Enabling improved water and environmental management in an irrigated river basin using multi-agent optimization of reservoir operations

Abstract

In Colorado's Lower Arkansas River Basin (LARB), inefficient irrigation and canal seepage contribute to salinization and waterlogging of irrigated lands and to stream-aquifer pollution. The geographic information system (GIS)-based river basin management model River GeoDSS is applied to further explore best management practices (BMPs) earlier determined to remedy these agro-environmental impacts. Unfortunately, BMP benefits are offset by altered irrigation return flows which change historical downstream river flows, threatening compliance with water rights and the Arkansas River Compact. Compensation is possible through optimal sizing and operation of a dedicated reservoir storage account. Multi-agent optimization combines a metaheuristic mutation linear particle swarm optimization (MLPSO) with a fuzzy rule-based system to produce generalized operational policies along with optimal storage sizing, to enable BMPs while satisfying legal constraints. A storage account making up less than 5% of available reservoir capacity can be operated with rules that enable implementation of even the most aggressive BMPs.