Reconstruction Of Solar Induced Chlorophyll Fluorescence In Forest And Grassland At Regional Scale Based On Multisource Remote Sensing Data

Forest and grassland are two major carbon sink ecosystems in terrestrial ecosystems,and their photosynthetic intensity and carbon sink capacity can be evaluated by detecting their solar-induced chlorophyll fluorescence(SIF).Since the SIF retrieved directly from satellite observations has the problems of low spatial resolution,discontinuity or low temporal resolution,some vegetation indices and meteorological factors are used as predictors to reconstruct SIF products.However,vegetation index and meteorological factors have different contributions to photosynthetic intensity at different life stages of vegetation.In addition,unlike vegetation indices,some meteorological factors have relatively low spatial resolution,and their observations are not always available.The purpose of this study is to use fewer and easier-to-obtain predictors to reconstruct SIFs with a spatial resolution of 500 m in forest and grassland areas considering growth stages.The main research contents and conclusions are as follows:(1)Selection of predictor combinations.Six forest and grassland regions in low,middle and high latitudes were selected,and the correlation between SIF and normalized difference vegetation index(NDVI),enhanced vegetation index(EVI)and land surface temperature(LST)was compared.The results show that EVI and LST have a stronger correlation with SIF,and their contributions to SIF are different in different life stages of forest and grassland,which provides a basis for reconstructing SIF only using EVI and LST.(2)Applicability of SIF regression models in forests and grasslands,as well as in time and latitude.This thesis studies the correlation between R~2and the relative standard deviation(RSD)of EVI in full latitude forests on a monthly scale.It is found that the greater the dispersion of EVI data,the more advantageous it is to construct a SIF regression model.Therefore,in order to seek the maximum dispersion of EVI data,this study proposes three data combination methods:global scale monthly regression,regional scale seasonal regression,and regional scale monthly regression.The results show that full latitude forests(January-May,October-December)are suitable for monthly regression on a global scale,with R~2ranging from 0.56 to 0.76;The mid latitude forest growth season(March-June)and the high latitude forest growth season(April-July)are suitable for regional scale seasonal regression,with R~2of 0.60 and 0.63 for the two;Low latitude grasslands(June-October),mid latitude grasslands(July-August),and high latitude grasslands(July)are suitable for regional scale monthly regression,with R~2ranging from0.45 to 0.65.Increase the R~2of SIF regression for mid latitude forests in June from 0.34 to0.60,and increase the R~2of SIF regression for high latitude forests in June and July from0.38 to 0.63.The R~2of SIF regression for full latitude forests from October to December and January to April was increased from-0.01 to 0.58 and 0.56 to 0.76 by expanding the spatial span of the dataset.The above results indicate that this thesis has found a more suitable spatiotemporal data combination method for SIF regression models of forests and grasslands in time and latitude.(3)The EVI and LST coefficients in the SIF model were compared interannually.The results showed that the coefficients of SIF model changed similarly in the same month between different years of forest.The similarity of EVI was corr=0.744,and the similarity of LST was corr=0.820.However,in the grassland area,EVI has no regularity in the change of year and month,and the similarity corr=-0.085.LST has a similar trend in the same month of different years,and the similarity corr=0.752.These results are expected to support high-resolution SIF reconstruction in forest and grassland regions based on EVI and LST predictors.(4)Validate the reconstructed SIF using gross primary production(GPP)data.The results show that the R~2values are more than 0.90,which proves the prediction ability of EVI and LST for 500 m spatial resolution SIF,and verifies that the performance of the reconstructed SIF model is highly correlated with the dispersion of EVI.RMSE was used to evaluate the difference between the aggregated SIF and GOME-2 SIF data.The results show that the RMSE of the reconstructed forest fluctuates around 0.100,while that of the grassland is less than 0.100.This shows that our model performs well and outperforms other models using the same predictor.