A new non-LTE retrieval method for atmospheric parameters from mipas-envisat emission spectra

Pergamon www.elsevier.com/locate/asr

Adv. SpaceRes. Vol. 27, Nos 6-7, pp. 1099-1104, 2001 © 2001 COSPAR.Publishedby ElsevierScienceLtd. All fights reserved Printed in Great Britain 0273-1177/01 $20.00 + 0.00 PII: S0273-1177(01)00169-7

A NEW NON-LTE RETRIEVAL METHOD FOR ATMOSPHERIC PARAMETERS FROM MIPAS-ENVISAT EMISSION SPECTRA B. Funke 1, M. L6pez-Puertas 2, G. Stiller 1, T. v. Clarmann 1, and M. HSpfner 1

1lnstitut far Meteorologic und Klimaforschung, Forschungszentrum Karlsruhe GmbH and Universitiit Karlsruhe, Postfach 3640, D-76021 Karlsruhe, Germany 2Instituto de Astroffsica de Andalucia, CSIC, Apartado Postal 3004, 18080 Granada, Spain ABSTRACT The Michelson Interferometer for Passive Atmospheric Sounding (MIPAS) is a high-resolution limb sounder on board the European polar platform ENVISAT, scheduled to be launched in 2001. A large number of atmospheric trace gases relevant to stratospheric ozone chemistry and global change are expected to be retrieved from the IR spectra covering a wide spectral range. While operational data analysis under responsibility of the European Space Agency is limited to conditions of local thermodynamic equilibrium (LTE), the analysis of limb radiances affected by nonLTE is left to scientific institutions. In this paper we present an innovative non-LTE retrieval method as part of the MIPAS semi-operational data processor developed at the Institut ffir Meteorologic und Klimaforschung (IMK). The new approach enables the treatment of vibrational, rotational, and spin non-LTE as well as a dependence of the non-LTE state distribution on the retrieval target quantities. In a case study, the method has been tested for its application to the non-LTE analysis of 5.3 pm MIPAS radiances. The fundamental to-vibrational band of nitric oxide emitting at 5.3 pm shows strong non-LTE effects arising from vibrational excitation of stratospheric NO and superposed thermospheric non-LTE emissions. A conventional non-LTE retrieval approach using ab initio vibrational temperatures of NO cannot be applied due to rotational and spin non-LTE in the thermosphere, and the dependence of stratospheric vibrational temperatures on the NO abundance itself. The ability of the developed non-LTE inversion tool to retrieve stratospheric NO abundances is demonstrated by retrieval simulations. The further application of this method to the simultaneous retrieval of NO and kinetic temperature in the thermosphere and the retrieval of important non-LTE process parameters has also been tested. © 2001 COSPAR. Published by Elsevier Science Ltd. All rights reserved.

INTRODUCTION The retrieval of atmospheric parameters from IR limb emission spectra under conditions of non-local thermodynamic equilibrium (non-LTE) requires independent information on the populations of emitting ro-vibrational states, since the non-LTE state distribution is not related to kinetic temperature by Boltzmann's law as given under LTE. Non-LTE state distributions of infrared emitters can only be derived directly from measured data by retrieving excitation temperatures (Timofeyev et al., 1995), if independent spectral information on all relevant states exists. In particular, spectral information on the vibrational ground state population of the emitting gas requires nonlinear radiative transfer (i.e. saturated spectra) or strong absorption of a background source. Such conditions can only be achieved in the lower atmosphere (< 50 km), which is usually 'non-LTE free', or by use of occultation techniques. On the other hand, the a priori calculation of non-LTE populations by means of non-LTE models is only feasible if all atmospheric parameters affecting the state distributions are well known. Therefore, it is not an appropriate method to account for non-LTE in the retrieval of atmospheric parameters which themselves drive the non-LTE state distributions, such as collisional rate constants or kinetic temperature. Also, the retrieval of trace gas abundances cannot be performed using precalculated non-LTE populations in the presence of significant chemical productions of the target species (as in the case of NO, 03, etc.) or if there is a strong dependence of the atmospheric radiative field on the target species abundance (as in the case of CO2, CO, 03, etc.). In the first case, the relative fractions of nascent and thermal populations depend on the local emitter's abundance (see Fig. 1) and in the latter case non-local radiative

1099

1100

B. Funkeet al. 120

200

1O0 150

L

T

~6o

SO 60 40

50 20 0

-0.5 0.0 0.5 1.0 rel. choncje in fractional population of CO (1) [%]

-,3 -2 -1 0 rel. chonge in froctlonol population of NO (~) [ ~ ]

Fig. 1. Relative change in the fractional population of NO(~ = 1) due to an 10% increase of NO density in the entire altitude range shown.

Fig. 2. Relative change in the fractional population of CO(v = 1) due to an 10% increase of CO density at 3050 km (a) and 80-100 km (b).

coupling between different atmospheric layers induces a dependence of the state distribution at a given altitude on the emitter's abundance at other altitude regions (see Fig. 2). It is evident that the non-LTE retrieval of atmospheric parameters driving the state distribution of the measured species requires a new iterative, nonlinear inversion scheme which includes the calculation of non-LTE populations within the retrieval process. Such an innovative non-LTE inversion scheme has been realized at the Institut fib Meteorologie und Klimaforschung (IMK) and the Instituto de Astroflsica de Andalucia (IAA) within the framework of a semi-operational scientific MIPAS-ENVISAT d a t a processor development. The non-LTE inversion tool is foreseen for the retrieval of non-LTE affected trace gases such as NO, O3, and CO, kinetic temperature in the middle and upper atmosphere (70-160 km), and non-LTE specific parameters related to processes driving non-LTE populations of NO, 03, and CO:. The developed retrieval method is described in more detail in the next section. Subsequently, the application of this method to the non-LTE analysis of MIPAS 5.3 #m radiances dominated by NO(:II) infrared bands is discussed. A NEW NON-LTE RETRIEVAL METHOD The new non-LTE retrieval approach presented here is part of the IMK MIPAS data processor (Clarmann et al., 2000) which is based on a constrained nonlinear least squares algorithm with Levenberg-Marquardt damping. The major difference to conventional non-LTE retrieval methods is the integration of a non-LTE population model in the retrieval scheme in order to recalculate the non-LTE populations for the updated vector of retrieval parameters within each iteration step (see Fig. 3). Non-LTE population models are available for the most important non-LTE emitting atmospheric species such as CO2, 03, NO, and CO. Input parameters of the non-LTE model which are not target quantities of the retrieval are provided as a priori information. The forward calculation of radiative transfer is performed by the Karlsruhe Optimized Radiative Transfer Algorithm KOPRA (Stiller et al., 1998). This line-by-line forward model supports the treatment of vibrational, rotational and spin non-LTE. In order to avoid successive calls of the forward model, Jacobians, i.e. the partial derivatives Kj,i of the simulated spectra I~ ate at the measurements grid points j with respect to the retrieval parameter xi, are calculated by K O P R A in a quasi-analytical manner. In case of a significant dependence of the fractional non-LTE state populations ~"on the target quantity xi, the non-LTE model provides also the derivatives OF/Oxi which then are used for correcting the Jacobians by K j ,i = ~rOIC~tClOx ~ j / Wl'Z=const+

(OI]°~ClOr-')(O~'lOxd ,

(1)

Here, OIc~tc/OF are the partial derivatives with respect to the fractional populations calculated by KOPRA. The retrieval can be constrained by user-defined regularization terms of optimal estimation (Rodgers, 1976) or Tikhonov type (Tikhonov et al,, 1963). For reasons of efficiency and in order to reduce systematic errors, a miFig. 3. Simplified scheme of the non-LTE crowindow approach has been chosen, i.e. the retrieval is performed retrieval approach. See text for explanations, on small subsets of the measured spectra. Microwindows are selected during a preprocessing step using an optimizing selection tool (Echle et aL, 1999). The non-LTE inversion approach has been developed particularly for the retrieval of temperature and volume mixing ratio profiles from non-LTE emission spectra where a strong dependence of the emitter's state distribution on the retrieval target quantities or insufficient spectral information on relevant state populations of the emitting gas

Non-LTE RetrievalMethodfor AtmosphericParameters

1101

is encountered. Fhrthermore, the presented approach allows for the retrieval of non-LTE-specific parameters such as quenching or reaction rates, if spectral information on the non-LTE state distributions is given within the measured spectra. The determination of non-LTE-specific parameters by means of remote sensing techniques allows for validation or improvement of non-LTE model parameters and thus for a better characterization of non-LTE populations which in turn improves the retrieval of atmospheric state parameters from non-LTE emission spectra. This method has advantage over the direct retrieval of excitation temperature profiles, since only a small number of non-LTE model input parameters instead of a large number of excitation temperatures has to be retrieved. R E T R I E V A L OF A T M O S P H E R I C P A R A M E T E R S F R O M 5.3 p m M I P A S - E N V I S A T E M I S S I O N SPECTRA The retrieval of atmospheric parameters from NO(2II) emissions at 5.3 #m has to deal with strong non-LTE effects of NO ro-vibrational states in the entire altitude range (Funke and L6pez-Puertas, 2000). Stratospheric emissions are affected by vibrational non-LTE mainly due to chemical production of NO(v > 0) by NO2 photolysis leading to an enhancement up to 30% with respect to LTE emissions. As discussed above, the stratospheric NO state distribution is sensitive to the NO density, i.e., reducing the NO abundance by a factor of 2 would result in nonLTE emissions up to 60% higher than in the LTE case. In the thermosphere, NO is the most prominent infrared emitter. Populations of NO(v > 0) above approximately 110 km are in rotational and spin non-LTE. The rotational and spin state distribution of NO(v > 0) is characterized by a subthermal part due to rotational energy transfer by NO-O collisions (RT-processes) and a superthermal part due to chemical production by N+O2 -~NO+O. Rotational temperatures of the nascent population produced in the latter reaction are in the order of 3000 5000 K. Vibrational NO(v > 0) populations above 100 km are mainly driven by vibrational energy transfer due to NO-O collisions (VTprocesses). Thermospheric emissions contribute to MIPAS-ENVISAT spectra of stratospheric tangent heights up to 55%. Mesospheric NO emissions are hardly detectable by MIPAS-ENVISAT. The most important quantity which can be derived from 5.3 pm MIPAS spectra is the stratospheric NO volume mixing ratio (vmr) profile. Due to its importance to stratospheric ozone chemistry it plays a key role in the scientific analysis of MIPAS data. The retrieval of stratospheric NO requires a dedicated non-LTE inversion scheme as presented here under consideration of vibrational, rotational, and spin non-LTE. Further quantities which can be retrieved from 5.3/zm MIPAS spectra by means of the developed retrieval method are profiles of kinetic temperature, and NO vmr in the thermosphere. These parameters have to be derived simultaneously. Independent spectral information on kinetic temperature is given by the rotational structure of the NO fundamental band. Both parameters are important for the understanding of energetics, dynamics and chemistry of the thermosphere. Thermospheric NO measured by MIPAS would help to quantify downward transport of aurorally produced NO as a source of stratospheric NOz. Additional quantities to be retrieved from 5.3 #m MIPAS data are the chemical NO production rates due to NO2 photolysis in the stratosphere and due to N+O2 in the thermosphere. Independent spectral information is given in the first case by absorption features in the stratosphere and in the latter case due to the characteristic 'hot' nascent rotational distribution. Apart from a better characterization of NO non-LTE populations, the retrieval of chemical production rates would help to detect long-term variations of NO2 photolysis rates in the stratosphere and to better understand thermospheric NO production mechanisms. The NO non-LTE model included in the retrieval is a A-iteration-type model. Radiative transfer is calculated line by line with consideration of interfering species. The statistical equilibrium equation is solved for vibrational, rotational, and spin orbit degrees of freedom in an atmospheric altitude region from 0 to 200 km. All known processes affecting the non-LTE state distribution of NO are considered. A detailed description of the model is given in Funke and L6pez-Puertas (2000). Retrieval errors have been assessed for the above mentioned target quantities for midlatitude day and polar summer Table 1. Systematic error sources considered.

Error source Interfering species

Kinetic temperature

Line of sight Chem. rates VT-/VV-rates RT-rates

H20 03 N20 < 60km 60 - 100km > 100km LOS NO2+hu N+O2 NO-O NO-O2 NO-O,02,N2

Assumed uncertainty 15% 10% 15% 2K 20 K 50 K 300 m 14% 100% 28% 20% 20%

1102

B. Funke etal.

observational scenarios. Measured spectra have been simulated by KOPRA under consideration of MIPAS specific instrumental parameters (apodized spectral resolution of 0.05 cm -1, noise equivalent spectral radiance (NESR) of 3 n W / ( c m 2 sr cm -1) at 5.3 pm). As an example, a Tikhonov-type smoothing operator R = £LITL1 with first order differential operator L1 has been chosen. The regularization strength )~ has been optimized by minimizing total retrieval errors. Total retrieval error contributions are noise induced errors, systematic errors due to uncertainties of a priori parameters, and smoothing errors induced by smoothing constraints to the retrieved profile shape. Noise induced errors AStxi and systematic errors A ks y xi of retrieval parameter x~ are calculated by

i,i'

Ak xi =

[(KTS~-IK + R ) - I KTs~_IAkff ] i '

respectively, with Jacobians K , spectral covariances S~, and the perturbation of the measurement A kff due to variations of the a priori parameter k. Systematic error sources investigated are uncertainties of interference species abundances, kinetic temperature, line of sight, and non-LTE model input parameters such as chemical reaction, VT, and RT rates (see Table 1). The uncertainties of profile parameters have been assumed to be fully correlated in altitude. Smoothing errors AregXi have been calculated as

Areaxi = ~f[(A - 1)Sx (A - 1)T]i,i,

Sz,j = ai e.2 (--Az li-jl/'r,) ,

A = (KTS~-IK + R ) - I K T S y l K '

(3)

where A is the averaging kernel and 1 is unity. Since the climatogical variability of the target parameters is correlated in altitude, the parameter covariance matrix Sx is expressed in terms of a mean variability di and a scaling height 3'/of the profile parameter xi. For NO, di was estimated to be 100% and 7/ranges from 6 km below 60 km to 12 km above. For the kinetic temperature above 80 km, these parameters were set to 20% and 24 km, respectively. Three different retrieval cases have been studied (see Table 2 for details): (A) retrieval of stratospheric NO, (B) retrieval of the altitude independent mean production rate by NO2 photolysis, PN02, by means of a joint fit with NO, and (C) simultaneous retrieval of thermospheric NO, kinetic temperature, and the altitude independent mean chemical production rate by N+O2, PN+O2. A scalar additive instrumental offset and a frequency independent aerosol continuum profile below 30 km have been included to the retrieval parameter vector in all cases. Results

for case A

Weak stratospheric NO(2II) radiance contributions with a signal-to-noise ratio worse than 8 for midlatitude day and 12 for polar summer conditions considerably restrict the accuracy of the stratospheric NO retrieval. Total retrieval error can be reduced to 1 - 5 ppbv below 50 km using a regularization strength of A = 0.1. Apart from noise induced errors, there is a significant loss in accuracy due to systematic errors, particularly due to uncertainties of temperature in the upper troposphere/lower stratosphere, and assignment of line of sight. These errors are more pronounced under polar summer than midlatitude conditions. Thus, more accurate a priori information on LOS and Tkin would significantly improve the retrieval of stratospheric NO. Errors due to uncertainties of non-LTE model parameters are of minor importance (see Figures 4a and b). Small smoothing errors of less than 10% of the total error demonstrate the ability of the Tikhonov-type constrained retrieval to account for variations of profile shape in the whole altitude range (0 - 200 km). However, an accuracy of MIPAS-derived NO better than 100% is expected to be reached only between 30 and 45 km, while in the lower stratosphere and tropopause region relative retrieval errors are on the order of 1000%. Thus, averaging of derived d a t a from individual measurements is required for scientific data analysis. The presented non-LTE retrieval approach avoids significant model errors due to insufficient non-LTE modeling. A conventional non-LTE retrieval of NO using precalculated vibrational temperatures would underestimate stratospheric NO by up to 20% for midlatitude and 5% for polar summer conditions when using an initial guess NO profile 50% lower than the true profile (see Figures 5a and b). Neglect of rotational/spin non-LTE or chemical excitation by NO2 photolysis would result in retrieval errors of up to 150% of the assessed total retrieval error for midlatitude day conditions. Since these errors cannot be reduced by averaging derived profiles, consideration of these non-LTE effects significantly improves the accuracy of retrieved NO. Results

for c a s e B

The mean NO production rate by NO~ photolysis PN02 can be derived from a joint retrieval with the NO vmr profile with an accuracy of 8570 using a regularization strength A = 1.0. Significant random retrieval errors of 51%

Table

case A

B C

2. Investi/~ated

retrieval cases.

target quantity NO PNO2, NO Tki,~, NO, PN+02

additional retrieval parameter continuum (0-30 km), offset continuum (0-29 km), offset offset

measurement scanning range 5-70 km ( A = 3 km) 5-70 km ( A = 3 km) 80-152 km ( A = 3 km)

profile parameter range 0-200 km ( A = 3 km) 0-200 km ( A = 3 km) 80-200 km ( A = 3 km)

Non-LTE Retrieval Method for Atmospheric Parameters

!I

o) Midlot day 70

i......... 12 i;'to, ........... NLT!c model

60

(~ ............

noise .... tot. s y s t e m a t i c . . . . . . . smoothing ___

\ .

i:

b) Polar summer

I L

............

\

Z ..... NLTE m o d e l

r'""

~

1103

_

~

.

%

noise

_

.........

- . tot. systematic , _ smoothing

vm¢

~ N O ~

..

5O

T 40

.