Predicted and actual results can be similar, but they often differ due to insufficient field investigation data, the complexity of the model parameters and spatio-temporal scale effects (Boardman, 2006; Ni et al., 2008).
Close agreement of modeled and measured results will afford some degree of validation, but it will not provide conclusive confirmation of the validity of the internal functioning of the model and, thus, of the predicted erosion rates. We assessed soil erosion in response to different land uses for slope cropland, grassland and forest, which we determined in relation to values obtained from 137Cs measurements made at ecological observation stations on the Loess Plateau (Fig. 9).
A significant relationship (P b 0.01) between the erosion rates determined by 137Cs measurements and the estimates by the RUSLE model was found (Fig. 10). The deviation of the regression coefficient of 0.7892 from 1.000 may be due to heterogeneity at the 1 km spatial resolution because of the mosaic of different vegetation and topography patterns.
The model's performance was also assessed based on the model efficiency (ME).
The ME in our study was 0.77, which was considered good (Vigiak et al., 2011).
Therefore, these findings show that the model applied in our study can provide valuable information for ecosystem restoration and soil erosion control.