Using SDP to optimize conjunctive use of surface and groundwater in China

A hydro-economic modelling approach to optimize conjunctive use of scarce surface water and groundwater resources under uncertainty is presented. Stochastic dynamic programming (SDP) is used to minimize the basin-wide total costs arising from allocations of surface water, head-dependent groundwater pumping costs, water allocations from the South-North Water Transfer Project and water curtailments of the users.

Each water user group (agriculture, industry, domestic) is characterized by fixed demands and fixed water allocation and water supply curtailment costs. The non-linear one step-ahead sub-problems are solved using a genetic algorithm (GA) that minimizes the sum of the immediate and future costs for given surface water reservoir and groundwater aquifer end storages.

The immediate costs are found by solving a simple linear allocation sub-problem, and the future costs are assessed by cubic interpolation in the total cost matrix from the following time step. The resulting total costs for all stages, reservoir states, and inflow scenarios are used as future costs to drive a forward moving simulation under uncertain water availability.

The use of a GA to solve the sub-problems is computationally more costly than a traditional SDP approach with linearly interpolated future costs, but is still computationally feasible and represents a clean and customizable method. The method has been applied to the Ziya River basin in northern China.

The basin is located on the North China Plain and is subject to severe water scarcity, which includes surface water droughts and groundwater over-pumping. The head-dependent groundwater pumping costs will enable assessment of the long-term effects of increased electricity prices on the groundwater pumping.

The optimization framework is used to assess realistic alternative development scenarios for the basin. In particular the use of electricity pricing policies to reach a sustainable groundwater table is investigated.