Multiple Linear Regression Model-Building To Predict Basin-Sediment Production Rates and Reservoir Capacity Loss Across The United States

Dam reservoirs across the nation are infilling with sediment, which reduces their water storage capacity and can hinder dam operations. However, there are no bathymetric surveys to constrain the volume of sediment and remaining capacity at the majority of 90,000 reservoirs across the United States. The goal of our study is to use data from sites that have been surveyed to detect trends in sedimentation that we can then apply to unsurveyed sites across the United States. We used repeat survey data (first and last survey) at 535 sites across the US to constrain sediment yield and basin-sediment production rates; most importantly, we accounted for upstream dams and changing trap efficiency (see Foster et al., SEDHYD 2022).

We are conducting a multiple linear regression analysis in Python to detect trends in the basin-sediment production rates for watersheds above our 535 reservoir sites. Drainage basins utilized in this study scale from between 3.7 km^2 and 2.8 x 10^6 km^2 across the contiguous US, spanning various climate zones and topographic regions. Examples of the environmental parameters we are testing include: precipitation, temperature, lithology, soil erodibility, latitude, land use, drought persistence, and sediment and hydrological parameters. In our preliminary data analysis thus far, drainage density is the only individual parameter that correlates well with sediment production for all 535 basins (R^2=0.66). We expected poor fits when considering the nation as a whole, as the factors influencing sediment production vary significantly across the contiguous US. Subdividing the dataset into IECC Climate Zones provides better correlations within the subsets. In IECC Zone 2, representative of hot climates, sediment production correlates best with basin-averaged precipitation (R^2= 0.64, n=20) whereas in Zone 7 (very cold), basin ruggedness (R^2= 0.55, n=11) correlates best with sediment production. By building individual and multiple linear regression models for each climate zone, we aim to identify the controlling variables for particular regions across the US; we will then use these controlling variables to develop predictive equations for basin-sediment production rates for watersheds within each zone. Combined with model assumptions for trap efficiency at individual dams, we can then calculate sedimentation and capacity losses at tens of thousands of unsurveyed reservoirs in the US. Importantly, this analysis will enable dam and reservoir managers to identify reservoirs in the upstream basin that have filled with sediment and may become sediment sources as their trap efficiency further reduces through time.