Sampling planning comprises one of the most critical steps for characterizing the spatial behavior of soil properties. However, the continuous management of agricultural areas influences the pattern of soil chemical, physical and biological attributes, which makes mapping difficult through sampling. The use of covariates is a possibility to assist with both sample planning and predictions. However, to obtain significant gains with these techniques, it is mandatory that the spatial variability structure of soils be understood. Therefore, this research aims to integrate several layers of information from the terrain, field management, and proximal and remote sensing as auxiliary information for sampling targeting and increasing the quality of predictions. For this, simulated data and sampling carried out in the field for chemical fertility and soil quality diagnosis will be used, evaluating different strategies for the allocation of sampling points. Geostatistical methods and machine learning algorithms will be employed to predict soil properties in unsampled locations. The results obtained by sample optimization and data interpolation techniques will be compared with regular sample grids and through validation samples.
News published in Agência FAPESP Newsletter about the scholarship: