Spatial Extension Of Satellite Solar-induced Chlorophyll Fluorescence Products

Solar-induced chlorophyll fluorescence（SIF）is an excited light phenomenon in which a chlorophyll molecule absorbs photons and re-emit when they return from the excited state to the unexcited state under natural light conditions.The SIF,as a photosynthetic concomitant,is important for the remote estimation of gross primary production（GPP）.In the past few years,SIF observations from space have become a disruptive innovation in assessing the spatial and temporal distribution of GPP.Compared with existing GPP estimation models,SIF is expected to break through fewer input parameters,reduce model uncertainties,and achieve accurate estimates of GPP.However,the majority of existing satellite SIF products suffer from either coarse spatial resolution or spatial discontinuity,which to some extent limits the ability of SIF to estimate GPP.Therefore,the study of the spatial extension of satellite SIF products from"spatial discontinuity and coarse spatial resolution"to"spatial continuity and fine spatial resolution"is of great significance to improve the application capability of SIF.This paper focuses on the scientific problem of spatial extension of satellite SIF,addresses the limitations of existing spatial extension methods,and carries out the analysis of key driving variables of SIF simulation model,the study of spatial extension based on TanSat SIF,the study of spatial downscaling based on GOME-2 SIF,and the spatial downscaling product application based on GOME-2 SIF these four aspects of research work.The main conclusions are as follows:（1）This paper explores the key driving variables of the SIF simulation model based on the SIF light use efficiency model through a combination of mechanistic analysis and empirical datasets,and finds that the normalized clear-sky SIF can be accurately simulated using reflectance（at red,near-infrared,blue,and green bands）,normalized vegetation index,cos（SZA）,and air temperature.Furthermore,adding cos（SZA）as the driving variable can greatly improve the accuracy of the model,with R²improving from 0.60 to 0.70 and RMSE decreasing from 0.30 to 0.26 m W m^-2 nm^-1 sr^-1.（2）Based on the key driving variables,the spatial extension model established by using the discontinuous Tan Sat satellite SIF combined with the random forest model can realize the SIF from"spatial discontinuity"to"spatial continuity".The random forest model validation results show that the R² equals to 0.69,0.70 and 0.73,and the RMSE equals to 0.33,0.30 and 0.27 m W m^-2 nm^-1 sr^-1 for 2017-2019,respectively;the cross-check results of the spatial extended SIF product and the TROPOMI satellite SIF show high consistency(R²=0.81,RMSE=0.21 m W m^-2 nm^-1 sr^-1 and Bias=0.18 m W m^-2 nm^-1 sr^-1);the spatial extended SIF products are in good agreement with the tower-based SIF observations(R²=0.57,RMSE=0.16 m W m^-2 nm^-1 sr^-1,under clear-sky;R²=0.50,RMSE=0.08 m W m^-2 nm^-1 sr^-1,under all-sky).（3）A GOME-2 satellite SIF spatial downscaling model is established to downscale the original 0.5°coarse spatial resolution SIF to 0.05°spatial resolution,using the fine resolution weighting coefficients simulated by the random forest model,and a spatial downscaled product with spatio-temporal continuity and a spatial resolution of 0.05°is developed.Model validation results show that the method preserves the original SIF information as much as possible while enhancing spatial details(R²≥0.74,RMSE≤0.10 m W m^-2 nm^-1 sr^-1,for 2010);cross-validation results with the TROPOMI satellite SIF show that the spatial downscaled GOME-2 satellite SIF is in high agreement with it(R²=0.78,RMSE=0.12 m W m^-2 nm^-1 sr^-1,Bias=0.09 m W m^-2 nm^-1 sr^-1and fit line was close to the 1:1 line);the tower-based validation results show that the spatial downscaled GOME-2 satellite SIF products have a good accuracy(R²=0.72,RMSE=0.19 m W m^-2 nm^-1 sr^-1,under clear-sky;R²=0.64,RMSE=0.08 m W m^-2 nm^-1 sr^-1,under all-sky).（4）Using the flux observations from Ameri Flux,FLUXNET2015 and Chinaspec,the performance of GOME-2 satellite SIF estimation of GPP before and after spatial downscaling was compared and analyzed,and it is found that spatial downscaling helped to improve the ability of GOME-2 satellite SIF estimation of GPP,and the R² of SIF estimation of GPP model improving from 0.44 to 0.57 and Bias decreasing from0.40 to 0.34 mol C m^-2 d^-1.The main innovative contributions of this paper include:（1）Based on the SIF light use efficiency model to analyze the key driving variables of the SIF simulation model,a spatial extension model of Tan Sat satellite SIF based on random forest algorithm is proposed to solve the problem of spatial discontinuity of TanSat satellite SIF,and the first dataset of domestic Tan Sat satellite SIF products with spatio-temporal continuity is developed.（2）To solve the problem that the existing spatial extension products of GOME-2satellite SIF cannot well retain the SIF structural and physiological information,a new GOME-2 satellite SIF spatial downscaling model is proposed,which uses the predicted SIF dataset obtained by machine learning methods as weighting coefficients to redistribute the original GOME-2 satellite SIF,and achieves spatial smoothing while retaining the original SIF signals as much as possible.（3）To solve the challenge of direct validation of satellite SIF products,the first direct verifications of Tan Sat satellite SIF spatial extension product and GOME-2satellite SIF spatial downscaled product using tower-based SIF observations of different ecosystem are conducted,and the quantitative accuracy verification results of the products are obtained.This paper provides the theoretical basis,experimental analysis,modeling method support,and shared products for the spatial extension of SIF.The achievements are important for the estimation of gross primary production of terrestrial ecosystems and the accurate simulation of global carbon cycle.