Master's Thesis
Using Machine Learning and Multi-Source Data for Predicting Grain Maize Yield in Precision Farming
— 2025
Key information
Authors:
Supervisors:
Published in
November 26, 2025
Abstract
Precision Farming (PF) is a technology-driven approach that enhances agricultural sustainability through the efficient and responsible use of resources, aided by Artificial Intelligence (AI) and Remote Sensing (RS). This thesis integrated multi-source data, including RS data from the ESA Sentinel missions, along with their derived Vegetation Indices (VIs), soil and terrain data, and meteorological records, to predict final maize yield throughout a crop’s season. The prediction was carried out using Decision Tree (DT), Random Forest (RF), and Extreme Gradient Boosting (XGBoost) models, employing a random split between training and test sets, and a Leave-One-Year-Out (LOYO) approach. The work also included an optimisation approach to determine the optimal hyperparameters for the XGBoost model. Maize yield was predicted at three distinct periods of the maize growing season. The models demonstrated constant predictive capabilities, ranging from an R2 score of 0.80 for week 1 (sowing) prediction to an R2 of 0.82 in Week 22 (harvesting). The LOYO approach, on the other hand, showed a lack of generalisation, with the best R² of 0.39 for week 17 of 2022 and the worst of -1.03 for week 17 of 2023. These results highlight the importance of PF in agriculture to maintain or increase productivity while securing long-term environmental sustainability. Future improved versions of these models may enable precise management of inputs such as irrigation and fertilisers, helping reduce chemical runoff, greenhouse gas emissions and the over-exploitation of resources, while supporting soil and ecosystem health.
Publication details
Authors in the community:
Ana Rita Costa Ramos dos Santos
ist197337
Supervisors of this institution:
Ricardo Filipe de Melo Teixeira
ist149031
Fields of Science and Technology (FOS)
chemical-engineering - Chemical engineering
Publication language (ISO code)
por - Portuguese
Rights type:
Embargoed access
Date available:
January 9, 2027
Institution name
Instituto Superior Técnico