Light Use Efficiency Based Gross Primary Productivity Estimation And Uncertainty Analysis

Terrestrial ecosystems carbon recycle is an essential component of global carbon recycles.Gross primary production(GPP)is one of the major processes controlling land-atmosphere CO2 exchange and plays a significant role in global carbon balance.Continuous monitoring and accuracy estimation of the spatio-temporal variations of GPP is of great significance.However,uncertainty remains regarding the spatial distribution and temporal dynamics of GPP.It is of great value to quantify model uncertainties and improve the accuracy of model by making full use of multi-source and multi-scale datasets and combining the satellite remote sensing data and ground measured data in GPP modeling.In this thesis,a systematic study of GPP light use efficiency model uncertainties was conducted in a highly heterogeneous area across the north China agro-pastoral ecotone and U.S.Great Plains distributed with abundant C3 and C4 grasslands.This thesis researched the uncertain effects of model structures,model parameters,and model inputs(e.g.,data sources and scales)on GPP light use efficiency model uncertainty analysis.The primary conclusions and contributions of this paper are listed as follows:(1)This study first evaluated the influence of model structures on site level GPP modeling accuracy based on VPM model across the north China agropastoral ecotone,a highly heterogeneous area with eddy flux towers densely instrumented.Then developed optimized GPP modeling for different vegetation types.Different model structures(e.g.,FPAR,water scalar,and temperature scalar)impact the model performance differently for grasslands,croplands,and forests,so optimal structures should be selected for each vegetation types.Relatively optimized models were chosen for spatial and regional GPP estimation and comparison.For forests,the original VPM model was used(R2=0.85).For grasslands and croplands,the water scalar(1+LSWI)/(1+LSWImax)in VPM was replaced with 0.5+LSWI,making an improvement of R2 from 0.67 to 0.78 and 0.71 to 0.76,respectively.(2)Sobol's sensitivity analysis method was used to quantify the relative contribution of each model input(PAR,Temperature,EVI,and LSWI)to GPP estimation uncertainties for optimized models.And the uncertain effects of spatial resolutions,landcover maps,and meteorology datasets on regional GPP simulations at magnitude,spatial,and temporal patterns were investigated.From Sobol's sensitivity analysis,the GPP models were most sensitivity to PAR with the highest first order index and total order index,followed by EVI and temperature,and LSWI contributed least to model uncertainties.On the magnitude and spatial patterns,altering landcover types contributed most to GPP regional uncertainties,followed by spatial resolution and meteorology datasets.While on the temporal pattern,the effect of spatial resolution and meteorology were larger.It was concluded that landcover datasets were the most influential data sources.Errors would be introduced from the uncertainties in land cover classification to the regional GPP estimation through maximal light use efficiency.The scale effects of different spatial resolutions on GPP uncertainties were not neglectable.Changing the meteorology had only little effect on regional GPP estimation.Although PAR had the highest sensitivity index in GPP modeling,but the spatial GPP uncertainties were not obvious in study area for the high precision and small value variation range of PAR.(3)Based on three light use efficiency models(MODIS/EC-LUE/VPM),this study compared model accuracy and the difference of max light use efficiency between C3 and C4 grasslands across the U.S.Great Plains.The model accuracies at sites level from high to low were EC-LUE,VPM,and MODIS.The max light use efficiency for C3 and C4 grasslands acted differently for MODIS(C3,1.07 g C MJ-1;C4,1.26 g C MJ-1),EC-LUE(C3,1.10 g C MJ-1;C4,1.45 g C MJ-1),VPM(C3,1.24 g C MJ-1;C4,1.48 g C MJ-1).Therefore,C4 grasslands show a higher photosynthesis capability compared with C3 grasslands.Moreover,the difference of LUEmax between C3 and C4 grasslands for EC-LUE was larger than other models,indicating EC-LUE model was more sensitive to distinguish the photosynthesis capacity of C3 and C4 grasslands.(4)Regional GPP was estimated based on MODIS,EC-LUE,and VPM for different grassland plant function types,and the uncertainties introduced by adopting unique parameter for C3/C4 grasslands were quantified.Six grassland towers from WorldGrass Agriflux dataset were used to verify the performance of spatial GPP.The model precisions from high to low were VPM,MODIS,and EC-LUE.The accuracies of the models significantly improved by using separated parameters for C3/C4 grasslands.From the spatial distribution,using unique parameter for C3/C4 grasslands would overestimate or underestimate GPP.The overestimate region was located in the north and northeast C3 grasslands,and the southwest with C3-C4 grasslands rotation exhibited no obvious overestimation or underestimation.Overall,using unique parameter for C3 and C4 grasslands would overestimate the regional mean and sum grassland GPP in the U.S.Great Plains.This thesis synthetically analyzed the uncertain sources of light use efficiency based GPP simulations and their uncertain effects on GPP estimation.It improved our understanding of light use efficiency model structures and parameters,and the influence of data sources and scales on GPP modeling.This study will contributed to accurate estimation of GPP.