Mapping the successional stages of biological soil crusts at 3-m resolution in the Gurbantunggut Desert, China through hydration-induced spectral response

Biological soil crusts (biocrusts) are essential components of desert ecosystems and provide diverse ecological services that benefit both the environment and human society. Biocrusts exhibit varying ecological functions as they pass through several successional stages—cyanobacteria, lichens, semi-mosses, and mosses. Remote sensing has been widely applied to monitor the spatial and temporal distributions of biocrusts. However, previous efforts have focused primarily on identifying biocrusts while disregarding their distinct successional stages. Additionally, biocrusts remain dormant or inactive for most of the year, resulting in biocrusts at different successional stages with similar spectral characteristics, making them challenging to distinguish. Fortunately, biocrusts at different successional stages exhibit distinct spectral responses to hydration events. By leveraging imagery with high temporal (1-day) and spatial (3-m) resolutions from the PlanetScope constellation, this study attempts to map biocrust successional stages on a regional scale using transient spectral responses induced by a snowmelt event. We employed a two-stage mapping framework utilizing the random forest (RF) model. The aim of the first stage was to identify biocrusts, while the second stage was focused on mapping their distinct successional stages. The results showed that snowmelt induces noticeable changes in biocrust spectra, helping to distinguish between biocrusts and other background components and among different stages of biocrust succession. Our mapping framework achieved overall accuracies of 0.96 (252 out of 263 correctly identified samples) and 0.8 (85 out of 106 correctly identified) in the above two stages, respectively, highlighting its ability to delineate spatial patterns of successional stages across landscape and regional scales. This study lays a foundation for future in-depth exploration of desert ecosystem dynamics, including structure, ecological services, and responses to climate change and human activities. Furthermore, we suggest that event-induced spectral responses could improve classification accuracy, especially when spectral features are similar under general conditions.