Soil Property Mapping Using Fuzzy Membership Derived by Fuzzy c- Means (fcm) Clustering

This paper explores the use of fuzzy membership values generated by fuzzy c-means clustering (FCM) method to predict soil properties over space. A weighted average model was used on fuzzy membership to get soil properties. To validate the efficiency of this model, it was then compared with a multiple linear regression model between the soil property and terrain attributes. Four indices were calculated to evaluate the performance of these two models: correlation coefficient between predicted and observed values, mean absolute error (MAE), root mean square (RMSE) and agreement coefficient (AC). The research was tested in a watershed located in Heilongjiang province China. Two soil properties were chosen: A-horizon organic matter and soil depth. The results showed that the fuzzy membership weighted model produced reasonably better performance than the regression model by using the same modeling points, while linear regression models were limited in the study area. Although the R of regression functions were very high, the functions constructing by modeling points may not suit for other points of the area. Therefore, we can conclude that weighted average model using fuzzy membership was an effective way to predict soil properties, and it is more extrapolatable than the regression approach.