Integrating Categorical and Real-Time Monitoring Data to Optimally Model Water Quality for Waterbasins and Beyond

The application of water-quality models for resource management and control can be time consuming and expensive. Riverine models, which are based upon physical equations, are applied one site at a time and are limited in their ability to accurately reproduce observed hydrologic and water-quality behaviors. Complex land-use models, which typically use a large parameter set due to the large numbers of physical equations, can be applied across geographically expansive systems, but often under-perform at replicating the diversity of behaviors displayed in historical data. Both types of models are limited in their ability to incorporate all of the available data types that represent the circumstances and forces that characterize a particular system, including time series that exhibit hydrologic, water-quality, and meteorological dynamics from many locations; and categorical (static) variables that describe location-to-location differences such as spatial coordinates, basin characteristics, and neighboring land uses. Reducing large numbers of variables to a select subset to accommodate modeling tool limitations usually leads to model uncertainty, subjective decision-making, poor data and model integration, and delayed formulation and implementation of management decisions and plans.This paper describes a new approach that utilizes all available categorical and time-series data, without subjectivity, to empirically model hydrologic and waterquality behaviors across expansive regions. The approach employs a sequence of optimization algorithms that include: 1) signal processing to separate time-series behaviors into components that are ascribable to different forcing functions, 2) time-series clustering to aggregate monitored locations according to their behaviors, 3) non-linear, multivariate sensitivity analysis using multi-layer perceptron artificial neural networks (ANNs) to determine the relative importance of categorical variables at predicting site-to-site behavioral variability, and, 4) predictive modeling also using ANNs. The approach has many advantages as compared to traditional modeling approaches by being faster (and less expensive), more comprehensive in its use of available data, and more accurate in representing a system's physical processes. This approach has been used successfully to model stream temperatures across western Oregon and Wisconsin, and the water levels in the Florida Everglades.