中文摘要：

如何获取CO2浓度时空分布特征,是气候变化研究中的一个关键问题。本文基于中国碳卫星在吉林航飞试验区的地面观测数据,分析地表CO2浓度与环境变量的相关关系,运用多元线性回归与HASM高精度曲面建模相结合的方法,模拟航飞区地表CO2浓度分布格局。结果表明：CO2浓度空间分布受气象条件的影响较大,短波辐射是影响CO2浓度的重要因素;第1时段整体浓度最高,特别是在西部区域;第2时段CO2浓度高值区东移,呈现西低东高的分布特点;第3时段浓度空间分布与第2时段有类似的特征,但细节存在差异,且高值区缩小;精度对比显示在采样点较少及采样密度不大的情况下,HASM方法的模拟误差小于Kriging方法。因此,这种使用多元线性回归模型通过引入环境变量获得高分辨率趋势面,结合HASM模型进行修正残差提高模拟结果精度的手段,可作为模拟地表CO2浓度时空分布的有效方法。

英文摘要：

As an important cause of global warming, carbon dioxide concentration and its change has aroused worldwide concern. How to have an explicit understanding of the spatial and temporal distribution of carbon dioxide concentration is a crucial technical challenge for climate change research. In this paper, based on the in situ observation data set collected in the TanSat flight test area, the correlations between the carbon dioxide concentrations and the environmental variables are analyzed, and suitable environment variables can be selected to establish a regression equation, through which we obtain a preliminary trend of surface carbon dioxide concentrations. Then combining the multiple linear regression model and High Accuracy Surface Modelling （HASM）, the carbon dioxide concentrations with a high accuracy in the entire test area are produced. The results indicate that the spatial distributions of the carbon dioxide concentrations in the study area are significantly different between three periods, and the short-wave radiation is an important factor for the regression equation. Because of the high temperature and drought condition, the highest concentration appears in the first period especially in the western area. The second period has a different distribution on the carbon dioxide concentration comparing with the previous period, as in this period the high value region moves eastward, and making the concentration high in the eastern area but low in the western area. Both of the second and third periods have similar characteristics except that the high value region in the eastern area is reduced in third period. Moreover, statistical analyses show that the mean absolute error and the mean relative error of the predicted value of the HASM model are 9.8 ppm and 2.48% respectively, which are both lower than the errors produced using the Kriging method, therefore the HASM model remains to have higher simulation accuracy in a condition of few sampling points and low sampling density. Therefore a combined method o