Using Machine Learning Algorithms to Mapping of the Soil Macronutrient Elements Variability with Digital Environmental Data in an Alluvial Plain

Fine resolution spatial digital maps of soil macronutrients, which are an important factor in plant nutrition, are needed to support agricultural productivity. Digital soil maps obtained with high precision and accuracy are at the forefront of innovative technological initiatives to increase agricultural production. We had 91 topsoil observations, indices produced from satellite imagery, topographical variables produced from the DEM, and the CORINE land cover classes map which showed the effectiveness of agricultural activities for many years. Our first ultimate goal was to create digital soil maps with a spatial resolution of 30 m of various soil macronutrients (P, Ca, Mg, K). We compared three machine learning algorithms: multiple linear regression, support vector machine, and random forest algorithms. Our results showed that random forest and support vector machine algorithms achieved relatively high accuracy in predicting spatial distributions of soil properties affected by human activities. CORINE land cover classes map was identified as an important environmental variable in models for the production of phosphorus map, especially. Framework of soil science and sufficiency class values of nutrients can be created in a raster environment by using raster calculation tools of geographic information systems and it will allow a more effective use of maps.