Prediction of Evaporative Flux from Shallow Watertables Using Regression and Artificial Neural Techniques

Knowledge of the relationships between the watertable depth and evaporation rate from the bare soil is of immense importance in arid and semi arid area. The rise in watertable, land inundation, and soil salinization are inevitable aftermath of over-irrigation in suchenvironments. Evaporative flux was measured in 200 mm inside diameter PVC columns filled with the sandy loam, loam and clay loam soils, maintaining the watertable in them at a depth of 40, 60 and 80 cm. Evaporation from both the bare soil and exposed water surface, soil water content, and the maximum and minimum air temperatures were measured daily for 74 consecutive days. The TDR technique was used for the soil water content determination.Several nonlinear models were developed using benefiting from the gamma test (theWinGamma program), including the local linear regression, the 2-layer back propagation,conjugate gradient descent, and the BFGS neural network. Validity of these models wasascertained using the root mean square error, the mean absolute error, and the largercoefficient of determination. The models satisfactorily predicted the measured evaporativeflux.