Combining optical and microwave remote sensing for assessing gullies in human-disturbed vegetated landscapes

[Display omitted] •Products SOIL_ABU, TSAVI, and Color Index are very promising for gully assessments.•SAR, optical texture and water/vegetation indices were helpful.•The method achieved 89% overall accuracy by the percentage of exposed soil.•Our methodology is useful to guide recovery plans for tropical gullies. The accurate assessment of the gully is key to stopping soil loss, especially in agricultural landscapes. This study aims to combine freely distributed remote sensing data for the evaluation of gullies located in a tropical watershed with a history of cattle production. Eighty-four vectorized gullies were defined in the Pirizal River basin, part of the highly eroded Upper Taquari (Brazil). We examined 56 variables from Sentinel-1/2 and ALOS-PALSAR-1 datasets, including SAR products and optical products like textures, water, vegetation, and terrain indices. Following a correlation analysis, 19 variables were selected for mapping in a Random Forest classifier by considering samples of active (soil) and stabilized (vegetation) pixels. The method reached an overall accuracy of 89%, in which soil abundance was responsible for 44% of the overall importance in the classification. Optical indices and texture products outperformed SAR products, whose importance represented only 14%. In the studied river basin, about 63% of the gullies were found stabilized, 30% in the process of stabilization, and only 7% active. The method proved effective, low cost, and promptly replicable to general river basins with gullies, mainly those in the tropics where vegetation has a significant role in soil loss control.