MODELING WATER-RETENTION CURVES OF SANELY SOILS USING NEURAL NETWORKS

Citation
Mg. Schaap et W. Bouten, MODELING WATER-RETENTION CURVES OF SANELY SOILS USING NEURAL NETWORKS, Water resources research, 32(10), 1996, pp. 3033-3040
Citations number
34
Categorie Soggetti
Limnology,"Environmental Sciences","Water Resources
Journal title
ISSN journal
00431397
Volume
32
Issue
10
Year of publication
1996
Pages
3033 - 3040
Database
ISI
SICI code
0043-1397(1996)32:10<3033:MWCOSS>2.0.ZU;2-F
Abstract
We used neural networks (NNs) to model the drying water retention curv e (WRC) of 204 sandy soil samples from particle-size distribution (PSD ), soil organic matter content (SOM), and bulk density (ED). Neural ne tworks can relate multiple model input data to multiple model output d ata without the need of an a priori model concept. In this way a high performance black-box model is created, which is very useful in a data exploration effort to assess the maximum obtainable prediction accura cy. We used a series of NN models with an increasing parametrization o f input and output variables to get a better interpretability of model results. In the first two models we used the nine PSD fractions, ED, and SOM as input, while we predicted the nine points of the water rete ntion curve. These NNs had 12 input and 9 output variables, predicting WRCs with an average root-mean-square residual (RMSR) water content o f 0.020 cm(3) cm(-3). After a few intermediary models with increasing parametrization of PSD and WRC using (adapted) van Genuchten [1980] eq uations we arrived at a final NN model that used six input variables t o predict three van Genuchten [1980] parameters resulting in a RMSR of 0.024 cm(3) cm(-3). We found saturated and residual water contents to be unrelated to the PSD, ED, or SOM, therefore the saturated water co ntent was considered to be an independent input variable, while the re sidual water content was set to zero. Sensitivity analyses showed that the PSD had a major influence on the shape of the WRC, while ED and S OM were less important. On the basis of these sensitivity analyses we established more explicit equations that demonstrated similarity relat ions between PSD and WRC and incorporated effects of SOM and ED in an empirical way. Despite the fact that we considered a large number of l inear and nonlinear variants these equations had a weaker performance (RMSR: 0.029 cm(3) cm(-3)) than the NN models, Proving the modeling po wer of that technique.