Prediction of groundwater drawdown using artificial neural networks

Area of Science:

• Hydrogeology
• Artificial Intelligence
• Environmental Science

Background:

• Traditional groundwater drawdown assessment methods (pumping tests, field experiments) are time-consuming and costly for large areas.
• Accurate groundwater drawdown prediction is crucial for sustainable water resource management.

Purpose of the Study:

• To develop and validate an artificial neural network (ANN) based methodology for predicting groundwater drawdown in an alluvial aquifer.
• To identify key factors influencing groundwater drawdown and integrate them into the ANN model.

Main Methods:

• Utilized field measurements from 250 piezometric wells in northern Iran.
• Employed artificial neural network (ANN) models, including modular neural networks (MNN), with data divided into training, cross-validation, and test sets.
• Input factors included groundwater depth, precipitation, evaporation, transmissivity, elevation, distance to sea/recharge, population density, and extraction within a 1000m radius.

Main Results:

• The modular neural network (MNN) demonstrated superior performance, achieving high R-squared values for training (0.96) and testing (0.81).
• The optimized ANN model accurately mapped annual groundwater drawdown across the entire aquifer, with an accuracy assessment yielding an R-squared of 0.8.
• The methodology proved effective in modeling and predicting groundwater drawdown.

Conclusions:

• The developed ANN methodology offers a time-efficient and cost-effective alternative to traditional methods for assessing groundwater drawdown over extensive regions.
• This approach is applicable for groundwater drawdown prediction in the studied alluvial aquifer and can be adapted to similar hydrogeological settings globally.