering-based method, thanks to the advancements of spaceborne SAR sensors and image processing techniques, many new approaches based on interferometric SAR (InSAR) and polarimetric SAR (PolSAR) have been developed since the launch of ERS-1 in 1991 to monitor snow cover under both dry and wet snow conditions. Critical auxiliary data including DEM, land cover information, and local meteorological data have also been explored to aid the snow cover analysis. This review presents an overview of existing studies and discusses the advantages, constraints, and trajectories of the current developments.