Application of artificial neural networks to the design of subsurface drainage systems in Libyan agricultural projects

Study region: The study data draws on the drainage design for Hammam agricultural project (HAP) and Eshkeda agricultural project (EAP), located in the south of Libya, north of the Sahara Desert. The results of this study are applicable to other arid areas. Study focus: This study aims to improve the prediction of saturated hydraulic conductivity (Ksat) to enhance the efficacy of drainage system design in data-poor areas. Artificial Neural Networks (ANNs) were developed to estimate Ksat and compared with empirical regression-type Pedotransfer Function (PTF) equations. Subsequently, the ANNs and PTFs estimated Ksat values were used in EnDrain software to design subsurface drainage systems which were evaluated against designs using measured Ksat values. New hydrological insights: Results showed that ANNs more accurately predicted Ksat than PTFs. Drainage design based on PTFs predictions (1) result in a deeper water-level and (2) higher drainage density, increasing costs. Drainage designs based on ANNs predictions gave drain spacing and water table depth equivalent to those predicted using measured data. The results of this study indicate that ANNs can be developed using existing and under-utilised data sets and applied successfully to data-poor areas. As Ksat is time-consuming to measure, basing drainage designs on ANN predictions generated from alternative datasets will reduce the overall cost of drainage designs making them more accessible to farmers, planners, and decision-makers in least developed countries.

Automated summary

The study examines drainage design data from agricultural projects in southern Libya, using Artificial Neural Networks (ANNs) and Pedotransfer Function (PTFs) to predict saturated hydraulic conductivity (Ksat). Results show that ANNs provide more accurate predictions than PTFs, and can be successfully applied in data-poor areas for drainage system design.