Identifying Scale-Dependent Snow patterns from learned fusion of Multi-modal, Multi-resolution satellite observations

Snow accumulation and melt dynamics govern water availability and flood timing across high-latitude catchments, yet observational gaps constrain understanding of scale-dependent hydrologic processes. Traditional snow monitoring relies on sparse gauge networks or coarse satellite products, preventing observation of sub-catchment patterns critical for hydrologic connectivity. Recent advances in Sentinel-1 SAR (10m, all-weather) and Sentinel-2 optical (10m, 20m, 60m) constellations offer transformative observational capabilities, yet systematically exploiting their complementary information for continuous fine-resolution snow monitoring across cloud-prone regions remains an operational challenge.

We present an innovative all-weather snow monitoring system that fuses Sentinel-1 SAR backscatter (sensitive to snow wetness and surface properties) with Sentinel-2 optical imagery (discriminating snow from clouds and bare ground) to deliver 10m resolution fractional snow cover estimates across boreal Finland. This fusion approach explicitly addresses the fundamental limitation of optical-only monitoring: persistent cloud contamination prevents observations during critical winter periods in high-latitude regions. Our methodology incorporates quality-aware atmospheric corrections (cloud masks, aerosol optical thickness, water vapor) to extract reliable snow information despite challenging atmospheric conditions.

A data-driven multi-resolution framework bridges the critical scale gap between fine-resolution satellite observations (10m) and operational hydrologic models requiring catchment-aggregated snow states. The system learns scale-dependent aggregation and disaggregation functions directly from observations, preserving fine-scale spatial patterns essential for understanding snow redistribution by wind, sublimation, and terrain-driven processes. This approach captures heterogeneity at forest-canopy scales while remaining compatible with distributed hydrologic model architectures.

Operational validation demonstrates that the system achieves physically realistic snow patterns with spatially coherent uncertainty estimates that appropriately elevate at snow-land boundaries where hydrologic transitions occur. These calibrated uncertainty bounds are critical for risk-informed water management and probabilistic flood forecasting, enabling downstream hydrologic models to appropriately weight observational constraints.

Key scientific innovations: (1) Demonstrated feasibility of all-weather snow monitoring by effectively combining complementary SAR and optical signatures, overcoming the cloud-cover limitation that constrains optical-only approaches during 60-80\% of winter days in boreal regions. (2) Developed a principled multi-scale learning framework that explicitly captures scale-dependent aggregation and disaggregation properties, bridging satellite observations and hydrologic model requirements. (3) Resolved sub-catchment snow heterogeneity previously masked in operational products (MODIS: 500m, VIIRS: 375m), enabling new insights into snow redistribution and hydrologic connectivity across fragmented landscapes. (4) Quantified spatial structure in prediction uncertainty, enabling probabilistic hydrologic forecasting that appropriately reflects observational constraints.

This next-generation observational capability addresses critical scientific and operational data gaps: calibrating distributed snow models at relevant scales, improving melt timing predictions through continuous all-weather depletion monitoring, validating snow-pack simulations in data-sparse headwater regions, and quantifying snow-climate feedbacks across heterogeneous landscapes. The framework's transferability to pan-Arctic and mountain regions demonstrates how integrating complementary space-based observations through data-driven fusion unlocks fine-scale process understanding previously limited by observational constraints, advancing our capacity for water security assessment and climate adaptation planning in snow-dependent regions.