Edwin A. Roehl

The use of data-mining techniques for developing effective decision support systems: a case study of simulating the effects of climate change on coastal salinity intrusion

Abstract Natural-resource managers and stakeholders face difficult challenges when managing interactions between natural and societal systems. Potential changes in climate could alter interactions between environmental and societal systems and adversely affect the availability of water resources in many coastal communities. The availability of freshwater in coastal streams can be threatened by saltwater intrusion. Even though the collective interests and computer skills of the community of managers, scientists and other stakeholders are quite varied, there is an overarching need for equal access by all to the scientific knowledge needed to make the best possible decisions. This paper describes a decision support system, PRISM-2, developed to evaluate salinity intrusion due to potential climate change along the South Carolina coast in southeastern USA. The decision support system is disseminated as a spreadsheet application and integrates the output of global circulation models, watershed models and salinity intrusion models with real-time databases for simulation, graphical user interfaces, and streaming displays of results. The results from PRISM-2 showed that a 31-cm and 62-cm increase in sea level reduced the daily availability of freshwater supply to a coastal municipal intake by 4% and 12% of the time, respectively. Future climate change projections by a global circulation model showed a seasonal change in salinity intrusion events from the summer to the fall for the majority of events.

Using "big data" to optimally model hydrology and water quality across expansive regions

This paper describes a new divide and conquer approach that leverages big environmental data, utilizing all available categorical and time-series data without subjectivity, to empirically model hydrologic and water-quality behaviors across expansive regions. The approach decomposes large, intractable problems into smaller ones that are optimally solved; decomposes complex signals into behavioral components that are easier to model with sub-models; and employs a sequence of numerically optimizing algorithms that include time-series clustering, nonlinear, multivariate sensitivity analysis and predictive modeling using multi-layer perceptron artificial neural networks, and classification for selecting the best sub-models to make predictions at new sites. This approach has many advantages over traditional modeling approaches, including being faster and less expensive, more comprehensive in its use of available data, and more accurate in representing a systems physical processes. This paper describes the application of the approach to model groundwater levels in Florida, stream temperatures across Western Oregon and Wisconsin, and water depths in the Florida Everglades.