Accurate flow prediction is a primary goal of hydrological modeling studies, which can be affected by the use of varying rainfall datasets, autocalibration methods, and performance indices. The combined effect of three rainfall datasets — Fifth generation of European ReAnalysis (ERA-5), Gridded meteorological data (gridMET), Global Precipitation Measurement Integrated Multi-satellitE Retrievals (GPM IMERG) — and three autocalibration techniques — Dynamically Dimensioned Search (DDS), Generalized Likelihood Uncertainty Estimation (GLUE), Latin Hypercube Sampling (LHS) — on SWAT+ river flow prediction was measured using three evaluation metrics — Nash Sutcliffe Efficiency (NSE), Kling Gupta Efficiency (KGE) and coefficient of determination (R 2) — for two watersheds in North Carolina (Cape Fear, Tar Pamlico) using the Soil Water Assessment Tool Plus (SWAT+) model. Five parameters in the SWAT+ model, cn2, revap_co, flo_min, revap_min, and awc, were found to be significantly sensitive under all combinations for both watersheds. Simulated flow varied more with the change in rainfall than the calibration technique used. We discovered that GPM IMERG gave the best results of the rainfall datasets, followed by ERA-5 and gridMET. We observed that the NSE score is more sensitive to different combinations of rainfall datasets and calibration techniques than the KGE scores. SWAT+ underperformed in the prediction of base flow for the groundwater-driven watershed. Overall, we recommend using the GPM IMERG rainfall dataset with the GLUE optimization technique and KGE performance index for optimal flow simulations. The results from this study will help hydrological modelers choose an optimal combination of rainfall dataset, autocalibration technique, and performance index depending on watershed characteristics.

Private groundwater wells have the potential to be an unmonitored source of contaminants that can harm human health for millions of people throughout the United States. Developing models that predict potential exposure to contaminants, such as nitrate, could guide sampling efforts and allow the residents to take action to reduce their risk. Machine learning models have been successful in predicting nitrate contamination using geospatial information such as proximity to nitrate sources or soil type, but previous models have not considered meteorological factors that change temporally. In this study, we test random forest (regression and classification) and linear regression models to predict nitrate contamination of wells using rainfall and temperature records over the previous 180-days. We trained and tested models for (1) all of North Carolina, (2) each geographic region in North Carolina, (3) a three-county region with high density animal agriculture, and (4) a three-county region with a low density of animal agriculture. All regression models had poor predictive performance (R2 = 0.04) for all areas tested. The random forest classification model for the coastal plain region showed fair agreement (Cohen’s kappa = 0.23) when trying to predict whether contamination occurred. All other classification models had slight or poor predictive performance. Our results show that temporal changes in rainfall and temperature alone are not enough to predict nitrate contamination in most areas of North Carolina but show potential in the coastal plain region.