Spatial modeling of groundwater depth fluctuations using co-active neuro-fuzzy inference system (CANFIS) and geographic information system (GIS)

Groundwater drawdown and reduction of groundwater resources are serious problems in the water supply. Unfortunately, sufficient and accurate data on the groundwater depth fluctuations values are not available in all plains of the world. In the present study, a co-active neuro-fuzzy inference system (CANFIS) network and geographic information system (GIS) were used to simulate the annual groundwater drawdown values on the southern coasts of the Caspian Sea (Iran). For this purpose, the annual drawdown values in 250 piezometric wells as the output variable were evaluated. Further, the effective factors in groundwater depth fluctuations were evaluated as inputs of the CANFIS network, including water table depth, annual precipitation, annual evaporation, the transmissivity of aquifer formations, topography, distance from the sea, distance from water sources, population density, and withdrawal values in the influence radius of the study wells. The modeling process was performed in three stages: training, cross-validation, and test. According to the results, the CANFIS network has a high performance in modeling groundwater drawdown ​​(training R-sqr = 0.97; test R-sqr = 0.82). Then, the inputs of the tested network were prepared in the GIS in raster format for the study plain. The tested network was used to simulate the annual fluctuations of groundwater depth ​​in the plain and the simulation results were presented in the GIS as an annual groundwater drawdown map. Overlap and comparison of the observed values ​​of groundwater drawdown and the simulated values ​​in the map indicate the high performance of the CANFIS network (R-sqr = 0.81). Therefore, the used methodology can be applied to simulate the annual drawdown of groundwater depth in plains without groundwater depth data.