Evaluation of the classical statistical, deterministic and geostatistical interpolation methods for estimating the groundwater level

Abstract

Proper management of groundwater resources depends on identifying, modeling and predicting the level of groundwater in the plains for long-term planning and optimal use of the potential of water. To achieve continuous and integrated maps and predict unknown values, interpolation methods can be used. In this research, the classic statistical interpolation methods (including nearest neighbor, natural neighbor, moving average, and triangulation with linear interpolation), deterministic interpolation methods (including inverse distance to a power, radial basis functions and local polynomial) and geostatistical interpolation method (kriging) were used to estimate the groundwater level of Naqadeh plain in the northwest of Iran in the period of 2001–2016. The cross-validation technique was applied to evaluate the accuracy of the various methods, and three indices—the determination coefficient (R2), the root mean squared error and mean absolute error—were used to compare the interpolation methods. The results of geostatistical analysis indicated that the groundwater level is regionalized variable and there is a high spatial structure ratio between groundwater level data. The best-experimented variogram is a Gaussian model with a correlation coefficient of 0.981. According to the results of cross-validation method, the geostatistical interpolation method with the highest accuracy and the minimum error and the classic statistical interpolation methods with the least accuracy and maximum error were introduced as the optimal and inappropriate methods, respectively.