Artificial Neural Networks for DRAINMOD-Based Drainage Design in Data-Poor Arid Agricultural Regions

Introduction: Agricultural drainage is essential for managing waterlogging and salinity in irrigated arid regions, but drainage design is often limited by the lack of reliable soil and weather datasets. This study evaluated whether artificial neural networks (ANNs) can support DRAINMOD-based drainage design in data-poor agricultural areas by estimating key model inputs that are expensive or difficult to measure directly.

Methods: The study was conducted using archived soil, weather, crop, and drainage data from two irrigated agricultural projects in southern Libya: Eshkeda Agricultural Project and Hammam Agricultural Project. ANN-predicted saturated hydraulic conductivity (Ksat) and reference evapotranspiration (ET0) were integrated into DRAINMOD and compared with simulations based on measured input data. Several input-data scenarios were evaluated to assess their effects on simulated drain spacing, relative crop yield, irrigation depth, drainage discharge, and economic return over a 30-year simulation period.

Results: The results showed that ANN-derived inputs can provide a practical alternative when measured Ksat and complete meteorological datasets are unavailable. Simulations using ANN-predicted Ksat and ET0 produced drainage design outputs that were generally comparable to simulations based on measured data. In contrast, less accurate input estimates increased uncertainty in drain spacing and relative yield predictions, which could lead to either over-design and unnecessary installation costs or under-design and increased risk of waterlogging and salinity stress.

Conclusions: Integrating ANNs with DRAINMOD offers a promising Hydro-AI approach for improving drainage design in data-poor arid regions. This approach can help reduce data-collection barriers and support more cost-effective, evidence-based drainage planning.