Monitoring rainfall events in desert areas using the spectral response of biological soil crusts to hydration: Evidence from the Gurbantunggut Desert, China

Abstract

Biological soil crusts (biocrusts) are essential biotic components in desert ecosystems, yet they are vulnerable to climate change, especially changes in precipitation. Unfortunately, the scarcity and unreliable performance of mainstream precipitation products in desert regions has prevented further research on the responses of desert ecosystems to climate change scenarios. By taking advantage of the spectral sensitivity of biocrusts to hydration, this study aims to examine the potential to monitor rainfall events using the spectral response of biocrusts. First, BRDF harmonization enhanced the rainfall-induced spectral response. Then, a detectability analysis was performed to clarify how influential factors affect the spectral response. Finally, a random forest (RF) model was developed to derive the temporal dynamic and spatial pattern of detectable rainfall events. The results show that rainfall events induce significant changes in biocrust spectra, especially within some spectral bands (i.e., SWIR, Red, and Blue) and indices (i.e., BSCI and NDVI). A rainfall event is detectable if the precipitation amount is >6 mm and recorded by satellites within 24 h. The RF model performs better (F1 score = 0.78) than the mainstream precipitation products (F1 score between 0.19 and 0.52) in temporal monitoring. Additionally, the model can delineate the spatial pattern of rainfall events well. These findings imply that this phenomenon has great potential for application in rainfall monitoring in desert areas, which could greatly contribute to understanding the mechanisms of desert ecological processes on a large scale.