Banca de QUALIFICAÇÃO: DAVID LOPES MACIEL

Uma banca de QUALIFICAÇÃO de DOUTORADO foi cadastrada pelo programa.
STUDENT : DAVID LOPES MACIEL
DATE: 05/06/2026
TIME: 10:00
LOCAL: Via - auditório virtual usando GOOGLE MEETING - https://meet.google.com/edn-gcrf-ojk
TITLE:

HYBRID MODELING: ARTIFICIAL INTELLIGENCE APPLIED TO DETERMINING CO2 FLOW IN SOYBEAN CULTIVARS IN THE LEGAL AMAZON.

KEY WORDS:

CO2 flux; Hyperspectral remote sensing; Artificial intelligence.

PAGES: 155
BIG AREA: Outra
AREA: Multidisciplinar
SUMMARY:

This thesis sought to develop and evaluate a hybrid computational model to estimate CO2 flux in soybean cultivated areas in the Brazilian Legal Amazon, integrating hyperspectral remote sensing, biophysical modeling using SVAT/ISBA schemes, and machine learning. The research was structured based on the hypothesis that the integration of artificial intelligence, orbital hyperspectral remote sensing, and biophysical modeling would allow the estimation of CO2 flux in tropical agricultural systems, especially in areas where direct field measurements are scarce or operationally limited. The study used images from the EO-1 Hyperion sensor, a soybean mask derived from MapBiomas, meteorological variables from the conventional INMET station in Canarana, Mato Grosso, Brazil, code 83270, and computational routines developed in Python to organize the analytical pipeline. Spectral indices and biophysical variables related to the canopy condition of the studied crop were calculated, and the data were subsequently integrated into a simplified physical formulation to estimate the net CO2 exchange of the ecosystem. The XGBoost algorithm was then applied as a residual correction component, seeking to adjust patterns not represented by the reduced physical parameterization. The results of the simplified physical model showed RMSE = 1.30 gCO2 m−2 h−1, MAE = 1.11 gCO2 m−2 h−1, and R2 = -0.04, indicating limitations in reaching the adopted biophysical reference. The hybrid approach presented RMSE = 0.013 gCO2 m−2 h−1, MAE = 0.008 gCO2 m−2 h−1, and R2 = 0.9999 within the scope of the adopted temporal validation. These values indicate greater statistical adherence to the modeled reference; however, the results obtained should be interpreted as internal evidence of methodological consistency, without characterizing independent observational validation. Therefore, the contribution of this thesis lies in the formulation of a reproducible and traceable pipeline for a hybrid model applied to the study of carbon fluxes in tropical agricultural systems, articulating hyperspectral observation, biophysical interpretation, and artificial intelligence. The proposed approach may support future strategies for environmental monitoring and low-carbon agriculture, provided that it is accompanied by progressive external validations using independent observational data.

 

COMMITTEE MEMBERS:
Presidente - 1466944 - FABRICIO MORAES DE ALMEIDA
Interno - 3063165 - DIEGO HENRIQUE DE ALMEIDA
Interna - 2061659 - ROSALINA ALVES NANTES
Externo à Instituição - RODRIGO CAIADO DE LAMARE
Externa à Instituição - CLAUDIA BRAZIL MARQUES
Externo à Instituição - ANTONIO CARLOS DUARTE RICCIOTTI
Externo à Instituição - ADJALMA CAMPOS DE FRANCA NETO