Nutrient balances, river loads, and a counterfactual analysis to determine drivers of Mississippi River nitrogen and phosphorus loads between 1975 and 2017

This data release includes data processing scripts, data products, and associated metadata for a study investigating trends in Mississippi River (MR) nitrogen and phosphorus loads to the Gulf of Mexico. This data release consists of three main components: 1) Nitrogen and phosphorus balances, which account for major nutrient inputs (fertilizer, manure, waste water treatment facility effluent, atmospheric deposition, weathering and nitrogen fixation) and outputs (crop harvest and removal and gaseous emissions of nitrogen). Annual balances are estimated for the Mississippi River Basin, which covers 2,887,854 square kilometers for the time period 1950 to 2017, 2) Weighted Regression on Time Season and Discharge (WRTDS) river loads, which account for delivery of nutrients to the Gulf of Mexico for one river monitoring site. River Loads were estimated for 6 water quality constituents: Total Nitrogen (TN), Total Phosphorus (TP), Nitrate, Ammonium, Orthophosphate, and Suspended Sediment for the time period 1975 - 2017. We used the flow normalized (FN) load and the "actual" (i.e. true condition) load estimated by WRTDS. Annual river loads and trends were estimated for the MR Outlet (MRO), a location near the mouth of the MR but upstream of the Old River Control Structure which diverts approximately 1/3 of the flow from the MR into the Atchafalaya River, and 3) We completed an impact evaluation, framed using a counterfactual approach, which is a technique that formally compares what actually happened to what would have happened under different conditions. Prior to the counterfactual analysis, we developed a multiple linear regression model to predict TN and TP load changes over time. We modeled annual FN river loads as a function of current nutrient balances, lagged nutrient balances, and a latent variable representing the aggregate effect of other potential causal factors. The trend attribution aimed to develop a regression model capable of capturing the simultaneous influence of contemporary nutrient balances (lag 0 term), lagged nutrient balances reflecting contributions from legacy nutrients retained in the watershed (higher lag terms), and the aggregate effect of potential latent processes affecting retention on changing nutrient loads over time (Year terms). Contemporary (current year) nutrient balances, which matched the same year as the river load data, and legacy (historical) N and P balances, which were sequentially shifted in one-year increments up to a lag of 25 years, were considered as independent variables. For the historical balances, 2017 would use the balances in 2016 through 1992 for the 1-year through 25-year lags, respectively. 25-years was the maximum lag, as 1950 was the first year that we had nutrient balance data. The most efficient model explaining TN loads included Year, Year2, Lag2, Lag4, Lag9, and Lag11 terms. The most efficient model explaining changes in TP river loads included Year, Year2, Year3, and Lag4 terms. We examined two different counterfactual scenarios, using hypothetical inputs to the calibrated TN and TP regression models. For Counterfactual A, the hypothetical inputs were current and lagged nutrient balances held constant at 1975 levels through 2017, and the Year terms were the same as the original inputs. The objective of holding the nutrient balance inputs constant was to investigate how river nutrient loads might have changed between 1975 and 2017 in the absence of any variability in nutrient balances after 1975. For Counterfactual B, the hypothetical inputs were the latent Year term held constant at 1975 levels through 2017, and the current and lagged nutrient balance inputs were the same as in the original inputs. The objective of holding the Year input constant at 1975 was to investigate how river nutrient loads might have changed between 1975 and 2017 in the absence of any variability in latent processes, potentially including best management practices implementation, watershed buffering capacity, and other factors. The impact analysis compared the mean annual counterfactual analysis results to the mean original regression results for the time period 2013 to 2017. The original regression results refer to the predicted river loads estimated from the calibrated regression model using the original data. The master list of input and output variables for all three of these components is available in the "MasterListVariables_MRB.xlsx" file.