Model Performance Results in Myrtle Beach, SC Using Virtual Beach and R Regression Software

Daily forecasts of beach water bacteria levels have been developed and automated by a beach water quality forecast team. With support from the Southeast Coast Ocean Observing Regional Association (SECOORA), R software and a variety of data sources were used to model daily bacteria levels in beach swimming waters in Myrtle Beach, SC. Modeled (predicted) water quality results are then shown for beach locations via a website and mobile device app. While R provides a robust set of tools for use in forecast modeling, the software has an extensive learning curve and requires skilled statistical interpretation of model results, which may reduce application of the approach in other areas. To address some of these concerns, the Environmental Protection Agency (EPA) created the “Virtual Beach” software package. Virtual Beach was developed to allow robust predictive models to be created without a long learning period. The forecast team and EPA were interested in comparing model performance using both R and Virtual Beach outputs to evaluate the utility of the more user-friendly Virtual Beach. Predictive models were developed and performance was analyzed using both packages. Recommendations were made based on ease of use and several performance measures. Model results indicate the two software packages yield comparable outputs in terms of performance. However, Virtual Beach tends to create better bacterial concentration predictions with more robust model forecasts, while R tends to produce more flexible model options and outputs.

Keywords: