DELINEATION OF HOMOGENEOUS ZONES BASED ON GEOSTATISTICAL MODELS ROBUST TO OUTLIERS

• Autor(a) principal: Barbosa, Danilo Pereira
• Data de Publicação: 2019
• Outros Autores: Bottega, Eduardo Leonel, Valente, Domingos Sárvio Magalhães, Santos, Nerilson Terra, Guimarães, Wellington Donizete
• Tipo de documento: Artigo
• Idioma: eng
• Título da fonte: Revista Caatinga
• Texto Completo: https://periodicos.ufersa.edu.br/caatinga/article/view/7497
• topic: Robust statistics. Precision agriculture. Apparent soil electrical conductivity. Spatial variability. fuzzy k-means.; Estatística robusta. Agricultura de precisão. Condutividade elétrica aparente. Variabilidade espacial. Fuzzy k-means.

Measures of the apparent electrical conductivity (ECa) of soil are used in many studies as indicators of spatial variability in physicochemical characteristics of production fields. Based on these measures, management zones (MZs) are delineated to improve agricultural management. However, these measures include outliers. The presence or incorrect identification and exclusion of outliers affect the variogram function and result in unreliable parameter estimates. Thus, the aim of this study was to model ECa data with outliers using methods based on robust approximation theory and model-based geostatistics to delineate MZs. Robust estimators developed by Cressie–Hawkins, Genton and MAD Dowd were tested. The Cressie–Hawkins semivariance estimator was selected, followed by the semivariogram cubic fit using Akaike information criterion (AIC). The robust kriging with an external drift plug-in was applied to fitted estimates, and the fuzzy k-means classifier was applied to the resulting ECa kriging map. Models with multiple MZs were evaluated using fuzzy k-means, and a map with two MZs was selected based on the fuzzy performance index (FPI), modified partition entropy (MPE) and Fukuyama–Sugeno and Xie–Beni indices. The defined MZs were validated based on differences between the ECa means using mixed linear models. The independent errors model was chosen for validation based on its AIC value. Thus, the results demonstrate that it is possible to delineate an MZ map without outlier exclusion, evidencing the efficacy of this methodology.