Handling Large Decision Variables in Multi-Objective Groundwater Optimization Problems: Aquifer Parameter-Based Clustering Approach

Number of decision variables (DVs) significantly impacts the convergence of multi-objective groundwater simulation-optimization problems (MO-GSOPs). Previous studies of reducing DV by decomposition methods based on the proximity between the pumping wells have yet to assess its implication on the Pareto fronts. This study introduces a novel approach to clustering known as aquifer parameter-based clustering. This work aims to decrease the number of wells involved in MO-GSOPs via clustering based on essential aquifer properties that govern groundwater flow, including initial head, recharge, top elevation of the aquifer layer, and hydraulic conductivity. The simulation-optimization model solves the objectives of maximizing pumping discharge and river-aquifer (R-A) exchanges. The resulting Pareto fronts are compared in terms of convergence and diversity. The analysis reveals that initial head-based clustering exhibits superior performance, leading to a significant increase in hypervolume (46%) and a decrease in the inverted generational distance (22%) compared to distance-based DV clustering. Comparison between results shows that aquifer parameter-based clustering has superior optimal results overall than traditional clustering based upon Euclidian distance. Furthermore, the discharge variation resulting from the parameter-based clustering is examined at the commune level. Notably, Chazey Sur Ain, located near the river, experiences a substantial increase in discharge (12659.13 m3/d), while communes situated near the study area’s boundary, namely Douvres and Jujurieux, observe marginal discharge increases (500 m3/d and 130.33 m3/d, respectively).