Generating Artificial WIPs Using Machine Learning Methods to Explore Management Responses to Land Use Scenarios

Keywords: Chesapeake Bay Watershed, Watershed Implementation Plans, Best Management Practices, Land Use and Land Cover, Machine Learning Model
Abstract Type: Paper Abstract

Abstract

Agricultural activities, urban and suburban land use, stormwater runoff, and wastewater treatment plants contribute significant amounts of nitrogen, phosphorus, and sediment to the Chesapeake Bay This leads to algae bloom growth that creates low oxygen zones that harm aquatic life. To address nitrogen, phosphorus, and sediment pollution in the Chesapeake Bay, jurisdictions across the Chesapeake Bay Watershed periodically develop watershed implementation plans (WIPs), which contain suites of best management practices (BMPs). These plans are necessarily responsive to land use land cover (LULC) and nutrient load sources and provide a blueprint for how practices will be implemented across the watershed. To understand how these plans might respond to future land use scenarios and, therefore, changes in landscape characteristics, we developed a multi-scale, machine-learning-based model that generates artificial WIPs. The model integrates local geographical context and past state-level decisions to generate representative plans under new scenarios, providing insights into the potential futures of the Bay and Watershed. These results aid decision-makers in understanding how plans and priorities might change under future policies, transportation scenarios, and growth projections. Accordingly, this research offers valuable insights into the spatial patterns of management practices that collectively protect the health of the Chesapeake Bay ecosystem.