Optimized groundwater drawdown in a subsiding urban mining area

This study establishes the first real-world application of evolution strategies for solving a groundwater management problem. In an urban coal mining area in the Emscher and Rhine Basin of Northwestern Germany the groundwater table rises relative to subsiding ground and threatens local infrastructure and basements of buildings. The active extraction system, which consists of one highly productive horizontal and twelve vertical wells that pump more than 500 m 3/h, is revised by combining groundwater model and algorithmic optimization procedure. By capitalizing on the robustness and self-adaptivity of evolution strategies, both fixed and moving well formulations are solved. It is shown that well layout can be improved by automatic optimization even though it has been previously soundly configured by experts. The total pumping effort can be noticeably reduced while complying with the drawdown targets given at 24 different locations in the study area. Savings increase if new well positions are considered. For example, one additional well yields a 9% reduction of the total extraction rate. We also investigate the relevance of the spatially variable drawdown targets and demonstrate how those targets that mainly control the optimized well layouts can be identified by varying the penalty function. It is revealed that there is huge potential for additionally reducing the extraction rate if one or more of these individual targets could be resigned, for example as a result of technical construction or land use changes. A reduction of more than 25% has been estimated for giving up the most notable constraining target. This way, by testing the significance of given constraints, algorithmic optimization may guide the re-formulation of the original optimization problem in order to conceive new groundwater management scenarios that ultimately lead to an increased efficiency of the well field. This procedure is similar to a chance-constraint approach, efficient with CMA-ES, but can be adopted in any other combined hydrological simulation–optimization problem.