Pre-processing data to predict groundwater levels using the fuzzy standardized evapotranspiration and precipitation index (SEPI)

• Published in: Water resources management 2017-11, Vol.31 (14), p.4433-4448
• Main Authors: Naderianfar, Mohammad, Piri, Jamshid, Kisi, Ozgur
• Format: Article
• Language: English
• Subjects: Atmospheric precipitations; Atmospheric Sciences; Catchment area; Catchment scale; Civil Engineering; Components; Computer simulation; Correlation; Data processing; Earth and Environmental Science; Earth Sciences; Environment; Evapotranspiration; Fluctuations; Fuzzy logic; Gauges; Geotechnical Engineering & Applied Earth Sciences; Groundwater; Groundwater data; Groundwater levels; Hydrogeology; Hydrologic data; Hydrology/Water Resources; Kriging interpolation; Linearity; Mathematical models; Measuring instruments; Neural networks; Precipitation; Principal components analysis; Rain; Rain gauges; Rainfall; Regression analysis; Statistical methods; Time; Water supply
• ISSN: 0920-4741; 1573-1650
• DOI: 10.1007/s11269-017-1757-8

Due to sudden declines in groundwater levels in Neyshabur Plain, one of the most important parts of water supply management programs at the catchment scale is to accurately predict the groundwater level fluctuations. In this paper, the rainfall data from 22 rain gauges and evapotranspiration stations during the period of 1974–2015 were used to find the cumulative effects of rainfall and evapotranspiration on fluctuations in groundwater levels. First, using the Hargreaves-Samani method, the modified evapotranspiration was calculated on the plain. Using the Kriging method, the average amount of precipitation and evapotranspiration of the reference plant was also calculated. Then, employing the fuzzy logic, the fuzzy standardized evapotranspiration and precipitation index (SEPI) was produced. The correlation results between SEPI indicator and fluctuations in groundwater levels showed that the long-term time scales had greater correlations. Thus, the correlations for the time scales of 30, 36, 42, 48, 54 and 60 months were respectively obtained as 0.56, 0.68, 0.71, 0.69, 0.59 and 046. These six parameters were used for principal components analysis (PCA) and the selection criteria (SC) index was used to select the properties affecting every component. The ranking results of testing local linear regression with PCA (LLR-PCA) and dynamic local linear regression with PCA (DLLR-PCA) models, Broyden, Fletcher, Goldfarb, Shanno algorithm with PCA (BFGS-PCA) neural network and Conjugate Gradient-PCA indicated that the DLLR model with three main components had the best performance so that the values of R 2 , RMSE, MBE and MAE were obtained as 0.84, 0.215, 0.028 and 0.162, respectively. The results generally showed that due to severe linearity between SEPI indicator and its time scales, the use of PCA is essential for simulating fluctuations of the groundwater levels.