Crop biophysical parameter estimation via AI-enhanced hyperspectral remote sensing

Hyperspectral remote sensing has become a key approach for retrieving crop parameters related to canopy structure, physiological status, biochemical composition and productivity. However, many current methods remain centered on task-specific empirical fitting between spectral features and target parameters. Their transferability is limited when crop growth varies across phenological stages, observation scales, sensor conditions and planting systems. Hyperspectral crop-parameter retrieval is moving toward an intelligent inversion paradigm, which is jointly supported by spatio-temporal-spectral representation, mechanistic constraints and trustworthy decision-making. Accordingly, we synthesize recent progress in crop spectral-response mechanisms, satellite-airborne-ground coordinated observation, spatio-temporal consistency processing, artificial intelligence (AI)-driven retrieval models and mechanism-data fusion. Particular attention is given to the integration of hyperspectral features with crop-growth processes and canopy radiative transfer. We also examine transferable spatio-temporal-spectral representations, applicability-domain assessment and uncertainty quantification. Future advances should move beyond improving the accuracy of multi-spatio-temporal observations and instead progress toward mechanism-informed intelligent retrieval that integrates crop-growth knowledge with deep learning. This approach should explicitly represent crop growth processes, spatial heterogeneity and applicability constraints, thereby supporting more reliable crop monitoring, growth diagnosis, yield prediction and decision-making in complex agricultural systems.