Model-based optimization of agricultural drainage design: comparing orthogonal design, range analysis, and surrogate-assisted evolutionary optimization

Artificial drainage is essential for agricultural productivity, but current design methods struggle to incorporate changing climate conditions and evolving policies and economic constraints. We propose a model-based optimization framework employing orthogonal design, range analysis, numerical simulations, and economic cost assessment. To benchmark the simplified approach, a Gaussian Process surrogate model (trained on the 16 orthogonal treatments) coupled with NSGA-II multi-objective evolutionary optimization was implemented to search for Pareto-optimal designs in continuous space. The framework was evaluated through a case study in Eastern China. The orthogonal design reduced the required simulations by 75% (64 to 16). The identified optimum (12 m spacing, 140 cm depth, 3 cm surface roughness) was corroborated through a continuous-space NSGA-II search, where the Pareto front knee point achieved a model-validated water table depth of 86.82 cm vs. 87.93 cm for the orthogonal optimum. The GP surrogate predictions agreed well with the numerical model (mean absolute error 4.5 cm, 4.4%). Global sensitivity analysis identified the drainage efficiency weight (67%) as the most influential factor. Systematic optimization methods, from orthogonal designs to surrogate-assisted evolutionary algorithms, can effectively identify optimal drainage designs while substantially reducing simulation effort, bridging the gap between optimization theory and practical drainage engineering.