Delineation of Soil Management Zone Maps at the Regional Scale Using Machine Learning

Abstract

Applying fertilizers to soil in a site-specific way that maximizes yields and minimizes environmental damage is an important goal. Developing soil management zones (MZs) is a suitable method for achieving sustainable agricultural production. Thus, this work aims to investigate MZs delineated based on the different soil properties using machine learning methods. To achieve these, 202 soil samples were collected at the agricultural land of pomegranate, pistachio, and saffron. A “random forest” model was applied to map soil properties based on environmental covariates. The predicted “Lin’s concordance correlation coefficient” values in validation soil properties varied from 0.65 to 0.79. The maps indicated low amounts of soil organic carbon, available potassium, available phosphate, and total nitrogen in most of the region. Furthermore, the study identified four different MZs according to relationships between soil properties and environmental covariates. Generally, the ranking of zones in terms of soil fertility was MZ4 > MZ1 > MZ3 > MZ2 based on the investigated soil properties and the soil quality (SQ) map. The five grades of SQ (i.e., very high, high, moderate, low, and very low) indicated that there was heterogeneous SQ in each MZ in the study area. There were 1.65 ha identified in MZ4 with very low SQ. This result is important in determining the amount of fertilizer to add to the soil in the different areas. It confirms the need for more specific regional management of agriculture lands in this region.