SPATIAL ESTIMATION OF SOIL MOISTURE AND SALINITY WITH NEURAL KRIGING
The study was carried out with 107 measurements of volumetric soil water content (SWC) and electrical conductivity (EC) for soil profile (0-30 cm) and the estimating accuracy of ordinary kriging (OK) and back-propagation neural network (BPNN) was compared. The results showed that BPNN method predicted a slightly better accurate SWC than that of OK, but differences between both methods were not significant based on the analysis of covariance (ANOVA) test (P >0.05). In addition, BPNN performed much better in EC prediction with higher model efficiency factor (E) and ratio of prediction to deviation (RPD) (E=0.8044 and RPD=3.54) than that of OK (E=0.7793 and RPD=0.39). Moreover, a novel neural kriging (NK) resulting from the integration of neural network (NN) and ordinary kriging (OK) techniques was developed through a geographic information system (GIS) environment for obtaining trend maps of SWC and EC. There was no significance between results of NK and OK through trend maps. Comparing with OK, NK gives better spatial estimations for its great advantage of establishing spatial nonlinear relationships through training directly on the data without building any complicated mathematical models and making assumptions on spatial variations.
KeywordsSoil Moisture Artificial Neural Network Soil Water Content Ordinary Kriging Back Propagation Neural Network
- Ferreyra, R.A., Apezteguia, H.P., Sereno, R, Jones, J.W.. Reduction of soil water spatial sampling density using scaled semivariograms and simulated annealing. Geodenna, 2002, 110: 265–289.Google Scholar
- Kitamura, Y., Yano, T., Honna T., Yamamoto S. and Inosako, K.. Causes of farmland salinization and remedial measures in the Aral Sea basin—Research on water management to prevent secondary salinization in rice-based cropping system in arid land. Agricultural Water Management, 2006, 85: 1–14.CrossRefGoogle Scholar