ArcticShield: A Lightweight Geospatial Framework for Permafrost Thaw Risk Assessment Using NASA POWER Surface Temperature Telemetry

Permafrost degradation across high-latitude regions represents an accelerating consequence of Arctic amplification, with significant implications for infrastructure stability, hydrological systems, and global carbon feedback mechanisms. Existing monitoring approaches, including borehole measurements and satellite-based deep learning pipelines, are constrained by high computational overhead, dependence on centralized cloud infrastructure, and limited deployability in remote field environments.

This paper presents ArcticShield, a lightweight geospatial framework designed to estimate permafrost thaw risk using surface temperature telemetry retrieved from the NASA Prediction of Worldwide Energy Resources (POWER) dataset. The framework employs a deterministic thermal heuristic that combines temporally aggregated surface temperature, a calibrated thermal offset approximating the zero-curtain effect, and regional soil moisture proxies to compute a normalized thaw risk score. Parameter calibration is performed using a regularized mean squared error objective against reference subsidence observations from documented thaw sites.

The system is implemented as an interactive, edge-capable analytical tool optimized for Windows ARM64 hardware, incorporating local SQLite caching to ensure responsiveness under intermittent connectivity. Experimental evaluation across three representative Arctic sites — the Batagaika Megaslump, Drew Point coastline, and Old Crow Flats — demonstrates stable risk estimates with an aggregate RMSE of 6.42 on a normalized 0–100 scale. Cached query execution achieves latency under 100 milliseconds, supporting real-time interactive use.

ArcticShield demonstrates that meaningful permafrost thaw risk assessment is achievable under strict computational and connectivity constraints, complementing large-scale modeling efforts by providing timely, interpretable situational awareness for field-oriented applications.