Kriging and cokriging for spatial interpolation of rainfall in Espirito Santo State, Brazil

Kriging and cokriging for spatial interpolation of rainfall in Espirito Santo State, Brazil
Alexson de Mello Cunha1,Gerson Rodrigues dos Santos2, Eliana de Souza1;Filipe Silveira Trindade3; Elpídio Inácio Fernandes Filho2; João Luiz Lani2; Michelle Milanez França1

1 Postgraduate Course on Soils and Plant Nutrition, Soils Department, Viçosa Federal University.CEP 36570-000. Viçosa (MG).
2.University Viçosa Federal University, CEP 36570-000. Viçosa (MG). (,,,,,,

Abstract: Kriging has become a widely used method for spatial distribution of hydroclimatic variables. The aim of this work was to evaluate the simple kriging and cokriging on the spatial distribution of rainfall in Espírito Santo state, Brazil. Data model elevation and distance from the sea, obtained from sampling points in regular and irregular grids were used as covariates in co-kriging. Also, we used average annual rainfall of 108 rain gauge stations geographically distributed in the state. The evaluation of methods and variables was based on cross-validation, considering the errors of the predicted values, adjusting the first degree linear regression model of the observed values as a function of the estimated values and the coefficient of determination. In general, the regular grid sampling of the covariates showed a slight trend toward better prediction accuracy compared to the irregular grid. Also, co-kriging provided more accurate results than kriging, which was verified by the average absolute errors, ranging from 7.066% to 7.924%, and by the root mean squared errors, varying from 109.6 mm and 120.4 mm. The results suggest that, wherenever possible, it is better to use cokriging and regular grids for sampling.

Keywords: elevation, coastline, geostatistic, spatial prediction

CunhaAccuracy2012.pdf327.32 KB