MAMAP – a new spectrometer system for column-averaged methane and carbon dioxide observations from aircraft : instrument description and performance analysis

Carbon dioxide (CO2) and Methane (CH4) are the two most important anthropogenic greenhouse gases. CH 4 is furthermore one of the most potent present and future contributors to global warming because of its large global warming potential (GWP). Our knowledge of CH 4 and CO2 source strengths is based primarily on bottom-up scaling of sparse in-situ local point measurements of emissions and upscaling of emission factor estimates or top-down modeling incorporating data from surface networks and more recently also by incorporating data from low spatial resolution satellite observations for CH4. There is a need to measure and retrieve the dry columns of CO 2 and CH4 having high spatial resolution and spatial coverage. In order to fill this gap a new passive airborne 2-channel grating spectrometer instrument for remote sensing of small scale and mesoscale column-averaged CH 4 and CO2 observations has been developed. This Methane Airborne MAPper (MAMAP) instrument measures reflected and scattered solar radiation in the short wave infrared (SWIR) and near-infrared (NIR) parts of the electro-magnetic spectrum at moderate spectral resolution. The SWIR channel yields measurements of atmospheric absorption bands of CH 4 and CO2 in the spectral range between 1.59 and 1.69 μm at a spectral resolution of 0.82 nm. The NIR channel around 0.76 μm measures the atmospheric O 2-A-band absorption with a resolution of 0.46 nm. MAMAP has been designed for flexible operation aboard a variety of airborne platforms. The instrument Correspondence to: K. Gerilowski (gerilows@iup.physik.uni-bremen.de) design and the performance of the SWIR channel, together with some results from on-ground and in-flight engineering tests are presented. The SWIR channel performance has been analyzed using a retrieval algorithm applied to the nadir measured spectra. Dry air column-averaged mole fractions are obtained from SWIR data only by dividing the retrieved CH 4 columns by the simultaneously retrieved CO 2 columns for dry air column CH4 (XCH4) and vice versa for dry air column CO2 (XCO2). The signal-to-noise ratio (SNR) of the SWIR channel is approximately 1000 for integration times (tint) in the range of 0.6–0.8 s for scenes with surface spectral reflectances (SSR)/albedo of around 0.18. At these integration times the ground scene size is about 23 ×33 m2 for an aircraft altitude of 1 km and a ground speed of 200 km/h. For these scenes the actual XCH4 or XCO2 dry air column retrieval precisions are typically about 1% (1 σ ). Elevated levels of CH4 have been retrieved above a CH 4 emitting landfill. Similarly the plume of CO2 from coal-fired power plants can be well detected and tracked. The measurements by the MAMAP sensor could enable estimates of anthropogenic, biogenic and geological emissions of localized intense CH 4 and CO2 sources such as anthropogenic fugitive CH 4 emissions from oil and gas industry, coal mining, disposal of organic waste, CO2 emissions from coal-fired power plants, steel production or geologic CH 4 and CO2 emissions from seepage and volcanoes. Appropriate analysis of the measurements of MAMAP potentially also yields natural CH 4 emissions from less intense but extensive sources such as wetlands. Published by Copernicus Publications on behalf of the European Geosciences Union. 216 K. Gerilowski et al.: MAMAP – a new spectrometer system for CH 4 and CO2: instrument description

design and the performance of the SWIR channel, together with some results from on-ground and in-flight engineering tests are presented.The SWIR channel performance has been analyzed using a retrieval algorithm applied to the nadir measured spectra.Dry air column-averaged mole fractions are obtained from SWIR data only by dividing the retrieved CH 4 columns by the simultaneously retrieved CO 2 columns for dry air column CH 4 (XCH 4 ) and vice versa for dry air column CO 2 (XCO 2 ).The signal-to-noise ratio (SNR) of the SWIR channel is approximately 1000 for integration times (t int ) in the range of 0.6-0.8s for scenes with surface spectral reflectances (SSR)/albedo of around 0.18.At these integration times the ground scene size is about 23 ×33 m 2 for an aircraft altitude of 1 km and a ground speed of 200 km/h.For these scenes the actual XCH 4 or XCO 2 dry air column retrieval precisions are typically about 1% (1 σ ).Elevated levels of CH 4 have been retrieved above a CH 4 emitting landfill.Similarly the plume of CO 2 from coal-fired power plants can be well detected and tracked.The measurements by the MAMAP sensor could enable estimates of anthropogenic, biogenic and geological emissions of localized intense CH 4 and CO 2 sources such as anthropogenic fugitive CH 4 emissions from oil and gas industry, coal mining, disposal of organic waste, CO 2 emissions from coal-fired power plants, steel production or geologic CH 4 and CO 2 emissions from seepage and volcanoes.Appropriate analysis of the measurements of MAMAP potentially also yields natural CH 4 emissions from less intense but extensive sources such as wetlands.

Introduction
Carbon dioxide (CO 2 ) and methane (CH 4 ) are the two most important anthropogenic greenhouse gases (GHG) contributing to climate change (IPCC, 2007(IPCC, , 2001; Wuebbles and Hayhoe, 2002).In addition CH 4 plays an important role in the chemistry cycle of the atmosphere (Rice et al., 2003;Wuebbles and Hayhoe, 2002;IPCC, 2001).Despite their importance, our knowledge about their natural and anthropogenic sources (and sinks) has significant gaps (IPCC, 2007;Wuebbles and Hayhoe, 2002).This arises in part because of the difficulty in estimating the highly spatially and temporally variable natural and anthropogenic atmospheric source emissions (IPCC, 2007(IPCC, , 2001;;Watson et al., 1990).
Up to now, flux estimates of CH 4 and CO 2 of current global, synoptic, and mesoscale 3-D chemical transport and climate models (CTM and CM) are based on either bottomup or top-down approaches.Bottom-up flux estimates of anthropogenic sources are typically compiled by national authorities by the assessments of economic statistical data or by emission factor estimates using a variety of procedures (NRC, 2010).For bottom-up flux estimates of natural sources, ground-based microscale measurements are collected from a variety of different techniques, such as closed chamber and eddy covariance methods (Sachs et al., 2008, and references therein).Emission and flux estimates obtained by these techniques are typically assigned to specific soil/vegetation types and then are spatially extrapolated to meso and synoptic scales using, for example, a global vegetation index, derived from satellite imaging data (Takeuchi et al., 2003).
In contrast global, synoptic, and mesoscale top-down emission estimates are based on precise and accurate atmospheric in-situ concentration measurements of the relevant gases from surface networks, tall towers, helicopters, aircrafts, and trains (Dlugokencky et al., 1995(Dlugokencky et al., , 2005;;Winderlich et al., 2010;Matsueda and Inoue, 1999;Jagovkina et al., 2000;Oberlander et al., 2002;Nisbet, 2005;Miller et al., 2007;Kort et al., 2008).These measurements are then inverted by inverse models to estimate flux rates between the surface and the atmosphere (Jagovkina et al., 2001;Chen and Prinn, 2006;Bergamaschi et al., 2005Bergamaschi et al., , 2007)).Natural and anthropogenic bottom-up flux estimates are often input into the inversion calculations.Based on the measured data the anthropogenic and natural fluxes are modified during the inversion calculation (Chen and Prinn, 2006;Bergamaschi et al., 2005Bergamaschi et al., , 2007;;Jagovkina et al., 2000), in a way that the simulated atmospheric concentrations better match the observations.As a result of the coarse density of the surface observation network, information about surface fluxes distant from the network is still not well defined and ambiguous (Villani et al., 2010;Bréon and Ciais, 2010;NRC, 2010).Especially discrimination of the different source types remains still inaccurate (NRC, 2010).
These new remote sensing data were incorporated for the first time to estimate the annual CH 4 surface fluxes at a resolution of several degrees using top-down inverse modeling (Bergamaschi et al., 2007(Bergamaschi et al., , 2009)).Recently a new mission for GHG observations, the GOSAT satellite with the Tanso-FTS on board (Yoshida et al., 2008), was launched successfully in January 2009.The footprint of the Tanso-FTS instrument is 10 km in diameter with a typical gap for nominal mode of about 160 km (adjusted recently to 320 km) between observations.Because SCIAMACHY's and GOSAT large typical footprint (i.e. 60 km × 30 km for SCIAMACHY and 10 km diameter for GOSAT), and the large gaps between the typical measurements of GOSAT single local emissions cannot be accurately resolved in the currently available satellite data (Bréon and Ciais, 2010).Therefore the contribution of small "hot-spot" areas and single facilities is not sufficiently resolved with the existing ground-based and satellite observational systems (NRC, 2010).Future satellite missions like OCO-2 with its reduced footprint sizes of 3.4 km 2 will partly overcome this problem for CO 2 (Crisp et al., 2009).In contrast CH 4 will not be sufficiently resolved in currently approved future satellite missions.
The deficiencies in our current knowledge of point sources and "hot-spot" areas emerge a clear need for the development of new measurement techniques to improve top-down estimates and constrain regional and local emissions (NRC, 2010).These techniques are needed to extend the coverage and facilitate the integration of existing global systems and address the up-scaling issue.In this respect, airborne passive and active remote sensing techniques offer potentially a unique set of opportunities, as they combine coverage with high spatial resolution.The measured data ideally need to be of an accuracy and precision to yield on inversion the CH 4 and CO 2 emissions from less intense but extensive and larger scale sources (and sinks), such as wetlands.As a threshold the accuracy and precision of the data yield on inversion significant constraints on local hot spot emissions to separate them from the less intense but extensive larger scale sources (and sinks) and thereby allowing an improved estimate of both.
Until recently, there was a lack of dedicated airborne instrumentation with the capability of measuring CH 4 and CO 2 dry atmospheric columns (mole fractions) within or above the planetary boundary layer (NRC, 2010) with sufficient relative accuracy and precision required for those applications.Active CH 4 systems measuring column concentration (not to confuse with mole fractions) designed for pipeline leakage detection and monitoring (Meyer et al., 2006;Zimig and Ulbricht, 2006;Ershov, 2007) are typically limited to altitudes below 300 m while recommended operation altitudes range typically around 100 m.Due to the altitude limitation those systems are actually not well adopted for local scale atmospheric applications requiring atmospheric dry column measurements up to et least the height of the planetary boundary layer which can reach in summer a height of up to 2000 m.It is reported in literature that active DIAL pipeline monitoring instruments reach a CH 4 column concentration threshold sensitivity during flights at typical flight altitudes of 100 m in the range of 80 ppm m1 (for ground based laboratory measurements) (Meyer et al., 2006) and 100 ppm m (airborne at 0.5 s measurement time) (Ershov, 2007).Recent developments of high altitude DIAL systems for airborne GHG dry column measurements are ongoing in the framework of future satellite developments but not yet available for field application (NRC, 2010).
For passive systems designed for atmospheric applications like the airborne SWIR FTS developed for GOSAT validation and calibration (Suto et al., 2008) dry air column precisions and "in-flight" detection limits have not yet been published.Passive instruments for gas leakage monitoring (like reported by Meyer et al., 2006) based on a compact 1/4 m polychromator working in the 1.60 µm to 1.68 µm spectral range have a reported detection limit of the measured column concentration of 800 ppm m (ca.5% total column, referred to a 1013 hPa normalized total atmospheric thickness of about 8580 m, see Sect. 4.3.4).Roberts et al. (2010) also reported recent successful trials adapting retrieval algorithms for hyperspectral imaging data of the Airborne Visible Infrared Imaging Spectrometer (AVIRIS) to map CH 4 emissions over strong marine geological CH 4 sources reaching theoretical detection limits for measurements in solar glint of 36 ppb CH 4 total column increase (corresponding to 300 ppm m) for the resulting CH 4 maps.For smaller surface spectral reflectance (SSR) (i.e.<0.22) the detection limit degrades proportionately with the decreasing radiance (Roberts et al., 2010).It is expected that CO 2 and CH 4 retrieval from terrestrial, low spectral resolution hyperspectral data is much more demanding as a result of the need to model accurately the SSR to reduce the uncertainty in the retrieved CH 4 and CO 2 column concentrations.For both passive systems (gas leakage monitoring an hyperspectral) airborne dry air column precisions have not been reported either.
In order to close the existing gap not accommodated by currently available instrumentation and to validate spacebased measurements on local and meso scales (i.e. from SCIAMACHY and GOSAT) a team from the University of Bremen and the Helmholtz Centre Potsdam, German Research Centre for Geosciences (GFZ), have developed an airborne spectrometer system, capable of direct and quantitative remote column-averaged measurements of atmospheric CH 4 and CO 2 .This system, named the "Methane Airborne MAPper" (MAMAP), is designed for flexible operations on a variety of airborne platforms, and is described in detail below.MAMAP is equipped with a down-looking telescope for nadir observations and an up-looking light inlet for zenith observations.The MAMAP aircraft instrument was designed to measure total vertical columns of CH 4 and CO 2 on small spatial scales (<250 m) with a precision and relative accuracy equal or better than 1-2% (threshold <2%, target <1%) dry atmospheric column for altitudes of at least the height of the planetary boundary layer (i.e. up to 2000 m) over land for typical albedo/SSR of about 0.18.This precision/relative accuracy requirement corresponds to a column enhancement of about 150-300 ppm m below the aircraft which places limitations on the measurable target emission strengths (see Sect. 5).The initial instrument precision threshold specifications were inferred primarily from flux estimates of strong anthropogenic and geologic methane emitters with large emission uncertainties.The emitters considered include for instance large landfills with CH 4 emissions equal or greater then 5-10 kt CH 4 yr −1 (EPER, 2004), fugitive emissions from entire gas fields/deposits like reported by Jagovkina et al. (2000, and references therein) with emission estimates between 2 and 10 Mt CH 4 yr −1 for an area of about 1.8 × 10 10 m 2 , emissions from coal mining facilities reported to reach up emissions >20 kt CH 4 yr −1 on local scales (EPER, 2004), or geologic emissions from marine methane bearing sediments in arctic regions were atmospheric concentrations (measured 2 m above ground) can reach values between 3 and 8 ppm with respect to the atmospheric background concentration of 1.85 ppm over areas with the extend of several square kilometers (Shakhova et al., 2007(Shakhova et al., , 2010)).The instrument precision goal requirements of ≤ 1% dry atmospheric column allows measurements of localized emitters or single facilities with even smaller emissions (i.e.<5 kt CH 4 yr −1 , see Sect. 5).
Due to the large uncertainties of CH 4 emissions of different local anthropogenic and geologic sources (NRC, 2010) measurements with MAMAP have also the potential to topdown constrain bottom-up estimates from those types of sources (see Sects. 5 and 6).For instance Chambers and Strosher (2006a,b) reported 4-9 times higher emissions for Canadian refineries and natural gas processing plans when measured with ground based DIAL than calculated by emission factors of the single components.Börjesson et al. (2000) found that, for the same landfill, CH 4 emissions were a factor of 4 higher, when estimated using a tracer gas technique, than that bottom-up estimated from closed chamber measurements.Anthropogenic Methane emissions from fossil fuel industry (coal, natural gas and oil) and landfills are estimated to be responsible for about 167 (86-274) Mt CH 4 yr −1 with respect to total global natural CH 4 wetland emissions estimated to be about 100 (91-232) Mt CH 4 yr −1 compared to total global CH 4 emissions (anthropogenic and natural) estimated to be about 503 (410-660) Mt CH 4 yr −1 (Wuebbles and Hayhoe, 2002).
Even though the goal and threshold requirements for MAMAP were initially inferred primarily for CH 4 they will also allow measurements of strong anthropogenic CO 2 emitters with typical emissions between 1-30 Mt CO 2 yr −1 like coal fired power plants (as demonstrated in Sect.3), steelworks, cement production etc. (EPER, 2004) or geologic emitters like volcanoes (Mörner and Etiope, 2002).Ackerman and Sundquist (2008) compared powerplant (PP) emission data bases and found that the absolute difference of the emissions of individual coal-fired PPs in the USA is typically about 20%.Thus instruments like MAMAP can potentially be used for independent monitoring (for inversion modeling of PP emissions with MAMAP measurements and associated uncertainties, see also Krings et al., 2011).Reaching the goal precision and relative accuracy of ≤1% will potentially allow CO 2 measurements and top-down emission estimates for major cities having even larger uncertainties of the emissions (NRC, 2010).Appropriate analysis of the measurements of MAMAP potentially also yields natural CH 4 emissions from less intense but extensive sources such as wetlands.
In the following the MAMAP instrument design and the performance together with some results from on-ground and in-flight engineering tests are presented.The analysis is focused primarily on the most important instrument parameters i.e. signal to noise ratio (SNR), associated precision and spectral stability.Furthermore the analysis is limited to the SWIR cannel and associated XCH 4 or XCO 2 products derived from SWIR data only.A discussion of systematic effects and the overall uncertainty of the data products will be given in a separate publication (i.e.Krings et al., 2011).Alternative retrieval strategies incorporating O 2 data from NIR measurements and associated concurrent XCH 4 and XCO 2 products will be discussed elsewhere.
This manuscript is organized as follows: in Sect. 2 the MAMAP instrument is described and its specifications are given.The current version of the MAMAP retrieval algorithm, which has been used to assess the on-ground and inflight instrument performance, is briefly explained in Sect.3. In Sect. 4 the results of the MAMAP instrument performance analysis are presented and discussed.In Sect. 5 relevant CH 4 and CO 2 emission targets are discussed.In Sect.6 first results are shown from the analysis of flights near or over anthropogenic CO 2 and CH 4 emission sources.Finally, a summary is given in Sect.7.

Description of the MAMAP instrument
The MAMAP instrument comprises two thermally stabilized grating spectrometer systems having a focal length of F = 300 mm and a f-number of f/3.9.One (non-imaging) spectrometer system measures in the SWIR over the spectral region at 1590-1690 nm to enable simultaneous retrieval of CO 2 (1590-1620 nm) and CH 4 (1630-1750 nm) columns.The second push-broom imaging spectrometer system measures in the NIR over the spectral region between 756-769 nm for the detection of the oxygen (O 2 ) absorption using the O 2 -A band.The retrieved O 2 columns can be used to convert the greenhouse gas columns into dry-air columnaveraged mixing ratios (see Sect. 3).
Both spectrometers have two independent telescopes, collecting electromagnetic radiation, and pointing towards nadir and upwards to the zenith-sky direction (Figs. 1 and 2).A fold-mirror allows switching between the nadir and zenithsky modes of operation.This permits sequential measurement of the diffuse up-welling and down-welling radiance.Optionally, cosine diffuser plates or a combination of cosine diffuser plates plus glass fibres and collimator optics can be installed on-top of the zenith sky telescopes.In this configuration the instrument can perform direct solar irradiance measurements and diffuse down-welling (ir)radiance measurements in zenith-sky mode.From these measurements total and/or partial vertical columns can be retrieved as will be described in Sect.3.
The instrument is designed to operate at altitudes of more than 20 km from pressurized cabins and up to 4 km in non-pressurized cabins allowing measurements to above the convective boundary layer (CBL).Furthermore, it achieves a ground scene of <300 m (along the flight track) on high altitude aircrafts with cruise speeds around 900 km h −1 .Depending on cruise speed, pixel sizes range typically between <50 and 150 m on slower (<400 km h −1 ) propeller aircrafts.

The CH 4 /CO 2 SWIR spectrometer
A single non-imaging grating SWIR spectrometer is utilized to simultaneously retrieve CH 4 and CO 2 .This spectrometer uses a special F = 300 mm modulation transfer function (MTF) optimised aspheric doublet lens (manufactured by ZEISS) for nadir observations and a single spherical lens (F = 300 mm) for zenith-sky observations.The instantaneous field of view (IFOV) of the SWIR spectrometer is 1.34 • across track (CT) and 0.02 • along the flight track (LT).
The sensor head of the SWIR spectrometer is a modified linear photodiode array camera from Princeton Instruments (OMA-5, LN-1024) using a linear extended InGaAs 1024 pixel focal plane array (FPA) as detector.The detector is cooled with liquid nitrogen to −120 • C to minimise detector noise.With a pixel pitch of 25 µm, this array covers 25.6 mm in the spectral direction of the spectrometer's focal plane.In combination with a 600 grooves mm −1 spectrometer grating, a spectral window of 97.3 nm can be covered with a spectral resolution of about 0.82 nm (FWHM).At this resolution a sampling of approximately 8.6 detector pixels per FWHM is achieved.The coverage of a 97.3 nm spectral window permits the simultaneous measurement of CH 4 and CO 2 absorption bands with the same detector.
The detector/camera non-linearity was provided by the manufacturer and stated to be <1%.The slit function of the system on different detector positions was measured with a single line from a spectral line source (SLS) while rotating the motorized grating slightly in different positions.The FWHM was then fitted by two Gaussian line shapes and slightly adjusted while minimizing the fit residuum (see Sect. 3).Spectral calibration was inferred by shift end squeeze until optimal matching of the absorber features can be achieved (see Sect. 3).Due to the high sampling of 8.6 detector pixels per FWHM sufficient good results can be achieved by the approach.The system was currently not absolutely calibrated.Nevertheless throughput of the system was inferred with limited accuracy from laboratory tests and from manufacturer throughput data of the single optical components.
The dark signal of the sensor was reduced from ∼600 fA to below 60 fA measured at an optical bench temperature of 25 • C by modifying the camera head.With the full well capacity of a single detector pixel of about 4.4 Me − this dark signal reduction yields a theoretical signal to noise ratio (SNR) of 1000 and higher over land (surface albedo/SSR = 0.18, detector exposure time t exp ∼0.6 s) and a SNR of larger than 350 over water (albedo/SSR ∼0.01, t exp ∼3-5 s).Details concerning the SNR estimates are given in Sect. 4. To prevent detector saturation for high albedo/SSR scenes at nominal operation, the exposure time for a single readout is typically reduced by a factor of 10 (from 580 ms to 58 ms) over land and to 1-2 s over water.Over land, bursts of typically 10 single detector-readouts are collected and stored.All single readouts of each burst are then co-added in order to reach an appropriate SNR.This mode of operation is referred to as "co-added burst mode".For a flight altitude of 1 km, a surface albedo/SSR of 0.18 and a flight speed of ∼200 km h −1 (e.g., Cessna 207 aircraft), a co-added ground pixel size of 23.4 m (cross track, CT) × 33 m (along track, LT) can be achieved for a total co-adding/integration time of ∼0.6 s.

The O 2 -A-band NIR spectrometer
A push-broom imaging NIR spectrometer system, operating in the 756 to 769 nm spectral range detects O 2 .It uses two F = 80 mm lens doublets for nadir and zenith sky operation.
The IFOV is 5.85 • (CT) × 0.072 • (LT).In the CT direction, the IFOV is approximately 4 times larger than the IFOV of the SWIR spectrometer.This larger IFOV was chosen to allow characterization of the surrounding scene.In imaging direction (CT) the NIR IFOV is subdivided in 85 discrete pixels.During post processing, approximately 1/4 of these 85 detector pixels is software binned to a window for each detector reading.In this manner, a single 1-D-spectral readout can be created which can be matched to the IFOV of the (1-D) SWIR spectrometer.To optimally match both IFOV (see Fig. 3) the binned NIR-IFOV window can be moved in CT direction during post processing, until optimum co-alignment to the SWIR IFOV is achieved.Due to both spectrometers small IFOV in the LT (i.e., flight-) direction small misalignments of both slits in LT direction are not critical as the LT pixel size is defined primarily by the total co-adding/integration time of each burst.Sufficient good co-boresighting of both slits in LT direction was achieved by moving a small source inside the IFOV of both systems and by imaging a slit in infinity onto both input lenses of the system.Both spectrometer slits were then aligned with respect to the imaged slit.Sufficiently good LT synchronisation of the pixels for both spectrometers is achieved by electronic synchronisation of the exposure times and the detector readouts.
The NIR O 2 -A spectrometer system uses an E2V 512 × 512 pixel frame transfer (FT) CCD with a pixel pitch of 16 µm × 16 µm as detector.The detector/camera nonlinearity provided by the manufacturer is stated to be <0.3%.The detector is cooled to −30 • C by thermoelectric coolers, to minimise the detector noise.To speed up the readout time and reduce the data output, 6 pixels in imaging and 2 pixel in spectral direction are hardware binned, resulting in an array of 85 (spatial) and 256 (spectral) pixels with a pixel size of 96 by 32 µm.The readout of the FT-CCD is fully hardware synchronized with the readout of the SWIR detector by trigger pulses.The FT-CCD covers 8.192 mm × 8.192 mm of the -push-broom -imaging spectrometers focal plane.In combination with a 1200 grooves mm −1 spectrometer grating, a spectral window of 13 nm can be imaged on the detector, resulting in a spectral resolution of ∼0.46 nm (FWHM).With the (binned) pixel size of 32 µm (in spectral direction) a sampling of ∼9 pixels per FWHM is achieved.Respectively, hardware binning of 6 pixels in imaging direction, divides the NIR spectrometers field of view in the mentioned 85 pixels (CT).For an albedo/SSR of ∼0.18, detector fillings in the range of 50-70% are achieved for an exposure time of ∼0.6-0.8 s.With the full well capacity of 1560 ke − for a single hardware binned (6 × 2) detector pixel, theoretical SNR values of ∼1000 per (binned) pixel can be achieved.In order to prevent detector saturation for high albedo/SSR scenes, the single readout exposure time is also reduced by a factor of 10 (as for the SWIR detector).Thus, a typical burst of 10 single detector-readouts are collected and stored (and later co-added).
The (6 × 2) binned pixel SNR was further increased by additional binning in across flight direction on the cost of spatial resolution.For example, CT software binning of all pixels of the NIR channel within the IFOV of the SWIR channel (i.e.∼1/4 of the 85 NIR spectrometer pixels, see Fig. 3), will result in a theoretical SNR of more than 4000.

Pointing and image navigation
To monitor the position and the pointing of the instrument, MAMAP is equipped with a Garmin 5 Hz GPS and a Microstrain 3DM GX1 gyro-system.The readout of the position data of both devices is fully synchronized with the readout of the spectrometers.Additional position logging systems can be synchronized to MAMAP via external triggers.The MAMAP system also contains a triggered 640 × 480 pixel 1/4 interline CCD colour camera (type: DFK 21BF04) for image acquisition which is synchronized with the spectrometer sensors.The CCD camera is equipped with a 25 mm lens and has the IFOV of 7.2 • (CT) × 5.7 • (LT).This optical control helps to optimise the pointing knowledge of the MAMAP system.In this manner, pointing information towards different ground surface types can link CH 4 and CO 2 column information with potential CH 4 or CO 2 sources.

System control, data acquisition and power supply
Each SWIR and NIR spectrometer system uses a separate ultra slim, fan-less Advantech S123T panel PC for data acquisition, management of housekeeping data and the spectrometer (zenith/nadir) fold mirror control.For control and data acquisition of the CCD camera, the GPS and the gyro system, a separate embedded PC is used.To avoid data damage over 3000 m flight altitude in unpressurised aircraft cabins all computers were equipped with flash disk devices for data storage.Thermal stabilization of the system was performed by two "off the shelf" digital PID controller units in combination with thermoelectric heaters.The system is powered through an uninterruptible power supply (UPS) and an optional 28 V DC to 220 V AC converter permitting the instrument to be operated either on 28 V DC or 220 V AC.The UPS has a GILL 28 V/43 Ah sealed lead acid battery, allowing the instrument to be operated up to 2 h without external power.
The whole system including all controllers and the batterybuffered power supply fits into two standard (DLR-Falcon) aircraft racks (556 mm × 650 mm × 968 mm each).The weight of each rack is approximately 120 kg.The first rack contains the camera controllers, the spectrometers and telescopes with the thermal control unit.The second rack contains the panel and the embedded PC's for the spectrometer control and data acquisition and the UPS system (Fig. 4).It has to be noted that MAMAP is currently not optimised w.r.t.mass, and there is potential for some significant mass reductions if required.
The instrument has been designed for flexible operation on-board a variety of airborne research platforms (e.g., Dornier 228, Dassault Falcon, Cessna Caravan, Basler-DC3 Polar-5 etc.) and provides all needed synchronisation signals for external Gyro and GPS logging systems.The sensor parameters are summarised in Table 1.

MAMAP retrieval algorithm and the determination of data products
The objective of the retrieval algorithm is to invert MAMAP spectra to derive the CH 4 , CO 2 and O 2 -total or partial -vertical columns and the CH 4 and CO 2 column-averaged dry air mixing ratios, XCO 2 and XCH 4 .For the retrieval, measurements of dark signal and pixel to pixel gain corrected (nadir) radiance spectra are used.The measurements and the target quantities being similar with those of the SCIAMACHY satellite instrument on ENVISAT (Buchwitz et al., 2005a,b;Schneising et al., 2008Schneising et al., , 2009)).The derivation of the absolute column amounts (in units of number of molecules per unit area, e.g., molecules cm −2 ) into column-averaged dry air mixing ratios (in ppm for CO 2 and ppb for CH 4 ) requires knowledge of the corresponding "dry air column", i.e. the total number of molecules in the observed atmospheric column, neglecting water molecules.For the interpretation of the MAMAP measurements the column-averaged mixing ratios are the preferred to the absolute columns, because of their much weaker dependence on the changes of surface topography/pressure and flight altitude.There are several approaches to estimate the dry air column needed for the conversion of the greenhouse gas columns into column-averaged mixing ratios: i. by the use of the simultaneous measurements of the oxygen (O 2 ) column retrieved from spectral measurements of the O 2 -A band (located at 760 nm) analog to the method described in Schneising et al. (2008), for SCIAMACHY column-averaged CO 2 retrieval, CH 4 /CO 2 -SWIR-spectrometer O 2 -NIR-spectrometer F = 300 mm temperature stabilized grating spectrometer system (f/3.9)F = 300 mm temperature stabilized push broom imaging grating spectrometer system (f/3.9)Grating: 600 grooves/mm −1 Grating: 1200 grooves mm −1 Detector: LN cooled 1024 pixel InGaAs FPA Detector: 512×512 pixel CCD Sensor, TE cooled, 6 pixel binned in imaging direction and 2 in spectral direction Spectral range: ∼1590-1690 nm Spectral range: ∼756-769 nm Spectral resolution: ∼0.82 nm FWHM Spectral resolution: ∼0.46 nm FWHM Spectral sampling: ∼8.6 pixel/FWHM Spectral sampling: ∼9 pixel/FWHM Detector-SNR: ∼1000 at ∼0.6-0.8 s integration time (10 detector readouts co-added, surface albedo/SSR 0.18) Detector-SNR: >4000 (binned) at ∼0.6-0.8 s integration time (10 detector readouts co-added, 1/4 of the 85 spatial rows binned, surface albedo/SSR 0.18) IFOV: ∼1.Precision requirement: Goal: better than 1% of the total CH 4 over CO 2 column-averaged dry air mixing ratio with respect to the atmospheric background; threshold: better than 2% (precision is defined as the random error of the retrieved CH 4 and CO 2 columns due to instrument noise) Size: 2 "Falcon" standard racks, 556 mm × 650 mm × 968 mm each Weight: ∼120 kg (each rack) Power consumption: ∼600-800 Watt at nominal operation, <1000 Watt at warm-up ii. by using another, well mixed gas whose mixing ratio is well enough known and varies less than the trace gas of interest (e.g. by using the simultaneously retrieved CO 2 column for normalizing the retrieved CH 4 columns to obtain the column averaged mixing ratio of CH 4 , Frankenberg et al., 2005;Schneising et al., 2009) (or vice versa for CO 2 ), and iii. by using external information on surface pressure obtained from, e.g., meteorological analysis by analogy to the method described in Barkley et al. (2006), for column-averaged CO 2 retrieval.However this is problematic since knowledge of surface pressure on a scale of a few meters would be required.
The advantage of the first approach is that the mixing ratio of O 2 in dry air is well known (20.95%) and constant up to about 100 km and comprises 99.99% of the atmosphere.However, differences in the radiative transfer of the electromagnetic radiation through both absorption and scattering result in the path of radiation through the atmosphere being dependent on wavelength.Consequently, the presence of scattering by aerosols, cirrus or other clouds, gives a somewhat different light paths around 760 nm in comparison to 1.6 µm, as the phase function for particle scattering depends on wavelength (see Schneising et al., 2008Schneising et al., , 2009;;Schneising, 2009, for a discussion of this approach).In order to use the O 2 band for the determination of XCO 2 and XCH 4 scattering needs to be explicitly accounted for.An alternative approach for the determination of XCH 4 at least in regions where diurnal or spatial CO 2 variations are small is to assume that the CO 2 is effectively constant and well mixed compared to CH 4 .As the relevant relatively weak absorptions of both gases occur spectrally close to one another, the path of the electromagnetic radiation is similar for CO 2 and CH 4 .For this reason, one of the XCH 4 data products is retrieved from SCIAMACHY (Frankenberg et al., 2005;Schneising et al., 2009) in this way.For this approach to be valid for the scene, CO 2 must be significantly less variable than CH 4 .For the SCIAMACHY's large ground pixel size (30 km × 60 km) this is reasonable, but could be problematic for the much smaller MAMAP ground pixel size.In several circumstances also a vice versa approach can be applied where larger CO 2 variability is expected with respect to CH 4 , e.g. for strong anthropogenic CO 2 emitters (see Sect. 6).
In summary all three methods could be used for MAMAP.The method, which performs best, depends on the target and the validity of the assumptions and the effort made to account for cloud and aerosol within the retrieval algorithm.
In this manuscript we focus on results obtained in the SWIR (1.6 µm) channel of MAMAP.To assess the instrument performance we have developed an initial version of a retrieval algorithm for MAMAP.In the following we present a short characterization of this algorithm.A more detailed description will be given elsewhere (Krings et al., 2011).
The MAMAP retrieval algorithm, used in this study, is derived from the Weighting Function Modified Differential Optical Absorption Spectroscopy (WFM-DOAS) retrieval algorithm (Buchwitz et al., 2000), referred to as the WFMD/M retrieval algorithm.The WFM-DOAS retrieval algorithm has been developed for and applied successfully to the retrieval of CH 4 and CO 2 vertical columns from SCIAMACHY nadir spectra (Buchwitz et al., 2005a,b;Schneising et al., 2008Schneising et al., , 2009)).Similar to WFM-DOAS, the WFMD/M retrieval algorithm uses a least-squares fitting procedure to minimise the difference between the logarithm of a simulated radiance spectrum with that measured.The simulated spectrum and the derivatives ("Jacobians") of this spectrum with respect to a change of atmospheric parameters are computed with the radiative transfer model (RTM) SCIATRAN (Rozanov et al., 2005) using the HITRAN data base (Rothman et al., 2005) and the solar spectrum from Livingston and Wallace (1991).These derivatives are called weighting functions.In addition to the geophysical fit parameters (i.e.CO 2 , H 2 O, temperature for CO 2 retrieval and CH 4 , CO 2 , H 2 O, temperature for CH 4 retrieval respectively), a low order polynomial in the spectral domain is used to account for all smoothly varying spectral parameters, which are not explicitly modelled or inadequately known.These parameters include, for example, the MAMAP absolute radiometric calibration function, aerosol scattering, absorption parameters and the surface spectral reflectance.Finally also an odd-even correction fitting an alternating function (−1, 1, −1, 1, ...) is applied.The odd-even effect is caused by the multiplexer design of the SWIR detector in combination with the tilted illumination.For all retrievals a standard background aerosol scenario as in Schneising et al. (2008) is used and a constant albedo of 0.18 (Lambertian surface) was applied as first approximation, assuming that most biases introduced by these assumptions will be small in the resulting ratios.
The spectral calibration of the system was obtained by shifting and squeezing the measured spectra with repect to the RTM simulated spectrum while minimizing the residuum.Once the optimum spectral calibration parameter were obtained only shifting was performed in the future data processing.Slit function parameters were optimized in a similar way.The measured slit function was fitted by two Gaussina line shape functions.After that the FWHM was adjusted while minimizing the residuum for the CO 2 and CH 4 windows respectively.Once optimum slit function parameters were obtained, no further adjustment of the FWHM was applied to the subsequent processing of the data.
Figure 5 shows an example result for a WFMD/M analysis of a single spectrum, recorded by the MAMAP SWIR channel.The absorption features of CH 4 (Fig. 5 left) and CO 2 (Fig. 5 right) are clearly visible in the MAMAP spectrum.Interfering gases in the CH 4 fitting window (left) are CO 2 and H 2 O.In the CO 2 fitting window (right) only H 2 O interferes.Also fitted is the shift of the temperature profile (only shown for the CO 2 fitting window).The retrieved CH 4 profile scaling factor (PSF) is 0.989 ± 0.014.The retrieved CO 2 PSF is 0.991 ± 0.022.The residual ("RES") is shown in the bottom panels and is the difference between the MAMAP spectral measurements and the fitted radiative transfer model.The root-mean-square (RMS) of the residual is ∼0.6% for both fitting windows.As can be seen, the fit residual is not only determined by measurement noise but also contains systematic features.This is currently attributed to wavelength calibration errors, slit function uncertainties, uncertainties of the spectroscopic line parameters or spectral structures of the white lamp calibration source.
As a result of the correlation between weighting functions of different altitude layers, the MAMAP retrieval is not height sensitive and weighting functions are integrated over the entire profile.Thus, the retrieval output PSF always indicates an altitude averaged change in the column concentration.For example a PSF of 1.01 means that the retrieved column is 1% higher than the vertical column which has been assumed for the radiative transfer simulations.During a flight of the MAMAP instrument, significant concentrations changes are expected below the aircraft, resulting from significant changes in surface emission fluxes.Inspection of the averaging kernels (i.e. the sensitivity of the retrieved parameter as a result of a perturbation of the true column) (Fig. 6) shows a striking difference of about a factor of two below and above the aircraft.This difference is explained by the fact that for a nadir viewing instrument electromagnetic radiation coming from the sun passes through the absorber below the aircraft twice: once before and once after surface reflection (or surface scattering).SCIATRAN accounts for the actual light path in the weighting functions for each layer.The current retrieval algorithm only uses the co-added weighting functions from all height layers.Hence a column averaged PSF will always overestimate the real concentrations in the total column, because the averaged weighting functions are smaller than the weighting functions below the aircraft.To account only for an increase or decrease in CO 2 and CH 4 concentrations compared to background below the aircraft while leaving the column above unchanged, the original profile scaling factors (PSF) is multiplied by a conversion factor c (Table 2) derived from RTM simulations.The resulting new column scaling factor (CSF) as the scaling factor for the total column increase or decrease, assuming that all changes in concentrations of CH 4 occur below the aircraft and the CO 2 concentration is constant -can be calculated from: The conversion factor c depends on geometry (aircraft altitude, solar zenith angle), atmospheric distribution of the according trace gas, and the surface albedo/SSR.Assuming an average mixing ratio of 1774 ppb for CH 4 , the concentration change in the total column below the aircraft C is then estimated by the following equation: A similar calculation can also be performed for CO 2 .Currently the WFMD/M retrieval algorithm is modified to invert directly changes below the aircraft.MAMAP zenith sky data are used to compare to nadir data, to derive column changes and to validate the assumption of an unchanged column above the aircraft.

Performance assessment of the SWIR channel
The instrument performance of the SWIR channel of MAMAP has been evaluated in two different ways: (i) by estimation of the signal-to-noise ratio (SNR) for each measured spectrum (Sect.4.1) and (ii) by estimation of the instrument precision (Sect.4.2).Time series of single detector readout and co-added burst mode measurements obtained for different operation conditions (on ground and in-flight) are analysed and compared in Sect.4.3.In Sect.4.4 the spectral stability of the system is examined.The on-ground measurements were carried out in May 2006 on the campus of the University of Bremen.The in-flight measurements were performed onboard the AWI Polar-5 aircraft and the FU-Berlin Cessna 207 aircraft during campaigns in August 2007 and November 2008.In the following sections the terms precision, accuracy, relative accuracy and measurement uncertainty are used.These are defined as follows: -Precision includes all random errors in the measurement and the retrieval resulting from detector shot noise, random illumination effects and other random effects.As systematic fast varying (near random) albedo/SSR effects can not be separated from the other fast random effects, they also are accounted for in the precision.
-Accuracy includes all systematic errors in the measurement and the retrieval resulting from aerosols and clouds, uncertainties resulting from insufficient knowledge of the slit function shape, errors resulting from the a priori profile and temperature information, the spectroscopic line parameters, SZA effects, the flight altitude and other systematic errors.-Measurement uncertainty is defined as the sum of both, systematic and random errors.

Signal to noise ratios (SNRs)
The SNR of MAMAP determines (in combination with spectral resolution and spectral sampling) to a large degree the achievable precisions with these type of spectrometers.
The SNR for the SWIR band of MAMAP was evaluated in two different ways: (i) by an estimate via simulations (Sect.4.1.1)and (ii) by a SNR estimate from real measurements (Sect.4.1.2). Results for both estimates are later compared and discussed in Sect.4.3 for different operation conditions.

SNR computations based on simulations
A theoretical noise (N) estimate for single detector readouts of the MAMAP system was inferred by applying the MAMAP instrument model simulation to a radiative transfer model spectrum.The noise for each detector pixel was calculated from the resulting shot noise of the estimated detector signal, the shot noise of the detector dark signal, the readout noise of the detector and the analog to digital converter.The simulated SNR for each detector pixel -SNR(sim) -was calculated by dividing the calculated detector signal (S) by the calculated noise (N ).  and simulated SNR values -SNR(sim) -for different operation conditions are averaged over all detector pixels of the entire fitting windows of CH 4 and CO 2 , and summarized in Table 3.In nominal operation, the MAMAP instrument utilises the so called "co-added burst mode".In this mode the instrument acquires a burst of a programmable number of n single spectra (typically n = 10).All acquired spectra of each burst are co-added during subsequent processing (Sect.2).Assuming a Gaussian error distribution for the single detector readouts, the simulated co-added burst mode (BM) signal-tonoise ratio -SNR BM (sim) -of each co-added measurement was calculated from the simulated single readout SNR(sim) by:

SNR estimates from real data
For comparison with the modelled values, the SNR of MAMAP was estimated from real data.The SNR has been calculated from the individual fit residuum ("RES i ", see Sect. 3) of each single measurement for the two fit-windows used for CH 4 and CO 2 .As RES i of each i-th measurement contains systematic features, the mean residuum for all measured spectra of the processed data set has been calculated and subtracted from the individual RES i spectra to remove the systematic components contained in RES i : After subtraction RES i contains shot noise, detector noise, noise effects resulting from varying spectral structures of the measured (spectral) radiance, noise effects due to tilted illumination of the detector and noise effects resulting from inhomogeneous illumination.The latter result from changes of the instrument slit function, induced by keystone and smile effects of the optical system of the MAMAP spectrometer in combination with inhomogeneous illumination of the slit.
Even-odd effects are induced by tilted illumination of the linear InGaAs detector due to the multiplexer design.
To estimate the SNR of each measurement, first the standard deviation (SRES) of the resulting new fit residuals (RES i ) has been retrieved for each single spectra: The SNR of each measurement was then estimated by the reciprocal value of the standard deviation: For comparison with the simulated values, also the mean SNR values SNR(ret) 1...n and mean measured signal values S(meas) were calculated for a selected set of measurement sequences and summarized in Table 3 (see Sect. 4.3).For co-added burst-mode (see Sect. 4.1.1)the retrieved SNR -SNR BM (ret) -has been estimated in a similar way as for the single detector readouts but using the burst averaged residua.
The results are summarized in Table 4.

CO 2 and CH 4 retrieval precisions
The instrument precision which can be reached by a grating spectrometer system can be theoretically evaluated as performed in Sect.4.2.1 or retrieved from real data sets as performed in Sect.4.2.2.Results for both estimates are compared for different operation conditions in Sect.4.3.

Theoretical retrieval precisions
The theoretical retrieval precision of MAMAP for CH 4 and CO 2 was estimated from the corresponding simulated SNR (Sect.4.1.1)and from the instruments spectral resolution and spectral sampling.The solution of the WFMD/M algorithm is based on a least squares approach of the following form: where K denotes the weighting function matrix, y denotes the wavelength dependent difference between measurements and model, and x the parameters to be retrieved.The error is expressed by .With the inverse measurement covariance matrix C −1 y derived for the simulated SNR the weighted least squares solution can be written as: with the corresponding parameter covariance matrix: The diagonals give the variance of the parameters.Hence the simulated profile scaling factor precision PSFP can be calculated (for 1 σ ) as: and Results for the individual PSFP(sim) calculated for different instrument operation conditions are summarized in Table 3. Accordingly, the simulated precision for the profile scaling factor ratios PSFRP(sim) can be calculated from the Gaussian error propagation and the individual simulated PSF precisions -PSFP CH 4 (sim) and PSFP CO 2 (sim) -of each gas: Assuming a Gaussian error distribution for the single measurements obtained by MAMAP, the simulated co-added burst mode (BM) PSFR precision -PSFRP BM (sim) -of each co-added measurement can be calculated from the simulated single readout PSFR precisions -PSFRP(sim) -and the number of co-added measurements as:

Precision estimates obtained using real data
To estimate the instrument precision for a given data set, the individual CH 4 and CO 2 columns were processed for each single measured spectra with the WFMD/M retrieval algorithm as described in Sect.3. To account for small systematic offsets caused for instance by insufficient knowledge of the slit function used for the fit procedure, each series of profile scaling factors (PSF i ) was first normalized for each gas by the mean value of all measurements of the processed data set.The resulting normalized profile scaling factors NPSF i can be calculated as: To account for path differences caused by topography and movements of the plane the (normalized) CH 4 /CO 2 profile scaling factor ratio (PSFR i ) was calculated as discussed in Sect.3: To account for slow SZA and atmospheric variations the CH 4 /CO 2 PSFR were additionally high-pass filtered.The retrieved profile scaling factor ratio precision PSFRP(ret) was then calculated as standard deviation of the PSFR i over the whole investigated data set: This procedure was repeated for a set of measurement sequences for different operation conditions of the instrument  as described in the next section.The obtained results from these sequences are summarized in Table 3 and compared to simulated values (see Sect. 4.2.1).For co-added burst-mode (see Sect. 4.1.1)the retrieved profile scaling factor ratio precision PSFRP BM (ret) has been estimated in a similar way as for the single detector readouts, but using the burst averaged profile scaling factors PSF BM instead.Results for co-added burst mode obtained from one data set are summarized in Table 4.

Single exposure SNR and precision for different operation conditions
To estimate the SNR and the PSFR precision (PSFRP) of the MAMAP instrument under different operating conditions, a data set of seven measurement sequences has been selected, including static on-ground and dynamic in-flight measurements.The in-flight sequences have been subdivided into measurement sequences over surfaces containing primarily homogeneous and surfaces containing primarily inhomogeneous distribution of the measured radiance (see Table 5).These sequences produce primarily homogeneous and variable inhomogeneous illumination conditions on the spectrometers slit, respectively.In addition, sequences of zenith sky in-flight measurements were investigated for which the slit was illuminated uniformly.In the following results obtained for the different operation conditions will be presented and discussed.For all sequences CH 4 and CO 2 SNR values, profile scaling factors (PSF) and associated PSFR precisions have been retrieved for single detector readouts as described in Sect.4.2.For comparison also the associated simulated SNR values and PSFR precisions were computed as described in Sects.4.1 and 4.2.All results of these calculations are summarized in Table 3.By intercomparison of the different results, potential error sources affecting the instrument performance can be identified or excluded.

Ground based measurements
To investigate the instrument's performance under vibrationfree static conditions, two types of on-ground measurements have been performed.In a first set-up the nadir telescopes of the spectrometer have been pointed towards a group of trees.This configuration is referred to as "pseudo" nadir or "sun-illuminated target" (SIT) configuration.The trees were located on the campus of the University of Bremen, in approximately 250 m distance to the MAMAP spectrometer.
In this configuration a measurement sequence was acquired with a single readout exposure time of 148 ms at ∼70% detector saturation.The illumination conditions (signal levels) were nearly constant due to (nearly) clear sky conditions, in contrast to typical in-flight conditions, where the signal varies as a result of changes in the surface albedo/SSR.In a second on-ground set-up, scattered light zenith radiance measurements (measurements of the down-welling diffuse radiance) were performed by pointing the MAMAP zenith telescopes directly into the sky.The data were analyzed using a LOWTRAN aerosol background scenario for the RTM simulation delivering good results for the SNR and precision estimates with the described approach (see Sect. 4.2.2).The single readout exposure time for the acquired sequence was 700 ms at ∼56% detector saturation and clear sky conditions.
The measurements have not been absolutely calibrated but have been corrected for dead and bad pixels and dark signal.The spectra were also normalized using a white light source (WLS) spectrum to account for pixel to pixel gain variations and etalons.The spectrometers were temperature stabilized being heated to ∼26 • C for an outside temperature of ∼15-20 • C.
After processing of both measurement sequences a RMS of the fit residuum (Sect.3) in the range of 0.5-0.6% was achieved (i.e.very similar to the RMS obtained for well filtered in-fight measurements, see Fig. 5).
Figure 8 shows precision and RMS results obtained for the on-ground sequences after processing.On the left, the results from the sun illuminated target (SIT) measurements performed on 11 May 2006 are displayed.On the right, the results from ground-based scattered zenith radiance measurements performed on 10 May 2006 are shown.The top panel shows the normalized PSFR precisions for CH 4 /CO 2 .The second panel shows the measured signal in [BU] (for 16 Bit ADC) with dark signal subtracted.The bottom panel shows the SNR estimate of each measurement for each gas (CH 4 black, CO 2 red symbols) as described in Sect.4.1.2.
It is assumed that the precision and the SNR of the MAMAP instrument under static conditions on ground is primarily dominated by the shot noise of the measured signal, the dark signal shot noise and readout noise of the detector and front-end electronics.Vibrations or changes in the illumination conditions of the spectrometers can be omitted.Therefore good agreement between measurements and model simulations is expected.
For the SIT measurements (as expected) a very good agreement between simulated and retrieved SNR values can be achieved for single detector readouts.The simulated SNR(sim) values for CH 4 and CO 2 are 1584 and 1605 vs. the retrieved SNR(ret) values of 1566 and 1651.The respective CH 4 /CO 2 simulated vs. retrieved PSFR precision for 148 ms exposure time was 0.27% -PSFRP(sim) -vs.0.33% -PS-FRP(ret).Thus also for the precision, good agreement between model simulations and measurements can be obtained (Table 3).
For the scattered zenith sky radiance measurements, the mean SNR is in good agreement (SNR(sim) = 1319 vs. SNR(ret) = 1181 for CH 4 and SNR(sim) = 1338 vs. SNR(ret) = 1290, respectively for CO 2 ).In contrast, the simulated and retrieved CH 4 /CO 2 PSFR precisions deviate by a factor of two (0.33% -PSFRP(sim) -vs.0.63% -PSFRP(ret), for 700 ms exposure time).The origin of this difference is not yet identified.It may possibly be induced by atmospheric variations caused by turbulences and light path differences caused by aerosol scattering inside the measured air masses.Another possibility is absorption of liquid water or ice in aerosol and cirrus clouds, which results in a broad band absorption at the short wavelength end of the channel but is not explicitly accounted for in the retrieval.In general measurements in scattered zenith sky radiance geometry are likely to be more affected by atmospheric variations than measurements in SIT ("pseudo" nadir) geometry where the solar radiation is scattered/reflected primarily by the target opposed to the atmosphere.
Figure 7 shows a time series of SIT measurements under slightly variable atmospheric conditions.On the left results from single readout processing are shown.The top panel shows PSFs obtained for both gases (CH 4 black, CO 2 red symbols).The second panel shows the CH 4 /CO 2 PSFR.The bottom panel shows the measured signal in [BU] with dark signal subtracted.On the right, in analogy to the nominal in-flight burst mode (Sect.4.3.4)results for the same time series with 10 co-added readouts are displayed.For the latter, the simulated burst mode PSFR precision -PSFRP BM (sim) -can be calculated from the single readout precision as (see Sect. 4.2.1): Comparison with the second panel on the right shows reasonable agreement of the PSFRP BM (sim) with the measured values.The systematic impact of atmospheric variability on the CH 4 and CO 2 PSF accuracy variation is attributed to thin clouds.This exceeds the precision of the single measured values.Improving the retrieval to identify and account explicitly for thin cloud effects will further improve the accuracy of the retrieval.

Airborne single readout measurements over homogeneous scenes
To investigate the instrument performance under airborne conditions, the spectrometer rack (containing the SWIR and NIR spectrometer systems) was attached with 6 antivibration mounts to an aluminium aperture plate.This aperture plate contains two 10 mm thick wedged Suprasil aperture windows with a diameter of 180 mm.The aperture plate itself was directly attached with screws to the structure of the plane (Fig. 4).Apart from vibrations, etalons from the Suprasil aperture windows and the spectrometer itself, spectral shifts caused by thermal gradients inside the spectrometers optical bench and effects from inhomogeneous slit illumination influence the in-flight measurements.To separate instrumental and vibration effects from illumination effects, first nadir measurement sequences over water and land with nearly homogeneous distribution of the measured radiance were investigated.It is thereby assumed that measurement sequences with smaller variations of the measured radiance (i.e.detector filling) will produce more homogeneous slit illumination conditions than measurement sequences where strong variations occur.To avoid detector saturation during airborne operation, the detector is operated in nadir mode typically at ∼10-20% of the total full well capacity (∼6000-13 000 [BU] at 16 bit resolution) corresponding to surface albedo/SSR over land in the range between 0.10-0.20.The exposure time of each readout for these albedos/SSRs over land was typically in the range of 60-100 ms, depending on solar zenith angle.Respectively over water (with typical albedos/SSRs of ∼0.01) exposure times in the range between 0.6 s and 1 s were applied.In Figs. 9 and 10 same plots as for ground based measurements (Fig. 8) are shown, but for dynamic in-flight conditions in nadir observation mode.are taken within the United States (7 November 2008).Both measurements are performed on board the AWI Polar-5 aircraft.On the left side of both figures (Figs. 9 and 10) measurement sequences for homogeneous and on the right side measurement sequences for inhomogeneous surface radiance distributions are presented.Beside nadir measurements also airborne zenith irradiance measurement sequences were investigated.For this the instrument observes the upper hemispheric downwelling radiance and solar irradiance through a set of 4 transmissive Spectralon diffuser plates.The incoming radiation is fed from the diffusers to the instrument via glass fibers.Two lenses are imaging the fibers via the zenith optical path (Fig. 1) directly onto the slit of the spectrometer.In this mode of operation, the slit is always homogeneously illuminated and no additional spectral structures (as for instance due to features in the SSR) exist.
On the left, Fig. 11 shows measurement sequences taken by the zenith optical path of the instrument.The measurement is performed on board of a Cessna-207 aircraft.The readout to readout variation of the measured irradiance is very similar to that observed on ground (i.e.SIT measurements, Fig. 8).
From these results it can be concluded that the in-flight measurements were barely affected by vibration effects or fast changing etalons.Assuming concentration changes to only occur in the CH 4 column below the aircraft and the CO 2 column as constant, the total column precision for CH 4 -   CP % (CH 4 ) -can be estimated from the PSFR precision and the conversion factor c for CH 4 (see Sect. 3) as: For an aircraft altitude of 4500 m, SZA of 40 • and an albedo/SSR of 0.01 the CH 4 single readout column precision CP % (CH 4 ) for measurements over water with homogeneous radiance (WHO) can be estimated to ∼0.82% (for exposure time = 1 s, albedo/SSR of 0.18).For the same plane altitude and SZA the single readout CH 4 column precision over land surfaces with homogeneous radiance (LHO) can be estimated to ∼0.78% (for exposure time = 58 ms).An according conversion factor can be applied in case of CO 2 .

Airborne single readout measurements over inhomogeneous scenes
To investigate effects of inhomogeneous illumination of the slit, airborne nadir measurement time series taken over water and land surfaces with inhomogeneous upwelling radiance were analyzed.The simulated and retrieved mean SNR values over water with inhomogeneous radiance deviate approximately by a factor of two (SNR(sim) = 1047 vs. SNR(ret) = 532 for CH 4 and SNR(sim) = 1063 vs. SNR(ret) = 598 for CO 2 ).The according PSFR precisions were 0.42% (simulated) vs. 2.74% (retrieved) and deviate by a factor of ∼6.5.
The degraded performance under inhomogeneous illumination conditions is attributed primarily to smile and keystone effects of the spectrometer system's optical bench, combined with an inhomogeneous illumination of the slit.This assumption is supported by ZEMAX ® optical design program end to end simulations of the optical system of MAMAP, showing that inhomogeneous slit illumination leads to variations of the slit function shape and position.Such variations induce errors in the retrieval.Other factors like even-odd effects of the used linear InGaAs detector also caused by inhomogeneous illumination (i.e.even-odd effects due to tilted detector illumination) can be accounted for by the WFMD/M algorithm and are believed to play a minor role.Effects such as small spectral features of the earthshine spectral reflectance can also not be completely excluded as reason.All single readout time series results obtained for the different operation conditions are summarized in Table 3.
To minimize effects of inhomogeneous illumination of the slit, a modification of MAMAP's optical bench has been initiated.After this modification it is expected that the instrument will reach the same or similar performances as over surfaces with homogeneous radiance distribution.
Assuming concentration changes to only occur in the CH 4 column below the aircraft and the CO 2 column as constant, the CH 4 single readout total column precision for a plane altitude of 4500 m, SZA of 40 • , albedo/SSR of 0.01 and exposure time of 1 s can be estimated to be ∼2.00%(see Table 2 for conversion factors) for water with inhomogeneous radiance (WIH).For the same plane altitude and SZA the single readout CH 4 column precision over land surfaces with inhomogeneous radiance (LIH) can be estimated to be ∼1.98% of the total column (for exposure time = 58 ms and albedo/SSR ∼0.18).

MAMAP nominal co-added burst mode
In this section, the MAMAP precision for the nominal burst mode of operation over land targets will be discussed.Over land targets typically bursts of 10 measurements were acquired and co-added to one measurement (see Sect. 2.1) to reach an appropriate SNR (i.e.SNR ≈ 1000).The retrieved burst mode SNR -SNR BM (ret) -and burst mode PSFR precision (PSFRP BM ) have been estimated in a similar way as for the single detector readouts (see Sects.4.2 and 4.3), using the burst averaged residua and PSFR values, respectively.
The simulated and retrieved burst mode mean SNR values and PSFR precisions are summarized in Table 4.These were calculated for the same measurement series over land surfaces with inhomogeneous radiance (LIH) as described for single readouts in Sect.4.3.3.In this section, only the worst case scenario (i.e.precision over surfaces with inhomogeneous radiance distribution) is investigated.
Assuming a Gaussian error distribution for the measurements, the burst mode PSFR precision and burst mode SNR can be derived for inhomogeneous targets (LIH) also in a indirect way.This was done by multiplication (i.e. for SNR calculation) or division (i.e. for precision calculation) of the retrieved single readout SNR (SNR SR (ret)) and single readout precision values (PSFRP SR (ret)) (Sect.4.3.3)with the square root of 10 (for n = 10 measurements per burst).
The resulting indirectly derived mean burst mode SNR (SNR BM From the comparison of both directly derived and indirectly calculated SNR values and PSFR precisions (SNR BM (ret) vs. SNR BM (ret) and PSFRP BM (ret) vs. PSFRP BM (meas) ) with the simulated values it is evident that random Gaussian error distribution can not be assumed for the single measured spectra.
This result supports the attribution of the random PSF errors observed with MAMAP to the inhomogeneous illumination of the spectrometer's slit and the variations in slit function position and shape, caused by keystone and smile effects.Ground structures like edges parallel to the flight direction can produce similar deviations in all 10 measurements of one burst.Therefore random Gaussian error distribution of the retrieved column errors can not be assumed.Spectral features in the earthshine spectral reflectance cannot be completely excluded either.In contrast, it is assumed that shot noise and noise of the readout electronics and the detector should produce random Gaussian error distributions like demonstrated for ground based measurements.
From these findings it can also be expected that the undergoing modification of the optical bench of the MAMAP instrument for reducing inhomogeneous slit illumination effects can lead to a significant improvement of the instrument's burst mode SNR and precision characteristics.Assuming that the instrument can then reach SNR and precision values similar to those achieved for single readouts over homogeneous land targets (LHO, Table 3), it can be estimated for burst mode, that SNR values in the order of SNR BM (meas) = 538• √ 10 = 1713 for CH 4 and SNR BM (meas) = 588• √ 10 = 1859 for CO 2 remain feasible.The according feasible burst mode PSFR precision may reach a factor of ∼5 better values than actually achieved (i.e. up to PSFRP BM (ret) = 1.10%/ √ 10 = 0.35%).Assuming again concentration changes to only occur in the CH 4 column below the aircraft (and the CO 2 column as constant), the actual MAMAP CH 4 co-added burst mode total column precision (CP BM,% ) over inhomogeneous land surfaces (LIH) can be estimated from the PSFR precision of ∼1.74% to be ∼1.23%(1 σ ) for a plane altitude of 4500 m, integration times of 580 ms, SZA of 40 • and an albedo/SSR of 0.18.For a background concentration of 1774 ppb this corresponds to an enhancement of 21.8 ppb of the total column.
By calculating the equivalent total column light path L e through the atmosphere as L e = g/(ρg) ≈ 8576 m (for p = 1013 hPa, ρ = 1.2041 kg m 4 , g = 9.81 m s −12 ) the total column precision of MAMAP is converted to ppm m for comparison with other instruments (SZA 40 • , albedo/SSR 0.18): For a typical SZA of 40 • the estimated column precision of 187 ppm m is inside the instrument threshold requirement, which is to measure the total column concentration in nadir with a precision of 1-2% with respect to the atmospheric background, corresponding to a precision of ∼150-300 ppm m.For a lower plane altitude of 850 m the 1 σ precision becomes 142 ppm m.
After the planned modification of the instruments optical bench to account for inhomogeneous slit illumination, a precision of the total column for a plane altitude of 850 m and integration time t int = 600 ms of approximately = 4.22 ppb (≈ 33 ppm m) is predicted.
In that case the MAMAP CH 4 total column uncertainty variation will not be limited by the precision of the instrument.For a total column precision below 0.5%, it can be expected, that the accuracy variation induced by atmospheric effects (i.e.light path differences for CH 4 and CO 2 caused by scattering and absorption of aerosols and clouds and variations of the albedo/SSR and refractive index of the atmosphere) will dominate the overall uncertainty variation.

Spectral stability
To investigate further the attribution of the degradation of the precision of MAMAP data products between inhomogeneous and homogeneous ground scenes to the inhomogeneous slit illumination and related model simulations, the short term spectral shifts in the observations over surfaces having respectively homogeneous and inhomogeneous radiance were analysed.It is assumed that any impact of the aircraft vibration for the different measurement series are similar.A comparison of short term spectral shifts derived by non-linear least squares fitting for inhomogeneous and homogeneous water surfaces shows a change of up to a factor of two: spectral shifts being 0.0049 nm for homogeneous scenes as compared to 0.0101 nm for inhomogeneous scenes above water.For land surfaces a similar deviation was observed: the spectral shift being 0.0180 nm for homogeneous compared to 0.0364 nm for inhomogeneous land surfaces.The different exposure times of 1 s over water and 58 ms over land and the difference in inhomogeneity distribution between land and water surfaces are attributed to the differences in the observed shifts.Thus the factor of two decrease in spectral stability in both cases (for land and for water) is consistent with a degraded performance of MAMAP being explained by inhomogeneous illumination effects of the spectrometers slit.For comparison, additionally to the airborne nadir measurements spectral shift results from ground based radiance and airborne zenith sky irradiance measurements are summarized in Table 6.MAMAP achieves for the fastest exposure times over homogeneous land targets (LHO), where the spectral shifts are expected to be dominated primarily by vibrations, a spectral stability of ∼1/46 of the FWHM of 0.82 nm (i.e.close to the required value of 1/60 of the FWHM).

MAMAP targets
Achieving 1-2% total column precision (and relative accuracy) for the data products XCH 4 and XCO 2 from a single ground scene measurement (corresponding to 150-300 ppm m -1 σ -column change for CH 4 and 33 000-66 000 ppm m for CO 2 below the aircraft; see Sect.4), is challenging for an airborne passive SWIR remote sensing instrument.Nevertheless, such a performance puts limitations on the target emissions which are suitable to be detected.In the following we will make an estimate based on the actually achieved MAMAP total column precision of ≈ 1%.Thereby it will be assumed, that a total column relative accuracy (neglecting a constant bias) of < ≈ 1% can be achieved on small scales (i.e.several kilometers) for clear sky atmospheric conditions when no aerosols are produced by the source itself.The assumption is justified by the theoretical error analysis performed on XCH 4 SCIAMACHY data (see for WFM-DOAS XCH 4 from SCIAMACHY also Schneising et al., 2008Schneising et al., , 2009;;Schneising, 2009).A detailed relative accuracy analysis including also aerosols produced by the source itself will be presented in a separate publication (Krings et al., 2011).
Because of the large background concentrations of CH 4 and CO 2 , an emission source must have an appropriate emission strength and horizontal extent in order to build up a column enhancement which can be detected.
In the following we discuss the conditions under which MAMAP can obtain information about supposed target emissions.For this purpose we use a simple model to relate the surface flux F of a given gas to the relative change of its vertical column V /V over a distance l along the main wind direction.In this section two cases are considered: (i) a range of targets from a well isolated source region of the size of a MAMAP footprint up to an area observed by MAMAP in several minutes, i.e. ∼25 m up to a few 10 km.Possible targets are landfills, seeps, fugitive emissions of gas/oil industry, power plants, steelworks, coal mines and (mud) volcanoes, (ii) an effectively homogeneous source region with an extension larger than ∼50 km such as extended wetlands.
For an isolated source region, time independent (static) meteorological conditions can be assumed as a first approximation for short periods of time where measurements are taken over and near the source.These conditions are characterized by clear sky over the measurement area, small atmospheric variations of the cloud and aerosol optical thickness, small solar zenith angle variations and nearly constant wind speeds.For such conditions, systematic accuracy variations in the CH 4 /CO 2 column caused by light path differences induced primarily by an inadequate modelling of the scattering of electromagnetic radiation by aerosols and optically thin cirrus clouds (Schneising et al., 2009;Schneising, 2009) is assumed to be small and was hence neglected in a first approximation (Krings et al., 2011).Depending on source size and the assumption that background aerosol and cirrus clouds are smoothly varying, the impact of slow systematic changes in the accuracy of the measured CH 4 /CO 2 column mixing ratio can optionally be further minimised by high-pass filtering the data.Strong local sources produce short term or small scale concentration changes for the flight path compared to changes produced by variations in background aerosols and (thin) cirrus clouds.For such conditions it can be assumed that the detection limit of MAMAP is primarily dominated by the instrument precision rather than by variation of the accuracy when no aerosol is produced by the source itself.
Assuming a constant wind speed u in the horizontal +x direction in the layer of interest and a mean flux F as first approximation, the resulting enhancement V of the vertical column V can be estimated as follows: V = F •t, where t is the accumulation time, which characterizes the time available for an air column to accumulate CH 4 when it moves over the target.For a target with extension l it follows t = l/u.The relative increase of the vertical column over the target is given by V /V = F /V •l/u.The smallest detectable flux F min for a given situation is then given by: For the estimation we assume a constant wind speed of u = 2 m s −1 and all concentration changes to occur below the aircraft, a horizontal extent of the emission source of l ≈ 400 m (e.g. a landfill) and the required total column enhancement V /V equal to ∼1% total column precision (∼150 ppm m for CH 4 and 32 600 ppm m for CO 2 ) of MAMAP for inhomogeneous scenes.With a typical value for the CH 4 background column of V = 10 g CH 4 m −2 (3.6 × 10 19 molecules cm −2 ) the smallest detectable flux F min can then be calculated to be 1.8 g CH 4 m −2 h −1 .At a constant wind speed of 2 m s −1 the product of flux F and source extension l is the main parameter which can be calculated for 1% CH 4 enhancement to be F min •l = 0.01•10 g CH 4 m −2 •2 m s −1 = 0.2 g CH 4 m −1 s −1 .Equivalent calculations can also be performed for CO 2 .
Over the globe, many different sources exist with fluxes exceeding the above calculated detection limits.For instance CH 4 emissions of landfills with organic waste and temporal coverage, equipped with gas recovery can reach mean CH 4 fluxes in the order of 2-4 g CH 4 m −2 h −1 (Fellner et al., 2003;Börjesson et al., 2000), i.e., larger than the actual detection limit of MAMAP.Furthermore Judd (2004) report CH 4 fluxes between 148 and 445 g CH 4 m −2 h −1 for the Coal Oil Point (COP) seep measured with tents for an area of 1800 m 2 near Santa Barbara, California.Miscellaneous strong dry seeps and mud volcanoes exist also in other parts of the world for instance at the Black Sea shelve offshore Georgia (Judd, 2004), in eastern Azerbaijan (Etiope et al., 2004) and Indonesia (Chakraborty and Anggraini, 2009).Strong seepage can also occur over shallow or submerged gas hydrates in arctic regions (Shakhova et al., 2010;Bowen et al., 2008).Other localized sources like fugitive emissions as the result of oil and gas well exploration and utilization by oil and gas industry can produce atmospheric fluxes strong enough to be detected with MAMAP as well.Jagovkina et al. (2000, and references therein) estimated for an area of ∼ 1.8•10 10 m 2 near Yamal in Russia a mean flux of 1-2 g CH 4 m −22 d −1 .The corresponding detection limit of MAMAP for such an area (e.g.extend of l ≈ 50 km) can be calculated to be in the range of F min ≈ 0.35 g CH 4 m −2 d −1 which is well below the reported values.Similar estimates can be performed for refineries and gas processing plants with fugitive emissions in the range of ∼140-300 kg CH 4 h −1 (Chambers and Strosher, 2006a,b).Emissions from oilsands tailings settling basins with flux estimates between 0.1-4.8g CH 4 m −2 h −1 (Siddique et al., 2008) also outrun the smallest detectable fluxes.Beside CH 4 also anthropogenic and geologic CO 2 emission from volcanoes (Mörner and Etiope, 2002), coal fired power plants, steelworks, cement production etc. (EPER, 2004) clearly exceed the detection limit of MAMAP.
These estimates show that MAMAP has the potential to detect strong local CH 4 and CO 2 emissions and corresponding gradients as shown in the next section.Under certain circumstances (i.e.knowledge of wind) also corresponding fluxes can be more accurately estimated or constrained when appropriate patterns are flown (Krings et al., 2011) as current flux uncertainties for several anthropogenic and geologic sources can reach values of up to 50-100% (NRC, 2010).After finalizing the modification of the optical system of MAMAP, it can be expected that the smallest detectable flux limit F min can be improved significantly.In addition appropriate flight strategies allowing further averaging of the observations can also improve the precision.Thus detection of weaker localized sources can be expected in the future.
Many important CH 4 sources emit significantly smaller fluxes of CH 4 compared to the values reported above (for localized sources).Siberian wetlands emit typically in the order of ∼20 mg CH 4 m −2 d −1 on average (Sachs et al., 2008) up to ∼200 mg CH 4 m −2 d −1 for summer seasons (Bohn et al., 2007).The approach used above can also be applied to estimate the detection limits required for extended regions of less intense source emissions assuming that the region is sufficiently homogeneous.For a strong summer CH 4 flux of F = 200 mg CH 4 m −2 d −1 for Western Siberian wetlands, a constant wind speed of u = 2 m s −1 , and a minimum detectable column enhancement of V /V = 1% (corresponding to 150 ppm m below the aircraft), a required accumulation distance of l ≈ 86 km can be calculated.In order to estimate gradients typically a larger distance is required (i.e.min.3 times of the accumulation distance, ∼250 km).This simple method to estimate the range of expected CH 4 total column changes emanating from these type of sources requires sufficiently stable conditions during the period of the aircraft measurement.When this requirement is not fulfilled, more complex regional chemical transport modelling (Jagovkina et al., 2000(Jagovkina et al., , 2001) ) is needed.It has to be noted that for extended sources the smallest detectable flux F min is restricted primarily by accuracy variations (i.e. the relative accuracy) and not by the precision of the instrument.A detailed discussion on the impact of aerosols, albedo, thin clouds and other effect on the accuracy/relative accuracy of MAMAP data products will be given in Krings et al. (2011).

First results from measurements over localized emissions sources
In order to test the MAMAP sensitivity to score emissions and to compare with results obtained in Sects.4.3.1 and 5 flights over localized targets have been performed in summer 2007.In the following, results from flights over a target with poorly known CH 4 emission rate (i.e. a landfill) and targets with well characterized CO 2 emission rates (i.e. two power plants) are presented.The flights were performed with a Cessna 207 aircraft, operated by the Free University of Berlin.The flights over the well characterized CO 2 targets focused on the coal fired power plants Jänschwalde and Schwarze Pumpe (located near Berlin, Germany).In the following we present results obtained with the SWIR channel for nadir observations only.Zenith observations are to be included in the retrieval algorithm in a subsequent studies.
The measurements were analyzed with the WFMD/M retrieval algorithm (see Sect. 3) using radiative transfer simulations performed with the SCIATRAN radiative transfer model (Rozanov et al., 2005).
The conditions of flight were such that the aircraft flew at 850 m altitude performing direct nadir observations.By computing the CO 2 /CH 4 ratio of the measured and retrieved profile scaling factors for the CO 2 sources (assuming that CH 4 is constant) and vice versa (CH 4 /CO 2 ) for the CH 4 source (assuming that CO 2 is constant), any line-of-sight errors, induced by rolling of the aircraft and lack of knowledge of altitude/ground distance were neglected (see Sect. 3).Each flight of each day was processed for one fixed altitude and one fixed solar zenith angle.Systematic effects in the measured columns caused by solar zenith angle changes, changes in altitude and low frequency cirrus cloud variations are minimised by applying an 80 point high-pass filter to the co-added burst mode data (see previous section).Compared to the extend of the investigated sources this filter removes primarily low frequency variations not caused by the emissions of the source.
For the RTM simulation, temperature, pressure, and water vapour vertical profiles corresponding to the US Standard Atmosphere, a constant albedo/SSR of 0.18 and a solar zenith angle of 40 • (as calculated for the time of the overfligts) have been used.Clear sky (cloud free) conditions have been assumed although some partial cirrus covers have been reported during the flights.
Figures 12 and 13 show retrieved and normalized CO 2 /CH 4 PSF ratios of MAMAP measurements of the power plant overflights near Berlin performed on 26 July 2007.
The target shown in Fig. 12 was the power plant Jänschwalde, operated by Vattenfall.This coal-fired power plant emits approximately 24.9 Mt CO 2 yr −1 (EPER, 2004).On 26 July 2007 the plant emitted 56.6 t CO 2 min −1 (Dietmar Heinze, Vattenfall Europe Generation AG & Co. KG, Cottbus, Germany, personal communication, 2008).We estimated (depending on wind speed) that this emission roughly corresponds to a CO 2 total column increase (which is about 8 × 10 21 molecules cm −2 for a surface pressure of about 1000 hPa and a CO 2 mixing ratio of 380 ppm) of few percent over and near the power plant (see Bovensmann et al., 2010).
The flight pattern has been chosen such that the aircraft was crossing several times the plume.The wind direction, which was almost perpendicular to the flight track, was clearly visible by the small steam clouds over the cooling towers (see photo Fig. 12).As can be seen in Fig. 12, elevated atmospheric CO 2 originating from the power plant and transported in wind direction, can clearly be detected with MAMAP.The elevated CO 2 is readily observed in the small map (on the right) showing the CO 2 PSF retrieved by the WFMD/M algorithm.The CH 4 PSF (the small map below) does not show such a clear pattern.This is as expected as there are no known local strong sources of CH 4 near the power plant.Also as expected, the ratio of the CO 2 to CH 4 profile scaling factors produce a smoother pattern, as light path errors due to, e.g., not yet considered changes of the aircraft rolling and distance to ground, cancel to a large extent when the CO 2 /CH 4 PSF ratio is computed.The approximately 3% enhancement in the CO 2 /CH 4 PSF ratio over the power plant shown in Fig. 12 is attributed to elevated CO 2 .Assuming the changes of CH 4 as small and CO 2 variations to occur only below the aircraft the total column increase C TC of CO 2 is estimated from the 3% enhancement of the PSF ratio by: which is in agreement with values obtained by Gaussian plume model simulations (Krings et al., 2011;Bovensmann et al., 2010).These results show that MAMAP reaches the initial sensitivity requirements very well.For more details on power plant emission measurement with MAMAP and associated inversion of fluxes see also Krings et al. (2011).
Figure 13 shows similar results as Fig. 12 but for overflights at the power plant Schwarze Pumpe, which has an output of approximately 10.9 Mt CO 2 yr −1 (EPER, 2004).On the 26 July 2007 Schwarze Pumpe had an output of 26 t CO 2 min −1 as reported by Vattenfall.The wind speed during both overflights was in the range of ∼2.5-5.0 m s −1 .The figure shows that enhanced CO 2 values can clearly be observed downwind of the source with MAMAP.
Figure 14 shows an overflight transect measured over the landfill Vorketzin on 26 July 2007.During the transect wind speed was measured nearby the landfill.The mean wind speed was estimated to be in the order of 3 m s −1 from south, south-south-west direction.The path length of the accumulation of the air-mass over the landfill body is estimated to about ∼450 m.
The anomaly in the retrieved normalized CH 4 /CO 2 PSF ratio during the transect was in the range of +1-2% (Fig. 14).With a SZA of ∼ 40 • and an aircraft altitude of ∼850 m and the assumption that the observed anomaly (of 2%) is mainly due to the increase of the CH 4 concentration below the aircraft, the corresponding enhancement in total column (for albedo/SSR of ∼0.18) can be estimated to: = 0.0107 ≈ 1.11% total column increase Assuming movement of the air-mass with constant speed (3 m s −1 ) over the landfill body and that the adjacent wet-lands produce much less CH 4 (typically <60 mg CH 4 m −2 d −1 ) than the landfill body itself, the CH 4 total column increase of ∼0.56-1.11%(after 450 m) leads to a rough estimate of the mean emission rate for the upper central landfill area within ∼1.24-2.48g CH 4 m −2 h −1 or larger.For similar types of landfills (with organic waste, equipped with temporal covers and gas recovery systems) mean flux rates to the atmosphere in the order of 2-4 g CH 4 m −2 h −1 are reported (Fellner et al., 2003;Börjesson et al., 2000) as discussed in the previous section.
To verify the CH 4 /CO 2 proxy approach for estimating landfill emissions and exclude any impact from landfill CO 2 emissions on the retrieved XCH 4 total column increase it was assumed that the estimated CH 4 fluxes of ∼1.24-2.48g CH 4 m −2 h −1 were induced only by 25% of the total landfill CH 4 production.Thus the fraction of 3.72-7.14g CH 4 m −2 h −1 was oxidized and emitted as CO 2 corresponding to a CO 2 flux to the atmosphere of 10.21-20.42g CO 2 m −2 h −1 .With a wind speed of 3 m/s and a CO 2 atmospheric background concentration of 8.192 × 10 21 molecules cm −2 it can be estimated, that the resulting CO 2 total column increase will be <0.01%after 450 m and therefore could be neglected in the calculation.
To also exclude albedo/SSR dependent offsets as origin of the CH 4 column increase, an empirical assessment of the data from the whole flight has been performed.From the assessment a linear equation was derived by linear regression as first approximation of the dependency between the retrieved difference ( PSFR(CH 4 /CO 2 )) of the normalized CH 4 /CO 2 PSF ratio, and the variation of the upwelling radiance R in binary units (BU).The correlation coefficent of the linear fit to the measured data (≈ 0.2) was not very high but gives a first guess for the signal (and also for the albedo/SSR) dependency of the retrieved CH 4 /CO 2 PSFR also including other signal dependent effects like nonlieratity of the detector.
The difference in the measured upwelling radiance during the landfill transect was in the order of R ≈ 2000-6000 BU (centre -side variations during the transect).Assuming that the observed radiances R (and corresponding detector fillings) are proportional to variations of the albedo/SSR, the resulting absolute value of the albedo/SSR dependent offset of the normalized CH 4 /CO 2 PSF ratio for the landfill transect can be estimated to be: PSFR(CH 4 /CO 2 ) ≈ 0.076-0.228%.The albedo/SSR dependent total column variation ( AC TC,% ) is then calculated to ≈ 0.041 − 0.122% offset of the total column.Thus the albedo/SSR dependent variation in the retrieved CH 4 total column is smaller than the retrieved total column increase of ∼0.56-1.11%CH 4 during the transect.

Summary
The Methane Airborne MAPper (MAMAP) is a new type of passive airborne remote sensing instrument, which measures the back scattered electromagnetic radiation in the spectral regions of the CH 4 , CO 2 and O 2 atmospheric absorptions.On inversion using differential optical absorption spectroscopy (DOAS), the reflected (and scattered) solar radiation in the NIR and SWIR spectral range yield the column and dry column amounts of CH 4 and CO 2 and the ratio of CH 4 to CO 2 .The instrument has been designed for flexible operation on board of several airborne platforms (e.g., Cessna 207, Cessna Caravan and AWI Polar-5 aircraft).
The spectral resolution of the instrument is ∼0.82 nm for CH 4 and CO 2 detection (between 1590 nm and 1690 nm) and 0.46 nm for the detection of O 2 (∼760 nm).In the current version of the WFMD/M retrieval algorithm simultaneously retrieved CO 2 columns are used to estimate the dry CH 4 air columns and to account for line-of-sight errors.For overflights over strong CO 2 sources (i.e.power plants) also the vice versa approach is applied and the retrieved dry CO 2 air columns are calculated by normalization with the respective measured CH 4 air columns.To test the instrument performance different ground based and airborne measurements on different aircrafts and on the campus of the university of Bremen have been performed.The "on-ground" tests demonstrate that the instrument is able to measure and retrieve with high precision CH 4 /CO 2 profile scaling factor ratios (PSFRs) in sun illuminated target (SIT) "pseudo" nadir configuration.
In this measurement geometry the instrument observed the radiance of a group of trees illuminated by the Sun.For these measurements a precision of ∼0.33% of the CH 4 /CO 2 PSF ratio (at 0.148 s exposure times) can be achieved for each single detector exposure.The precision is defined as the standard deviation (1 σ ) of the retrieved PSF ratios.The according SNR for each single exposure was in the range of SNR ≈ 1600 (Sect.4.4).This is in good agreement with model simulations where shot noise, dark signal shot noise and readout noise of the detector are the main contributors.
A second ground test demonstrated that the instrument is able to directly measure scattered down-welling hemispheric radiance.With a spectrometer f-number of f/3.9, single detector exposure times in the order of 0.7-1 s have been achieved in zenith sky geometry.For these measurements a single readout precision of 0.63% of the retrieved CH 4 /CO 2 PSF ratios and SNR values of ∼1200 have been accomplished.These values were close to the results obtained by instrument model simulations (Sect. 4.4).
It was demonstrated that at an altitude of 4500 m MAMAP reaches an average single readout precision of the retrieved CH 4 /CO 2 PSF ratios in the range of ∼1.10% (for 0.058 s exposure time) over land targets with homogeneous radiance distribution.Assuming that changes within the concentration only occur in the CH 4 column below the aircraft, this corresponds to a ∼0.78% CH 4 total column variation.Over water (at 4500 m altitude, albedo/SSR 0.01) with homogeneous radiance an average single readout precision of the retrieved CH 4 /CO 2 PSF ratios of 1.12% (for 1 s exposure time) can be obtained (corresponding to ∼0.82% CH 4 total column variation).Both PSFR precision estimates are very close to the simulated values of 0.71% for homogeneous land (LHO) and 0.83% for homogeneous water targets (WHO) as summarized in Table 3.The MAMAP precision over these types of targets is limited primarily by shot and detector noise and not affected by vibrations.The estimated airborne shortterm spectral stability for these targets is well within the requirements, even though the stability is a factor of ∼5.5-20 decreased compared to ground based observations (see Table 6).
For airborne measurements over targets with inhomogeneous surface radiance, the instrument achieves a single readout precision of the retrieved PSF ratios of 2.8% (at 0.058 s exposure time) for land (LIH) and 2.74% (at 1 s exposure time) for water (WIH).Assuming changes in the CO 2 column as constant and CH 4 changes to occur only below the www.atmos-meas-tech.net/4/215/2011/Atmos.Meas.Tech., 4, 215-243, 2011 aircraft (altitude of 4500 m), this corresponds to singe readout total column precisions of ∼1.98% for land (albedo/SSR ∼0.18) and ∼2.0% for water (albedo/SSR ∼0.01).These estimated precisions are a factor of 3.9 (land)-6.5(water) lower compared to model simulations.The degraded performance for inhomogeneous illumination is attributed to smile and keystone effects of the spectrometer system's optical bench, combined with the resulting inhomogeneous illumination along the slit.Minor effects, like small spectral features in the earthshine spectral reflectance, can not be completely excluded as a reason.The assumption that the inhomogeneous slit illumination is primary responsible for the degraded performance was supported and confirmed by ZEMAX ® optical design program's end-to-end simulations of the optical system of MAMAP.The assumption is furthermore also supported by the fact, that degraded short-term spectral stability over targets with inhomogeneous radiance can be observed, as compared to those with homogeneous radiance.The origin of this stability degradation can not be traced back to vibrations or thermal changes of the optical bench.A modification of the spectrometer's optical bench incorporating a specially designed spatial scrambler unit is proposed to reduce spectral shifts and slit variations.It is expected that the precision for the CH 4 /CO 2 PSF ratio for each single readout can be reduced from currently 2.8% to less than 1.5%.Under the assumption that the precision is barely affected by small spectral structures of the SSR, single readout precisions for the PSFR in the order of 1% remain feasible as can be demonstrated for homogeneous targets (i.e.PSFRP = 1.1% for LHO and 1.12% for WHO, corresponding to CH 4 total column precisions of ∼0.75% for LHO and ∼0.94% for WHO).
For airborne operation in co-added burst mode (BM) the instrument achieved a CH 4 /CO 2 profile scaling factor ratio precision PSFRP BM over land targets with inhomogeneous radiance (LIH) of ∼1.74% (10 measurements co-added) as summarized in Table 4.For a flight altitude of 4500 m this corresponds to a ∼1.23%CH 4 total column precision (∼190 ppm m), assuming all changes below the aircraft and CO 2 as constant.
For the total co-adding time of 0.6-0.8s and a cruise speed of ∼200 km h −1 , ground scenes lengths along the flight track (LT) of 33-44 m are achieved.Overall the instrument achieves the target precision of 1-2% total column (corresponding to 150-300 ppm m below the aircraft) for the target ground scene lengths of <200 m (LT) over land for typical albedos/SSR of ∼0.18.After modification of the optical bench, total column precisions 1% for ground scene lengths (LT) < 200 m are predicted.In that case it can be assumed that the MAMAP CH 4 total column uncertainty variation will no longer be limited by the precision of the instrument but by the relative accuracy.
In 2007 several flights were performed over anthropogenic targets.It has been demonstrated that MAMAP is able to measure elevated levels of CO 2 downwind from coal-fired power plants.Flights over a landfill with organic waste indicated anomalies in the retrieved CH 4 /CO 2 PSFRs (see Sect. 6) within the range of ∼ 1-2%, corresponding to a ∼0.56-1.11%concentration increase of the total CH 4 column below the aircraft.From this measured anomaly and by knowledge of the wind speed, estimates of the expected fluxes were made using a simple model.With these calculations it can be estimated that the mean emission rate of the landfill for the upper central area must be in the range of 1.24-2.48g m −2 h −1 or larger (see Sect. 6).
Using models of the emission, it can also be demonstrated that the achieved instrument precision of ∼1% total column at the high spatial resolution enables the CH 4 emissions from strong local sources to be quantified.Such local sources comprise geological sources such as dry seeps and mud volcanoes, the destabilization of shallow gas hydrates, anthropogenic emissions from landfills with organic waste and fugitive emissions from oil and gas industry (i.e.well drilling and abandoned gas wells, oil sand tailings settling basins, emissions from gas and oil processing and gas compression and transport).In addition strong local CO 2 sources such as coal-fired power plants and direct and sub areal emissions from volcanoes can be measured and characterized.Measurements of the emissions from strong and large areal sources such as rice paddies, tropical and Siberia wetlands, will become feasible for periods of large emissions but requires appropriate weather conditions, flight patterns and data averaging strategies.Under stable atmospheric conditions, MAMAP measurements (obtained from SWIR data only) can potentially be used for micro-, meso-and synoptic scale validation of CH 4 /CO 2 column ratios obtained from daily CH 4 and CO 2 chemical transport model simulations, and for validation of satellite measurements.However this first requires validation of the MAMAP measurements itself.
Furthermore, MAMAP also serves as a test bed for future high spatial resolution greenhouse gas imaging sensor developments for airborne and space instrumentation, as for example the CarbonSat concept (Bovensmann et al., 2010).

Fig. 1 .
Fig. 1.Sketch of the MAMAP SWIR and NIR spectrometer modules.Both spectrometers have two separate light intake telescopes pointing towards nadir and zenith-sky directions for measurements of nadir and zenith radiances.A mirror enables switching between both modes.For zenith irradiance measurements zenith optics can be equipped optionally with glass fibers and transmissive diffuser optical inlets (not shown in this sketch).

Fig. 4 .
Fig. 4. The MAMAP spectrometer rack mounted on the apertureplate (front) carrying the optics and the spectrometer systems and the MAMAP auxiliary rack (back) carrying the controllers, power converters and the buffer battery (both racks mounted on the RWE Cessna caravan aircraft).
34 • across track (CT) × ∼0.02 • along track (LT) IFOV: ∼5.85 • across track (CT, divided into 85 pixel) × ∼0.072 • along track (LT) Spatial resolution: At 1 km flight height, ground speed 200 km h −1 the co-added ground pixel size is ∼33 m along track over land (surface albedo/SSR 0.18) and larger for lower albedo/SSR surfaces.Across track the pixel size is ∼ 23.4 m Spatial resolution: At 1 km flight height, ground speed of 200 km h −1 the co-added ground pixel size is ∼33 m along track over land (surface albedo/SSR 0.18) and larger for lower albedo/SSR surfaces.Across track the pixel size is ∼102.2m (divided by 85)

Fig. 5 .
Fig. 5. WFM-DOAS fits in the spectral regions used for CH 4 (left) and CO 2 (right) retrieval.The MAMAP spectral measurements have been made on 7 November 2008 during the flight from Oshawa to Wilmington, USA, on board of the Polar-5 aircraft of the Alfred-Wegener-Institute (AWI), Bremerhaven, Germany.Left:The top panel shows a MAMAP nadir spectrum (grey symbols) and the solid line the fitted linearized radiative transfer model.The bottom panel shows the fit residuum, which is the difference between measurement and simulation after the fit (the root-mean-square, RMS, of the fit residuum, RES, is 0.64%).The second panel shows details of the methane fit.The solid line is the scaled derivative of the radiance with respect to a change of the methane vertical column.The retrieved scaling factor for the methane vertical profile is 0.989 ± 0.014, i.e., the retrieved columns is 1.1% lower than the vertical column which has been assumed for the radiative transfer simulations.The grey symbols show the "methane fit residuum", which is identical with the black curve except that the spectral fit residuum has been added.The third and the fourth panel show the corresponding results for the interfering gases CO 2 and H 2 O. Right: similar as left figure but for the CO 2 fitting window.

Fig. 6 .
Fig.6.Averaging kernels for the MAMAP instrument depending on solar zenith angle (SZA) and surface albedo/SSR for a nadir measurement at 850 m flight altitude.The figure shows the striking difference of the averaging kernels below the instrument compared to above the instrument.This difference is due to the fact that for a nadir viewing instrument light coming from the sun passes through the absorber below the aircraft twice (once before and once after reflection at the surface).Left: averaging kernels for CH 4 ; right: averaging kernels for CO 2 .

Fig. 7 .
Fig. 7. Analysis of time series of MAMAP single detector readout CH 4 and CO 2 measurements (Left: for the 11 May 2006, Bremen stationary "on-ground" sun illuminated target, SIT, measurements, right: for the 10 May 2006 "on-ground" zenith sky scattered radiance, ZRG, measurements).Left: top panel: normalized CO 2 /CH 4 profile scaling factor ratios (PSFR i ).Second panel: maximum radiance (detector filling) in binary units [BU] as measured with MAMAP and digitalized with 16 Bit ADC.Bottom panel: estimated signal-to-noise ratios (SNR) of the single measurements.Right: as on left side but for the zenith sky scattered radiance measurements (only measurements with >3000 [BU] are processed).

Fig. 8 .
Fig. 8. Analysis of time series of MAMAP CH 4 and CO 2 measurements for slightly variable atmospheric conditions due to thin clouds for the 11 May 2006 stationary "on-ground" sun illuminated target (SIT) measurements.Left: top panel: normalized single readout CH 4 (black) and CO 2 (red) profile scaling factors (PSF).Second panel: normalized single readout CH 4 /CO 2 profile scaling factor ratio (PSFR).Bottom panel: maximum radiance (detector filling) in binary units [BU] as measured with MAMAP and digitalized with 16 Bit ADC.Right: as on left side but for 10 readouts co-added to one measurement.

Fig. 9 .
Fig. 9.As Fig. 8 but for in-flight nadir measurements over water; left: scene with homogeneous reflectance distribution (target WHO); right: scenes with inhomogeneous reflectance distribution (target WIH).

Fig. 11 .
Fig. 11.As Fig. 8 but for in-flight measurements in zenith geometry and for nadir measurements over targets with inhomogeneous reflectance distribution; left: analysis of time series of single detector readout MAMAP CH 4 and CO 2 dynamic "in-flight" zenith sky irradiance measurements (solar + hemispheric).Measurements performed over the MAMAP fiber optical inlet equipped with 4 transmissive Spectralon diffuser plates (target ZIR).Measurements conducted on 2 August 2007.Top panel: normalized CO 2 /CH 4 profile scaling factor ratio (PSFR).Second panel: maximum irradiance (detector filling) in binary units [BU] as measured with MAMAP and digitalized with 16 Bit ADC.Bottom panel: estimated signal-to-noise ratios (SNR) of each measurements; right: as on left side but for co-added burst mode nadir measurements over land targets with inhomogeneous reflectance distribution (target LIH).Co-added measurements derived from measurements shown in Fig. 10 (right) but each burst of 10 measurements co-added and averaged (only measurements with >3000 [BU] are processed).

Fig. 12 .
Fig. 12. Left: normalized MAMAP CO 2 /CH 4 profile scaling factor ratio (PSFR BM ) retrieved from measurements acquired on 26 July 2007 over the power plant Jänschwalde (black cross) located north of Cottbus (south-east of Berlin) in Eastern Germany, right: photo automatically taken during the flight over the power plant (top), dimensionless CO 2 profile scaling factors (middle), and dimensionless CH 4 profile scaling factors (bottom).All values shown in a given map as part of the flight have been scaled with a constant factor such that the scaled values of the whole flight are close to unity (green).The data have been smoothed using a seven point moving average and high-pass filtered with a 80 point high-pass filter.Gaps are due to the quality filtering (shown are only measurements where the spectral signal was larger than 3000 counts after dark signal correction and the root-mean-square (RMS) of the fit residuum, i.e. the relative difference between measurement and model after the fit, is better than 1%).The CO 2 output of the power plant during the overflight was 56.6 t min −1 (Dietmar Heinze, Vattenfall Europe Generation AG & Co. KG, Cottbus, Germany, personal communication, 2008).

Fig. 13 .
Fig. 13.As Fig. 12 but for the power plant Schwarze Pumpe located near the power plant Jänschwalde.The CO 2 output of the power plant was 26 t min −1 during the overflights (Vattenfall Europe Generation AG & Co. KG, Cottbus, Germany, personal communication, 2008).

Fig. 14 .
Fig. 14.Normalized CH 4 /CO 2 profile scaling factor ratio (PSFR BM ) showing anomaly in the upper part of the landfill Vorketzin located near Berlin in Eastern Germany.Data acquired with MAMAP on 26 July 2007.The data have been smoothed using a three point moving average.Gaps are due to the quality filtering: shown are only measurements where the spectral signal was larger than 3000 [BU] after dark signal correction and the rootmean-square (RMS) of the fit residuum (relative difference between measurement and model after the fit) is better than 1%.The dashed line shows the landfill body.To remove low frequency variations data were filtered with a 80 point high-pass filter.

Table 2 .
MAMAP total column conversion factors c for retrieval output profile scaling factors (PSF), assuming all deviations from mean column occurred below the aircraft.All conversion factors were computed based on SCIATRAN RTM simulations.-Relativeaccuracy (accuracy variation) includes all systematic error variations in the measurement excluding constant systematic biases caused for instance by insufficient knowledge of the slit function shape, etc.

Table 3 .
Simulated and measured MAMAP SWIR single readout detector performance for different reference targets including on-ground and in-flight measurements.The table includes the exposure time t exp (column 3), the simulated and measured signal in [BU] (column 4 and 5), the simulated and retrieved SNR (column 6 and 7), the simulated profile scaling factor precision (PSFP) for each gas (column 8) and the simulated and (from measurements) mean retrieved profile scaling factor ratio precision (PSFRP) calculated from the CH 4 /CO 2 PSF ratios (column 9 and 10).
(Bovensmann et al., 2010)ng of MAMAP, a model was developed, based on the CarbonSat instrument model(Bovensmann et al., 2010), which is similar in concept to that used for SCIAMACHY.For comparison, simulated signals (in [BU])

Table 4 .
MAMAP in-flight SWIR detector performance over inhomogeneous land targets (LIH) for co-added burst mode.The table include the simulated and retrieved SNR of the instrument for 10 coadded burs mode (BM) measurements (column 4 and 5) and also the simulated and (from measurements) retrieved CH 4 /CO 2 PSF ratio precision (PSFRP BM ) in column 6 and 7.

Table 5 .
MAMAP reference targets data set selected for precision estimates.

Table 6 .
MAMAP SWIR relative single readout detector spectral stability (readout to readout) as retrieved by best estimate of the WFMD/M algorithm for different reference targets for a slit function FWHM of 0.82 nm for different dynamic in flight and static on ground measurements.The different stability values are not directly comparable because of different exposure times during the measurements.Comparison of the shift values for different conditions give only a raw estimate for short time spectral shifts caused by vibrations and short time spectral shifts caused by inhomogeneous illumination of the slit induced by keystone and smile effects of the spectrometer.