Optimization of Piezometers Number for Groundwater Level Prediction Using PCA and Geostatistical Methods

Network Optimization of existing Piezometers number for estimating groundwater levels, since it reduces maintenance cost and saves the charge of data acquisition, is an important step. The main goal of this research is to optimize the number of piezometers to estimate groundwater levels in the unconfined aquifer of Tabriz plain. That was accomplished at three main stages. At the beginning, ground water levels in 75 piezometers of the unconfined aquifer of Tabriz plain were obtained and the performances of five different interpolation methods were evaluated. At this stage, results of the Universal Kriging method with Root Mean Square Error (RMSE) of 1.12 m was selected as an appropriate surface. At the second stage, two methods; including Cross-Validation Error and Principal Component Analysis (PCA) were applied to remove some points. Finally, with respect to RMSE of the created surfaces, the two applied methods were compared. The results showed that by assuming reasonable error threshold (11%) in PCA method, 30 points were removed and the RMSE of the estimated surface increased by14 percent. In the cross validation method by removing 30 other points, the RMSE of the created surface increased by 16.5 percent. Therefore, RMSE changes using PCA compared to cross-validation error method were less, and therefore PCA result was more valid. The results indicated that, by eliminating 30 piezometers from the groundwater level monitoring network of the unconfined aquifer of Tabriz plain, the suitable accuracy for groundwater level estimating could be achieved by increasing the accuracy of water level measurement in the rest of the piezometers that would cause the saving of time and cost.