Application of genetic algorithm technique to inverse modeling of tide–aquifer interaction

Abstract

Modeling of tide–aquifer interaction plays a vital role in the management of coastal aquifer systems. A novel and robust methodology is presented in this paper for estimating aquifer parameters of coastal aquifers from tide–aquifer interaction data using tide–aquifer interaction model and genetic algorithm (GA). Two stand-alone computer programs were developed to optimize hydraulic diffusivities of unconfined and confined coastal aquifers at multiple sites using GA technique and tide–aquifer interaction model and considering two approaches (‘lumped tidal component approach’ and ‘multi-tidal component approach’). Five sets of real-world tide–aquifer interaction data at two sites of an unconfined aquifer and one set of tide–aquifer interaction data at three sites of a confined aquifer were used to demonstrate the efficacy of the methodology. The analysis of the GA-based inverse modeling results indicated that the ‘multi-tidal component approach’ yields more accurate and reliable hydraulic diffusivities for the unconfined aquifer (RMSE = 0.0129–0.0521 m, NSE = 0.70–0.97, and d1 = 0.91–0.99) as well as for the confined aquifer (RMSE = 0.0204–0.0545 m, NSE = 0.95–0.97, and d1 = 0.99) compared with the ‘lumped tidal component approach’. A comparative evaluation of data-size revealed that the short-duration datasets of the unconfined aquifer provide more reliable estimate of hydraulic diffusivity than the long-duration datasets. Further, it was found that the spring and neap tidal data yield unreasonable values of hydraulic diffusivity with considerably high values of RMSE and very low values of r 2, NSE, and d1, thereby suggesting that spring and neap tidal data are not suitable for aquifer parameter estimation. Overall, it is concluded that the GA-based tide–aquifer interaction model following ‘multi-tidal component approach’ is the most efficient tool for estimating aquifer parameters of unconfined and confined aquifers from tide–aquifer interaction data. The developed methodology is also applicable to other coastal basins of the world irrespective of hydrogeological settings.