Developing a predictive environment-based model for mapping biological soil crust patterns at the local scale in the Sahel

Several studies have demonstrated the great range of possibilities offered by remote sensing in identifying, estimating and mapping biological soil crust (BSC) patterns, i.e. a feature recognised to play major functions in drylands. However those techniques are suitable mainly where BSC patterns are abundant ( > 30%) and vegetation cover low (< 10%), otherwise reflectance values matched different levels of BSCs mixed with vegetation and bare soil surfaces. This study developed an alternative methodology in mapping BSC presence in areas with a wide range of BSC cover associated with different mosaics encompassing vegetation and bare surfaces in the Sahel. Data were collected during intensive field surveys and remote sensing imagery of two typical Sahelian watersheds in western Niger (Banizoumbou and Tamou). Statistical methods were used to explore relationships between BSC occurrence and abundance and key environmental factors (rainfall, land use, land cover, vegetation, physical crusts). A predictive model of BSC spatial distribution was developed based on logistic regressions. This model allowed predicting and mapping BSC occurrence in areas where BSC cover ranged from 0 to 65% at Tamou (15% in average) and 1 to 48% at Banizoumbou (4% in average) and where vegetation cover ranged from < 1% to > 75%. Predicted values were obtained with an overall accuracy of 77.7% (kappa = 0.54), classifying the model as good and discriminant. This work is the first step in assessing the local scale ecological functions of BSC. Further work is needed for extrapolation at the regional scale in order to provide a useful tool for ecological surveys and for predictions of soil surface dynamics related to global changes in dryland areas.