Combining Spectral Libraries and 3-D Radiative Transfer Modeling to Improve the Simulation of Phenology in Semi-Arid Grasslands

Semi-arid grasslands and savannas are essential for human well-being, and are particularly affected by climate change effects. Remote sensing data offers the possibility to survey the evolution of these ecosystems through the estimation of vegetation traits that quantify ecosystem function and health. However, physically based methods such as radiative transfer models (RTMs) have been shown to have larger uncertainty in simulating semi-arid grasslands’ radiometric response due to their high spatio-temporal heterogeneities, including failing to account for the response of nonphotosynthetic vegetation (NPV) throughout the phenological cycle. As such, this work proposes a three-dimensional modeling framework which integrates NPV optical libraries, compiled from in situ measurements at the study site, to represent different phenophases within the discrete anisotropic radiative transfer model. To improve our understanding of the target ecosystems, and relate it with the modeling results and uncertainties, we quantified the spatio-temporal spectral and vegetation traits variability in a semi-arid savanna located in southwestern Europe. The results showed that the more arid phenophases presented the highest traits variability, in contrast to the spectral variability patterns, which generally achieved maximum levels during the biomass peak. However, the proposed RTM model was able to capture the main spectral features of a semi-arid grassland, as the simulations compared well against field, airborne and satellite observations. Spectral angle mapper (SAM) angular distances between simulated and observed spectra ranged between 4.5° and 8.8°, considering all scales and phenophases assessed. Lower SAM values (1.5° to 2.1°) were obtained in the shortwave infrared region, which demonstrates an adequate simulation of the senescent herbaceous and litter cover. The proposed modeling framework has the potential to improve vegetation trait estimation through proximal and remote sensing and to better protect these valuable ecosystems from climate-induced disturbances such as droughts or wildfires.