Temperature-Based Feed-Forward Backpropagation Artificial Neural Network For Estimating Reference Crop Evapotranspiration In The Upper West Region

Abstract The potential of modeling the FAO Penman-Monteith FAO-56 PM method for computing reference crop evapotranspiration ETo using feed-forward backpropagation artificial neural networks FFBANN with minimal measured climate data such as with the air temperature maximum and minimum was investigated using local climatic data from the Wa Meteorological weather station. Three FFBANN models were developed and trained with the Levenberg-Marquardt algorithm and the early stopping approach. These three FFBANN models are temperature-based models and have the same input variable as the established temperature-based empirical methods the Hargreaves Blaney-Criddle and the Thornthwaite methods. A comparative study was carried to see how these FFBANN models performed relative to the other three established temperature-based empirical methods using the FAO-56 PM method as the benchmark. In general the FFBANN models outperformed these established methods in estimating the ETo and should be preferred where only measured air temperature maximum and minimum is the variable available for estimating the reference crop evapotranspiration.