Delineamento de zonas de manejo por imagens suborbitais, orbitais e variáveis de solo

Abstract

Precision Agriculture is constantly concerned with coherent use of natural resources, which makes it key in optimizing management, as it considers the spatial variability of an area and allows its segmentation into subregions (Management Zones – MZ). One of the possible ways to evaluate such variability is by using spectral data, i.e., via orbital, sub-orbital, and terrestrial sensors. Among the suborbital sensors, Unmanned Aerial Vehicles (UAVs) enable non-intrusive surveys with high level of detail and at lower cost. Thus, this work proposes to create MZ focused on temporal analysis on a short period (one harvest) from three datasets: visible vegetation indexes from UAVs and Sentinel-2, and soil data; comparing the MZs with yield maps. In this study, one area was monitored throughout two harvest cycles (soybean, during the 2018/2019 crop, and maize, crop of 2019), collecting UAV images, field data (soil sampling and yield). MZs were designed for the subsets through spectral data (VANT and Sentinel-2) and soil sampling using hierarchical clustering with spatial constraints weights, evaluating different spatial contiguity matrices (Queen, Rook, and KNN). Among the visible vegetation indexes ExG, GLI, GRVI, RGBVI, and VARI, GRVI and VARI achieved higher results. MZs created by UAV or Sentinel-2 for both crops achieved higher performance with two groups; and for soil data, by four groups. None of the MZs shared any similarities regarding their cultures’ productivity. Visible vegetation index achieved poor results for MZ creation, regardless of the sensor employed (VANT or Sentinel-2).