Climate variables mapping, as intense rainfall, is very important to environmental management. Although, statistical tools for spatial interpolation should be analyzed and characterized. This paper aims to analyze models and methods of semi-variogram modeling applied to intense rainfall with duration time of 20, 60, 360 and 1440 minutes and 5, 50 and 100 years of recurrence and consequently, giving support for its mapping, using kriging, in Minas Gerais State. Exponential, Spherical and Gaussian semi-variogram models were tested based on Weighted Minimum Square (WMS) and Maximum Likelihood (ML) methods, using GeoR software. For the best model and method evaluation was considered the mean absolute error produced by cross-validation. For mean error similarity, it was considered the spatial degree of dependence and smaller nugget effect. Visual adjustment of semi-variogram was also analyzed to complete the selection. Exponential model was predominant in nine of twelve situations, followed by Gaussian model in two situations and Spherical for just one. Weighted Minimum Square was the best adjust method in all situations. These results have indicated the exponential model adjusted by Weighted Minimum Square to intense rainfall mapping for Minas Gerais State conditions.
Spatial dependence structure; intense rainfall; geostatistics
