Modeling system for predicting enterococci levels at Holly Beach

Remote sensing algorithm for retrieving global-scale sea surface solar irradiance

This paper presents a new remote sensing (RS) algorithm for retrieving instantaneous sea surface solar irradiance (SR) by using the XGBoost (XGB) package in RStudio and available remote sensing data along with ground-truth solar irradiance data. By means of XGB, the new RS algorithm, called LSU model, was structurally built with nine key RS parameters, including photosynthetically available radiation (PAR); instantaneous PAR (iPAR); water leaving reflectance Rrs at wavelengths 412, 443, 469, and 488 nm; angstrom; aerosol optical thickness (aot_869); and latitude that represent major sources and sinks of solar irradiance, as model input variables. Among the nine parameters, the most important four parameters are PAR, iPAR, latitude, and aot_869. It was found that the sea surface SR is highly affected by conditions in both the atmosphere and the seawater. The aot_869 is by far the most important factor describing the effects of the atmospheric absorption and scattering of SR before reaching the sea surface. The most important factors describing the effects of seawater characteristics on solar irradiance are PAR, iPAR, and latitude. Comparisons with existing SR models indicate that LSU model is scientifically sound due to the use of major source and sink factors of SR as model input variables. LSU model is also technically accurate due to its fine resolution (1×1 km) and overall best performance in predicting instantaneous SR. More importantly, LSU model is globally applicable as it can be utilized to obtain global-scale SR data for any day, any time, and anywhere in the world.

Deep learning-based ensemble modeling of Vibrio parahaemolyticus concentration in marine environment

Vibrio parahaemolyticus (V.p) is a marine pathogenic bacterium that poses a high risk to human health and shellfish industry, yet an effective regional-scale nowcasting model for managing the risk remains lacking. This study presents the first regional-scale model for nowcasting the level of V.p in oysters in the marine environment by developing an ensemble modeling approach. The ensemble modeling approach involves the integration of genetic programming (GP) and deep artificial neural networks (DNN)-based modeling. The new approach was demonstrated by developing three GP-DNN ensemble models for predicting the V.p level in North Carolina, New Hampshire, and the combined region. Specifically, GP was employed to establish nonlinear functions between the V.p level and antecedent conditions of environmental variables. The nonlinear GP functions and current conditions of individual environmental variables were then utilized as inputs into a DNN model, forming a GP-DNN ensemble model. Modeling results indicated that the GP-DNN ensemble models were capable of predicting the V.p level with the correlation coefficient of 0.91, 0.90, and 0.80 for North Carolina, New Hampshire, and the combined region, respectively, demonstrating the impact of distinct environmental conditions in the local areas on accuracy of the combined regional-scale model. Sensitivity analysis results showed that sea surface temperature and sea surface salinity are the two most important environmental predictors for the abundance of V.p in oysters, followed by water level, pH, chlorophyll-a, and turbidity. The findings suggested that the GP-DNN ensemble models could be utilized as effective predictive tools for mitigating the V.p risk.

Numerical Modeling of Microbial Fate and Transport in Natural Waters: Review and Implications for Normal and Extreme Storm Events

Degradation of water quality in recreational areas can be a substantial public health concern. Models can help beach managers make contemporaneous decisions to protect public health at recreational areas, via the use of microbial fate and transport simulation. Approaches to modeling microbial fate and transport vary widely in response to local hydrometeorological contexts, but many parameterizations include terms for base mortality, solar inactivation, and sedimentation of microbial contaminants. Models using these parameterizations can predict up to 87% of variation in observed microbial concentrations in nearshore water, with root mean squared errors ranging from 0.41 to 5.37 log10 Colony Forming Units (CFU) 100 mL−1. This indicates that some models predict microbial fate and transport more reliably than others and that there remains room for model improvement across the board. Model refinement will be integral to microbial fate and transport simulation in the face of less readily observable processes affecting water quality in nearshore areas. Management of contamination phenomena such as the release of storm-associated river plumes and the exchange of contaminants between water and sand at the beach can benefit greatly from optimized fate and transport modeling in the absence of directly observable data.

Geostatistical Prediction of Microbial Water Quality Throughout a Stream Network Using Meteorology, Land Cover, and Spatiotemporal Autocorrelation

Predictive modeling is promising as an inexpensive tool to assess water quality. We developed geostatistical predictive models of microbial water quality that empirically modeled spatiotemporal autocorrelation in measured fecal coliform (FC) bacteria concentrations to improve prediction. We compared five geostatistical models featuring different autocorrelation structures, fit to 676 observations from 19 locations in North Carolina's Jordan Lake watershed using meteorological and land cover predictor variables. Though stream distance metrics (with and without flow-weighting) failed to improve prediction over the Euclidean distance metric, incorporating temporal autocorrelation substantially improved prediction over the space-only models. We predicted FC throughout the stream network daily for one year, designating locations "impaired", "unimpaired", or "unassessed" if the probability of exceeding the state standard was ≥90%, ≤10%, or >10% but <90%, respectively. We could assign impairment status to more of the stream network on days any FC were measured, suggesting frequent sample-based monitoring remains necessary, though implementing spatiotemporal predictive models may reduce the number of concurrent sampling locations required to adequately assess water quality. Together, these results suggest that prioritizing sampling at different times and conditions using geographically sparse monitoring networks is adequate to build robust and informative geostatistical models of water quality impairment.