Monitoring Forest Foliar Moisture Using Sentinel-2 Reflectance and Radiative Transfer Model Inversion

Foliar moisture content (FMC) of forests canopies is important for understanding biodiversity, including animal habitat quality for forest-dependent fauna and forest fire risk. In this study, we modelled spatiotemporal variation in FMC using optical remote sensing. We inverted the PROSPECT and GEOSAIL Radiative Transfer Models (RTM) on Sentinel-2 satellite reflectance data of 20 m ground resolution to retrieve FMC, and then predicted FMC at sub-continental scales using a random forest (RF) regression emulation of the RTM model. An RTM look-up-table, from earlier work, was filtered by ecological criteria from 24 sites sampled along precipitation and canopy cover gradients and was inverted by using the spectral angle, as a merit function. The RF emulator allowed efficient computation and presentation of an extensive and fine-scale forest FMC data cube of interest in animal, fire and plant ecology. Our RTM-based, predictions of forest and woodland FMC had a root mean square error (RMSE) of 19.9% of dry matter content and explained more than 60% of variance (r ² = 0.62). This represents an improvement over previous models using reflectance data of coarser spatial resolution, particularly in terms of explained variance (r ² =0.17 and RMSE=32%). The emulator model achieved similar performance to the RTM with slightly larger error (RMSE = 21.77%) and smaller explained variance (r ² = 0.54). Overall, both models performed best in forests, and woodlands of moderate to high canopy density (> ∼0.75 LAI). This study demonstrates that FMC can be monitored at spatial resolutions that allow intra- or inter-landscape patterns to be resolved surpassing previous capabilities.