Conjunctive Management of Surface Water and Groundwater for Agricultural Use using ANN and Metaheuristics
Abstract
In the study area (the Osan watershed, Central South Korea), the water use for agriculture is unbalanced: surface water extractions are nearly six times higher than groundwater withdrawals. To satisfy future agricultural water demands, the conjunctive use of surface and groundwater is indispensable. Research has been carried out on using the simulation-optimization (S-O) approach to plan and manage conjunctive water use (Shi et al., 2012; Ashu and Lee, 2021). In this study, the trained artificial neural network (ANN) was employed as a simulation model, and a metaheuristic optimization scheme (Jaya Algorithm (JA) or genetic algorithm (GA)) was used to solve a water allocation problem.
As for an ANN model for simulation, the Levenberg-Marquardt training technique and a single-layer architecture with three hidden layers were used for the study using MATLAB neural network fitting tools.The objective of the S-O model is to minimize the agricultural water shortage unless the water demand exceeds the total available water. The ANN model outputs are optimized to attain the best conjunctive management practice subject to various constraints. The objective function is represented as Minimize W = ∑(Demand - Supply)2 + θ subject to SWE ≤ SWEmax GWE ≤ GWEmax ∑Δh ≤ Δhmax Where θ is the acceptable threshold violation, 𝑆𝑊𝐸 is the surface water volume supplied (MCM), 𝐺𝑊𝐸 is the extracted groundwater volume, and Δℎ is the volume of groundwater level change. The best-fit neural network structures were determined for each subarea based on the R2, RMSE, MSE, and MAPE values. The groundwater level was successfully predicted using the ANN model and could be used to facilitate water management decision-making. The S-O model results show that the total water supply can reach nearly 80% of the water demand with JA and 70% with GA. The simulation-optimization model used in the study provides the best possible solution to the water shortage minimization problem with sustainability issues. Summarily, our study shows that groundwater levels can be successfully predicted using ANN. The S-O model determined the best conjunctive practice in order to reduce the gap between water supply and demand with groundwater level constraints. The S-O approach might reduce water shortages in the future and increase water efficiency to nearly 80% for JA and 70% for GA.- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2022
- Bibcode:
- 2022AGUFM.H42L1436A