Effect of Mississippi River discharge and local hydrological variables on salinity of nearby estuaries using a machine learning algorithm

Abstract Estuarine water salinity is important to coastal ecosystems and the economy. To examine the potential effect of the Mississippi River freshwater discharge on water salinity of other estuaries, we used the machine learning Cubist algorithm to model salinity of three estuaries (Apalachicola Bay, Weeks Bay, and Grand Bay) of the northern Gulf of Mexico (GoM). The models were trained — both with and without the Mississippi River discharge as input — to examine its effect on the salinity of other estuaries along with other variables including wind, discharge, and water depth. By including the Mississippi discharge, the Root Mean Squared Error (RMSE) of the model training sets decreased by 0.81, 1.06 and 1.73 for the Apalachicola Bay, Weeks Bay, and Grand Bay, respectively, while the decrease of the model testing sets was 0.20, 0.64, and 0.84, respectively, using the practical salinity scale. The rank of predictor importance of the Mississippi discharge increased as the modeled estuary proximity increased to the Mississippi outlet. The results showed that the Mississippi River discharge can affect the water quality of other estuaries along the northern GoM, although the variation of the salinity explained by this flow was smaller than that explained by local river flow (except for the Grand Bay where there is no local river flow) and local wind.