Spatial Modeling of Categorical Attributes Using Indicator Simulation and Soft Information With Uncertainty Analyses

Abstract

This work explores a methodology to apply indicator geostatistical simulation approaches to geospatial modeling of categorical attributes using hard and soft information. Uncertainty analyses of the predictions are performed to evaluate the quality of classifications. Sample points of a categorical attribute are considered as the hard, or primary, information while a categorical map is used for determine the soft, or the secondary, information. The soft information is incorporated in the indicator simulation procedure as prior mean values, taken from a probability distribution function, related to the hard data. The prior mean values are then updated via indicator simulation to account for the hard data available in their neighborhoods. To illustrate the methodology a case study is presented with samples of soil texture classes, as the hard data, and with classes of a soil map determining the soft information. These data are gathered from an experimental farm of agriculture researches. Uncertainty analyses of the results show that the use of soft information, along with the hard data, allows one to find out new specific regions of higher and lower uncertainties. The highest uncertainties regions should be considered as candidates for future resampling.