Flowering-Stage Indicators and Environmental Variables for Mango Yield Prediction: A Systematic Review of Vision-Based and Remote-Sensing Approaches

Mango (Mangifera indica) is among the most economically significant fruit crops of the tropics, and its final yield is decided very largely during the flowering window. Growers, traders and planners would all benefit from knowing the size of the crop before it arrives, yet the estimates they work with are still produced by walking the orchard and forming a judgement. Such judgements are slow, they vary from one observer to the next, and they cannot be scaled. Over the past five years a substantial body of work has attempted to replace them, using computer vision on ground imagery, unmanned aerial and satellite remote sensing, and machine-learning models built on climatic and physiological predictors. This paper reviews that literature systematically. Studies published between 2020 and 2025 were screened against explicit inclusion and exclusion criteria and the retained work was organised into three thematic clusters: vision-based flowering detection, remote-sensing yield estimation, and hybrid models combining spectral, climatic and phenological inputs. The review finds that the detection problem is largely solved, with flower and panicle detectors now matching trained human observers, while the prediction problem is not: the reported correlations between measured flowering intensity and final yield remain weak, typically in the region of R² ≈ 0.19 to 0.28. The gap does not lie in perception. It lies in the absence of longitudinal, region-specific datasets that follow the same trees from flowering through to harvest, and in the fact that flowering is almost never modelled jointly with the environmental variables that determine whether flowers become fruit. The paper concludes by setting out the requirements that a credible mango yield-prediction framework would have to meet, particularly under Indian semi-arid conditions.