静态浮箱法中不同拟合模型对水-气界面通量估算的影响

水域生态系统是温室气体的重要来源，静态浮箱法（SFC）是目前广泛采用估算水-气界面气体通量的方法之一，然而在利用线性回归（LR）拟合模型估算水-气界面气体通量时存在低估问题。采用密闭式静态浮箱-温室气体分析仪系统，观测了在三峡库区香溪河库湾水-气界面之上通量箱内甲烷和二氧化碳气体浓度的变化，对监测得到的数据进行拟合分析。对比分析了线性回归（LR）、二次回归（QR）和指数回归（ER）三种模型的拟合效果，结果表明非线性回归模型可以更好地拟合静态浮箱内气体浓度随时间的变化情况，从而得到更接近真实值的水-气界面气体通量。认为采用线性回归（LR）模型拟合水-气界面气体通量得到的结果在很大程度上被低估了，该方法得到甲烷和二氧化碳气体通量估算值平均约占实测值的77%和80%。此外，大多数情况下指数拟合模型的效果要优于二次回归拟合模型，而二次回归模型拟合系数没有物理意义，从理论上不适用于气体通量的估算。 关键词： 静态浮箱法；气体通量；线性拟合；非线性拟合

#br# Influence of Different Fitting Models on Gas Fluxes Estimation AcrossWater air Interfaces with Static Floating Chamber#br#

Aquatic ecosystem is the most important sources of greenhouse gases. However, the uncertainty in estimating the gas flux of water gas interface has not yet been paid attention to by using the linear regression (LR) model in the static floating box method (SFC).This study in Xiangxi Bay of Three Gorges reservoir, using closed static floating box greenhouse gas analyzer continuous and real time observation of the gas water interface of methane and carbon dioxide, and through software for fitting the monitoring data fitting. The comparison and analysis of the linear regression (LR), two regression (QR) and exponential regression (ER) fitting results of three kinds of models, the results show that it can better fit the static floating box gas concentration change with time nonlinear regression model, to get closer to the real gas water interface gas flux. This paper argues that the use of linear regression (LR) model to simulate the air interface of the gas flux of the obtained results are largely underestimated, the method of methane and carbon dioxide gas flux estimates averaged about 77% and 80% of the measured value. Moreover, in most cases, the effect of the exponential fit model is better than that of the two fit model, while the two regression model has no physical meaning in terms of the fitting coefficients, and is not applicable to the estimation of gas flux theoretically. Key words：Static floating chamber; Gas flux estimation; Fitting model；Nonlinear fitting