Effect of polyacrylamide applying on soil–rock mixture detachment and prediction of detachment capacity using machine learning

Abstract Soil detachment capacity ( D c ) is a key parameter for characterizing the soil erosion process. Polyacrylamide (PAM) mitigates soil erosion, but the mechanism by which it acts on soil–rock mixtures is unclear. This study investigated the impact of applying PAM on detachment of soil–rock mixtures and predicted D c using machine learning models. Small‐sample scouring tests were conducted in a flume with a 30° slope, under flow discharges of 4, 8, 12, 16, and 24 L·min −1 ; gravel content of 0%, 10%, 30%, 50%, and 70%; and PAM (anionic type, molecular weight 12 million, degree of hydrolysis 20%) application rates of 0, 1, 2, 3, 4, and 5 g·m − 2 . When flow discharge was lower than 16 L·min −1 , the best D c inhibition effect was achieved by applying 4 g·m −2 PAM rate. From 16–24 L·min −1 , the optimal application rate of PAM for D c inhibition varied according to gravel content: 3 g·m −2 for gravel content of <50% and 4 g·m −2 for gravel content of 50%–70%. PAM primarily influenced D c indirectly by enhancing shear strength, but as gravel content increased, PAM effect on shear strength reduced. At 30% gravel content, the soil–rock mixture was more stable, and D c remained consistently low. The extreme gradient boosting model trained using four parameters (PAM application rate, gravel content, shear strength, and stream power) outperformed multiple regression equations when used to predict D c .