Modeling Optimal Stomatal Behavior Of Crops Based On Machine Learning And Remote Sensin

The slope（g₁）of stomatal conductance to photosynthesis is an important parameter in the optimal stomatal behavior theory-based stomatal conductance model.Although multiple studies modeled the spatial variations in g₁,disclosing its variations over environmental gradients and different plant functional types.However,the above methods are still not accurate enough on a global scale,as they do not consider the temporal variations in g₁.To address this issue,improve the accuracy of remote sensing simulation of gross primary productivity（GPP）and latent heat flux（LE）of farmland at a global scale and large area with g₁ parameters.This research established a remote sensing（RS）based photosynthesis-evapotranspiration coupling model（SCOPES-Crop）to explore the seasonal variation and spatial and temporal distribution characteristics of g₁,and model g₁ for C₃ and C₄ crops based on the established model.The main innovations and contributions of this paper are as follows:（1）The Medlyn model of stomatal conductance based on optimal stomatal behavior theory was integrated into the simulation of cropland carbon and water budgets based on RS,and a remote sensing-based model termed SCOPES-Crop was constructed,and the model parameters were optimized at a global scale.（2）This research,the Ensemble Kalman Filter（En KF）to assimilate tower based GPP LE of 17 cropland flux sites,to derive the temporal variations in g₁ for C₃ and C₄crops.Results showed g₁ to rise rapidly in spring and summer,and then decline in autumn.The value of g₁ reached the lowest value and remained stable in wintertime.（3）This research,the feedforward artificial neural network（FANN）and RS variables are used to model g₁.FANN-based modeling of g₁ showed R（RMSE）=0.81(1.94 k Pa^0.5)and 0.90(0.70 k Pa^0.5)for C₃ and C₄ Crops,respectively,for the testing dataset.The estimates of GPP and LE using FANN-derived g₁ at the 17 flux sites were improved as compared to that using fixed g₁.The mean values of site-level R（RMSE）for GPP and LE simulated using FANN-derived g₁ are 0.92(1.8 g C m^-2 d^-2)and 0.85(22.5 W m^-2),respectively.（4）Remote sensing raster data were used to verify the reliability of the FANN model in the North China Plain,and the spatial and temporal distribution characteristics of g₁ in the North China Plain were analyzed.The results showed that g₁ values of summer maize（2 to 10）in the North China Plain simulated by FANN were higher than those of summer maize（0 to 4）generated by En KF data assimilation,and were consistent with the range of values of winter wheat（0 to 10）.Temporally simulated g₁values for winter wheat and summer maize in the North China Plain are consistent with the seasonal variation trend of g₁ generated by En KF assimilation.