Efficiency assessment of some machine learning algorithms for predicting the Qanat's discharge

Qanat is an ancient man-made underground construction designed to access groundwater and supply water in arid and semi-arid regions. The ability to predict Qanat's discharge is indispensable for achieving sustainable water resource management goals. But predicting the water flow of Qanat is a complex task, given the many influencing factors at play. The main goal of this study is to evaluate support vector machines, random forests, and decision trees as popular ML algorithms for modeling Qanat's discharge. To carry out the present study, data and information of 76 Qanats were collected in the Mashhad-Chenaran plain, Iran. Then, effective factor of Qanat's discharge include type of Qanat, the depth of the mother well, the length of the Gallery, the groundwater table, the gradient of groundwater table, hydraulic conductivity, specific yield of aquifer, and annual precipitation in the area were collected for each Qanat. The results of modeling showed that in the training and test steps, the RF model exhibited the highest performance (training: R2 = 0.92 and NS = 0.88 testing: NS = 0.71, R2 = 0.82). Finally, it can be stated that crucial factors controlling the water flow of Qanats in the study area include the Qanat's geometry and the specific yield of the aquifer.