Assimilation Of MODIS LAI Time Series In Bamboo Forest And Its Application In Carbon Flux Simulation

Subtropical forest ecosystems play an important role in the global carbon cycle and the carbon sequestration functions has been challenging the traditional knowledge about the main functional field of carbon sequestration located in the temperate forests in Europe and America.Bamboo forest is an important component of forest ecosystems,which has a strong carbon sequestration capacity and play an important role in the global carbon cycle.The leaf area index(LAI)is one of the most important physiological parameters of terrestrial vegetation.LAI has close relationship with photosynthesis,transpiration,water use,and the formation of ecosystem productivity,and is always considered as an important parameter and indicator in researches of carbon and water cycling and energy exchange of terrestrial ecosystems.LAI,which is obtained through remotely sensed data and used as input data to drive ecosystem carbon cycle model,to achieve cross-scale simulation of carbon cycle process,and to reflect the spatial distribution and dynamics of carbon budget in regional(even global)scale,has been a hot point and major direction in future in estimation of forest ecosystems carbon cycle.However,due to the impact of cloud cover,aerosol,snow cover and sensor failure,the many existing satellite-based LAI products are characterized with high noise,low accuracy,large fluctuation in time series,which cannot correctly reflect the process of plant growth continuity,and thus constrain the global application of LAI products.Therefore,obtaining high-accuracy LAI time series is very important when researching the carbon cycle characteristics of subtropical forest ecosystems.Based on the extraction of remotely sensed information from bamboo forest,this paper firstly integrated the reflectance data,physical biochemical parameters,MODIS reflectance and LAI products of bamboo forest by data assimilation filtering algorithm,using the PROSAIL model as a bridge between the LAI dynamic model and the MODIS observation reflectivity,and the bamboo forest MODIS LAI data assimilation system was constructed.Then,on the basis of data assimilation system optimization,the MODIS LAI products were assimilated to obtain high accuracy LAI time series products.Finally,the LAI assimilation products of bamboo forest were used to drive the Boreal Ecosystem Productivity Simulator(BEPS),and the carbon fluxes of bamboo forest ecosystem in Zhejiang Province were simulated.The main study contents are listed as follows:1.Mapping of Bamboo information based on multi-source remote sensing data.Based on the data of MODIS and Landsat,combined with the ground survey data,the maximum likelihood method was used to classify four land cover types of woodland,urban,water and farmland types from Zhejiang MODIS data.Then,the abundance of bamboo forest was extracted by mixed pixel decomposition,and the spatial distribution of bamboo forest in Zhejiang Province was delineated.2.Study on MODIS LAI assimilation method of bamboo forest.Firstly,the parameters of the PROSAIL model were optimized based on the measured reflectance,and the canopy reflectivity was simulated.Secondly,the PROSAIL model,coupled with Dual Ensemble Kalman filter and particle filter methods,were used to establish the LAI assimilation system of bamboo forest,and the MODIS LAI time series data was assimilated from 2014 to 2015 in two flux sites,respectively.Finally,comparing these two assimilation methods,the better assimilation method was selected to assimilate LAI of the bamboo forests in Zhejiang Province.3.Study on the application of LAI assimilation produce in the simulation of carbon flux model of bamboo forest ecosystem.The BEPS model was driven by LAI assimilation,and the carbon flux was simulated from 2011 to 2014.Compared with observed carbon fluxes,the better assimilated LAI was selected to assimilate the temporal and spatial carbon fluxes of the bamboo forest ecosystems in Zhejiang Province.There are several conclusions from this study as follows:1.Based on multi-source remote sensing data,the extraction accuracy of distribution information of bamboo forest was high,which provided an important basic data for the LAI assimilation of bamboo forest and the simulation of carbon flux in bamboo forest ecosystem.The classification total accuracy of Zhejiang Province forests in 2012 and 2014 were 93.18% and 92.01%,respectively,and the Kappa coefficient were 90.50% and 88.83%,respectively,which laid the foundation for further extraction of bamboo forest information form forest land.The coefficient of determination(R~2)between remote sensing estimated area of Zhejiang province in 2012 and the forest inventory area reached 0.81,the root mean square error(RMSE)was 7016 hm2,and the absolute bias(aBIAS)was 5605 hm2.The R~2 between remote sensing estimated area of Zhejiang province in 2014 and the forest inventory area reached 0.81,RMSE was 7727 hm2,and aBIAS was 5242.4 hm2.2.The optimized PROSAIL model can simulate the canopy reflectance of bamboo forest with high precision.the leaf reflectance modeled by PROSPECT 5 was correlated significantly with the observed leaf reflectance,with R~2 greater than 0.99 and RMSE of 2.7%-3.2%.This indicated that the model was well optimized and that the modeled leaf reflectance could be used for simulating canopy reflectance using the 4SAIL model.R~2 of the simulated canopy reflectance against the observed data higher than 0.99 and low RMSEs 1.9%-2.5% also indicated high accuracy of the canopy reflectance modeled by the 4SAIL model.Therefore,the LAI could be assimilated by coupling the modeled canopy reflectance and LAI dynamic model with the data assimilation algorithm.3.Dual Ensemble Kalman filter and particle filter can greatly improve the accuracy of MODIS LAI products,but the particle filter method has higher precision of LAI products.Compared with observed LAI,the relationship between non-assimilated LAI and observed LAI was much smaller,with R~2 lower than 0.25,RMSE and a BIAS were greater than 1.00.The relationship between LAI calculated using data assimilation and observed LAI was very significant,with R~2 greater than 0.85.RMSE and aBIAS were inferior than 0.42.LAI assimilated using the Particle filter showed the following with respect to observed LAI: R~2 increased by 9.2%,RMSE reduced by 33.3%,and a BIAS reduced by 36.1%.LAI was assimilated for a bamboo forest in Zhejiang province using Particle filter,and the assimilated results were consistent with the overall temporal and spatial distribution of the bamboo forest,assimilated LAI for the four seasons is highest in summer,the autumn is the second,the winter is the lowest of the four seasons.4.Based on the BEPS model driven by assimilated LAI products using the particle filter,carbon fluxes of bamboo forest ecosystem can be achieved with high simulation accuracy,and carbon fluxes of bamboo forest ecosystem were simulated by using this method from 2011 to 2014 in Zhejiang Province.Compared with the results of the carbon cycle simulated by non-assimilated LAI driving BEPS model,relationship between simulated carbon fluxes by using particle filter algorithm and observed carbon fluxes data were very significant,with the R~2 increased by 6.4%-45.8%,RMSE reduced by 22.1%-29.9%,and a BIAS reduced by 25.3%-33.8%.Compared with simulated carbon fluxes by Dual ensemble Kalman filter assimilated LAI driving BEPS model,relationship between simulated carbon fluxes by the particle filter and observed carbon fluxes data were almost the same,but RMSE reduced by 1.75%-4.03%,and aBIAS reduced by 1.59%-4.30%.