A Mixed Application of Geographically Weighted Regression and Unsupervised Classification for Analyzing Latex Yield Variability in Yunnan, China

This paper introduces a mixed method approach for analyzing the determinants of natural latex yields and the associated spatial variations and identifying the most suitable regions for producing latex. Geographically Weighted Regressions (GWR) and Iterative Self-Organizing Data Analysis Technique (ISODATA) are jointly applied to the georeferenced data points collected from the rubber plantations in Xishuangbanna (in Yunnan province, south China) and other remotely-sensed spatial data. According to the GWR models, Age of rubber tree, Percent of clay in soil, Elevation, Solar radiation, Population, Distance from road, Distance from stream, Precipitation, and Mean temperature turn out statistically significant, indicating that these are the major determinants shaping latex yields at the prefecture level. However, the signs and magnitudes of the parameter estimates at the aggregate level are different from those at the lower spatial level, and the differences are due to diverse reasons. The ISODATA classifies the landscape into three categories: high, medium, and low potential yields. The map reveals that Mengla County has the majority of land with high potential yield, while Jinghong City and Menghai County show lower potential yield. In short, the mixed method can offer a means of providing greater insights in the prediction of agricultural production.