Influence Of Land Cover Data On Regional Carbon And Water Fluxes Simulation

Carbon and water cycles are the bridge of cycle of the geosphere, biosphere and atmosphere, and also are the coupled ecological processes in terrestrial ecosystems. Land cover is an indispensable parameter of carbon and water cycle models about terrestrial ecosystems. The capacities of exchanging energy, momentum and CO2between land surface and atmosphere are markedly different with the land cover types, so the accuracy of the land cover data determines the reliability of the simulation results of carbon and water models at a large extent. At present, there are five global land cover datasets (UMD, USGS, GLOBCOVER, GLC2000and MODIS global land cover dataset) to support the global change research and the extracting of other land surface parameters (such as leaf area index (LAI) and clumping index, etc.) based on remote sensing. Because of the differences in classification systems and methods, and data used to produce these datasets, these five global land cover datasets exhibit certain inconsistencies.To evaluate the applicability of these five land cover datasets in simulating the ecological and hydrological processes in typical regions, these five land cover datasets were employed in conjunction with the MODIS reflectance, meteorology data, and soil data to inverse carbon and water fluxes (GPP (Gross Primary Productivity), NPP(Net Primary Productivity), and ET (evapotranspiration)),during2001-2010, using the remote sensing-based BEPS ecological model, in the Poyang Lake Basin, Jiangxi Province, China. The simulated GPP, NPP, and ET results were validated on site and watershed scales, to assess the impact of land cover on the inversion of carbon and water fluxes quantitatively in the regional ecosystems. The spatial and temporal distributions of GPP, NPP, ET and soil water were analyzed and their sensitivities to the model input data and parameters were investigated at the end of the paper. This paper is under the support of the "863" key project issue "Application and demonstration of land cover data in the surface process simulation model". From this research following findings and conclusions were drawn:(1) Inconsistencies exhibiting in these five land cover datasetsConversions were done for USGS, UMD, GLC2000, and GLOBCOVER land cover datasets to make them compatible with the IGBP/MODIS classification and coding systems. There were significant differences in statistical characteristics and spatial distributions among these five land cover datasets. The total proportions of five plantations in GLC2000, GLOBCOVER, and MODIS were57.25%,52.16%and43.93%, respectively, and were more consistent with the sixth forest survey result, which was55.86%, than USGS and UMD land cover datasets. The proportion of farmland in USGS land cover dataset, achieving38.15%, was much larger than the second land survey result. Grassland accounted for the largest proportion in UMD land cover dataset, and its proportion was46.69%. These five land cover datasets showed a similar spatial distribution, but were rather different in the south.(2) The applicability of different land cover datasets used to simulate carbon and water fluxes on sites and watershed scalesThe simulated GPP, NPP and ET had the same seasonal variations with the observed data at Qianyanzhou station. daily GPP and daily ET inversed by GLC2000and MODIS land cover datasets had the highest consistency with the observed data, approaching the precision of71.28%and71.92%, respectively. It indicated that the BEPS model perfermaced well on simulating seasonal variations of GPP, NPP and ET, and responsed timely to abnormal climatic conditions.Consistencies between the ET simulated using these five land cover datasets and the ET calculated by precipitation and runoff data were different in different watersheds.The cropland NPP inversed by GLOBCOVER and MODIS land cover datasets had higher spacial consistency than cropland NPP inversed by UMD, GLC2000and USGS land cover datasets on the county scales. The R2(N=44) were0.46,0.43,0.32,0.22and0.18respectively.(3) Temporal and spatial distribution of carbon and water fluxes simulated by these five different land cover datasetsGPP, NPP, ET and soil water inversed by these five land cover datasets showed roughly the same spatial variation. From north to south, the daily average GPP, NPP and ET significantly increased and soil water decreased. But there were large differences among the statistical characteristics of GPP, NPP and ET simulated by five land cover datasets. Annual GPP and ET simulated by MODIS land cover dataset averaged during2001and2010were the highest among these five simulated results, and they were1090.67g C m-2a-1and566.25mm, respectively, while annual GPP and ET simulated by UMD land cover dataset were the lowest and only were860.08,412.53g C m-2a-1and515.61mm. Annual NPP simulated by GLC2000land cover dataset were the highest and it was590.91g C m-2a-1, while annual NPP simulated by UMD land cover dataset was the lowest and only was412.53g C m-2a-1(4)The relationship between the precipitation/radiation and the simulated GPP, NPP, and ETA significant negative correlation was observed between the annual GPP, NPP and ET, and the annual precipitation, as well as the soil water and the annual solar radiation. However, there was a significant positive correlation between the annual GPP, NPP and ET, and the annual solar radiation, and also between the soil water and the precipitation.