Evaluating the canopy structure dynamics model for maize phenology prediction using Sentinel-2

Near real-time (NRT) acquisition and accurate prediction of key phenological stages in maize are essential for optimizing irrigation decisions and field water management. The shape model fitting (SMF) approach, based on remote sensing technology, has been widely used for phenological stage detection due to its high accuracy and comprehensiveness. However, existing NRT crop phenology monitoring models are often constrained to specific regions or crop types, with validation primarily focused on temporal scales. Systematic evaluations of these models’ applicability across different regions and crop varieties remain insufficient. Moreover, there is a lack of consensus on the effectiveness of various vegetation indices (VIs) for extracting key phenological stage information and their applicability in phenological inversion. This study integrates an enhanced canopy structure dynamics model (CSDM) with the SMF approach, leveraging Sentinel-2 satellite data to assess the role of different VIs in enhancing the precision of key phenological stage identification and to evaluate the model’s applicability across diverse regions and crop varieties. By analyzing VIs data from two maize varieties cultivated on four farms in Heilongjiang Province, China, we identified nine key phenological stages (seeding, emergence, development of stem, heading, flowering, development of fruit, ripening, senescence, and end of season). The results showed that while different VIs exhibited varying sensitivities and responsiveness to environmental changes at different phenological stages, the enhanced model consistently achieved high predictive accuracy, with RMSEs for most key phenological stages remaining under five days. Additionally, the model exhibited robust fitting performance for varieties with similar thermal time requirements and achieved high accuracy across different regions. It provided stable predictions during early phenological stages, with minor deviations in later stages, primarily attributed to variations in accumulated thermal time rates. In summary, this study systematically evaluated the applicability of the enhanced CSDM-SMF model for maize phenology prediction based on Sentinel-2 data from three perspectives: VI selection, regional differences, and varietal adaptability. The findings provide a more comprehensive understanding for applying this model in academic research and contribute to improving the accuracy of agricultural monitoring and management practices.