Predicting Olive Yield in Mediterranean Climate Zones of Turkiye Using Remote Sensing and Artificial Neural Networks: A Case Study of Mugla Province

Abstract

Cagatay Nalcaoglu, Durdane Hilal Nalcaoglu, Aynur Dogru and Sezginyaprak

This study focuses on predicting olive yield in the Mugla province of Turkey—one of the country’s major olive production regions— using remote sensing data and artificial neural networks (ANN), a machine learning approach. The research integrates multi- source data, including Sentinel-2 and MODIS satellite imagery (NDVI, LST, GPP), meteorological data from the Turkish State Meteorological Service, and soil parameters from the Soil Grids database. These multidimensional datasets were used to train and evaluate an ANN-based model to predict annual olive yield at the district level between 2020 and 2024. The ANN model demonstrated high predictive performance, with a test R2 of 0.82, RMSE of 0.18 t/ha, and MAE of 0.12 t/ha, outperforming alternative models such as XGBoost. The results confirmed strong positive correlations between NDVI and GPP with yield, and a negative correlation with LST. The model outputs offer valuable tools for agricultural planning, climate adaptation strategies, and spatially targeted interventions. This research contributes a novel, high-resolution, district-level modeling approach using ANN for perennial crops, providing insights for both data-driven agricultural policy and smart farming practices in Mediterranean environments.

PDF