Reconstruction Of Satellite SIF With High Spatiotemporal Resolution And Its Potential For Photosynthetic Productivity Monitoring

Solar-Induced Chlorophyll Fluorescence(SIF)is a long-wave spectral signal in the range of 650-850 nm emitted by plant chlorophyll molecules under natural light conditions,which is a pathway for energy release after plants absorb photosynthetically active radiation,excluding photosynthesis and heat dissipation.As a companion product of photosynthesis,SIF can reveal the real state of plant photosynthetic physiology better than traditional remote sensing reflectance index,and is considered as an indicator of vegetation photosynthetic activity.With the successful retrieval of SIF from satellite sensors,satellite SIF provides a new method for directly detecting gross primary production(GPP)of terrestrial vegetation.However,satellite SIF products suffer from spatiotemporal discontinuities and low spatial resolution,which limit it to monitor global and regional GPP at finer scales.Therefore,in order to enhance the continuity and resolution of satellite SIF products,the development of high spatiotemporal resolution reconstruction methods for satellite SIF is of great significance for monitoring ecosystem carbon cycle at fine scale.Accordingly,this thesis focuses on the limitations of satellite SIF products in spatiotemporal continuity and spatial resolution,and develops the model of high spatiotemporal resolution reconstruction of satellite SIF to achieve the scientific goal of upgrading satellite SIF products from "kilometer-level,spatial and temporal discontinuity" to "one hundred-meter,daily seamless",and improves the application potential of SIF satellite products for monitoring GPP at fine scale.Taking the TROPOMI SIF products with 2600 km width and 7.5 km×3.5 km~14 km(to 5.5 km×3.5 km~14 km after 2018)resolution as the research object,we use multi-source satellite,meteorological,and tower observation data to conduct three aspects of research work,including developing the spatiotemporal expansion method,establishing the spatial downscaling method and evaluating the potential of the reconstructed SIF product for monitoring GPP.The main conclusions are as follows:(1)In this thesis,using MODIS reflectance,vegetation index,reanalysis of meteorological data and other spatiotemporal continuous driving variable datasets,based on a random forest model,we evaluate the effects of driving variable combinations and spatiotemporal constraints on model prediction accuracy,establish a spatiotemporal extension model with spatiotemporal window-constraints,develop global daily seamless SIF product at0.05° resolution from 2018 to 2020 product(SDSIF),and validate reconstructed SIF product using original TROPOMI SIF and tower-based SIF.Three cross-validation averaged model prediction results for the 2019 test sample show that the test accuracy of the spatiotemporal extended model with subcontinent and month constraints(R2 = 0.928,RMSE = 0.0596 m W/m2/nm/sr)is better than the accuracy of the general model(R2 = 0.913,RMSE = 0.0653 m W/m2/nm/sr).The validation results based on the original TROPOMI SIF show that the spatial gap of the original TROPOMI SIF is filled by the reconstructed SIF product,with an average R2 of 0.75 for the temporal correlation between them for 80% global regions and a residual between them within ± 0.081 m W/m2/nm/sr.The validation results based on tower-based SIF show that the reconstructed SIF products fill the temporal gap and have a better agreement with the in-situ SIF than the original TROPOMI SIF(the average R2 was improved from 0.467 to 0.744 at five sites).(2)In this thesis,we developed a SIF product at five hundred-meter high resolution(BFSIF)over China by establishing a spatial downscaling model correcting for spatial cross-scale bias using high spatial resolution multi-source driving variable datasets,based on random forest algorithm and bias correction.And we validated SIF product by using TROPOMI SIF and tower-based SIF data.The results show that the total testing model accuracy is high(R2 = 0.886,RMSE = 0.0716 m W/m2/nm/sr)in 2019.The validation results based on the original TROPOMI SIF show that the spatial downscaling SIF products have finer spatial details,correct the bias resulting from predicting SIF product at fine reolution by using the model at coarse resolution,and have higher linear correlation with the original TROPOMI SIF(R2 improves from 0.873 to 0.898).The validation results based on the tower-based SIF show that the spatial downscaling SIF products have better linear correlation with the in-situ SIF compared to the directly predicted SIF by using the downscaling model(R2 is improved from 0.814 to 0.948 at the Daman site).(3)In this thesis,we compared the potential of reconstructed SIF products and the original TROPOMI SIF for monitoring the photosynthetic productivity by using auxiliary data such as high-spatial resolution satellite GPP and flux measured GPP data,and assessed the advantages of the reconstructed SIF products for monitoring photosynthetic productivity.The spatiotemporal extension SIF products filled spatial gap of the original TROPOMI SIF and showed stronger linear correlation with the tower-based GPP than the original TROPOMI SIF(R2 is improved from 0.701 to 0.811 at the Daman site).The spatial downscaling SIF products have finer spatial detail and the spatial distribution pattern of it is more consistent with the high-resolution satellite GPP products than vegetation indices such as NIRv.Spatial downscaling SIF product has higher temporal correlation with satellite GPP data than the original TROPOMI SIF products.In this thesis,in order to solve the problems of spatiotemporal discontinuity and insufficient spatial resolution of satellite SIF,we develop the spatiotemporal extension and spatial downscaling methods of TROPOMI SIF,produce high spatiotemporal resolution SIF products,and evaluate the application of satellite SIF for estimating global and regional GPP at fine scale.The results of this study can be used for estimating global and regional scale GPP at fine scale.The research results can provide methods and data support for monitoring photosynthetic carbon of vegetation at global and regional scales.