Automated calibration of stomatal conductance models from thermal imagery by leveraging synthetic images generated from Helios 3D biophysical model simulations.

Stomatal conductance (gs) is indicative of plant carbon dioxide uptake via photosynthesis and water loss via transpiration, making it a crucial plant biophysical trait. Direct measurement of gs is labor-intensive and usually not scalable to large fields. Using manual measurements to estimate parameters of gs models is even more labor-intensive and prone to sampling errors. This study aimed to develop an automated pipeline for gs measurement and model calibration using thermal imagery data, which not only disentangles the impacts of genotype-specific stomatal traits and environmental conditions but also enables the prediction of gs in new environments. The methodology involved using simulated thermal imagery data generated from a 3D biophysical model to train a machine learning model that could be applied to real thermal images to predict stomatal model parameters and gs itself. The method was evaluated by comparing predictions against manual gs measurements, all of which were not part of the model training process, as the model was trained against only simulated images. When compared against manual gs measurements using a porometer, the prediction R2 was 0.7, which is likely comparable to the accuracy of the manual porometer-based gs measurements (relative to a leaf gas exchange system). The developed pipeline enables high-throughput gs model parameter calibration and gs estimation.