Characteristics and predictive models of hillslope erosion in burned areas in Xichang, China, on March 30, 2020

Wildfires often greatly aggravate hillslope erosion, which is a complex, highly dynamic and constantly changing water-soil interaction process. In this study, the remote sensing interpretation, field investigation, in-situ soil erosion pins experiments, and laboratory test were used to examine the controlling factors of post-fire hillslope erosion after the Xichang Fires occurred on March 30, 2020, in southwest Sichuan Province, China. We developed an empirical model to predict predicting post-fire hillslope erosion, and evaluated the model performance at watershed scale combined with the Sediment Delivery Ratio model (SDR). The controlling factors of hillslope erosion (fire severity, soil properties, rainfall, and topography) were collected. The correlations between rainfall events-based hillslope erosion depth (HED) and the controlling factors were examined. The results show that the fire severity has the strongest correlation, followed by soil properties, rainfall, and topography. The new predictive models were developed using multiple linear and non-linear (power law relationship) regression based on the normalized differenced Normalized Burn Ratio (ndNBR), soil erodibility (K), accumulated rainfall erosivity (∑R) and topographic factor (LS). Then, the methods were evaluated by comparing the predicted and observed values on the testing subset based on the statistical coefficients including R-squared, root mean squared error (RMSE) and discrepancy ratio (DR). The results suggest that the developed empirical model based on multiple non-linear (power law relationship) has a better performance (R-squared: 0.600, RMSE: 1.948), and the discrepancy ratio of 87.2% data range from 0.5 to 2. The model was used to estimate the spatial distribution of hillslope erosion depth across the entire study area, and the sediment yield at watershed scale (WSY) combined with the SDR model. The comparison between the observed sediment mobilized by debris flow and the predicted WSY shows that the model combined with the SDR showed an acceptable performance at watershed scale.