A genetic algorithm for resizing and sampling reduction of non-stationary soil chemical attributes optimizing spatial prediction

Abstract

Aim of study: To evaluate the influence of the parameters of the geostatistical model and the initial sample configuration used in the optimization process; and to propose and evaluate the resizing of a sample configuration, reducing its sample size, for simulated data and for the study of the spatial variability of soil chemical attributes under a non-stationary with drift process from a commercial soybean cultivation area. Area of study: Cascavel, Brazil Material and methods: For both, the simulated data and the soil chemical attributes, the Genetic Algorithm was used for sample resizing, maximizing the overall accuracy measure. Main results: The results obtained from the simulated data showed that the practical range did not influence in a relevant way the optimization process. Moreover, the local variations, such as variance or sampling errors (nugget effect), had a direct relationship with the reduction of the sample size, mainly for the smaller nugget effect. For the soil chemical attributes, the Genetic Algorithm was efficient in resizing the sampling configuration, since it generated sampling configurations with 30 to 35 points, corresponding to 29.41% to 34.31% of the initial configuration, respectively. In addition, comparing the optimized and initial configurations, similarities were obtained regarding spatial dependence structure and characterization of spatial variability of soil chemical attributes in the study area. Research highlights: The optimization process showed that it is possible to reduce the sample size, allowing for lesser financial investments with data collection and laboratory analysis of soil samples in future experiments.