Active layer thickness (ALT) is an essential climate variable for monitoring permafrost degradation, whose deepening can lead to increased greenhouse gas emissions, altered hydrology and ecology, infrastructure damage, and a positive climate feedback. Quantifying ALT spatial heterogeneity remains challenging due to the influence of localized variations in terrain, microclimate, snow/soil properties, vegetation cover, and surface disturbances. It is also unclear how local ALT patterns (e.g., sub-meter to 10 m) and mechanisms scale up to broader landscape footprints (e.g., 10–1000 m) represented from global satellite observations and Earth system models. We assessed ALT spatial heterogeneity in the Arctic-foothills tundra within the Northern Slope of Alaska through intensive field sampling over four 90 m × 90 m plots, combined with multi-source remote sensing and machine learning (ML). Analysis using field observations and ML revealed that vegetation, surface wetness, subsurface rocks, and micro-topography exert strong influence on 5-m ALT variations, whereas terrain controls dominate (~65 % contribution) at coarser 10-m spatial resolution. By leveraging cm-level optical-infrared drone imagery, we further generated 0.1-m ALT maps over a larger 5 km × 5 km region and examined ALT scaling effects. Our analysis showed a quadratic relationship in scale-dependent uncertainties, characterized by a rapid increase in uncertainties at the sub-meter level (e.g., RMSE normalized by the standard deviation of 0.1 m ALT climbed by ~10 %), followed by another 10 % increase from 1 m to 30 m resolution, and more conservative error increase (~5 %) from 30 m to 1,000 m scale. Our study allows for improved interpretation of remote sensing and process-based ALT simulations for the changing Arctic by clarifying scale-dependent uncertainties and underlying mechanisms.