Normalization of Sentinel 2 data over tropical forests to nadir BRDF adjusted reflectance

Abstract

Accurate surface reflectance retrieval in tropical forests remains challenging due to strong directional reflectance effects (BRDF) and limited observation opportunities caused by persistent cloud cover. Here, we present a sensor-adapted, physically based BRDF correction framework specifically designed for Sentinel-2 imagery over dense tropical forests. Using space-for-time substitution to calibrate Ross–Thick Li–Sparse (RTLS) models across central African forests, we derive Sentinel-2–specific parameters for nine spectral bands and apply a rigorous validation scheme based on spatiotemporally independent data.We demonstrate that our original calibration approach is not only feasible but necessary, given the amplitude of directional effects on raw forest reflectance – and the limited performances of MODIS-based correction models. The RTLS model, despite being originally developed for coarser resolutions, remains highly effective at S-2 scale, with mean relative percent absolute differences (MRPAD) reductions by up to 70% compared to uncorrected data, and by more than 50% on average compared to MODIS-derived BRDF parameters. Validation across continents confirmed the generalizability of the proposed coefficients.More than ten years after the launch of Sentinel-2, and as tropical forest monitoring becomes increasingly critical for climate and biodiversity applications, fully exploiting the sensor's spatial and spectral capabilities requires robust BRDF correction. To facilitate operational use, the proposed tools are distributed as both a Google Earth Engine script and an R package, enabling scalable and reproducible directional correction.