A regional flow duration curve-based approach for predicting sediment yield at ungauged sites

Sediment yield estimates in river basins are essential for studies related to river morphology, water quality modeling, development of erosion control and management plans, and design of water control structures (e.g., dam, barrage). In data-scarce scenarios, calibration and validation of numerical models for estimating sediment yield become challenging. A regional FDC (Flow Duration Curve)-based methodology is proposed for predicting the sediment yield at ungauged locations in river basins. Its effectiveness was investigated through Jackknife cross-validation experiment on the frequent flood-prone Mahanadi basin, considering daily records of 13 sediment and flow monitoring stations for the time period 1980-2019. A set of 34 catchment-related attributes derived based on morphology, climate, landuse, and location were considered. Potential attributes influencing flows in the basin were identified as those having significant correlations with flow quantiles corresponding to 15 chosen exceedance probabilities (P=0.1, 0.5, 1, 5, 10, 20, 30, 40, 50, 60, 70, 80, 90, 95, and 99). The identified attributes include (i) catchment area, (ii) number of streams, (iii) minimum elevation, (iv) basin relief, (v) the percentage of area classifiable as grassland, and (vi) longitude. To arrive at sediment yield at an ungauged location, first FDC corresponding to the location is derived using regression relationships fitted between each of the 15 flow quantiles and the potential attributes of gauged sites in the region. The relationships were developed using the best subset regression analysis. Subsequently, the daily discharge and daily sediment time series at the ungauged location were derived from the FDC. For this purpose, the rating curve parameters for the ungauged site were obtained from its neighboring sites in the attribute space, through a proposed strategy. The performance of the proposed approach in predicting the discharge and sediment time series was found to be effective when assessed in terms of various performance measures, which included Nash-Sutcliffe efficiency, and Kling-Gupta efficiency.