SINGLE BOOTSTRAP APPROACH WITH GEOGRAPHICALLY WEIGHT REGRESSION MODELING USING PARTICLE-SIZE FRACTION

The challenge of the high need for soil spatial data information has led to the rapid development of spatial modeling for soil attributes in the last few decades. Soil texture is an essential attribute that determines the direction of soil management and must be modeled accurately. However, on the other hand, soil texture is a soil attribute that is relatively difficult to model because it is a compositional data set. The difficulty that arises from this compositional data set is the limitation of constant quantities; namely, the sum of the fractions of sand, silt, and clay must be 100%. Through DEM data, topographical variability can be obtained so that it will be a predictor or independent variable in predicting soil texture. In addition, Geographically Weighted Regression (GWR) was also used in this study to pay attention to the effect of spatial heterogeneity. It uses the bootstrap method with the GWR model to overcome bias in the model parameters. Residual bootstrap is a bootstrap method that is applied to the residual resampling process. The aims of this study: (1) To establish a soil texture prediction model using GWR with a single bootstrap approach, (2) To test the model's reliability in predicting surface soil texture. The results of this study are in the form of a prediction model and a map of the spatial distribution of PSF on surface soil which can later be used as a basis for determining sustainable soil management and supporting precision agriculture.