SUGARCANE PRODUCTIVITY PREDICTION FROM SAR IMAGES IN HIGH SPATIAL RESOLUTION

• Grant number: 23/12735-7
• Support Opportunities: Scholarships in Brazil - Scientific Initiation
• Start date: February 01, 2024
• End date: January 31, 2025
• Field of knowledge: Agronomical Sciences - Agronomy
• Principal Investigator: José Paulo Molin
• Grantee: David Lopes Lima
• Host Institution: Escola Superior de Agricultura Luiz de Queiroz (ESALQ). Universidade de São Paulo (USP). Piracicaba , SP, Brazil

Abstract

The productivity maps obtained by equipment embedded in harvesters are a precision agriculture technology of great importance for monitoring the variability of Brazilian sugarcane fields, representing an essential variable for Differentiated Management Units (UGDs). However, as a result of technological, logistical and operational barriers, most farmers opt for productivity estimates based on empirical processes rather than productivity maps. In this sense, remote sensing has been widely explored as it allows the identification of variability in a non-destructive way and at high spatial and temporal resolution through vegetation indices. Based on this assumption, the present work aims to develop predictive models of sugarcane productivity through images obtained by Synthetic Aperture Radar (SAR) imaging sensors onboard the Sentinel-1 satellite, which do not suffer from climate fluctuations. Therefore, the Vegetation Optical Depth (VOD) vegetation index will be used, known for showing a high correlation with the water content and biomass of plants; essential parameters in sugarcane productivity. The predictive model will be developed by supervised machine learning algorithms using satellite images and real productivity data at high spatial resolution, identifying the most influential factors. The statistical parameters for evaluating the prediction capacity will be carried out by R², RMSE and Index of Agreement Modified (dmod). Therefore, it is proposed to develop a model capable of predicting sugarcane productivity in high spatial resolution before harvesting, so that the information generated can assist and optimize harvest logistics; offer alternatives to the productivity maps obtained by productivity monitors and compose a satisfactory variable for the characterization of UGDs.