Machine Learning Approaches for Assessing Avocado Alternate Bearing Using Sentinel-2 and Climate Variables—A Case Study in Limpopo, South Africa

Alternate (irregular) bearing, characterized by large fluctuations in fruit yield between consecutive years, remains a major constraint to sustainable avocado (Persea americana) production. This study aimed to assess the potential of satellite remote sensing and climatic variables to characterize and predict alternate bearing patterns in commercial orchards in Tzaneen, Limpopo Province, South Africa. Historical yield data (2018–2024) from 46 ‘Hass’ avocado blocks were analyzed alongside Sentinel-2 derived vegetation indices (NDVI, GNDVI, NDRE, CIG, CIRE, EVI2, LSWI) and flowering indices (WYI, NDYI, MTYI). Climatic predictors including maximum temperature (Tmax), minimum temperature (Tmin), vapour pressure deficit (VPD), and precipitation were incorporated. Five machine learning algorithms—Random Forest, XGBoost, CATBoost, LightGBM, and TabPFN—were trained and tested using a Leave-One-Year-Out (LOYO) approach. Results showed that VPD, Tmin, and Tmax during the flowering period (July–September) were the most influential variables affecting subsequent yields. TabPFN achieved the highest predictive accuracy (Accuracy = 0.88; AUC = 0.95) and strongest temporal generalization. Spectral gradients between flowering and early fruit drop were lower during “on” years, reflecting stable canopy vigour. These findings demonstrate that integrating remote sensing and climatic indicators enables early discrimination of “on” and “off” years, supporting proactive orchard management and improved yield stability.