FouryearsofNOAA-10 retrieved C02concentrations. - PowerPoint PPT Presentation

Four�years�of�NOAA-10 retrieved C02�concentrations. Application�to�AIRS�simulations Alain�Chedin� AIRS�science�team�meeting��Sept.�2002

FIRST�GLOBAL�MEASUREMENT�OF�MID- TROPOSPHERIC�CO 2 �FROM�NOAA�POLAR�SATELLITES� � THE�INTERTROPICAL�ZONE�� � A�neural�network�approach� � -� Multilayer�perceptron�(Rumelhart,�1986)� � -� Two�hidden�layers� � -� Inputs:�satellite�observations�(HIRS�and�MSU)� � -� Outputs:�� -� difference� between� true� mixing� ratio� and� a� reference� value�for�CO 2 �(354�ppm)� -� difference� between� true� and� reference� CO 2 � brightness� temperatures�for�HIRS�channels� � q true �-�q ref� HIRS� � and� (T B � true �-�T B � ref ) HIRS � MSU� Channels� 2�hidden�layers� Trained�on�the�TIGR�data�set�with�variable�q CO2 � � Noised�T B �

FIRST�CO2�OBSERVATION�FROM�SPACE�:�NOAA�POLAR�SATELLITES Time�period�covered :�07/87-07/91�– Monthly�maps 5°x5°�- Intertropical�zone Examples of�results for�4�months (02/90;�04/90;�08/90;�11/90) IPSL/LMD/ARA * *�Paper�in�preparation (A.�Chédin,�S.�Serrar,�N.�A.�Scott;�2002)

The�mean�seasonal�cycle�for�10�latitude�bands�between�25N�and�25S left:�as�retrieved�by�NOAA-10�(07/87-07/91) right:�as�observed�by�commercial�aircraft�(1994-1998;�Matsueda�et�al.,�2002)

Application to AI RS AI RS channels select ion : t he OSP met hod For each pressure level : 1-Select t he channels t hat best cover t he level (by t he CO2 J acobian). 2-Rej ect t he channels present ing a low ∆ T B (< 0.05K). CO2 signal 3-Comput e t he STI rat io : int erf erence H2O, N2O, CO, O3, ε and Ts signals 4-Order t he channels and remove t he channels present ing a t oo low rat io. Crevoisier et al., 2002. Submitted to Q. J. R. Meteorol. Soc.

The 43 channels obt ained wit h t he OSP met hod. Spectral location of the 43 channels CO2 signal 15µm 4.3µm 21 channels 22 channels

The 43 channels obt ained wit h t he OSP met hod. CO2 Jacobians of the 43 channels St rat osphere 10 channels Tropopause Troposphere 33 channels

St and-alone approach f or CO2 ret rieval. Based on a Neural Net work approach. 43 AI RS T B 43 ∆ T B 4 surf ace T B ∆ q CO2 (AMSU-A T B ) 2 hidden layers • Trained on t he TI GR dat a set wit h variable q CO2 . • Noised T B . We have t rained t he NN f or only one scan angle and f or several set s of input s and out put s.

First result s… . Training set : 872 Tropical sit uat ions. I nput s : 43 AI RS T B and 4 surf ace T B . Convergence of t he NN Noise 1.79 ppmv Noise/3 No Noise 0.77 ppmv 0.28 ppmv

First result s… . Training set : 872 Tropical sit uat ions. I nput s : 43 AI RS T B and 4 surf ace T B and 9AMSUA-A T B. Convergence of t he NN Without AMSU With AMSU 1.79 ppmv 1.71 ppmv

First result s… . Training set : 872 Tropical or 742 Temperat e sit uat ions I nput s : 43 AI RS T B and 4 surf ace T B . . Convergence of t he NN Tropical Temperate 1.98 ppmv 1.79 ppmv