نموذج شبكة عصبية اصطناعية لتقدير المحتوى المائي عند السعة الحقلية و عند نقطة الذبول الدائم في الترب الصحراوية

Abstract EN

An artificial neural network (ANN) pedotransfer function (PTF) for estimating volumetric soil water contents at field capacity (FC) and permanent welting point (PWP) of arid soils from easily obtained bulk density and particle-size distribution data was developed and its validity was tested using data from arid Saudi Arabia.

The PTF was calibrated using 145 samples and its validity was tested using an independent dataset consisting of 45 samples.

The sampling scheme was designed to cover soils at the great group level from the main physiographic regions of Saudi Arabia.

The ANN technique was combined with the bootstrab technique to generate uncertainty estimates.

Experimental performance of the PTF on the training data (accuracy) and the independent data (reliability) was evaluated quantitatively using mean error (ME, cmJ cm'J), root mean squared error (RMSE, cmJ cm*3) and coefficient of determination (R2).Values of ME, RMSE and R2 for the training dataset are 0.0004, 0.01 and 0.88, respectively, for FC and 0.0001, 0.0057 and 0.90, respectively, for PWP.

For the validation dataset, the values of ME, RMSE and R2 are 0.0002, 0.009 and 0.89, respectively, for FC and 0.014, 0.007 and 0.90, respectively, for PWP.

Such experimental performance indicates that the PTF is both accurate and reliable when applied to arid soils.