High-resolution data for mapping the spatio-temporal variability of sugarcane fields

• Autor(a) principal: Canata, Tatiana Fernanda
• Data de Publicação: 2021
• Tipo de documento: Tese
• Idioma: eng
• Título da fonte: Biblioteca Digital de Teses e Dissertações da USP
• Texto Completo: https://www.teses.usp.br/teses/disponiveis/11/11152/tde-15092021-120624/
• topic: Agricultura de precisão; Índices de vegetação; Nuvem de pontos; Point cloud; Precision agriculture; Remote sensing; Sensoriamento remoto; Vegetation index

Data-driven solutions have been more common in agriculture, mainly, with the progress of technology to optimize the operations at the field. Precision agriculture and remote sensing techniques have enabled the data acquisition of agronomical variables and more accurate diagnostics than using traditional methods. The crop monitoring for sugarcane area is commonly associated with regional scale or classification of sugarcane areas based on satellite imagery. Some alternative techniques can support the assessment of the spatio-temporal variability of the fields using three-dimensional (3D) sensing data provided by LiDAR (Light Detection and Ranging) technology, which have contributed to the site-specific crop management considering the canopy height or volume variation. Researchers have applied LiDAR data and aerial images in small areas, which indicated the requirement of understanding better the data acquisition and processing for large-scale applications, such as sugarcane areas in Brazil. This study aims to investigate the potential of 3D sensing data and satellite imagery to map the spatio-temporal variability of sugarcane fields before harvesting. Chapters 1 and 2 introduce the topic of the thesis and highlight the state of the art of the concepts used in the research. Chapter 3 describes the applied methods for data processing of 3D sensing data and aerial images in order to assess the spatio-temporal variability of a commercial sugarcane field. The sugarcane yield map was generated by a sensor-system that measures the mass flow based on the difference of hydraulic pressure of the chopper system in the harvester. The plant height of sugarcane fields was obtained with an ALS (Airborne Laser Scanning) in the final crop production cycle for two consecutive seasons. The point cloud generated, followed by the data filtering, enabled to obtain the canopy height model (CHM) as an information to investigate the association between the spatial variation of crop height and the yield map. A moderate relationship was found between the CHM and the yield map, which demonstrated the potential of high-resolution data to identify the spatial variability at the field level. Chapter 4 explores the use of orbital images for sugarcane yield mapping using reflectance data and vegetation index over the crop cycle for three consecutive crop seasons. The yield prediction models were developed by integrating satellite images, machine learning and multiple linear regression. The regression using Random Forest (RF) showed greater accuracy, since the non-linearity of the dataset was observed, and the spectral bands showed a lower error in estimation of yield. Crop vigor mapping using time-series analysis of satellite imagery supports the identification of the spatial variability of sugarcane fields. The results of this research demonstrated the potential applications of high-resolution data for guiding complementary diagnostics and local interventions in agricultural systems, since it indicates the crop height or vigor variation in large-scale before harvesting.