Validation of Rapid and Low-Cost Approach for the Delineation of Zone Management Based on Machine Learning Algorithms

Abstract

Proximal soil sensors are receiving strong attention from several disciplinary fields, and this has led to a rise in their availability in the market in the last two decades. The aim of this work was to validate agronomically a zone management delineation procedure from electromagnetic induction (EMI) maps applied to two different rainfed durum wheat fields. The k-means algorithm was applied based on the gap statistic index for the identification of the optimal number of management zones and their positions. Traditional statistical analysis was performed to detect significant differences in soil characteristics and crop response of each management zones. The procedure showed the presence of two management zones at both two sites under analysis, and it was agronomically validated by the significant difference in soil texture (+24.17%), bulk density (+6.46%), organic matter (+39.29%), organic carbon (+39.4%), total carbonates (+25.34%), total nitrogen (+30.14%), protein (+1.50%) and yield data (+1.07 t ha−1). Moreover, six unmanned aerial vehicle (UAV) flight missions were performed to investigate the relationship between five vegetation indexes and the EMI maps. The results suggest performing the multispectral images acquisition during the flowering phenological stages to attribute the crop spatial variability to different soil proprieties.