Gaussian transformation methods for spatial data

Abstract

Data gaussianity is an important tool in spatial statistical modeling as well as in experimental data analysis. Usually field and experimental observation data deviate significantly from the normal distribution. This work presents alternative methods for data transformation and revisits the applicability of a modified version of the well-known Box-Cox technique. The recently proposed method has the significant advantage of transforming negative sign (fluctuations) data in advance to positive sign ones. Fluctuations derived from data detrending cannot be transformed using common methods. Therefore, the Modified Box-Cox technique provides a reliable solution. The method was tested in average rainfall data and detrended rainfall data (fluctuations), in groundwater level data, in Total Organic Carbon wt% residuals and using random number generator simulating potential experimental results. It was found that the Box-Cox technique competes successfully in data transformation. On the other hand, it improved significantly the normalization of negative sign data or fluctuations. The coding of the method is presented by means of a Graphical User Interface format in MATLAB environment for reproduction of the results and public access.