Shaping the future of citrus yield prediction with multimodal data fusion and machine learning: a systematic review

Abstract

Accurate yield forecasting enables growers and the citrus industry to optimize resource allocation, mitigate economic risks, and enhance supply chain efficiency. However, existing studies remain fragmented, with substantial variation in input features, data acquisition platforms, preprocessing strategies, and predictive algorithms, limiting comparability and the development of common best practices. This review systematically synthesizes the state of the art in citrus yield prediction from four perspectives: (i) input features influencing yield; (ii) data acquisition platforms and feature extraction methods; (iii) prediction models and analytical approaches; and (iv) complementary yield-related methods, including estimation, monitoring, and measurement. Comparative analysis of published studies highlights consistent trends, such as the importance of fruit counts and canopy spectral indices for short-term predictions, the critical but underexplored role of mid-term forecasting in bridging imaging and climate-based data regimes, the reliance on climatic and soil variables in long-term forecasting, and the increasing use of machine learning to capture complex, nonlinear interactions. Despite notable progress, critical challenges remain, including limited availability of long-term and large-scale datasets, occlusion and variability in image-based methods, and poor model transferability across orchards, varieties, and growing regions. To address these barriers, future research should prioritize standardized and harmonized feature sets, robust multi-source data integration, interpretable and hybrid modeling frameworks that incorporate biological knowledge, and scalable pipelines capable of supporting operational decision-making. By consolidating fragmented knowledge, this review provides both the current state of citrus yield prediction and a roadmap for advancing toward reliable, transferable, and sustainable forecasting systems.