Estimating sediment and nutrient delivery ratios in the Big Sunflower Watershed using a multiple linear regression model
Authors: N. Kannan, E. Osei, Y. Cao, A. Saleh
This study is part of an effort to analyze the nutrient load reductions obtained from current and future best management practices implementations in the Big Sunflower Watershed to meet the 45% nutrient reduction goal set for the watershed based on the US Environmental Protection Agency Science Advisory Board's (USEPA 2007) Gulf of Mexico hypoxia report. This paper describes the identification of dominant pollutant delivery mechanisms in the watershed, estimation of instream pollutant delivery ratios (DR) from subbasins to watershed outlet, and development of a tool to estimate changes in instream pollutant DR for what-if scenarios. The Big Sunflower Watershed is a 7,800 km2 intensively cultivated agricultural watershed in the State of Mississippi. The Comprehensive Environmental and Economic Optimization Tool (CEEOT) modeling system, consisting of the Soil and Water Assessment Tool (SWAT) and Agricultural Policy and Environmental Extender (APEX) models, was used to develop a multiple regression equation to estimate the sediment and nutrient DRs for this watershed. The models used 32 years of weather data from 1981 to 2012. The explanatory variables considered for the DR are distance to watershed outlet, flow, and pollutant loads leaving subbasins. They were chosen based on their strength of correlations and type of relationship with DR. Our results indicate that flow from each subbasin is the dominant factor affecting DR for this watershed. Together, the explanatory variables considered under the multiple linear regression framework were able to estimate sediment and nutrient DRs with satisfactory regression parameters. The R2 values for the regression relationship between the pollutant DRs and their counterparts estimated with multiple linear regression method were 0.8 for sediment, 0.96 for total nitrogen (N), and 0.9 for total phosphorus (P). The corresponding standard errors were 0.01 for sediment, 0.03 for total N, and 0.07 for total P. The explanatory variables were more strongly correlated to sediment DR than to nutrient DR. The tool developed to analyze changes in DRs for alternative scenarios appears to be useful for watershed managers.