The purpose of this study is improving the ability of Continuous PTFs used to predict the parameters of van Genuchten model. This study focused on medium-texture soils. For solving the fuzzy area on a triangular texture figure, K-Means cluster analysis was used to classify samples according to particle size distribution rather than texture. Multiple leaner regression was used to develop models which included all samples and classified samples. The result showed that the classified models could improve the prediction in parameter α, however, they could do so in parameter n. Observation of each model revealed that the parameter n was affected by clay content. This study further compared continuous PTFs which were developed according to region, and the result showed that the models which were developed based on particle size distribution had better prediction.We also proved that classifying samples is necessary before developing a model.
Key Words : Pedotransfer functions, van Genuchten model, K-Means cluster analysis, Multiple leaner regression.