Reconstruction Of Global Solar-induced Chlorophyll Fluorescence Remote Sensing Data With Canopy Correction

Gross primary productivity(GPP)of terrestrial ecosystems is an important part of the global carbon cycle.It is of great scientific significance to improve the estimation accuracy of GPP.Solar-induced chlorophyll fluorescence(SIF),an electromagnetic radiation emitted during photosynthesis,was an"effective probe"of plant photosynthetic rate,and showed a strong correlation with GPP.At present,there are a variety of global SIF products,but some of the existing SIF products have short time series and some have sparse spatial sampling,which are difficult to meet the application requirements.Therefore,some researchers proposed to reconstruct SIF remote sensing data by using MODIS reflectance and other data,so as to obtain SIF data with relatively high spatial and temporal resolution.Existing major SIF reconstruction data are all based on the original SIF data obtained by sensors.However,due to the influence of canopy interception and scattering,the SIF observed by satellite sensors at the top of the canopy only represents a part of the total excited SIF of all leaves in the canopy,and its relationship with GPP is not as direct as the total SIF of the canopy.At present,there is a lack of reconstructed data set of total canopy SIF.In view of this,this paper firstly conducted canopy correction and soil correction based on the canopy invariant theory to obtain the total canopy SIF observed by OCO-2 and TROPOMI sensors.Then,the use of machine learning algorithms,high space-time resolution of remote sensing and meteorological data,respectively,the two sensors(OCO-2 and TROPOMI)of three kinds of SIF data(sensor observations of the original SIF,the canopy total SIF,soil correction after the canopy total SIF)refactoring,generate six sets of2001 to 2020,the spatial resolution of 0.05°,eight days time resolution of the reconstructed SIF data set.The main conclusions of this paper are as follows:(1)In the training and validation stage of reconstructed SIF,the six sets of reconstructed data sets all have high modeling accuracy,and the R² of the verification set is between 0.74-0.86.The time series of reconstructed SIF at a single point can well reproduce the seasonal variation of SIF before reconstruction,indicating that the reconstruction of SIF in this paper is effective.(2)The relationship between the reconstructed SIF data and GPP was verified by using the vorticity correlation observation data.Two site selection schemes were designed according to vegetation type and spatial heterogeneity,and the correlation between site reconstruction SIF and site GPP at site scale was calculated.The results showed that the SIF and R² of OCO-2 and TROPOMI were both increased after canopy correction,and the increase range was between 0.02and 0.07.After the evaluation by vegetation type,the original SIF of savannas,deciduous broad-leaved forests,coniferous forests and grasslands had a strong correlation with GPP(R²>0.6).After the correction of canopy,the R²was further increased,and the improvement range was between0.02-0.07.The correlation between original SIF and GPP in farmland and evergreen broad-leaved forest was low(R² was the lowest 0.26),and R² increased by 0.02-0.12 after canopy correction.The above results indicate that the reconstructed SIF data set after canopy correction in this paper has a better correlation with GPP than the original SIF obtained by the sensor.(3)On a global scale,the consistency between the six data sets reconstructed in this paper and the FLUXCOM GPP product was compared.On average and maximum spatial distribution model,in this paper,six sets of data set and to reconstruct the FLUXCOM GPP similar:day average SIF in northern South America,central Africa and southeast Asia is higher,the maximum SIF in the corn belt areas of the United States,central Europe and China's northeast and southeast Brazil,and South America and other regions,a higher value in central Africa region.The correlation between the 6sets of reconstructed data sets and the Fluxcom GPP time series was calculated on a raster basis,and the spatial pattern distribution map of R² was drawn,showing that there was a high R² in most regions of the world except tropical rain forests.The seasonal trend of SIF is analyzed by taking SIF_ocoSC as an example,which shows the expected temporal and spatial variation in the global scale.In conclusion,based on SIF observation on the basis of the canopy correction,based on remote sensing data and machine learning method to build a global scale,high space-time resolution canopy SIF data sets,verified and compared with existing does not consider canopy correction of SIF refactoring data set(e.g.,GOSIF),the data set and eddy covariance observation of GPP has better linearity,thus can provide better for terrestrial ecosystem carbon cycle research data to support.