STGCN-WQ: A new Spatio-Temporal Graph Convolutional Network for predicting beach Water Quality

Polluted waters pose significant health risks to beachgoers. While monitoring Fecal Indicator Bacteria (FIB) is a slow process, predictive models can serve as valuable tools for beach management by facilitating timely public health advisories. However, previous studies often overlook the spatiotemporal characteristics of beach water quality in their predictive models. This study addresses this gap by introducing a new Spatio-Temporal Graph Convolutional Network for predicting beach Water Quality (STGCN-WQ). Additionally, we propose a Spatio-Then-Temporal (STT) imputation strategy to handle missing data, which first leverages spatial correlations among neighboring beaches to estimate missing values and subsequently applies temporal interpolation to refine predictions. This two-step approach improves robustness against both irregular sampling and data sparsity. The STGCN-WQ model is applied to 24 beaches along the southern shoreline of Lake Erie, collecting 18,519 FIB sample records from 2009 to 2020. Results indicate that the STGCN-WQ model achieves significant improvements in performance metrics, with F1 score and AUC value increasing by 78% and 19%, respectively, compared to the baseline “Persistence Method”, which solely relies on the most recent observation collected prior to the current day for nowcasting FIB conditions. This study provides valuable insights and new tools for effective beach water quality management.