Integrating remote sensing and crop modeling for enhanced sugarcane yield predictions in the main production region of Brazil

• Ano de defesa: 2024
• Autor(a) principal: Rosa, Juliano Mantellatto
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Biblioteca Digitais de Teses e Dissertações da USP
• Palavras-chave em Português: Saccharum officinarum L.; Assimilação de dados; Data assimilation; DSSAT; Ensemble smoother; Remote sensing; SAMUCA; Sensoriamento remoto; TED
• Link de acesso: https://www.teses.usp.br/teses/disponiveis/11/11152/tde-12032025-154152/

Sugarcane is a crucial raw material for producing sugar and ethanol, making effective strategies for monitoring its quantity and availability essential. Various methods have been developed for yield estimation, particularly those combining process-based crop models (PBMs) with data assimilation (DA). These methods leverage two different sources of information, each with its own uncertainties, to enhance yield estimation accuracy. However, a comprehensive framework for analyzing sugarcane data at regional scales using DA has yet to be established. This study aimed to improve sugarcane yield prediction accuracy in Brazil\'s primary production regions by integrating Leaf Area Index (LAI) data into a stochastic crop simulation model. The LAI data, sourced from Landsat 7 ETM+ sensor, were extracted at the pixel level for sugarcane blocks across São Paulo state. Approximately 167,000 remotely sensed LAI data points were assimilated into the SAMUCA model using a DA method known as Ensemble Smoother (ES) to minimize estimation uncertainties. Spanning the 2003-2013 period across São Paulo\'s sugarcane-growing regions, the study employed the Technology Extrapolation Domains (TEDs) to segment areas based on biophysical characteristics. Results show significant improvements in prediction accuracy: the DA method reduced the Root Mean Square Error (RMSE) from 43.98 Mg ha-1 to 17.83 Mg ha-1, an improvement of up to 59%, while the Mean Absolute Error (MAE) decreased from an average of 41.85 Mg ha-1 to 11.65 Mg ha-1 across the 22 TEDs. Notably, individual TEDs consistently showed reductions, with average RMSE reductions of 57% and MAE reductions reaching up to 96% in specific areas. These findings underscore the value of incorporating LAI data into operational yield estimation frameworks, with implications for regional-scale production resource optimization and strategic agricultural management. This approach establishes a robust foundation for regional crop yield estimation.