Deciphering sediment pathways: Novel integrated approaches for sediment source identification and vulnerability prediction by machine learning models in major dam catchments in Chota Nagpur plateau, India

Abstract

The declining storage capacity of major dam reservoirs in the Chota Nagpur Plateau is primarily attributed to excessive sedimentation from upper catchments. However, identifying sediment source zones and understanding sediment connectivity remains a challenge. This study introduces an integrated ‘RUSLE–IC–SDR-SWAT-SEH-ML’ framework to assess reservoir sedimentation by combining soil erosion hotspots (SEH) with sediment connectivity pathways. The methodology was applied to the Maithon, Panchet, and Tenughat dam catchments to evaluate sediment yield (SY) variations. The results indicate an increasing trend in severe soil erosion (SE) across all catchments, with Maithon exhibiting an increase from 0.74 % to 1.36 % (Δ0.31, R² = 0.65), Panchet from 1.78 % to 3.58 % (Δ0.285, R² = 0.52), and Tenughat from 0.92 % to 1.49 % (Δ0.7, R² = 0.69). The SWAT model estimated mean SY at 9.146 $t{ha}^{-1}{yr}^{-1}$ (Tenughat), 5.871 $t{ha}^{-1}{yr}^{-1}$ (Panchet), and 7.662 $t{ha}^{-1}{yr}^{-1}$ (Maithon). Machine learning analysis showed SVM as the best performer for Maithon and Tenughat, while RF was superior for Panchet (R² = 0.998, 0.994) in predicting SY vulnerability. The extent of well-connected areas increased from 14.45 km² to 17.69 sq.km in Maithon, 21.22 sq.km to 26.64 sq.km in Panchet, and 18.77 sq.km to 22.96 sq.km in Tenughat, indicating an intensifying risk of sediment input into the reservoirs. The Mantel test confirmed that key variables explained 90–97 % of SY variance across the catchments (p < 0.001). ANOVA results showed a statistically significant difference (p < 0.001) within the catchments, while the LSD post-hoc test revealed significant differences (p < 0.05) between Maithon and Panchet, as well as Panchet and Tenughat. The observed differences between the Panchet and Tenughat dams (p < 0.001) are attributed to the regulation of water and sediment flow in the Panchet Dam's upper catchment through the Tenughat Dam. Additionally, the Panchet Dam catchment exhibited the highest sediment yield due to extensive mining activities, leading to significant statistical differences compared to Maithon. This study underscores the importance of integrating sediment connectivity analysis and machine learning models for effective reservoir sediment management. The findings provide critical insights for sustainable soil and water conservation strategies in sediment-prone dam catchments.