Application of Multi-Layer Artificial Neural Networks for Forecasting Groundwater Level (Case study: Yolo County, California)

Groundwater resources are one of the primary sources of water supply. In recent years, the natural balance between fresh, and saline water due to over-exploitation has deteriorated and groundwater levels (GWLs) in parts of the world aquifers have turned negative. Today, mathematical and unique models used to predict and evaluate groundwater levels. In this study, two separate artificial feed-forward neural networks (ANN) employing backpropagation algorithms have been developed using two sets of groundwater level (GWL) data, to simulate groundwater level fluctuations. The recorded daily GWL data from 1992 to 2014, to be fed as training input to the ANN models. The model inputs are the number of months and the number of years (a logarithmic expression), and monthly GWLs are the model s outputs. Two of the selected models were trained with data from 4/1992 to 12/2012, and then data from 1/2013 to 9/2014 were used for the verification process. The model’s mean absolute errors were calculated as 0.51 and 0.66 (ft.), respectively and the prediction rate R for both models was calculated as 0.95. A significant advantage of the current study is its capability to predict the GWL, independent of parameters such as temperature or precipitation rate.