Application of a Genetic Algorithm for Estimating Recharge Potential of the Choushui Rover Alluvial Fan

As groundwater resources are vital to the regional water supply, protection of groundwater resources have become important issues. A systematic approach to locating high recharge areas is the first step in protection of groundwater. In a previous study proposed by the Central Geological Survey (CGS), the CGS used a factor-based approach to determine the spatial distribution of recharge potential (RP). Seven factors to define RP were selected, including “land usage”, “type of surface soil”, “drainage density”, “average annual rainfall”, “correlation between rainfall and variation of groundwater level”, “variation of unit aquifer storage”, and “hydraulic conductivity”. Features of each factor were transformed into factor scores, which ranged from 0 to 100. A high score implied high recharge capacity. The features of five factors (“drainage density”, “average annual rainfall”, “correlation between rainfall and variation of groundwater level”, “variation of unit aquifer storage”, and “hydraulic conductivity”) were continuous variables and could be linearly normalized as factor scores. The features of the remaining factors were discrete variables. In the abovementioned study, factor scores were subjectively assigned based on the experience, opinion, or judgment of the researchers. The factor scores of the seven factors were weighted and summed as the value of RP. The weighted values of various factors were based on a schematic sketch that showed the interactive influence of factors in regards to properties of recharge. The sketch, proposed by Shaban et. al. (2006), showed major and minor interactive influences. Shaban et. al. set the ratio of minor influence to major influence to 0.5. To increase objectivity in the calculation of RP, this study used a genetic algorithm (GA) to determine optimal scores for the two discrete factors. Because the spatial distribution of nitrate concentration (NC) was positively correlated with the spatial distribution of RP, the correlation coefficient of NC and RP was the validation target. The objective function of GA was maximization of the correlation coefficient. The decision variables were the factor scores of each feature. In this study, the ratio of minor influence to major influence varied in analysis of sensitivity. The ratios were set to 0 and 0.5 to compare the effects of different ratios. Results showed that the distribution of RP decreased from the proximal fan to the distal fan, which matched the opinion of experts. After comparing the original result with the optimal result, the correlation coefficient of RP and NC was found to be capable of increasing from 0.618 to 0.667 and from 0.621 to 0.688 respectively when the ratios were set to 0.5 and 0. This study proposes a systematic approach to increasing objectivity in calculation of RP. Although the correlation coefficient of NC and “drainage density” was very low, this was due to NC being caused by interaction between the recharged water and the fertilized land. Although quantity of recharge in areas with high drainage density was greater, recharged water from the river channel reduced NC in local areas. Therefore, the results indicated that NC might not be the best choice for validation of RP.