1.Anhui Institute of Optics and Fine Mechanics, Hefei Institutes of Physical Science, Chinese Academy of Sciences, Hefei 230031, China 2.University of Science and Technology of China, Hefei 230026, China 3.School of Electronic Engineering, Huainan Normal University, Huainan 232038, China
Abstract:In this paper, a near-infrared laser heterodyne spectrometer developed by the laboratory is used to investigate the inversion of greenhouse gas column concentration and approximately evaluate the system measurement errors based on the optimal estimation algorithm. Firstly, the spectral database and the calculation results from the reference forward model are compared with the ground-based FTIR results, thereby selecting the detection window, the corresponding laser and detector. Secondly, the optimal estimation concentration inversion algorithm based on the reference forward model is established, and the Levenberg-Marquardt (LM) iterative method is adopted to realize the inversion of the concentration and vertical distribution profile of atmospheric CO2 column in the whole layer, and the long-term observation comparative experiment is carried out to verify the feasibility of this algorithm. Finally, by simulating the selected detection window spectrum in different white noise, the approximate corresponding relationship between the system signal-noise-ratio (SNR) and CO2 column concentration measuring error is eventually obtained. This research is an indispensable theoretical calculation part of the detection system and will conduce to improving the application of laser heterodyne technology in atmospheric observations. Keywords:laser heterodyne/ measuring error/ the reference forward model/ CO2 column concentration
2019年3月12日在中国科学院合肥物质科学研究院(31.9°N, 117.166°E)采集的0.0073 cm–1高光谱分辨率近红外激光外差数据如图3所示, 采集模式为逐点扫描方式. 图 3 激光外差实验结果 (a) CO2的高分辨率的外差信号; (b) 实时跟踪的太阳光信号; (c) 激光器的DC信号 Figure3. Experimental results of the laser heterodyne: (a) The high-resolution heterodyne signal of CO2; (b) the sunlight signal tracked in real time; (c) the DC signal of the laser.
其中, S1为功率校准的激光外差信号, S2为激光外差原始信号, D为偏移量, B为DC信号. 最后, 借助波长计和HITRAN数据库对外差信号进行波长标定, 并与参考正向模型相同的频率间隔进行插值处理. 需特别指出的是, 由于激光电流的工作模式为逐点扫描模式, 利用波长计 (Bristol, 621B) 能够方便地进行实时的波长标定. 但是, 在实际测量过程中, 激光器的工作波长可能会有轻微的变化(如图4(a)所示). 为了准确识别该波长偏移, 如图4(b)所示, 通过移动真实信号, 计算RFM模型计算的参考信号和实际外差信号的相关系数, 其中移动步长可以用插值方法定义为任意小的长度(如0.0001 cm–1). 最后通过相关系数的最大值位置(如图4(b)所示)可以精确地确定波长偏移, 该方法在信噪比较差的情况下依然适用. 最终获得处理后的LHR频谱 (图7(a)中的黑点)以备下一步的数据反演. 特别指出, 对于太阳光功率波动较大的信号(大于平均太阳光功率的10%), 将不再做进一步的处理. 图 4 波数偏移校准原理 (a) 参考信号与实验信号之间的波数偏移; (b) 计算波数偏移的过程, 插图为相关系数的结果 Figure4. Principle of the wavenumber shift calibration: (a) Wavenumber shift between the reference signal and real signal; (b) the process of calculating the wavenumber shift. The inset shows the result of correlation coefficients.
图 7 LHR数据反演结果 (a) 实验和拟合LHR谱图以及迭代过程的收敛性 (插图); (b) 残差; (c) 和 (d)分别获取的CO2的先验和反演的垂直浓度分布图 Figure7. Inversion results of LHR data: (a) Experimental and fitted LHR spectrogram and convergence of iterative process (illustration); (b) residue; (c) and (d) obtained prior and inversion vertical concentration profiles of CO2, respectively.
其中n(h) 为高度h处的分子数密度, 整个大气层78 km, x(h) 为CO2在高度h处的体积百分比. CO2的柱浓度为8.2184 × 1021 molecule/cm2. 图8(a)显示了2019年3月14日的连续测量结果, $X_{\rm CO_{2}}$(表示为CO2的柱浓度)的标准偏差约为0.47 ppm. 本研究仅对上午11:00至下午13:00的测量结果进行了分析, 这是因为美国气象环境预报中心NCEP和中国气象数据服务中心CMCC提供的时间点为12:00, 且认为该段时间内大气参数是稳定的. 图 8 (a) 2019年3月14日从11:00至13:00的$X_{{\rm{CO}}_2} $的反演结果; (b) $X_{{\rm{CO}}_2} $的外差结果和GOSAT结果的时间序列 Figure8. (a) Retrieval results of the $X_{{\rm{CO}}_2} $ versus time from 11:00 to 13:00 on March 14, 2019; (b) time series of the LHR results and GOSAT results of $X_{{\rm{CO}}_2}$.
3.系统测量误差近似评估模型为了近似评估系统测量误差, 通过模拟正向计算谱叠加不同幅度白噪声, 得到SNR, 3σ, Max (平均值加上3σ)、Min (平均值减去3σ) 与白噪声幅度之间的关系; 通过改变CO2的廓线比例, 透过率谱最低点变化引起柱浓度变化; 以此Max, Min进行柱浓度插值, 从而得到了SNR和柱总量的差值(测量误差)间的近似对应关系. 首先, 对正向计算谱叠加不同幅值的白噪声以模拟不同的信噪比, 如图9(a), RFM模型加载0.05幅度的白噪声; 图9(b)为加载5000次幅值为0.05白噪声, 由波段吸收峰波谷位置的变化情况, 获得信噪比计算公式 图 9 (a) 叠加0.05幅度的白噪声后的透过率谱; (b) 叠加5000次后透过率谱最低点的变化 Figure9. (a) The transmittance spectrum after adding white noise with the amplitude of 0.05; (b) change range of the lowest point of transmittance spectrum after adding 5000 times.
${\rm{SNR}} = ({{T' - T''}})/{\sigma },$
其中$T'$和$T''$分别代表图9(a)透过率谱吸收峰的最大值和最小值, $\sigma $代表图9(b)的5000次数据的标准偏差. 模拟白噪声从幅度0.001到幅度0.050, 步长为0.001, 获得了SNR, $3 \sigma $, Max (平均值加上$ 3\sigma $), Min (平均值减去$3 \sigma $) 的变化, 如图10所示. 从结果分析得到$3 \sigma $ 和Max随着白噪声的幅度增加而增加, 而Min和SNR随着白噪声的幅度减少而减少. 图 10 白噪声幅度0.001变化到幅度0.050引起的$3 \sigma $, Max, Min和SNR的变化 Figure10. Changes of $ 3{\rm{\sigma }} $, Max, Min, and SNR caused by the white noise amplitude increasing from 0.001 to 0.050.
其次, 通过改变CO2的廓线比例, 透过率谱最低点也随之变化, 相应的柱浓度变化情况如图11(b). 利用图10中的Max插值获得柱总量1, Min插值获得柱总量2, 柱总量2减去柱总量1为柱浓度差值, 即为测量误差. 图 11 (a) 最小值与廓线变化关系; (b)柱总量变化与最小值关系 Figure11. (a) The relationship between the minimum value and the change of profile; (b) the relationship between the change of column total and the minimum.
最终建立近似SNR和柱总量的差值关系, 如图12所示. SNR越大, 测量误差越低. 2019年3月12日探测结果显示, 信噪比为365.55, 通过此模型获得该系统测量误差为0.44 ppm. 需说明的是, 在此仅考虑了信噪比对测量误差的近似评估, 其他影响如激光频率偏差、路径偏差以及仪器函数等的影响并未考虑, 后期将对该问题做进一步研究. 图 12 SNR与测量误差间的对应关系 Figure12. The relationship between SNR and measurement error.