Advanced image processing methods as a tool to map and quantify different types of biological soil crust

Biological soil crusts (BSCs) modify numerous soil surface properties and affect many key ecosystem processes. As BSCs are considered one of the most important components of semiarid ecosystems, accurate characterisation of their spatial distribution is increasingly in demand. This paper describes a novel methodology for identifying the areas dominated by different types of BSCs and quantifying their relative cover at subpixel scale in a semiarid ecosystem of SE Spain. The approach consists of two consecutive steps: (i) First, Support Vector Machine (SVM) classification to identify the main ground units, dominated by homogenous surface cover (bare soil, cyanobacteria BSC, lichen BSC, green and dry vegetation), which are of strong ecological relevance. (ii) Spectral mixture analysis (SMA) of the ground units to quantify the proportion of each type of surface cover within each pixel, to correctly characterize the complex spatial heterogeneity inherent to semiarid ecosystems. SVM classification showed very good results with a Kappa coefficient of 0.93%, discriminating among areas dominated by bare soil, cyanobacteria BSC, lichen BSC, green and dry vegetation. Subpixel relative abundance images achieved relatively high accuracy for both types of BSCs (about 80%), whereas general overestimation of vegetation was observed. Our results open the possibility of introducing the effect of presence and of relative cover of BSCs in spatially distributed hydrological and ecological models, and assessment and monitoring aimed at reducing degradation in these areas.