Applicability Evaluation Of GPP Model Based On Remote Sensing Data And Flux Observation Dat

To address the increasingly significant issue of global warming,various countries have proposed measures to reduce greenhouse gas emissions such as CO₂ while increasing the ability of ecosystems to absorb and fix CO₂.The Gross Primary Productivity（GPP）of vegetation is an important component of the carbon cycle in ecosystems,so quantitatively describing carbon flux in ecosystems and accurately calculating GPP are inseparable.Researchers from various countries have proposed various methods for estimating GPP for decades.Flux towers provide relatively accurate site-specific GPP data,while satellite remote sensing technology can provide global observation capabilities for GPP.How to estimate ecosystem GPP more accurately has become a current research focus.This study calculated the GPP of 134 flux sites in the Northern Hemisphere using MODIS data,Solar-induced Chlorophyll Fluorescence（SIF）data,and flux observation data,using the TG,GR,VI,AVM,and SCM models.Firstly,the vegetation index was filtered and reconstructed to reduce the impact of data quality on the model.Secondly,the accuracy and applicability of different models were evaluated based on GPP model accuracy and applicability evaluation indicators at different time scales（8-day,growing season,and interannual）in different ecosystems.Finally,the various factors that affect the accuracy and applicability of GPP models were analyzed,including the correlation analysis of environmental factors and the quantitative analysis of the effects of vegetation index and light conversion coefficient on the model.The main conclusions of the study include the following aspects:（1）The calculation results of EVI were described and analyzed,and the data smoothed by S-G method,Kalman filter and low-pass filter based on wavelet transform were compared.The data after Kalman filter was selected.Secondly,the light energy conversion coefficient（m）of each model in different ecosystems was calculated,and the parameters of vegetation index scalar and temperature scalar in different ecosystems were determined by AVM model.（2）On an 8-day time scale,the estimated GPP of each model has a correlation coefficient R²value between 0.58-0.74 with the measured GPP,and the simulation results are good.All models have small estimation bias between estimated GPP and measured GPP.Generally speaking,TG model has the highest estimation accuracy,followed by VI and GR models,while ACM and SCM models have low estimation accuracy.On a quarterly time scale,the simulation effect of GPP cumulative value of vegetation ecosystem estimated by model on site measured GPP quarterly cumulative value is biased high or low in estimating accuracy of GPP cumulative value for four quarters of each year.Among them,the fitting degree of GPP in the first and fourth quarters is low,while that in the second quarter is high and that in the third quarter is second.On an interannual time scale,different models estimate GPP annual cumulative values with large root mean square errors compared with flux tower observations of GPP,and their fitting performance is general.However,these models perform well in areas where annual point density of GPP station is high to some extent,indicating that these models can estimate it well within a certain range of GPP.（3）Different ecosystems have different sensitivities to different input data for estimating GPP.Among them Enhanced Vegetation Index（EVI）,SIF and Land Surface Temperature（LST）have great influence on estimating GPP in most ecosystems.At the same time clear sky index and PAR also have some influence on estimating GPP.The light energy conversion coefficient inverted by flux observation during vegetation growth period is generally high overall which makes simulation results larger than different ecosystems by various models.And using 70%site measured data to invert light energy conversion coefficient can significantly improve accuracy of five GPP estimation models in this study.There is little difference before and after filtering in TG,VI and GR models,only using filtered data estimation bias slightly smaller than evergreen broad-leaved forest,deciduous broad-leaved forest and evergreen coniferous forest three ecosystems.In AVM model,filtered data estimation result obviously better.