Improving Olive Yield Prediction Using Landsat Multispectral Data and Advanced Ensemble Learning in Tunisia

ABSTRACT Olive cultivation is a key agricultural activity in Mediterranean regions, yet its yield is highly sensitive to the impacts of climate change. Accurately predicting olive yield using optical remote sensing and data‐driven models remains a challenging task. In this study, we developed an efficient workflow to estimate olive yields in the Kairouan and Sousse governorates of Tunisia. Our approach involved extracting features from multispectral reflectance bands and vegetation indices derived from Landsat‐8 Operational Land Imager (OLI) and Landsat‐9 OLI‐2 imagery, combined with topographic data from a digital elevation model (DEM). These spatial features were integrated with ground‐truth observations collected through field surveys to construct a structured tabular data set. We then implemented an automated ensemble learning framework using AutoGluon to train and evaluate multiple machine learning models, optimise model combinations through stacking, and generate reliable yield predictions through five‐fold cross‐validation. The findings demonstrate strong predictive accuracy for both optical sensors, with Landsat‐8 OLI achieving an R ² = 0.8635 and an RMSE = 1.17 tons ha −1 , while Landsat‐9 OLI‐2 achieved an R ² = 0.8378 and an RMSE = 1.32 tons ha −1 . Our study presents a robust, scalable, and cost‐effective approach for olive yield prediction, with promising applicability for monitoring agricultural crop yields across diverse regions worldwide.