Interactive comment on “ Intercomparison of integrated IASI and AATSR calibrated radiances ”

Thank you to all of those involved, and to the referees for their helpful and insightful comments, which have been taken onboard. We have carefully considered the questions and suggestions of the referees, and detailed responses to these questions and suggestions are indicated. Reviewer comments are quoted using [...].Some changes in the manuscript have resulted from the referee’s suggestions, resulting in us submitting a revised paper for publication in ACP.


Introduction
The Infrared Atmospheric Sounding Interferometer (IASI) is a nadir sounding Fourier transform spectrometer flying operationally on the MetOp-A satellite, and which will operate also on the two subsequent satellites of the MetOp pro-Correspondence to: S. Illingworth (smi3@le.ac.uk) gramme; the resulting record of spectrally resolved radiances should span at least 15 years.Geophysically, the accuracy of these radiances is very important for a number of reasons.Firstly, these radiances have a direct impact on the derivation of air and surface temperature, humidity, and composition profiles for which IASI is already beginning to deliver interesting results.This atmospheric composition data will be significant for climate, as well as numerical weather prediction applications.Secondly, the spectrally resolved radiances can be used to study directly the long-term radiance change over decades (Harries et al., 2001), or to directly test radiation schemes within General Circulation Models (GCMs) (Huang et al., 2007).Finally, a long-term series of calibrated radiances can and will be used for the direct testing of the outputs of GCMs (e.g., Haskins et al., 1999;Brindley et al., 1999); calibrated radiances are effectively to be considered a geophysical product in themselves, as they represent Top of the Atmosphere (TOA) radiation at the appropriate wavelengths of the data.
There are two important consequences of developing climate time series of either radiance itself or derived geophysical products.Firstly, it is important to intercompare wellcalibrated, long-term instruments, where possible, thus providing independent verification of radiances.Secondly, it is a valuable and necessary exercise to undertake such tests as part of a long-term testing strategy which can evaluate trends in instrument calibration.This paper is directed towards the development of a methodology for the former, and the provision of first results for this aspect for the IASI instrument.The intercomparison studied here is between the calibrated and geophysically located, Level 1, TOA radiances from the IASI and the Advanced Along Track Scanning Radiometer (AATSR) on ENVISAT.
This paper is organised as follows: the description of the IASI and AATSR instruments used in the cross-calibration is provided in Sect. 2. Section 3 describes the crosscalibration technique used to radiometrically calibrate the S. Illingworth et al.: IASI validation IASI instrument and presents the results of this process.Section 4 concludes this study with some perspectives on this work, discussing what the results of the cross-calibration imply for the radiometric accuracy of the IASI instrument.
The IASI instrument is designed to be calibrated to an absolute brightness temperature of at least 1 K (objective 0.5 K).It is radiometrically calibrated using views of an internal hot blackbody, whose temperature can be modified and is carefully monitored (typically 293 K), a cold blackbody held at a temperature below 100 K (typically 95 K), and views of deep space.In addition, a non-linearity correction is performed for the Band 1 (B1) detector which is photoconductive.Details are given in Blumstein et al. (2004), which notes that accuracies of better than 0.25 K are shown by ground calibration, and that effects include the non-linearity of B1, as well as small deviations from perfect emissivity of the internal blackbodies and the ground calibration reference blackbodies.
The IASI data used in this paper are the Level 1C data which consist of geolocated and apodised spectra.The data used in this paper are the operationally produced version 2.0 data set, obtained from the UMARF Archive of IASI data.

AATSR
The AATSR, on the Environmental Satellite (Envisat), is used here as the reference inter-comparison instrument; the Envisat equator crossing time is 10:00 LST.The AATSR is a dual-view, scanning radiometer with a 1 km 2 footprint for the nadir view at the sub-satellite point.It has seven channels, including three thermal channels: the two split-window channels at 11 and 12 µm, and a third infrared channel at 3.7 µm, which is only used at night as it is contaminated by reflected solar radiation during the day.The filter functions for the 11 and 12 µm channels are shown in Fig. 2.More details about the AATSR instrument are given in Llewellyn-Jones et al. (2001); Smith et al. (2001);ESA (2008).
The AATSR instrument uses an exceptionally stable onboard calibration system for its infrared channels, resulting in a high intrinsic radiometric sensitivity and accuracy.Two targets ("hot" and "cold" at approximately 301 K and 263 K respectively) are viewed every scan for calibration, and the detectors are cooled to 80 K by a Stirling Cycle cooler to ensure that the radiometric noise of these channels at 270 K is <0.05K.The AATSR is radiometrically calibrated to a high accuracy; pre-launch calibration using high-accuracy external black bodies indicated that the AATSR BTs were within 30 mK of target temperatures for the 11 and 12 µm channels (Smith et al., 2001).The heritage of the instrument design is well proven since AATSR is the third instrument (for results for radiometric design and accuracy of the earlier ATSR-1 and ATSR-2 instruments see e.g., Mutlow et al. (1994)) founded on well-designed black bodies for radiometers (Mason et al., 1996).
In this paper, only the nadir-view AATSR data in the 11 and 12 µm channels are used as the channel at 3.7 µm does not provide both day and night data, and also the full AATSR bandpass does not fall completely within the IASI band pass.The AATSR Level 1B V2.0 data were obtained from the NEODC UK Archive of AATSR data.

Overview of concept
Sensor radiometric calibration involves well-known processes such as pre-flight calibration, on-board calibration and, in the case of optical sensors, vicarious calibration against well-known targets; these approaches are well documented (e.g, Dinguirard and Slater, 1999).For the thermal infra-red, pre-flight and on-board calibration are also standard.However the cross-calibration of satellite instruments against "well-known" targets has proved to be more of an issue, because scene temperature and atmospheric state variability are critical parameters, and a mismatch in time (for example), can cause significant differences in observed TOA radiances.Further issues arise because of differing spectral filter functions and differing views of the instruments to be compared.Some of these issues have been examined, for example, by Merchant et al. (2003), who developed radiance correction factors for the Visible Infrared Spin-Scan radiometer on-board the Japanese Geostationary Meteorological Satellite (GMS-5), using equivalent data from the ATSR-2 instrument.
The advent of spectrometers such as IASI and the Atmospheric Infrared Sounder (AIRS) has provided more scope for cross-calibration against radiometer TOA data, since the radiometer may be more easily compared using integration of the measured TOA spectral radiances of IASI/AIRS for co-located fields-of-view.A particularly attractive aspect of this comparison is that the integration can utilise exactly the  assumed spectral passbands of the radiometers, without the need for interpolation or radiative transfer modelling (see Sect. 3.2 for exact conditions).Geophysically, it enables TOA radiance measurements from radiometers, usually at higher spatial resolution, to be linked radiometrically to TOA spectrally resolved radiance measurements at lower spatial resolutions.This provides a nice match of characteristics for testing climate models, especially if such a combined system of TOA radiance measurements can form a long time series.
Preliminary work for such spectrometer-radiometer intercomparisons has been performed, (e.g., Qu et al., 2005;Wang and Cao, 2007), but so far such comparisons have not used an instrument like the AATSR, which has been designed for high radiometric accuracy.In this study we focus intentionally on comparing the IASI and AATSR instruments, for which we have a good knowledge of radiance calibration on the ground, but wish to intercompare the performance of the systems in space.In particular, we examine how well such a comparison can be carried out by focussing on the intercomparison for surfaces with well understood characteristics, in a manner analogous to the vicarious calibration of visible sensors.

Cross-calibration match-ups
Thermal infra-red TOA radiances vary strongly with wavelength, surface temperature and emissivity, atmospheric temperatures, atmospheric composition (particularly humidity), and clouds.Therefore, for the most accurate intercomparison of radiances, it is ideal to consider surfaces of largely uniform emissivity, closely matched in time and space, and observed with similar viewing angles.These factors are now considered in turn.Since thermal emission at 11 µm and 12 µm does not substantially change its characteristics diurnally, both day and night data are used in the comparison.
The AATSR thermal channels at 11 µm and 12 µm are designed to be able to determine surface temperature, particularly over the oceans, with very good accuracy.For the present purpose, scenes with both a uniform surface emissivity and small changes of surface temperature with time are ideal.Oceans satisfy both criteria in general (rather than land), and thick clouds may at least satisfy the first criterion but quite possibly the second.Therefore, this paper focusses firstly on clear sky ocean scenes and subsequently considers www.atmos-chem-phys.net/9/6677/2009/Atmos.Chem.Phys., 9, 6677-6683, 2009 results for cloudy scenes (see Sect. 3.3 for the classification system) over the ocean.Surface temperature (and cloud fields) change more rapidly over land than ocean, and the radiance is more view angle dependent over land, thus precluding the use of land data for accurate radiometric calibration work.
In order for a very good cross-calibration to be performed, the measured BTs for both instruments need ideally to be taken from the same geographic (latitude and longitude) location (co-located view), and measured at similar times.The AATSR achieves global coverage approximately once every three days, with a repeat cycle of 35 days (ESA, 2008).In comparison IASI achieves twice daily coverage with a 29 day repeat cycle (Camy-Peyret and Eyre, 1998).As the two instruments have equator crossing times that differ by only half an hour, there will be some measurements where the two orbits become coincident at approximately the same time.Fig. 1 shows that on 01/09/2007 the views of the two instruments coincided on both a spatial and temporal scale.The red box indicates an example area in which this occurs.The comparisons that are performed in this study consider all data that was taken on 01/09/2007, and for which there is a spatial and temporal coincidence, as outlined below.
Considering the temporal domain, the match ups were limited to only those occurring within 20 min, in order to limit the possibility that the cloud scene will have changed between the IASI and AATSR views of the same scene.Twenty minutes was chosen as a threshold value in order to give a reasonable time constraint allowing match-ups but ruling out the vast majority of cloud formation in the very homogeneous ocean scenes (see filters below) chosen for analysis.This threshold difference is similar to that adopted by Merchant et al. (2003) who applied a mean time difference between two co-located satellite data sets of 24 min.
In nadir mode the AATSR instrument scans between ±21.433 • , whilst the IASI instrument makes measurements with viewing angles ranging from −48.3 • to +48.3 • across the swath.So far we have considered all of the AATSR and IASI measurements, irrespective of the differences in the satellite viewing angle.A difference in the viewing angles between the measurements made by the two instruments will introduce an offset, caused by the different atmospheric path lengths that radiation from the measured scene must travel to the detector.This offset will principally depend on the amount of water vapour in the atmosphere: for a polar winter, this difference could be close to zero, whereas for a large angular difference in a tropical atmosphere, this difference could be as much as 0.4 K at 12 µm (Noyes, 2005).Due to this fact only match-ups where the absolute difference of the satellite zenith angles is less than 1 • are considered.

Radiance comparisons
Having defined the match-ups between the IASI and AATSR data sets, the IASI spectrally resolved TOA radiances were integrated to "AATSR-like" radiances, and then converted to radiometric brightness temperatures (BTs), which are the equivalent form of the AATSR TOA radiances.The AATSR spectral filter functions for the channels 11 µm and 12 µm are shown in Fig. 2. For the selected regions, the mean and standard deviation of the AATSR BTs within each IASI pixel were then calculated, and also the match-ups flagged for cloud and sea/land.The standard deviations are subsequently used to illustrate the homogeneity of the scenes and for the cloud scenes to perform a cross-calibration consistent with the clear sky comparisons.Further classification was performed using the flags associated with the AATSR data flags for land and cloud, since the AATSR instrument has the higher spatial resolution.The AATSR contains a detailed cloud-filtering system which labels each AATSR pixel as being either "clear sky" or "cloudy"; the cloud processor is described in Závody et al. (2000).The cloud and land filters of the AATSR data set within each IASI coincident view were used to separate the match-ups into different sub-groups which could be further analysed.IASI pixels containing only "land" AATSR pixels were assumed to have been taken over the land, whilst those containing only "sea" AATSR pixels were assumed to be have been taken over the sea; land pixels were not subsequently used in the analysis.Any IASI pixels which were deemed as being over both land and sea (such as a coastline) were also discarded, as they were too few in number to draw any significant statistical conclusions from.The IASI measurements over ocean were then separated into three further groups: data that was cloud-free (contained only clear sky AATSR pixels), data that was fully-cloudy (contained only cloudy AATSR pixels), and data that was fractionally cloudy (contained both clear sky and cloudy AATSR pixels).Thus each match-up between IASI was labelled with the AATSRlike IASI BT, the mean AATSR BT in the IASI pixel and the standard deviation of the AATSR BTs, the sea/land flag and a cloud group flag.

Cross-calibration results
Results for the differences between IASI equivalent BTs and the AATSR mean BTs are shown in Fig. 3, for each classification (clear sky, cloudy etc.) and for the 11 µm channel as a function of the AATSR BT standard deviation within each IASI pixel.The clear sky data show very small deviations (a mean standard deviation of 0.08 K for both the 11 and 12 µm comparisons) with much larger variations for the pixels containing fractional or cloudy pixels.The fractional and completely cloudy data have very large standard deviations, as well as mean differences, as expected.Hence we consider first the clear sky data.
The results for the cross-calibration over the 11 µm and 12 µm spectral domains, for measurements taken over the sea and in cloud free conditions, are shown in Fig. 4.This plot shows that there is a very strong correlation between the AATSR and IASI BTs, in both the 11 µm and 12 µm spectral regions, over a temperature range of just greater than 10 K.It also shows that the IASI instrument overestimates the BTs, compared to those measured by the AATSR instrument.For the 11 µm spectral region, a mean IASI-AATSR BT difference of −0.05 K can be observed, compared to 0.23 K in the 12 µm region.In both spectral regions, the mean BT difference is less than 0.3 K, with a small standard deviation, indicating that the IASI instrument is in agreement with the AATSR instrument to at least this level, in cloud-free scenes.Therefore the indication would be that IASI is likely to be meeting its target objective over ocean clear sky conditions.
The clear sky comparisons for this day are limited in their temperature range.However, the real criterion for conducting a comparison is a uniform scene rather than requiring the scene to be free of clouds.Therefore, the comparisons were extended to include fully-cloudy pixels where the AATSR BT standard deviations were less than 0.395 K for the 11 µm case, and less than 0.334 K for the 12 µm differences (these thresholds represent the maximum values of the clear sky standard deviations).As a further restrictive test of homogeneity for the scene, any IASI pixels which contained AATSR BTs which were greater than 3σ from the modal average of the AATSR BTs within the scene were rejected.Figure 5 shows the resulting relationship between the IASI and AATSR BTs for all such data points.It is immediately clear that the temperature range is now larger; the correlation coefficient is also improved, but this is as expected and is simply a consequence of the increase in temperature range.The mean differences between IASI equivalent BTs and AATSR mean BTs are now −0.05K for the 11 µm channel and 0.19 K for the 12 µm channel (error of 0.03 K on the mean for both; 66 points), with the agreement between the two instruments still remaining good to at least 0.3 K over this wide range of temperatures.
Based on the assumption that the AATSR is intrinsically radiometrically calibrated to an accuracy of 30 mK, these results would seem to indicate that the IASI instrument is itself able to measure BTs to an accuracy of better than 0.3 K.However, the comparison does rely on knowledge of the AATSR spectral filter functions.Currently there are some uncertainties in the exact knowledge of these filter functions to at least the 0.2 K level.Therefore, the current results are best viewed as indicating an absolute comparison of the two instruments to probably the 0.3 K level, and consequently an absolute calibration uncertainty for IASI of 0.3 K

Conclusions
Based on the cross-calibration results in this study, integrated IASI radiances agree very well with AATSR BTs at 11 µm and 12 µm, and indicate that IASI is meeting its target accuracy of 0.5 K accuracy and achieving accuracies of better than 0.3 K with 0.1 K accuracy demonstrated at 11 µm.The uncertainties on the mean differences are also relatively small as they are of the order of 0.03 K, and so the larger question concerns uncertainties in any bias terms.Of these, the largest requirement is to reduce the uncertainty in the AATSR spectral filter function because although AATSR is itself radiometrically accurate, the spectral filter function is needed in order to integrate the spectra radiances measured by IASI.This is currently under investigation and it should be possible in future to reduce the filter uncertainty and thus establish whether there are individual instrument biases or not at the 0.3 K level or smaller.
More generally, the methods developed in this study show that it is possible to intercompare satellite radiances or BTs with a high precision, and that the spectrometer-radiometer combination offers additional insights, e.g. the use of high spatial resolution information to characterise the homogeneity of the scene, and identify scenes suitable for provision of the most information on instrument bias.The use of cloudcovered pixels has been shown here to extend the BT range over which comparisons can be performed.Therefore sug- All match-ups have been selected over the oceans and represent homogenous conditions (see Sect. 3.4).For both the 11 µm and the 12 µm comparison there were a total of 66 IASI pixels which met this criteria, which in both cases was comprised of 41 clear, and 25 fully cloudy, IASI pixels.gestions that highly accurate Fourier transform spectrometers (FTS) could be flown in future to inter-calibrate satellite sensors, e.g.Anderson et al. (2004) and the Climate Absolute Radiance and Refractivity Observatory (CLARREO) mission, have some provenance.At the same time, it is also clear from this study that a very good overlap in time is required between the instruments to be inter-compared, and that there is a very tight requirement on knowledge of the spectral filter function for satellite radiometers.
Despite dealing with a limited set of data (one day of IASI measurements were analysed) this paper is very clear both on objective and conclusions: we present the first results on the IASI absolute radiometric calibration, which will likely be useful for further climate and/or chemistry applications.The results of the study depend more on the details of the comparison and methodology so that analysis of further days, whilst valuable, is not as important as establishing a quality comparison for one day.It can be seen from the standard deviations on the mean that the random errors are already quite small for one day of comparisons.This paper has also identified the chief aspects which are important for radiometric intercomparisons for climate purposes.A clear challenge for the future is to improve knowledge of the biases, so that careful use may be made of IASI integrated radiances for climate applications.First the analyses should be extended to a longer time series of data on an operational basis.Secondly, extension of the analyses to warmer temperatures, e.g. over land, would be interesting and would extend the temperature range over which the data could be used for climate applications with confidence.Finally, refinement of the AATSR spectral filter function will hopefully permit a tighter error constraint on the quality of the IASI data and consequently confirm a likely absolute calibration uncertainty for IASI of better than 0.3 K

Fig. 1 .
Fig. 1.The time since 00:00 on 01/09/2007 that the IASI and AATSR measurements were made.The first full AATSR swath of the day is shown, along with the first central scan-line of the IASI instrument; the start-time of the IASI and AATSR swaths being 00:35:55 and 00:32:18 respectively.The red box indicates one of the areas on this day that were selected for cross-calibration, chosen because of the similar temporal values and spatial coincidence between the two instruments.

Fig. 3 .
Fig. 3. Comparison of 11 µm IASI equivalent BTs minus mean AATSR BTs in each IASI pixel versus the standard deviation of the AATSR BTs in each pixel: (a) all pixels; (b) clear sky pixels; (c) cloudy (all pixels); (d) fractionally cloudy.All match-ups have been selected over the oceans.The small spread of AATSR BTs within each IASI pixel under clear sky conditions is clearly shown.

Fig. 4 .
Fig. 4. Comparison of 11 µm and 12 µm IASI equivalent BTs minus mean AATSR BTs in each IASI pixel: (a) 11 µm; (b) 12 µm.All match-ups have been selected over the oceans and represent clear sky conditions.For both the 11 µm and the 12 µm comparison there were a total of 41 clear IASI pixels.

Fig. 5 .
Fig. 5. Comparison of 11 µm and 12 µm IASI equivalent BTs minus mean AATSR BTs in each IASI pixel: (a) 11 µm; (b) 12 µm.All match-ups have been selected over the oceans and represent homogenous conditions (see Sect. 3.4).For both the 11 µm and the 12 µm comparison there were a total of 66 IASI pixels which met this criteria, which in both cases was comprised of 41 clear, and 25 fully cloudy, IASI pixels.