Assessing the impact of the temporal resolution of performance indicators on optimal decisions of a water resources system

Large scale water infrastructure serves multiple purposes including provisioning of freshwater, protection from floods, navigation, hydro-electricity, etc. Decision analysis for reservoir systems relies heavily on optimization techniques that identify optimal operational strategies using a dynamic systems model. Optimization requires the analyst to define performance indicators, or, objective functions, that aggregate performance across multiple time periods in a planning horizon. A question thus arises: does the temporal resolution at which objective functions are aggregated have a substantial impact on resultant optimal operational strategy? Here, we investigate this issue using three formulations of a multi-objective optimization problem for a proposed inter-basin water transfer in Southern India. The Borg multi-objective evolutionary algorithm (MOEA) is used to optimize five objectives: the reliability, resilience and vulnerability of demand satisfaction, reliability of maintaining minimum environmental flows, and reliability of preventing high flow exceedances downstream. We alter the temporal resolution at which the reliability of demand satisfaction is calculated by aggregating water deficits at fortnightly, seasonal and annual scale, resulting in three formulations of the decision problem. Pareto approximate strategies resulting from the fortnightly, seasonal, and annual reliability formulations result in annual demand deficits ranging from 82 to 172 Mm3, 79–152 Mm3, and 61–96 Mm3, respectively. This improvement in performance for annual reliability-based strategies exploits greater flexibility within the systems model to store water temporarily in secondary storage structures like lakes, ponds, check dams, in the command area serviced by the project. We also find that the choice of temporal resolution of objective function has a substantial influence on the combination operators preferred during the search phase of the Borg MOEA. Our results highlight the value of exploring objective functions at different temporal resolutions for reservoir optimization.