Symposium no. 04 Paper no. 41 Presentation: poster 41-1

The direct measurement of the water retention in soil at a defined matric potential is a tedious and hard action. To estimate it indirectly from other soil parameters, which are easily measured, statistical models have been developed, and nowadays they are also used artificial neural networks (ANN). The aim of this study has been to see if the ANN application improves the precision in water retention estimation in relation with some of most common statistical models. It has been used to measure data of water retention at -- 33 kPa and --1500 kPa. These data correspond to 309 soil samples of Castilla-La Mancha (Spain). Basic parameters of these samples were also known. For matric potentials at --33 kPa, the best results were obtained with a network with two hidden layers of two neurons each one with a mean error of 0.70%. For matric potentials at -- 1500 kPa, the best results were obtained with simple networks of one hidden layer with maximum 4 neurons, with a mean error of 0.30 %. In both cases the precision was slightly better than the one obtained with statistical methods