Study On Estimation And Uncertainty Of Terrestrial Ecosystem Productivity Based On RS And GIS

Terrestrial ecosystem carbon cycle is one of the core contents of global change research, while terrestrial ecosystem productivity is the important part of carbon cycle. The special variation characteristics of climate in Tibetan plateau play an important role in regional ecosystem carbon cycle. In this paper, we take the grassland transect in Tibetan plateau as the study region, and estimate the productivity of terrestrial ecosystem - ecosystem gross primary productivity (GPP) in different temporal and spatial scale and make a uncertainty analysis of the result through using remote sensing(RS), geographic information system (GIS)technology and ecological model. With the estimation of terrestrial ecosystem productivity in different time and estimation of spatio-temporal distribution patterns and the uncertainty evaluation of results as the research cores, the whole research is based on the eddy covariance flux data in stations of ChinaFLUX, and combines with the information of vegetation growth from remote sensing (RS)data, meteorology interpolationdata and the cover status information of surface. The following conclusions have been made in this paper.(1) Based on CO₂ flux data of eddy covariance, variation characteristics of gross primary productivity (GPP) in four flux observation stations were studied, which are an alpine meadow (HBBT), an alpine shrub meadow (HBGC), a swamp alpine meadow (HBSD) and a steppe alpine meadow at Dongxiong (DX). The results show that photosynthetic capacity of the alpine meadow is the highest, and the annual total GPP is 652.2g C/m². Daily-differencing approach is used to analyze the random error of CO₂ fluxes measurements. The results show that the distribution of random error follows more closely follows a double-exponential (Laplace), rather than a normal (Gaussian) distribution, capturing the high peak and thick tail, and the random error varies with environment variables, which violates the assumptions for the ordinary least squares fitting with normality and homoscedasticity, consequently, we introduce maximum likelihood method (MLE) for parameter optimization. Meanwhile, Monte Carlo (Bootstrapping) uncertainty method is used to assess the estimations of model parameters and GPP estimations. The relative uncertainty (RU) of the max photosynthetic efficiency (α) in HBGC, HBSD, HBBT and DX are 8.28%, 13.50%, 6.65% and 21.62% respectively. The relative uncertainty(RU) of GPP estimations in HBGC,HBSD,HBBT and DX are2.78%, 4.74%, 3.17% and 6.29%, respectively.(2) Spatial database of the grassland transect in Tibetan Plateau are constructed , containing the 8-day enhance vegetation indices(EVI) and 8-day land surface water indices(LSWI), which are obtained through computing surface reflectance of MODIS, and Harmonic analysis of time series (HANTS) is used to processing the abnormal data of vegetation. 8-day mean temperature and 8-day solar radiation are obtained through using spatial different interpolation model (AUSPLIN) to interpolate the meteorological data in stations of China state meteorological bureau.(3) The regional ecosystem GPP is simulated by VPM and spatial analysis method of ArcGIS, based on CO₂ flux data of eddy covariance to obtaining the key parameters of VPM, linked to vegetation indices data and spatial meteorology data. The results show that 2004 annual total of GPP reached to 52.38 T g C (1Tg=10¹²g) . When compared to the 8-day observation GPP and the GPP product of MODIS, the predicted GPP values agree reasonably well with observed values and the prediction accuracy is higher than the GPP product of MODIS,which proving that there is a development potential for VPM in the regional estimation of GPP.(4) The sensitivity analysis is used to determine the influence of input parameters and variables to the output results, the results depict that the optimal photosynthesis temperature (T_opt) is the most important influence. Error propagation model is used to evaluate the GPP estimation of grassland transect in Tibetan Plateau, based on the uncertainties of input variables and the result of sensitivity analysis. The results show that the GPP estimation uncertainties of most pixels are 25%～35%, the steppe alpine meadow ecosystem had the highest uncertainty, about 42.88%.