Modeling cover management factor of RUSLE using very high-resolution satellite imagery in a semiarid watershed

Vegetation cover is regarded as one of the most important protection measures for controlling soil erosion caused by water. Numerous articles have been published about the fact that more delicate, practical, and reliable estimations can be made through normalized difference vegetation index (NDVI) for calculating “cover management (C) factor” in the Revised Universal Soil Loss Equation (RUSLE), the most commonly recognized erosion prediction model worldwide. In this study, the C-factor map of the Tortum-North sub-watershed in the mountainous northeastern part of Turkey was estimated using NDVI values derived from the 50-cm resolution WorldView-2 satellite imagery. The C-factor values, collected from 55 sampling plots by measuring crown closure, canopy height, litter layer depth, and surface cover of the study area, were plotted against the NDVI values and then curved using the simple linear regression method. The resulting regression models (linear, cubic, exponential, growth) and five other best-known NDVI-related models from the literature (Knijff, Smith, Karaburun, De Jong, and Durigon) were compared using model diagnostic statistics ($$R_{{\text{adj}}}^{2}$$Radj2, RMSE, MAE, Mallows’ Cp) and information criterion statistics (Akaike’s information criterion, the Sawa’s Bayesian information criterion, Schwarz’s Bayesian criterion). The curve estimation results showed that the cubic model (R2 = 0.83, RMSE = 0.063), the Knijff et al. (1999)’s model (R2 = 0.85, RMSE = 0.059), and the linear model (R2 = 0.81, RMSE = 0.067) were the top three estimators of the C-factor. The least estimator of the C-factor was the growth model (R2 = 0.46, RMSE = 0.113). The residual analysis results showed that the cubic model performed well (total score of 57) by the best fitting of the overall regression model selection process. It was concluded that the C-factor estimation can be improved by the NDVI-based per-pixel approach using very high-resolution satellite imagery in the semiarid mountainous areas.