CONJUNCTIVE USE OF SURFACE AND GROUNDWATER IN COASTAL DELTAS USING SIMULATED ANNEALING AND NEURAL NETWORKS

The hydrology of Indian deltas is largely influenced by vagaries of monsoon rainfall that affects groundwater recharge during monsoon season and surface water availability during both monsoon and non-monsoon seasons. The irrigation here is well developed from diversion structures. However, often the demand is not met adequately during the non-monsoon season through surface water sources only, especially in the lower reaches of deltaic plains. Excessive groundwater extraction in these regions to meet the demand may lead to significant SWI. These twin issues of water availability and SWI in coastal and deltaic regions may be addressed in several approaches. Conjunctive utilization of surface and groundwater sources is one of the approaches. The demand at any point in space during a given time period may be met either from the surface source or groundwater source or from both. Similarly, if enough surface water is available during a particular time period, it can be used to recharge the groundwater reservoir at several demand centers, for use at a later time. This increases the groundwater levels so that pumpage at a later stage does not induce too much of SWI. A conceptual conjunctive use model is presented for a near-real deltaic aquifer system, irrigated from a diversion system, with some reference to hydro-geo- climatic conditions prevalent in the east coastal deltas of India. Water resources are sufficiently available in these regions under average monsoon rainfall conditions. However, their distribution in space and time has been ever challenging to water managers. The surface-water availability shows temporal fluctuations in terms of floods and droughts. The groundwater availability shows mainly spatial variability in terms of quality and quantity due to the hydro geologic setting, boundary conditions, and aquifer properties. The combined simulation-optimization model proposed in this study is solved as a nonlinear, non-convex combinatorial problem using a simulated annealing algorithm and an existing sharp interface model. The computational burden is managed within practical time frames by replacing the flow simulator with artificial neural networks (ANN) and using efficient algorithmic guidance.