ABSTRACT
Frequent rainstorms caused by climate change are causing significant stresses and impacts on karst zones and even global hydrological systems. However, few reports have focused on rainstorm sediment events (RSE) based on long series, high-frequency signals in karst small watersheds. Present study assessed the process characteristics of RSE and analyzed the response of specific sediment yield (SSY) to environmental variables using random forest and correlation coefficients. Management strategies are then provided based on revised index of sediment connectivity (RIC) visualizations, sediment dynamics and landscape patterns, and modeling solutions for SSY are explored through the innovative use of multiple models. The results showed that the sediment process showed high variability (CV > 0.36), and the same index had obvious watershed differences. Landscape pattern and RIC show highly significant correlation with mean or maximum suspended sediment concentration (p<0.01, |r|>0.235). Early rainfall depth was the dominant factor affecting SSY (Contribution = 48.15 %). The hysteresis loop and RIC infer that the sediment of Mahuangtian and Maolike mainly comes from downstream farmland and riverbeds, while Yangjichong comes from remote hillsides. The watershed landscape is centralized and simplified. In the future, patches of shrubs or herbaceous plants should be added around the cultivated land and at the bottom of the sparse forest to increase the sediment collection capacity. The backpropagation neural network (BPNN) is optimal for modeling SSY, particularly for running the variables preferred by the generalized additive model (GAM). This study provides insight into understanding RSE in karst small watersheds. It will help the region to cope with future extreme climate change and develop sediment management models that are consistent with regional realities.
