Novel Remote Sensing Approach for Estimating Biocrust Fractional Cover In Semiarid Ecosystems

Drylands are particularly vulnerable to climate change and land degradation, which is a major issue given these represent Earths largest biome and support the livelihoods of billions of people. Given their important role in terrestrial carbon and water cycles and their vulnerability to global change, it is imperative that we are able to monitor large-scale changes in dryland vegetation and soil cover types. However, most dryland classification algorithms have focused solely on detecting shrub cover (e.g., Brandt et al., 2020). Separating out the fractional cover (fCover) of other cover types in these sparse, heterogeneous ecosystems, including the biological soil crusts (biocrusts) that are characteristic of dryland ecosystems worldwide (Belnap et al., 2016) has been at the limit of what is possible given the spatial and spectral resolution of existing remote sensing (RS) data. However, with increasing availability of novel RS datasets, we are now entering an era in which the full spectrum of dryland cover types can be detected. In this study, we used unsupervised spectral unmixing and a novel data fusion method combining 1m hyperspectral and a canopy height model (CHM) derived from LIDAR data collected by the NEON airborne observation platform flown at the Santa Rita Experimental Range near Tucson, AZ in August 2018, to separately detect biocrust fCover from other vegetation and bare soil cover types. We validate the derived biocrust spectral signatures against hyperspectral signatures collected in the field for 14 key biocrust species (Yan et al., in prep). We show that biocrust fCover detection is possible even without CHM information and with only 5 spectral endmembers. Our results open up the possibility that upcoming hyperspectral satellite missions will be able to detect biocrust fCover at broad spatial scales without the need for a large collection of field data to train the classification algorithm.