Interpolation of Mexican soil properties at a scale of 1:1,000,000

Mexico maintains several databases with biotic or abiotic information that enable large scale-studies (for example, at a resolution of 1 km2); unfortunately, there is no information at this resolution for soil properties. The goal of this paper was to generate a set of soil variables to address this absence. We evaluated 4400 soil samples taken on Mexican territory. The following nine soil properties were evaluated for each sample: Ca, K, Mg, Na, organic C, organic matter, electrical conductivity, sodium absorption ratio, and pH. With the use of geostatistical methods, a layer was generated for each soil property after six different semivariance models were evaluated. The kriging model (simple, ordinary, or universal) was selected as the best semivariance method with a 10-fold cross-validation approach. Exponential, pentaspherical, and spherical models were selected with ordinary and universal kriging methods to spatially predict the soil properties. Five classes for each soil map were generated. The nine soil properties were classified primarily as slightly low or medium according to their distribution throughout Mexican territory.