Hybrid modelling framework for photosynthesis rate estimation from leaf to global scales: integrating the spectrum-based V25cmax simulator into the process-based model

Photosynthesis rate is the key element in the carbon cycle process. Accurate photosynthesis rate estimate hinges on the maximum carboxylation rate (V²⁵_cmax). The high uncertainty in deriving V²⁵_cmax has long hampered efforts toward the performance of the photosynthesis models from leaf to global scales. Recently studies suggest a strong relationship between spectral reflectance and V25 cmax,0. We proposed the spectrum-driven V²⁵_cmax simulator using deep learning methods and built the hybrid modelling framework for photosynthesis rate estimation by integrating the data-driven V²⁵_cmax simulator in the process-based model. The performance of hybrid photosynthesis models was evaluated at leaf, field and global scales. At the leaf scale, we developed a novel deep learning architecture, which incorporated spatial attention and prior knowledge of spectral indices calculation modules, to extract the V²⁵_cmax from leaf hyperspectral images. At field scale, we combined the high-resolution unmanned aerial vehicle (UAV) multispectral imagery and convolutional neural networks (CNN) to estimate V²⁵_cmax at the paddy field. At a global scale, we utilized a fully connected deep neural network (DNN) to construct the satellite multispectral-driven V²⁵_cmax model based on the FLUXNET2015 dataset. Our result showed that spectrum information can accurately estimate V²⁵_cmax. The hybrid framework fully extracts the information of all available spectral bands using deep learning to reduce parameter uncertainty while maintains the description of the photosynthetic process to ensure its physical reasonability. We also highlighted the significance of spatial heterogeneity for V²⁵_cmax estimation. This study provides new insight into monitoring photosynthesis rate across different spatial scales.