A new method to evaluate the accuracy of the sediment source mixing model

Soil erosion is the worldly environmental and ecological problems. How to accurately identify the source of the sediment is important for soil and water conservation. Fingerprint identification technology has been widely used in the extraction of sediment source proportion, but currently most of the methods used to evaluate the accuracy of the results are a-good-fit (GOF) or a mean absolute error (MAE). We propose a new method to evaluate the accuracy of the sediment source mixing model and quantitatively evaluate the two sediment source mixing models. A typical check dam in the Loess Plateau was used to evaluate the new method by combining field sampling and numerical simulation. Collins (C) and Modified Hughes mixing (M-H) models were used to quantitatively analyze the sediment sources in the dam-control watershed. The results showed that the optimum composite fingerprints were Mg, Cr, Ni, and TOC, and they had 97.2% discrimination ability. The contribution rates of sediment source from gully, farmland, grassland and branch ditch were 54%, 24%, 15% and 7%, respectively. The M-H mixing model had a higher comprehensive score (2.26) when compared with the C mixing model (2.20). The comprehensive evaluation method could provide an effective scientific theoretical basis for optimal allocations of water and soil conservation in small watersheds.