Understanding snow depth is crucial for water resource management, as it directly informs predictions of snowmelt timing. Accurate snowmelt timing helps communities that rely on snowmelt runoff prepare for changes in water supply. Remote sensing technologies, such as Light Detection and Ranging (LiDAR), enable precise snow depth estimation across spatial scales unachievable by traditional manual measurements and sensor networks. However, LiDAR measurements are limited in temporal frequency. Other remote sensing tools, like spaceborne synthetic aperture radar (SAR) and optical imagery, offer broader spatial coverage and higher temporal resolution than LiDAR, yet they lack precision and spatial resolution. Moreover, SAR and optical imagery are insufficient as standalone data sources for accurate snow depth estimation.
Machine learning, particularly deep learning, provides an opportunity to bridge these limitations by enabling accurate and precise SAR-and-imagery-derived snow depth estimation across unprecedented spatial and temporal scales. This study leverages aerial LiDAR data from the Airborne Snow Observatory as “ground truth” measurements of snow depth and uses SAR and optical imagery data from NASA’s Sentinel-1 and Sentinel-2 missions for regression analysis. We employ a supervised learning approach, utilizing an existing deep convolutional neural network (CNN) to estimate snow depth across the Western United States, where substantial aerial LiDAR data is available. We build upon this existing architecture by encoding snow depth sensor networks into the CNN model to test the benefits of incorporating additional highly accurate but spatially sparse snow depth measurements.
By refining this CNN with sensor network data, we expect enhanced accuracy in snow depth estimations from SAR and imagery-based observations. This improvement will support reliable, large-scale snow depth measurements that will ultimately enhance predictive capabilities in water resource management. The potential impact is magnified by the upcoming NASA-ISRO Synthetic Aperture Radar (NISAR) mission, which will provide high-resolution (5-10m) and frequent (12-day repeat cycle) global coverage. The integration of NISAR and other open-source satellite imagery with similar spatial and temporal granularity into this model could further amplify water resource sustainability and climate resilience research.