Comparing Two Applicative Criteria of the Soil Erosion Physical Model Concept

Abstract

Core Ideas The physical model concept allows testing of the reliability of soil loss predictions. This concept was applied by using the original and a new criterion in southern Italy. The new criterion was generally more restrictive than the original one. The physical model represented by a replicated plot has been suggested to be the best possible, unbiased, real world model to predict plot soil erosion. The aim of this investigation was to compare the original applicative criterion of the physical model concept proposed by Nearing with that later suggested by Bagarello et al. The comparison was performed by using three empirical soil erosion models (the Universal Soil Loss Equation [USLE], a modified USLE [USLE‐MM], and the Central and Southern Italy [CSI] model) and plot soil loss data collected at the experimental station of Sparacia, in Sicily (southern Italy). The investigation showed that (i) the new criterion was generally more restrictive, i.e., less prone to accept the similarity hypothesis between predictions and measurements, than the original one; (ii) the new criterion gave a similar number of acceptable predictions as the original one when absolute differences between measured and predicted soil losses by the replicated plot were associated with a frequency occurrence factor of 0.87; (iii) for both tested criteria, the percentage of acceptable predictions could be considered time independent, and consequently, the checked performances of the three soil loss models could be considered generally representative of their capability in predicting soil losses at the sampled site; (iv) with the exception of the calibrated USLE, the hypothesis that, according to the original criterion, an effectiveness coefficient greater and lower than 0.6 could be expected for a calibrated and uncalibrated model, respectively, was confirmed; and (v) the new criterion allowed establishment of an effectiveness coefficient of 0.12, which discriminates between uncalibrated and calibrated models.