Abstract

Since 1979, sensors on board the National Oceanic and Atmospheric Administration (NOAA) series of polar meteorological satellites have provided continuous measurements of the earth's surface and atmosphere. One of these sensors, the Television Infrared Observational Satellite (TIROS-N) Operational Vertical Sounder (TOVS), observes earth-emitted radiation in the infrared—with the High-Resolution Infrared Sounder (HIRS)—and in the microwave—with the Microwave Sounding Unit (MSU)—portions of the spectrum. The NOAA and National Aeronautics and Space Administration (NASA) Pathfinder program was designed to make these data more readily accessible to the community in the form of processed geophysical variables (temperature, water vapor, cloud characteristics, and so on) through the “interpretation” of the infrared and microwave radiances. All presently developed interpretation algorithms more or less directly rely on the comparison between a set of observed and a set of simulated radiances. For that reason, the accuracy of the simulation directly influences that of the interpretation of radiances in terms of thermodynamic variables. Comparing simulations to observations is the key to a better knowledge of the main sources of errors affecting either the former or the latter. Instrumental radiometric problems, radiosonde, surface data, and forward radiative transfer model limitations as well as difficulties raised by differences in space and in time of satellite and radiosonde observations (collocations) have long been studied in detail. Less attention has been paid to errors, presumed negligible, generated by the absence of consideration of main absorbing gases (CO2, N2O, CO, O3, and so on) atmospheric seasonal cycles and/or annual trends. In this paper, all important sources of variability of the observations and of the simulations are first reviewed. Then it is shown that analyzing, at different timescales (seasonal, annual), the departures between simulated and observed NOAA TOVS brightness temperatures reveals signatures of these greenhouse gases' concentration variations. Not only the shape of the seasonal variations (locations of the peaks) is in good agreement with what is presently known, but also their amplitude (peak-to-peak) matches relatively well the values predicted from a line-by-line radiative transfer model. Moreover, annual trends correspond very well with the known increase in concentration of gases such as CO2 or N2O, as a result of human activities. Limits of such an analysis are discussed: the most significant one finds its origin in the modest spectral resolution of the TOVS channels that integrate signatures from several absorbers and from many atmospheric layers. However, results from this work leave some hope to extract from these channels interesting information on CO2, N2O, and CO distributions. These results also strengthen the hope to improve greatly the knowledge of the global distribution of a variety of radiatively active gases with the coming second generation of vertical sounders such as NASA's Advanced Infrared Radiation Sounder (AIRS) or the CNES/Eumetsat Infrared Atmospheric Sounder Interferometer (IASI), both characterized by a much higher spectral resolution.

1. Introduction

Since 1979, sensors on board the National Oceanic and Atmospheric Administration (NOAA) series of polar meteorological satellites have provided continuous measurements of the earth's surface and atmosphere. One of these sensors, the Television Infrared Observational Satellite (TIROS-N) Operational Vertical Sounder (TOVS), observes earth-emitted radiation in 27 wavelength bands within the infrared and microwave portions of the spectrum. The overall mission of the TOVS instruments on board the NOAA Polar Orbiting Environmental Satellites is to provide continuous, global measurements of atmospheric temperature and moisture profiles, cloud and surface properties, etc., thereby creating a valuable resource for studying the climate of our planet. The NOAA–National Aeronautics and Space Administration (NASA) Pathfinder program was designed to make these data more readily accessible to the community in the form of processed geophysical variables. The Atmospheric Radiation Analysis group at the Laboratoire de Météorologie Dynamique (LMD), Palaiseau, France, was selected to process TOVS data into climate products (Path-B), and the time period from January 1987 to September 1995 has already been reanalyzed.

The TOVS consists of three passive vertical sounding instruments (Smith et al. 1979): the High-Resolution Infrared Radiation Sounder (HIRS-2), a radiometer with 19 channels in the infrared band and one in the visible band; the Microwave Sounding Unit (MSU), a microwave radiometer with four channels in the vicinity of 55 GHz; and the Stratospheric Sounding Unit (SSU), a pressure-modulated infrared radiometer with three channels near 15 μm. Only HIRS and MSU data are used here. Scan widths are approximately 2200 km wide, providing global coverage every 12 h. HIRS-2 measures atmospheric and/or surface emission in seven channels located around 15.0 μm, five located around 4.3 μm, one 11.0-μm window channel, and three 6.7-μm H2O channels. Surface and O3 emission is measured in one 9.6-μm window channel, surface emission and reflected solar radiation in two 3.7-μm window channels, and reflected solar radiation in one visible channel. The MSU measures atmospheric emission in three 55-GHz O2 channels and surface and atmospheric emission in one 55-GHz window channel.

The Improved Initialization Inversion (3I) method was developed at LMD for the purpose of interpreting NOAA TOVS observations in terms of meteorological and climate variables. It has been extensively discussed in the literature: see, for example, Chédin et al. (1985, 1994) and Scott et al. (1999). Like most physical retrieval methods, the 3I method estimates geophysical variables by minimizing the differences between a set of observed and a set of simulated brightness temperatures. As a consequence, systematic biases between simulated and observed brightness temperatures can be problematic, not only for the retrieval accuracy, but also for further analysis of the climate variability and evolution. As these biases may differ from satellite to satellite, spurious trends may result unless these biases can be identified and accounted for.

Differences between radiative transfer–simulated and satellite-observed brightness temperatures may find their origins in various sources:

  • lack of an accurate description of the atmospheric and surface states (surface temperature and emissivity, upper-stratospheric temperatures, water vapor profile, etc.) by the in situ radiosonde measurements;

  • lack of an accurate time–space collocation between radiosonde measurements and satellite observations, inducing random as well as systematic errors;

  • limitations of the forward radiative transfer model: airmass dependence of errors, relatively poor description of surface (land) emissivities, occurrence of unexpected events (e.g., the Pinatubo volcano eruption in June 1991), insufficient description of the distribution of the radiatively active gases, and so on; and

  • problems with the satellite radiometers: change over the lifetime of the platform, calibration errors, and so on.

All of these potential error sources are not equivalent. Actually, a forward radiative transfer model, constructed within the limits of our current theoretical knowledge and modeling capabilities may be understood as an “interpreter” between the thermodynamic “world” and the satellite radiometric world. Ideally, significant differences between model-simulated and satellite-observed brightness temperatures will tell the user that the model is wrong (and possibly, why) and by how much it deviates in its interpretation, opening the way to pertinent corrections. Problems occur when either of the model input data or satellite-observed data are wrong. In both cases, a good interpreter may be considered as bad and subsequently corrected although it is right. Poorly extrapolated stratospheric temperatures (thermodynamic world) or radiometer calibration errors (satellite radiometric world) are examples of such problems. Consequently, biases resulting from the direct comparison between modeled and observed brightness temperatures may be quite complex to explain. A more detailed examination of the change in time of these biases reveals seasonal variations and/or yearly trends. A tentative explanation of these behaviors is based on the fact that the forward radiative transfer model used here, like most such models, has fixed values of the main absorbing gas concentrations, whereas it is known that CO2, O3, and CO, for example, show important seasonal variations. Moreover, mainly due to human activities, CO2 or N2O atmospheric concentrations steadily increase in time. Then, the question is raised of whether TOVS is capable of identifying the signatures of the change in the concentration of such key greenhouse gases and, if yes, whether the results agree with the present observations and in situ measurements. The answer is important for two reasons. First, better accounting for these concentration variations in the forward model should result in an improvement in the accuracy of the variables retrieved from TOVS. Second, an appropriate reanalysis of the more than 20 yr of TOVS observations could bring new information on the global-scale evolution of important greenhouse gases. In addition, a positive answer should strengthen our hope to significantly improve our knowledge on the distribution of major greenhouse gases when the next generation of high spectral resolution infrared sounders such as NASA's Advanced Infrared Radiation Sounder (AIRS) and the CNES/Eumetsat Infrared Atmospheric Sounder Interferometer (IASI) come into use.

To answer the above questions, we analyze here differences between model-simulated and satellite-observed brightness temperatures over the operational time period of NOAA-10 (July 1987–September 1991), focusing attention on the channels sensitive to atmospheric CO2 concentration and to other greenhouse gases such as N2O or CO: the 15-μm and the 4.3-μm channels. Only some of the TOVS channels will be reviewed here. The present analysis excludes channels that are too sensitive to water vapor or to surface temperature and focuses on HIRS sounding channels 2–5 (channel 1 is excluded because measuring much too high in the stratosphere and even in the mesosphere) in the longwaves, and 13–15 in the shortwaves (channels 16 and 17 are excluded for the same reasons as channel 1). Window channels 8 (longwave) and 18 (shortwave), the ozone channel 9 (centered at 9.6 μm), and the microwave channels are used to help in interpreting the various signatures observed.

We first review the data sources [collocations between radiosondes and satellite observations and the forward radiative transfer model used here (section 2)]. The main sources of variability of the deviations between simulated and observed brightness temperatures for the channels considered, as well as their respective sensitivities, are then detailed in section 3. Channel-by-channel 12- and 3-month monthly running means of the above differences are presented and discussed in section 4. Conclusions and perspectives are given in section 5.

2. Data sources and model

Analysis of the variability of NOAA TOVS observations from comparisons with model simulations requires collecting an extensive archive of quality controlled collocations of in situ radiosonde and satellite observations, as well as using an efficient tool for simulating observed radiances as accurately as possible, keeping in mind that thousands of comparisons are necessary to convincingly sample the global-scale variability and to reach appropriate conclusions.

a. Satellite–radiosonde collocations: The National Environmental Satellite and Data Information Service (NESDIS) DSD5 archive

Since TIROS-N was launched in 1978, NOAA/NESDIS has collected and collocated satellite and radiosonde data. One of the most important characteristics of this so-called DSD5 archive is the space–time window. For the present study, the space window has been reduced from 300 to 100 km. The time window is variable: 3 h for land collocations with no time interpolation and 6 h for oceanic collocations with possible time interpolation from bracketing radiosoundings. The general characteristics of the resulting data archive are given and discussed in detail in Uddstrom and McMillin 1994 (hereinafter UMM94) and Uddstrom and Korpela (1999). The archive developed at LMD for the reanalysis of TOVS observations mixes information from the DSD5 archive and original level-1B TOVS observations.

1) Upper-air radiosonde data

These are directly extracted from the DSD5 archive with the requirements that temperature measurements be available at least up to 50 hPa (30 hPa in about one-half of cases), that water vapor measurements be available at least up to 300 hPa, and that surface pressure be not smaller than 950 hPa. Only collocations declared as clear in the DSD5 archive are considered.

Radiosonde data are interpolated (linearly in lnP) to the Automatized Atmospheric Absorption Atlas (4A) 40 atmospheric levels from the surface to 0.05 hPa (Scott and Chédin 1981).

Extrapolation to the highest 4A level (0.05 hPa) is done using the NESDIS temperature retrievals (based on SSU) from the DSD5 archive. Changes in the NESDIS retrieval methodology are expected to induce changes in the comparison between satellite-observed and model-simulated brightness temperatures for channels “looking” high in the stratosphere.

2) Satellite and surface data

The period covered here extends from July 1987 to September 1991 and uses observations from NOAA-10, a morning satellite. Morning satellites have local equatorial crossing times (LECTs) at 0730 and 1930; afternoon satellites have LECTs at 0130 and 1330. LECTs may be affected by important drifts: up to 0345 for NOAA-11, 75 min for NOAA-12, and only about 20 min for NOAA-10 (Christy et al. 1995; Prabhakara and Iacovazzi 1999).

Satellite data are extracted from the original level-1B Pathfinder TOVS archive and not from the DSD5 archive where they are corrected for limb and, for the windows, for water vapor attenuation (regression). In the present study, only the MSU channel data are NESDIS-limb corrected. Satellite observations with a viewing angle less than 40° are accepted.

Because the DSD5 archive does not contain observed surface temperature data, a satellite-derived substitute, extracted from the DSD5 archive, is used here. Several choices are available (UMM94): physical retrieval from the longwave window HIRS channel 8; 15-day running mean ocean skin surface temperature from the National Meteorological Center over sea; regressed skin surface temperatures from HIRS channels that are sensitive to the surface and to water vapor attenuation.

The choice made for the present study is the HIRS shortwave (4 μm) window channel 18 (regression) corrected for water vapor attenuation by NESDIS and archived in the DSD5 files. This choice has advantages and drawbacks. The main advantage is the high degree of transparency of the atmosphere (transmittance to the surface between 0.8 and 0.9) at this wavelength, contrary to the longwave window (channel 8) whose transmittance to the surface may be as low as 0.35 for humid tropical situations. Channel 18 may be much more easily corrected to provide an accurate estimate of the surface skin temperature. Drawbacks are its sensitivity to solar radiation, particularly for a satellite such as NOAA-11, due to its LECT in the early afternoon, but also for morning satellites like NOAA-10 and, potentially, its sensitivity to surface emissivity. The former may induce seasonal variations, a problem for the present study of variabilities, whereas the latter mostly induces a bias. The quality of this substitute is examined in section 3b.

Comparisons between the channel 18 (NESDIS regression corrected) substitute and climatologic sea surface temperature records show a cold bias of 1.5 K (UMM94). The regression is corrected accordingly.

b. The fast forward radiative transfer model

The radiance I(υ) reaching the satellite at the frequency υ is given by

 
formula

where B(υ, T) is Planck's spectral radiance, τ(υ, p) is transmittance between the top of the atmosphere and pressure level p, ɛsurf(υ) is surface emissivity, and T is temperature. Scattering and reflection on the earth's surface are neglected in Eq. (1).

A first attempt to increase the speed of transmittance and radiance derivations was made with the 4A model (Scott and Chédin 1981) relying on a line-by-line precomputed lookup table of monochromatic optical depths, followed by the integration of Eq. (1). Although dramatically faster than the conventional line-by-line models, the 4A model is unable to produce radiances as quickly as required when tens of thousands of atmospheric situations (collocations) have to be processed. Pattern recognition forms the basis of the very fast “Rapid Radiance Recognition” (3R) model (Flobert et al. 1986), which is dedicated to the computation of TOVS radiances from atmospheric thermodynamic profiles (temperature, water vapor, surface characteristics).

In the pattern recognition approach used here, each input (radiosonde measurement) is compared with all of the situations archived in the “Thermodynamic Initial Guess Retrieval” (TIGR) climatologic dataset (Chédin et al. 1985; Achard 1991; Escobar-Munoz 1993).

This (frozen) library of atmospheres consists of about 1800 situations [recently extended to 2300 situations for a better representation of the tropical regions (Chevalier et al. 1998)] selected by statistical methods out of 150 000 radiosonde reports. Clear-sky transmittances, radiances, and weighting functions for all TOVS sounding channels are precomputed for each situation in TIGR by the fast line-by-line 4A model. This model assumes constant mixing ratios for CO2 (355 ppmv), N2O (305 ppbv), CH4 (1.74 ppmv), and CO (82 ppbv). Calculations are performed for 10 viewing angles between 0° (nadir) and 60° (the maximum value for angular scanning), for 19 values of surface pressure (up to about 500 hPa) for elevated terrains, and for two surface types: land and sea. These results are also stored within the TIGR dataset. It is worth pointing out that TIGR is not sensitive to the relative quality of the radiosounding samples it contains but only to their representativeness and plausibility. Actually, it is sensitive to the quality of the relationship between thermodynamic quantities and radiative quantities. For this reason, great attention has been paid to the validation of the 4A model.

The situations in TIGR have been stratified by a hierarchical ascending classification into five airmass types (tropical, temperate, cold temperate and summer polar, winter polar, Northern Hemisphere very cold polar), depending on their virtual temperature profiles (Achard 1991; Chédin et al. 1994).

The comparison between the input situation and the situations archived in TIGR takes the characteristics of the collocation into account: in particular, surface pressure and type, satellite viewing angle, and a set of distances D are derived. The distance D between the input situation and any of the archived ones approximately gives the same weight to the temperature and water vapor profiles. The minimum value of D, Dmin, is obtained for the closest situation:

 
formula

where ΣT is the sum over the atmospheric levels between the top and the surface, and ΣH2O includes 6 layers between 200 mb and the surface. Here Tobs is the temperature profile of the input situation, Tclosest (qH2Oclosest) is the temperature (water vapor) profile of the closest situation, Stdv(T) [Stdv(qH2O)] is the standard deviation, for the airmass type concerned in the TIGR dataset, of temperatures (water vapor amounts) at each pressure level, and qH2O is the water vapor amount summed for each layer (g cm−2). Assuming the brightness temperature TB only depends upon the temperature and water vapor concentration profiles, TB may be expanded around the closest situation:

 
formula

where ∂TB/∂T is the partial derivative of the brightness temperature TB with respect to temperature T, and COR(H2O) accounts for the differences between the water vapor profiles of the input situation and of the closest situation.

The computation of the second term in Eq. (2) makes use of the Jacobians stored in the TIGR dataset. Regressions based on TIGR predictors and predictands are used to estimate the correction term COR(H2O). It appears as a linear combination of the differences in water vapor amounts, for the layers (up to six) considered, between the input and the closest situations. Each regression applies to a given airmass type and a given set of observing conditions (viewing angle, surface pressure, and emissivity). This correction term appears to be significant for 7 HIRS-2 channels: 6, 7, 8, 10, 11, 12, 19, and, to a lesser extent, channels 5, 13, and 14. It should be noted that airmass type attributed to the input situation is that of the closest situation found in TIGR. For the channels considered here, comparisons between 4A and 3R show almost no bias and standard deviations between 0.1 and 0.3 K (channel 5).

For each collocation between radiosonde and satellite measurements, 3R-simulated brightness temperatures are then compared with TOVS level 1B–observed values, channel by channel, and separately over land, over sea, for nighttime and daytime observations. Statistics are derived for three latitude zones: 60°–20°N, 20°N–20°S, and 20°–60°S, because of their different behaviors, in particular with respect to the seasonal variations of absorbing gases such as CO2, CO, O3, and so on.

In order to minimize the dependence of the performances of the forward model on airmass type (UMM94), differences between model-simulated and satellite-observed brightness temperatures are first centered with respect to the global mean of their own airmass type and then divided into classes, each corresponding to a combined index composed of the land/sea flag, the day/night flag, and the latitude zone. Central values are shown in Fig. 1, for two representative channels, HIRS 5 and 14. Like most sounding channels considered here, channel 5 centers do not significantly change with the airmass type or with the class index. Channels 14, and more notably 13 (not shown), display larger variabilities due to their sensitivities to surface conditions (surface temperature and emissivity).

Fig. 1.

Mean difference between model-simulated and satellite-observed brightness temperatures for (a) HIRS channel 5 and (b) HIRS channel 14. Mean values are computed for each airmass type and each latitude zone (index 1–3 for 20°N–20°S, 60°–20°N, 20°N–20°S, respectively). Almost no polar situations are identified between 60°N and 60°S; D = day, N = night

Fig. 1.

Mean difference between model-simulated and satellite-observed brightness temperatures for (a) HIRS channel 5 and (b) HIRS channel 14. Mean values are computed for each airmass type and each latitude zone (index 1–3 for 20°N–20°S, 60°–20°N, 20°N–20°S, respectively). Almost no polar situations are identified between 60°N and 60°S; D = day, N = night

Grouping of classes can also be done when appropriate, for example, day and night over land. The separation into three latitude zones is always preserved. Note that the limits 60°N–60°S in latitude greatly reduce the number of polar situations in the statistics. Three-month and 12-month monthly running means and their standard deviations are then produced. They are analyzed in the following sections.

3. Main sources of discrepancies between NOAA TOVS observations and forward radiative transfer model simulations

The main limitations of most radiative transfer models are reviewed below, when simulating satellite-observed radiances. For the channels considered here, significant sources of variability are the following:

  1. the seasonal and annual variations of the main absorbers (CO2, N2O, O3, CO) not taken into account in most forward models;

  2. errors in the input radiosonde data: “boundary” conditions in the upper stratosphere (extrapolation) and at the surface (temperature, emissivity);

  3. impact of the diurnal cycle due to the collocation time window and to satellite orbital time drifts (particularly for NOAA-11) that randomly or systematically modify the difference between the radiosonde time and the satellite overpass time;

  4. errors in the satellite radiometers: calibration errors, sensitivity to sun illumination, and so on; and

  5. unexpected events: the Pinatubo volcano eruption, that occurred in June 1991, did not seriously affect the NOAA-10 time series analyzed here.

These five potential sources of variability will now be reviewed. In the following, the 4A fast line-by-line model and a set of 42 representative situations (Garand et al. 2000) are used to compute the sensitivities of the HIRS and MSU channels to such or such variation. They are listed in Table 1.

Table 1.

Sensitivities (K) of HIRS and MSU channels to changes in various parameters. Sensitivities to actual changes (seasonal, interannual) may be deduced linearly

Sensitivities (K) of HIRS and MSU channels to changes in various parameters. Sensitivities to actual changes (seasonal, interannual) may be deduced linearly
Sensitivities (K) of HIRS and MSU channels to changes in various parameters. Sensitivities to actual changes (seasonal, interannual) may be deduced linearly

a. Seasonal and annual variations of main infrared absorbers

1) CO2

The global mean concentration of atmospheric CO2 is steadily increasing at a rate of 0.4% yr−1. As summarized by Masarie and Tans (1995), the CO2 surface concentration seasonal variation is large, 3%–4%, for the latitude zone 60°–20°N, moderate for 20°N–20°S, and almost zero for 20°–60°S (Fig. 2). Global temporal and spatial variations in the atmospheric increase in CO2 are from GLOBALVIEW-CO2 (1999). Sarmiento (1993) and Keeling et al. (1995) have shown the potential impact of El Niño events and, apparently even more important, of the Pinatubo eruption in June 1991 on the CO2 concentration. A dramatic downward anomaly was observed in 1992 and 1993, followed by a return to “normal” values.

Fig. 2.

The CO2 surface concentration seasonal variation for the latitude zones 60°–20°N, 20°N–20°S, 20°–60°S (after Masarie and Tans 1995)

Fig. 2.

The CO2 surface concentration seasonal variation for the latitude zones 60°–20°N, 20°N–20°S, 20°–60°S (after Masarie and Tans 1995)

Altitude also plays an important part in the seasonal variation of CO2. At 60°N, for example, the seasonal peak-to-peak variation of CO2 was about 15 ppm at surface, and only 5 ppm at 300 hPa, in 1993 (Schneider et al. 2000). See also Nakazawa et al. (1991), and Wofsy et al. (1994) for more details.

Longwave HIRS channels are sensitive to CO2 absorption, and their degree of sensitivity varies as seen in Table 1. Signs and amplitudes are due to the mean temperature gradient as “seen” by the channels: tropospheric channels become colder and stratospheric channels warmer. In between, variations are small and “erratic”, as for channel 3 with rather comparable contributions from the stratosphere and upper troposphere. In the shortwaves, only channel 15 is significantly sensitive to CO2 absorption. Channels 13 and 14 are only sensitive to N2O and CO absorptions (see below). These sensitivities significantly depend on the atmospheric temperature gradient: the steepest the slope, the highest the sensitivity. The induced range of variation of the mean value is parenthesized. Representative Jacobians (partial derivatives of the brightness temperatures with respect to CO2 concentration variations) are given in Fig. 3 for channels 2, 3, 4, 5, and 15, for a 3% increase of CO2 throughout the entire atmospheric column.

Fig. 3.

HIRS CO2-Jacobians for five channels (2–5 and 15) and three representative atmospheric situations: (a) tropical, (b) temperate, (c) cold temperate and summer polar. A CO2 increase of 3%

Fig. 3.

HIRS CO2-Jacobians for five channels (2–5 and 15) and three representative atmospheric situations: (a) tropical, (b) temperate, (c) cold temperate and summer polar. A CO2 increase of 3%

2) N2O

Global-average N2O concentrations have increased from about 300 ppb around 1976 to 312 ppb around 1996 (Elkins et al. 1996; Prinn et al. 1998) with substantial year-to-year variations: 0.75 ppb yr−1 in the 1980s, up to 1 ppb yr−1 in 1991, and down to 0.5 ppb yr−1 in 1995. The cause of this sudden decrease remains unclear although the impact of the Pinatubo eruption might be important (Schauffler and Daniel 1994).

Nitrous oxide is the major absorber for the shortwave channels 13 and 14 and is comparable to CO2 for channel 15. For a +4% variation of the N2O column around a mean value of 305 ppbv for the considered period, the (gross mean) variation in brightness temperature is −0.25 K (±0.08 K) for channel 13, −0.4 K (±0.2 K) for channel 14, and −0.35 K (±0.1 K) for channel 15, all of these channels having sensitivity peaks in the troposphere (below 350–400 hPa). Figure 4 shows representative Jacobians with respect to N2O variations for the three channels, for an increase in N2O of 4%. Seasonal concentration variations seem to be small and significantly less than 0.3% in both hemispheres, and in the mid- and lower troposphere (Khalil and Rasmussen 1983).

Fig. 4.

Same as Fig. 3 for the N2O-Jacobians and three channels (13–15). N2O increase of 4%

Fig. 4.

Same as Fig. 3 for the N2O-Jacobians and three channels (13–15). N2O increase of 4%

3) CO

Seasonal variations of the CO atmospheric concentration are very important, particularly in the Northern Hemisphere with maxima of about 200 ppb around March and minima of less than 90 ppb in July, flatter in the Southern Hemisphere with maxima of about 70 ppb in October and minima of about 50 ppb in spring (Novelli et al. 1992, 1998); see Fig. 5. Prior to 1990, CO in the lower troposphere increased at a rate of about 1% yr−1. A sharp decrease in CO was observed after 1991: the zonally weighted, global-average annual decrease was 2.3 ppb yr−1 and CO concentrations were, in 1998, 11% lower than those in 1991 (Novelli et al. 1998: see, in particular, Fig. 11 of this paper).

Fig. 5.

Three-dimensional representation of the latitudinal distribution of atmospheric CO in the marine boundary layer (figure from P. C. Novelli 2000, personal communication)

Fig. 5.

Three-dimensional representation of the latitudinal distribution of atmospheric CO in the marine boundary layer (figure from P. C. Novelli 2000, personal communication)

Fig. 11.

Latitude–month distribution of the number of collocations archived in the NESDIS DSD5 archive: (top) nighttime; (bottom) daytime

Fig. 11.

Latitude–month distribution of the number of collocations archived in the NESDIS DSD5 archive: (top) nighttime; (bottom) daytime

Carbon monoxide absorption significantly affects channels 13 and 14 brightness temperatures: 0.25 K (±0.08) and −0.1 K (±0.03), respectively, for a +40% variation in CO concentration. Figure 6 shows representative Jacobians with respect to CO concentration variations for these two channels, for a 40% increase in CO.

Fig. 6.

Same as Fig. 4 for the CO-Jacobians for a CO increase of 40%

Fig. 6.

Same as Fig. 4 for the CO-Jacobians for a CO increase of 40%

4) O3

Ozone seasonal variations and annual trends have been extensively measured and studied (see, e.g., World Meteorological Organization 1999; Logan 1999; Neuendorffer, 1996). Seasonal variations are significant in both hemispheres (see Fig. 7, McPeters and Labow 1996). Total ozone also shows sharp features that move rapidly in space and time. The 3-month window considered here smoothes out these signatures.

Fig. 7.

O3 total column (Dobson Unit) seasonal variation for the latitude zones 60°–20°N, 20°N–20°S, 20°–60°S (after McPeters and Labow 1996)

Fig. 7.

O3 total column (Dobson Unit) seasonal variation for the latitude zones 60°–20°N, 20°N–20°S, 20°–60°S (after McPeters and Labow 1996)

Trends in total ozone column vary, in the Northern Hemisphere, from −2% to −6% decade−1 with extrema that coincide with those of the seasonal cycles.

Ozone absorption influences the four longwave HIRS-2 channels considered here. An increase by 35% in the whole ozone column leads to a (gross mean) variation in brightness temperature of channels 2–5 of +0.08 (±0.05), +0.10 (±0.04), −0.10 (±0.04), −0.35 (±0.20), respectively. Examples of Jacobians with respect to ozone concentrations are shown in Fig. 8, for a 35% increase in O3.

Fig. 8.

Same as Fig. 3 for the O3-Jacobian and four channels (2–5). An O3 increase of 35%

Fig. 8.

Same as Fig. 3 for the O3-Jacobian and four channels (2–5). An O3 increase of 35%

b. Errors in the radiative transfer model input data

Radiosonde data are the main source of information for simulating brightness temperatures observed by satellite radiometers such as TOVS. They suffer from important limitations mostly due to the coupling between the radiosonde sensors and the radiative environment. Such errors vary from one sensor type to the next (Nash and Schmidlin 1987; UMM94). Luers and Eskridge (1998) analyze the limitations implied by the use of radiosonde temperatures in climate studies and review their sensitivities to the environment (day, night, solar angle, cloud cover, surface temperature, etc.). They indicate that climate trends (probably the most difficult quantitative conclusion to draw) can currently be estimated with good confidence for monthly averaged nighttime soundings, for various in situ sensors and with the proper correction model (see also Gaffen and Ross 1999).

Such models have been discussed in detail in UMM94 to which interested readers can refer. DSD5 collocations have in principle been corrected and are not expected to suffer from significant radiosonde radiative errors. Moreover, comparisons between model-simulated and satellite-observed brightness temperatures, separately for very homogeneous classes (air mass, day–night, land–sea, latitude zones) and for centered errors (global biases removed) should not significantly be affected by such problems.

More problematic is the extrapolation of the temperature profiles to upper levels: the situations contained in the DSD5 archive have to be extrapolated partly above 50 hPa and partly above 30 hPa. As explained in section 2, for the present study, and due to the lack of any other source of information, radiosonde profiles are extrapolated using the NESDIS retrievals included in the DSD5 archive. Reasons for this choice have been discussed in UMM94. A +1-K variation in the temperature profile above 50 hPa leads to a variation in the brightness temperatures of HIRS channels 2–5 of +0.6, +0.5, +0.20, 0.15 K, respectively, and of 0.15 K for MSU channel 4. As a consequence, a change in the retrieval methodology may have a direct consequence on the quality of the extrapolation: this is observed (see section 4b) in late 1988, at the time of the NOAA-11 launch, with the SSU instrument (UMM94) on board contrary to NOAA-10, and leading to an improvement of the retrieval mean accuracy. Diurnal phase errors were also observed due to the LECT difference between the two satellites, when one (NOAA-11) is used to complement the second (NOAA-10). See also section 4b.

Surface temperature raises another difficult boundary problem, particularly over land. The substitute used here, derived from the shortwave channel-18 observations (see section 2a), is expected to provide a poor estimate for daytime observations, particularly for high solar elevation angles (more important for NOAA-11 than for NOAA-10), due to solar radiation contamination, which is difficult to quantify and, thus, to correct for. The quality of this substitute may be quantitatively assessed by feeding the radiative transfer model (see section 2b) with this estimate of the surface temperature and comparing to the original level-1B observations. Standard deviations of about 0.2–0.3 K for channel 18 (however, the process is somewhat “incestuous”) and of about 0.6 K for channel 8, both over sea and land at night, are obtained. Mean errors are of the order of 0.2 to 0.3 K for the Tropics for channel 18 over sea, and about −1.5 K over land. For channel 8, mean errors range from 0.3 to 0.5 K over sea, and are around −0.5 over land. A large part of the mean error over land is due to the emissivity value chosen for the model: significantly too low for channel 18, slightly too low for channel 8. However, these results show that the estimated surface temperature is acceptable.

Among the channels considered here, channel 13 is the most sensitive to surface temperature. For a +1-K variation in the surface temperature, the channels 13, 14, 15, MSU2 and 5, display a variation in brightness temperature of (gross mean): 0.45, 0.30, 0.10, less than 0.10 (over land) and 0.04 K, respectively. All other channels are strictly not sensitive. As a consequence, an analysis of daytime biases for (at least) channels 13 and 14 is questionable when errors in the estimated surface temperature are expected to be large (several kelvins). However, nighttime surface temperature, as estimated by NESDIS from channel 18 observations, appears to be an acceptable trade-off for the present study (see section 2a).

Last, as expected because of the limited number of sites where it is measured, ozone distribution is poorly described in the DSD5 file: no profiles and total ozone contents, retrieved from HIRS channel 9, for less than 50% of the collocations during the NOAA-10 time period considered here. For these reasons, the ozone profile used in the forward model 3R is the one associated, in TIGR, to the obtained closest situation whose distance only depends on temperature and water vapor profiles. Consequently, on a seasonal point of view, it can be viewed as a random variable, (although it is in the right airmass class) particularly when looking at 3-month (or more) monthly running statistics. Indeed, HIRS channel 9 differences between model-simulated and satellite-observed brightness temperatures clearly reflect the seasonal cycle of ozone with a peak-to-peak amplitude of about 4 K in the Northern Hemisphere and 3 K in the Southern Hemisphere. These statistics give a clear view of where the signatures of the ozone seasonal variations are to be found. They are used here for the interpretation of channels 2–5 statistics.

c. Time collocation problems and diurnal cycle

NOAA sun-synchronous polar platforms sample the diurnal cycle twice a day at specific local times: roughly 0730 and 1930 for the so-called morning satellites (NOAA-10 and -12) and 0130 and 1330 for the afternoon satellites (NOAA-11). Radiosonde measurements are carried out at constant (with some flexibility) universal times: 0000, 0600, 1200, and 1800 UTC. Within the selected time collocation window, averaged over 3 months, the time difference between the radiosonde and the satellite observations may be significant: this is shown in Fig. 9, which displays, for DSD5 collocations, the mean value of the difference between the radiosonde and the satellite observation times for NOAA-10, over land at night. The isocontours give the number of collocations concerned. For the period July 1987–December 1988, for the latitude zone 20°–60°S mainly and marginally for the two others, and for a few collocations around July 1989 (zone 20°–60°S), this difference may reach 6 h. This period corresponds to the DSD5 “old” type 1 format (M. J. Uddstrom 2000, personal communication). Such differences may potentially result in significant diurnal cycle–induced trends. Similar, but even more complex problems are expected to occur over sea as the collocated radiosondes may have been time interpolated between bracketing true observations within a 6-h time window. For that reason, a figure similar to Fig. 9 over sea is not easy to reconstruct. This kind of problem no longer occurs after this time period for NOAA-10.

Fig. 9.

Monthly mean values of the difference between radiosonde and satellite observation times for NOAA-10, at night, over land, from Jul 1987 to Sep 1991: (a) 60°–20°N, (b) 20°N–20°S, (c) 20°–60°S. Isocontours give the number of collocations concerned

Fig. 9.

Monthly mean values of the difference between radiosonde and satellite observation times for NOAA-10, at night, over land, from Jul 1987 to Sep 1991: (a) 60°–20°N, (b) 20°N–20°S, (c) 20°–60°S. Isocontours give the number of collocations concerned

Systematic trends may also occur if the platform undergoes large drifts in its LECT, which has been the case for NOAA-12, and especially for NOAA-11. NOAA-10 experienced a drift of only 20 min (see section 2a), which therefore had a little influence here. These effects are expected to be greater over land because of a stronger diurnal cycle.

d. Impact of errors due to the satellite radiometers

Errors in the satellite-observed brightness temperatures, when not appropriately taken care of, can be interpreted as errors in the forward radiative transfer model and result in erroneous interpretations of the data in terms of atmospheric thermodynamic variables. A number of issues have been or continue to be analyzed by NESDIS (L. McMillin 2000, personal communication). An incorrect calibration of the radiometers is clearly an important issue and recent papers have drawn the attention on the MSU Sounder (see, e.g., Mo 1995 for MSU on NOAA-12). Again for MSU, Prabhakara et al. (1998) report on a diurnally dependent calibration problem.

Another critical issue, discussed by McMillin and Crosby (2000), is the solar contamination of shortwave HIRS channels and, in particular channel 15. The correction proposed is linear with respect to the solar zenith angle, with a large slope for NOAA-11, and a smaller one for NOAA-10, as expected. Since the statistics presented here are zonal, the potential errors should be small, but should depend on how regularly the collocations are distributed with latitude within each latitude zone. Other important investigated issues relate to the periodic noise observed in HIRS channels 1, 2, and 3, or to problems with the scanning mirror (L. McMillin 2000, personal communication). Although essential for an improved accuracy of the daily retrievals, they are not related to the present study of yearly trends or seasonal variations.

e. Unexpected events: The June 1991 Pinatubo volcano eruption

Mt. Pinatubo injected 20 megatons of sulfur dioxide, primarily into the stratosphere during its 1991 eruption in the Philippines. The conversion of SO2 into stratospheric sulfate aerosols caused a significant extinction in both the infrared and visible spectral regions. These particles were steadily removed and about three years after the eruption nearly all of the aerosols were gone (World Meteorological Organization 1999). During this period, significant modifications also occurred in the rate of increase of CO2, N2O, CO, and so on (see section 3a). Signatures of these changes are important for NOAA-11 and NOAA-12 (paper in preparation) but not for the period of time of NOAA-10 considered here.

4. Signatures of annual and seasonal variations of CO2 and other greenhouse gases in HIRS channels

Section 3 attempted to review all of the main potential sources of seasonal and annual variabilities of the TOVS channels, the list being certainly not exhaustive. Results and tentative explanations are now presented for the channels selected here: HIRS 2–5 and HIRS 13–15 and for the time period from July 1987 to September 1991 for NOAA-10.

Differences between modeled (using the 3R forward radiative transfer model) and observed (level-1B original Pathfinder TOVS archive) are first centered, over the whole period, with respect to the mean corresponding to their own airmass type (five classes: tropical, two midlatitude, and two polar), as determined from 3R (see section 2b). A 3- or 12-month monthly running mean is then applied, separately for the items corresponding to nighttime or daytime observations, over land or over sea, and to the three latitude bands: 60°–20°N, 20°N–20°S and 20°–60°S. For each channel and each latitude zone, separate statistics of the airmass-centered deviations between modeled and observed brightness temperatures for night–land, night–sea, day–land, day–sea are produced. Figure 10 shows 12-month monthly running standard deviations for representative combined classes and for HIRS channels 2, 5, 13, 14, and 15: they are almost constant for the time period considered.

Fig. 10.

Std dev of the airmass-centered 12-month monthly running mean differences between model-simulated and satellite-observed brightness temperatures for HIRS channels 2, 5, 13, 14, and 15: (a) 60°–20°N at night over land; (b) 20°–60°S at day over sea

Fig. 10.

Std dev of the airmass-centered 12-month monthly running mean differences between model-simulated and satellite-observed brightness temperatures for HIRS channels 2, 5, 13, 14, and 15: (a) 60°–20°N at night over land; (b) 20°–60°S at day over sea

As 12 combined classes will be considered (land, sea, night, day, and three latitude zones), only part of the results can be shown here, and the examples presented in the following have been selected for their representativeness: number of items in the statistics, structure of the variability. Figure 11 shows the number of items in the statistics (number of collocations on a monthly basis available from the DSD5 archive) for the whole latitude zone considered. The Northern Hemisphere appears as the one best represented.

a. HIRS shortwave tropospheric channels

HIRS shortwave channels 13, 14, and 15 are sensitive to N2O variations; channels 13 and 14 are also sensitive to CO variations; and channel 15 to CO2 variations. None of these channels is affected by extrapolation problems. Channel 13 is significantly sensitive to the surface temperature, channel 14 slightly sensitive, and channel 15 not sensitive at all (see sections 3a and 3b).

Because of its nearly total insensitivity to surface and upper-stratospheric boundary conditions, we first examine results for channel 15.

Because it is sensitive to both CO2 and N2O, the time series of the centered deviations (model simulated minus satellite observed) for channel 15 may be expected to do the following:

  • first, to show a slowly increasing 12-month monthly running mean (hereinafter CM-12: C for centered, M for mean, and 12 for the time period of reference) because of its tropospheric behavior and of increasing CO2 and N2O atmospheric concentrations when the model values are fixed; and

  • second, to show seasonal variations of CM-3 (3-month running mean) for the Northern Hemisphere zone (20°–60°N), because of the seasonal variation of CO2 concentration.

Results must generally be regarded cautiously when the number of items involved in the statistics is small (Fig. 11), and, in particular for channel 15 at day because of its sensitivity (although weak) to solar radiation (see section 3d). For an increase of 2% in CO2 and 1.3% in N2O during the satellite time period considered, the observed brightness temperatures are expected to decrease by 0.2 K (about 0.09 for CO2 and 0.11 for N2O), approximately. Figure 12 shows NOAA-10 channel 15 CM-12 time series for the Northern Hemisphere (20°–60°N). The trend for night/land observations (with a mean number of items of about 1500) is of the order of magnitude expected but irregular, whereas day–land seems to be too small and night–sea and day–sea too large. A possible solar contamination of channel 15 being smoothed out on centered yearly means, another explanation (if any) has to be found. Valuable information is provided by a comparison with the MSU channel 2, which integrates approximately the same midtropospheric layer but is only sensitive to oxygen absorption. Figure 13 shows MSU2 CM-12 time series for the same conditions as in Fig. 12. The well-known great stability of MSU should result in an almost constant zero mean if no other problems interfere. We have verified that the two above-mentioned boundary conditions (surface, upper stratosphere) have no impact.

Fig. 12.

NOAA-10 channel HIRS 15 12-month monthly running centered mean difference between model-simulated and satellite-observed brightness temperatures. Latitude zone: 20°–60°N; (a) nighttime, (b) daytime; over land (solid line), over sea (dashed line)

Fig. 12.

NOAA-10 channel HIRS 15 12-month monthly running centered mean difference between model-simulated and satellite-observed brightness temperatures. Latitude zone: 20°–60°N; (a) nighttime, (b) daytime; over land (solid line), over sea (dashed line)

Fig. 13.

Same as Fig. 12 for the channel MSU2

Fig. 13.

Same as Fig. 12 for the channel MSU2

The difference between radiosonde and satellite observation times thus appears to be the remaining potential source of problems. Figure 14 shows the mean time difference for the 20°–60°N zone over land, day and night (relative value: Fig. 14a; absolute value: Fig. 14b). Assuming that the departure of MSU2 CM-12 time series from zero is the signature of these time differences for a channel “observing” the midtroposphere, it may be expected that the difference between the time series of channels HIRS 15 and MSU2 CM-12 is the way to rid channel 15 from this source of error. Indeed, there is a good correspondence between channel 15 CO2 Jacobian (see Fig. 3) and the MSU2 weighting function (not shown). This difference is shown in Fig. 15 for the Northern Hemisphere (20°–60°N), which displays an almost constant increase up to the expected value of 0.2 K. This result may be confirmed by noting that the mean time difference between radiosonde and satellite observations obtained when grouping the land–sea and day–night items almost vanishes. It is then verified (Fig. 16), first, that the same grouping of data results in an almost constant zero bias for the MSU2 time series, and, second, that the corresponding MC-12 channel 15 time series regularly increases up to a total value of about 0.2 K. The number of items involved is also shown. Similar results are obtained for the two other latitude zones. From these results, it may be concluded that channel 15 CM-12 increase (slightly less than 50% due to CO2, slightly more than 50% due to N2O) is a good indicator of the sum of the yearly increases of CO2 (1.4 ppm yr−1) and N2O (0.7 ppb yr−1) atmospheric concentrations.

Fig. 14.

Mean [(a) relative, (b) absolute value] time difference between radiosonde and satellite observation times for the 20°–60°N latitude zone. Daytime (solid line) and nighttime (dashed line)

Fig. 14.

Mean [(a) relative, (b) absolute value] time difference between radiosonde and satellite observation times for the 20°–60°N latitude zone. Daytime (solid line) and nighttime (dashed line)

Fig. 15.

Difference between HIRS channel 15 (Fig. 12) and MSU2 (Fig. 13) CM-12 time series. Latitude zone 20°–60°N; nighttime; over land (solid line), over sea (dashed line). Similar results are obtained for daytime observations. The corresponding increase in CO2 concentration is also shown (right ordinate; Masarie and Tans 1995)

Fig. 15.

Difference between HIRS channel 15 (Fig. 12) and MSU2 (Fig. 13) CM-12 time series. Latitude zone 20°–60°N; nighttime; over land (solid line), over sea (dashed line). Similar results are obtained for daytime observations. The corresponding increase in CO2 concentration is also shown (right ordinate; Masarie and Tans 1995)

Fig. 16.

Same as Fig. 12 for HIRS channel 15 (dashed line) and MSU2 (solid line) for the latitude zone 20°–60°N, the classes day, night, land, and sea being merged. Number of items: dotted line and right ordinate

Fig. 16.

Same as Fig. 12 for HIRS channel 15 (dashed line) and MSU2 (solid line) for the latitude zone 20°–60°N, the classes day, night, land, and sea being merged. Number of items: dotted line and right ordinate

At a seasonal scale, the MC-3 channel 15 time series also display interesting features that seem to be related to the seasonal variation of CO2 concentration since N2O does not show any clear seasonal cycle. Figure 17 shows the results, grouping day and night, land and sea, for the 20°–60°N zone (Fig. 17a) and for the 20°–60°S zone (Fig. 17b), together with the corresponding number of items included in these 3-month monthly running mean statistics. Figure 17a reflects relatively well the CO2 seasonal cycle with maximum values approximately located in April–May, corresponding to the maximum values of the CO2 concentration and minimum values approximately located in September–October, close to the minimum values of the CO2 seasonal cycle. The mean surface peak-to-peak concentration variation of CO2 for this latitude zone is of the order of 13 ppm (see Fig. 2), and approximately corresponds to a 0.15-K variation in channel 15 brightness temperature. The value observed on Fig. 17a is somewhat smaller: this is due to the fact that this channel integrates layers where the CO2 peak-to-peak concentration variations are smaller (see section 3a). Seasonal oscillation irregularities are mainly seen before and after July 1989 where the number of items involved in the statistics dramatically drops down. Figure 17b, for the 20°–60°S zone, shows a small (less than 0.1 K) biannual oscillation, also seen on the land plus sea statistics at night (not shown), which is not easy to interpret: CO2 shows almost no seasonal cycle in the Southern Hemisphere, N2O has no documented cycle in either hemispheres, and similar signatures are observed nighttime or daytime. Small cyclic variations of the difference between radiosounding and satellite time observations on the one hand, and radiative transfer model errors due to cyclic variations of the collocation latitudes, on the other hand, may be the sources of these small oscillations.

Fig. 17.

HIRS channel 15 3-month monthly running centered mean difference between radiosonde-simulated and satellite-observed brightness temperatures, for the latitude zones (a) 20°–60°N and (b) 20°–60°S. Number of items in the statistics: dotted line and right ordinate. The classes day, night, land, and sea are merged

Fig. 17.

HIRS channel 15 3-month monthly running centered mean difference between radiosonde-simulated and satellite-observed brightness temperatures, for the latitude zones (a) 20°–60°N and (b) 20°–60°S. Number of items in the statistics: dotted line and right ordinate. The classes day, night, land, and sea are merged

To continue with the HIRS shortwave tropospheric channels, HIRS channel 14 is expected, first, to display a slow regular increase in its CM-12 due to its sensitivity to increasing N2O atmospheric concentration, up to about 0.15 K (gross mean) over the NOAA-10 lifetime, and, second, to show seasonal variations in its CM-3 because of the seasonal variation of atmospheric CO concentration. In addition, channel 14 is sensitive to surface temperature: we have checked that, at least for nighttime statistics, the surface temperature estimate (see sections 2a and 3b) is accurate enough not to create perceptible problems on the CM-3 time series when grouping land and sea–night observations. Although somewhat irregular because of small differences between radiosonde and satellite observation times (here, no MSU “equivalent” channel is available to eliminate the influence of the time shift), the total increase in the CM-12 time series amounts to about 0.15 K for both 20°–60°N and 20°–60°S latitude zones (see Fig. 18a), as expected.

Fig. 18.

HIRS channel 14 (a) 12-month and (b) 3-month monthly running means (CM-12 and CM-3 time series), at night over land and sea: 20°–60°N: solid line; 20°–60°S: dashed line

Fig. 18.

HIRS channel 14 (a) 12-month and (b) 3-month monthly running means (CM-12 and CM-3 time series), at night over land and sea: 20°–60°N: solid line; 20°–60°S: dashed line

Figure 18b shows the CM-3 time series (land and sea at night): low values associated with a rather sharp peak in July and high values in February–March in the Northern Hemisphere, low values in spring, and high values in October in the Southern Hemisphere are in good agreement with the known CO seasonal cycle while the mean peak-to-peak variation of about 0.3 K for the 20°–60°N zone and about 0.15 K for the 20°–60°S zone are of the expected order of magnitude (approximately 0.25 and 0.1 K, respectively) although somewhat larger, particularly if the CO cycle is flatter in altitude than it is at the surface. This could be due to remaining spectroscopic problems in the forward model. However, these results confirm that channel 14 is a good indicator of the year-to-year increase in N2O and of the seasonal variation of atmospheric CO.

Quite similar results are obtained from channel 13, the differences with channel 14 being in agreement with what is expected from their different sensitivities to N2O (smaller for channel 13 than for channel 14) and CO (larger for channel 13 than for channel 14). First, channel 13 is expected to display a slow regular increase of about 0.1 K in its CM-12 over the NOAA-10 period, in good agreement with what is really observed (see Fig. 19a) for the two zones, 20°–60°N and 20°–60°S. Here again, irregularities in the trends are mostly due to varying time differences between radiosonde and satellite observations. Second, channel 13 is expected to show seasonal variations of its CM-3, which are slightly larger than those for channel 14 due to its increased sensitivity to CO variations. Expected peak-to-peak values are about 0.5 K (gross mean) for a doubling of CO between February–March and July in the Northern Hemisphere and a value of 0.20–0.25 K for the Southern Hemisphere. These results are in relatively good agreement with those of Fig. 19b (see above comments on channel 14).

Fig. 19.

Same as Fig. 18 for HIRS channel 13

Fig. 19.

Same as Fig. 18 for HIRS channel 13

As a verification of the relatively good quality of the surface temperature substitute used here, and because channel 13 is quite sensitive to this variable, Fig. 20 shows the CM-3 time series for HIRS channel 8 (a window channel) for the two zones (20°–60°N and 20°–60°S), for nighttime observations over land and sea. No important signatures, capable of substantially modifying the above conclusions, may be seen from this figure. Between both channels 13 and 14, the former seems to be best suited for following CO variations and the latter for following N2O increase.

Fig. 20.

Same as Fig. 18 for HIRS channel 8 (midinfrared window)

Fig. 20.

Same as Fig. 18 for HIRS channel 8 (midinfrared window)

b. HIRS longwave tropospheric to stratospheric channels

HIRS longwave channels 2–5 are sensitive to CO2. They display different behaviors and different sensitivities. They are also sensitive to O3 (see section 3a). The shape of the temperature weighting functions of the longwave channels make them very (channel 2) to slightly (channel 5) sensitive to the quality of the upper-stratosphere temperature profile, and, consequently, to the quality of extrapolation (see section 3b). None of these channels is significantly affected by errors in the surface temperature, provided they are small enough, which is the case here.

Channel 5, a midtropospheric temperature channel, as well as an upper-tropospheric CO2 channel (see Fig. 3), is the most sensitive to CO2, but also to O3. This is clearly seen on channel 5 CM-3 time series in the Southern Hemisphere (20°–60°S) where no significant seasonal variations of CO2 occur. Figure 21 shows, superimposed, the CM-3 series of channels 5 and 9 (the HIRS ozone channel) for the classes night–land (Fig. 21a) and day–land (Fig. 21b). The large signatures of the ozone seasonal variations displayed by channel 9 (see section 3b for details) are reflected by channel 5, although with a time lag of about one month, which is almost certainly due to the different ozone response functions of both channels (see Fig. 8). In the Northern Hemisphere (20°–60°N), channel 5 responds to both the seasonal variations of CO2 and O3, which are relatively well in phase for the high peak values (O3 peaks slightly later) and less in phase for the low peak values (August–Sepember for CO2, November–December for O3). Figure 22 shows the channel 5 CM-3 time series for the class land/day plus night (which contains the largest number of items; see Fig. 11). This figure approximately reflects the mixed cycles of CO2 and O3, mostly for the high peak values, less clearly for the low peak values, particularly for the autumn of 1989. Seasonal oscillations due to CO2 are not expected to be large because the CO2 response function of channel 5, as it reaches substantially higher into the atmosphere than that of channel 15 (see Fig. 3), integrates layers that are much less affected by CO2 seasonal variations. What is observed is certainly more related to O3 than to CO2 seasonal variations.

Fig. 21.

HIRS channels 5 and 9 (ozone) 3-month monthly running means (CM-3 time series) for the latitude zone 20°–60°S over land at (a) night and (b) day. Channel 5: solid line; channel 9: dashed line

Fig. 21.

HIRS channels 5 and 9 (ozone) 3-month monthly running means (CM-3 time series) for the latitude zone 20°–60°S over land at (a) night and (b) day. Channel 5: solid line; channel 9: dashed line

Fig. 22.

HIRS channel 5 CM-3 time series, nighttime and daytime over land, for the latitude zone 20°–60°N

Fig. 22.

HIRS channel 5 CM-3 time series, nighttime and daytime over land, for the latitude zone 20°–60°N

Channel 5 CM-12 time series are expected to show a slow regular increase throughout the NOAA-10 time period analyzed here, up to a total of about 0.2 K (CO2 increase of 2%). Figure 23 shows the 20°–60°N CM-12 channel 5 time series over land, day and night. Superimposed, for the same conditions, is the CM-12 time series of the difference between channels 5 and MSU2, which, as already explained in section 4a for channel 15, is expected to cancel out the time difference between radiosonde and satellite observations (see section 4a). More regular than the former, the time-corrected time series increase amounts to about 0.1 instead of 0.2 K. Similar results are observed for other classes. A detailed analysis of channel 5 responses shows that upper-stratosphere temperature extrapolation problems are the main source for this deficit. As reported in UMM94 (see section 3b) and by Finger et al. (1993), important changes were made in the procedure used to extrapolate the temperature profiles archived in the DSD5 files; namely, a change from the statistical to a physical retrieval algorithm in late September 1988 and a change from the SSU on board NOAA-9 to that on board NOAA-11 in late October 1988.

Fig. 23.

HIRS channel 5 CM-12 time series (dashed line) and difference between the CM-12 time series of channels HIRS-5 and MSU2 (solid line) for 20°–60°N, over land, day and night

Fig. 23.

HIRS channel 5 CM-12 time series (dashed line) and difference between the CM-12 time series of channels HIRS-5 and MSU2 (solid line) for 20°–60°N, over land, day and night

Channel 2, whose temperature response function reaches relatively high into the stratosphere (around 60 hPa), clearly bears the signature of this evolution. Figure 24 shows the CM-3 channel 2 time series for the latitude zone 20°N–20°S (chosen for the greater coherence of tropical temperature profiles) and the class land/day and night: a sharp drop (≈0.6 K) is seen at the end of 1988 and at the beginning of 1989 (number of items is also shown in Fig. 24). According to the respective sensitivities of channels 2–5 to stratospheric temperatures, this drop of 0.6 K in channel 2 mean brightness temperatures approximately corresponds to a drop of 0.1–0.15 K in channel 5 brightness temperatures. Before this return to more realistic values, too high stratospheric temperatures have artificially increased model-simulated brightness temperatures, as well as their deviations from satellite observations. A correction of about +0.1 K to the trend seen in Fig. 23 (channel 5 minus channel MSU2 CM-12 time series) gives the expected value of 0.2 K. However, if this scenario seems to be quite convincing, it cannot preclude other possible, more intricate, explanations.

Fig. 24.

HIRS channel 2 CM-3 time series, over land, day and night, for the latitude zone 20°N–20°S (solid line); number of items involved in the statistics (dashed line and right ordinate)

Fig. 24.

HIRS channel 2 CM-3 time series, over land, day and night, for the latitude zone 20°N–20°S (solid line); number of items involved in the statistics (dashed line and right ordinate)

As a whole, since it is sensitive to both CO2 and O3 variations as well as to stratospheric temperatures, channel 5 does not appear to be a good indicator of CO2 variations (neither seasonal, nor annual) in the context of the NOAA-10 DSD5 archive (no ozone profiles, problematic stratospheric extrapolation).

Channel 4, and even more so, channel 3, which are less sensitive to both CO2, and O3, but are more sensitive to stratospheric temperatures, show a mixed tropospheric and stratospheric behavior that results in weak annual trends (CM-12 time series) and in CM-3 time series either resembling that of channel 5 (case of channel 4) or that of channel 2 (case of channel 3).

HIRS channel 2 as a CO2 stratospheric channel, is expected to display weakly decreasing CM-12 time series, less than 0.1 K over the NOAA-10 time period considered here. A much larger value is observed, as shown, for example, in Fig. 25 (20°–60°N, land–night) or Fig. 24. Actually, channel 2 CM-12 or CM-3 time series are dominated by the signal created by the varying stratospheric temperature extrapolation methods.

Fig. 25.

HIRS channel 2 CM-12 time series over land at night, for the latitude zone 20°–60°N

Fig. 25.

HIRS channel 2 CM-12 time series over land at night, for the latitude zone 20°–60°N

Last, among the HIRS longwave channels having a weak sensitivity to surface conditions or to water vapor absorption, only channel 5 could be a good indicator of the yearly increase and of the seasonal variations of CO2, provided accurate O3 and stratospheric temperature measurements are available for simulating the observed brightness temperatures.

5. Conclusions and perspectives

For the last 20 years, NOAA polar satellites have fulfilled their operational mission of observing the structure of the atmosphere in temperature and water vapor through the “interpretation” of the HIRS and MSU radiances. All presently developed interpretation algorithms more or less directly rely on the comparison between a set of observed and a set of simulated radiances. For that reason, the accuracy of the simulation directly influences that of the interpretation of radiances in terms of thermodynamic variables. Comparing simulations to observations is the key to a better knowledge of the main sources of errors affecting either the former or the latter. Instrumental radiometric problems, radiosonde, and forward radiative transfer model limitations, as well as difficulties raised by differences in space and in time of satellite and radiosonde observations (collocations) have long been studied in details. Less attention has been paid to errors, presumed negligible, generated by the absence of consideration of main absorbing gases (CO2, N2O, CO, O3, etc.) atmospheric seasonal cycles and/or annual trends.

The present study has shown that a detailed analysis, at different timescales (seasonal, annual) of the departures between simulated and observed NOAA TOVS brightness temperatures reveals signatures of these greenhouse gases concentration variations. Not only the shape of the seasonal variations (locations of the peaks) is in good agreement with what we presently know, but also their amplitude (peak-to-peak) matches relatively well the values computed from a line-by-line radiative transfer model. Moreover, annual trends correspond quite well with the known increase in concentration of gases like CO2 or N2O, as a result of human activities. Limits of such an analysis are in the number and distribution of the collocations, in an insufficient knowledge of conflicting variables (surface characteristics, upper-stratospheric temperatures, ozone profiles, etc.) and in the mixing of two gas signatures (CO2 and O3, N2O and CO, etc.). This results from the rather modest spectral resolution of TOVS channels (the ratio of the central wavelength to the channel half-width varies between 50 and 100) that integrate signatures from several absorbers and from many atmospheric layers. However, results from this work leave some hope to extract from the more than 20-yr archive of the TOVS observations interesting information on at least CO2, N2O, and CO (HIRS channel 11, not considered here, probably bears signature of methane, although certainly difficult to extract) that might bring significant constraints to biogeochemical greenhouse gas cycles.

Not relying on collocations between satellite and radiosonde observations, a promising approach is now being developed based on the coupling between HIRS infrared channels and MSU microwave channels. Because the former are sensitive to both atmospheric temperature and trace gas concentration, when the latter, measuring in the oxygen absorption band, are only sensitive to temperature, a differential approach, applied to coarse midtropospheric layers, may be carried out to extract the trace gas signal (Chédin et al. 2001, manuscript submitted to J. Geophys. Res.). This method will use all clear field of views available for each orbit, bringing a large number of items per month for the (relatively large) areas presently considered for the reconstruction of the atmospheric carbon dioxide distribution. Thinner atmospheric layers could be considered for NOAA-15 and NOAA-16 equipped with the Advanced Microwave Sounding Unit (AMSU A and B) that offers 20 channels instead of 4 for MSU.

These results also strengthen our hope to greatly improve our knowledge of the global distribution of a variety of radiatively active gases with the coming second-generation vertical sounders such as NASA's Advanced Infrared Radiation Sounder (AIRS) or the CNES/EUMETSAT Infrared Atmospheric Sounder Interferometer (IASI), both characterized by a much higher spectral resolution (at least 1200). Much smaller spectral intervals, specific of one absorbing gas, and much thinner atmospheric layers will be observed and their analysis made much simpler (one channel, one signature) than that of the present TOVS channels. Both instruments are also to be flown in association with AMSU.

Perhaps more than their basic mission (1-K temperature accuracy for 1-km-thick layers and 10% relative humidity accuracy for 1–2-km-thick layers), both instruments should open the way to accurately measure the atmospheric abundance of all essential greenhouse gases and, for most of them, to infer the most important features of their vertical distribution.

Acknowledgments

We are particularly happy to thank Michael Uddstrom and Larry McMillin for fruitful exchanges on the NOAA/NESDIS DSD5 archive, and John Christy for helpful comments. Many thanks are due to Ellen Brown and Cecil Paris for making the satellite–radiosonde collocation files available to us. We warmly thank Roy Jenne and Dennis Joseph from NCAR, Peter Topoly and Alex Kidd from NOAA/NESDIS, and George Serafino from NASA GSFC for having made the level-1B TOVS data available to us on the occasion of our involvement in the NOAA–NASA Pathfinder program.

REFERENCES

REFERENCES
Achard
,
V.
,
1991
:
Trois problèmes clés de l'analyse 3D de la structure thermodynamique de l'atmosphère par satellite.
Thèse de doctorat, Université Paris 7, 168 pp. [Available from Laboratoire de Météorologie Dynamique, Ecole Polytechnique, 91128 Palaiseau Cedex, France.]
.
Chédin
,
A.
,
N. A.
Scott
,
C.
Wahiche
, and
P.
Moulinier
,
1985
:
The improved initialization inversion method: A high resolution physical method for temperature retrievals from satellites of the TIROS-N series.
J. Climate Appl. Meteor
,
24
,
128
143
.
Chédin
,
A.
,
N. A.
Scott
,
C.
Claud
,
B.
Bonnet
,
J.
Escobar-Munoz
,
S.
Dardaillon
,
F.
Cheruy
, and
N.
Husson
,
1994
:
Global scale observation of the Earth for climate studies.
Adv. Space Res
,
14
,
155
159
.
Chevallier
,
F.
,
F.
Cheruy
,
N. A.
Scott
, and
A.
Chédin
,
1998
:
A neural network approch for a fast and accurate computation of a longwave radiative budget.
J. Appl. Meteor
,
37
,
1385
1397
.
Christy
,
J. R.
,
R. W.
Spencer
, and
R. T.
Mc Nider
,
1995
:
Reducing noise in the MSU daily lower-tropospheric global temperature dataset.
J. Climate
,
8
,
888
896
.
Elkins
,
J. W.
, and
and Coauthors
,
1996
:
Nitrous oxide and halocompounds.
Climate Monitoring and Diagnostics Laboratory—No. 23, Summary Report 1994–1995, D. J. Hofmann, J. T. Peterson, and R. M. Rosson, Eds., NOAA/ERL
.
Escobar-Munoz
,
J.
,
1993
:
Base de données pour la restitution de variables atmosphériques à l'échelle globale.
Thèse de doctorat, Université Paris 7, 190 pp. [Available from Laboratoire de Météorologie Dynamique, Ecole Polytechnique, 91128 Palaiseau Cedex, France.]
.
Finger
,
F. G.
,
M. E.
Gelman
,
J. D.
Wild
,
M. L.
Chanin
,
A.
Hauchecorne
, and
A. J.
Miller
,
1993
:
Evaluation of NMC upper-stratospheric temperatures analyses using rocketsonde and lidar data.
Bull. Amer. Meteor. Soc
,
74
,
789
799
.
Flobert
,
J-F.
,
N. A.
Scott
, and
A.
Chédin
,
1986
:
A fast model for TOVS radiances computation.
Proc. Sixth Conf. on Atmospheric Radiation, Williamsburg, PA, Amer. Meteor. Soc., 186–189
.
Gaffen
,
D. J.
, and
R. J.
Ross
,
1999
:
Climatology and trends of U.S. surface humidity and temperature.
J. Climate
,
12
,
811
828
.
Garand
,
L.
, and
and Coauthors
,
2000
:
Radiance and Jacobian intercomparison of radiative transfer models applied to HIRS and AMSU channels.
J. Geophys. Res., in press
.
GLOBALVIEW—CO2
,
1999
:
Cooperative Atmospheric Data Integration Project—Carbon Dioxide.
NOAA/CMDL, CD-ROM. [Also available online at ftp.cmdl.noaa.gov, Path: ccg/co2/GLOBALVIEW.]
.
Keeling
,
C. D.
,
T. P.
Whorf
,
M.
Wahlen
, and
J.
van der Plicht
,
1995:
:
Interannual extremes in the rate of rise of atmospheric carbon dioxide since 1980.
Nature
,
375
,
666
670
.
Khalil
,
M. A. K.
, and
R. A.
Rasmussen
,
1983
:
Increase and seasonal cycles of nitrous oxide in the earth's atmosphere.
Tellus
,
35B
,
161
169
.
Logan
,
J. A.
,
1999
:
An analysis of ozonesonde data for the lower stratosphere: Recommendations for testing models.
J. Geophys. Res
,
104
,
16151
16170
.
Luers
,
J. K.
, and
R. E.
Eskridge
,
1998
:
Use of radiosonde temperature data in climate studies.
J. Climate
,
11
,
1002
1019
.
Masarie
,
K. A.
, and
P. P.
Tans
,
1995
:
Extension and integration of atmospheric carbon dioxide data into a globally consistent measurement record.
J. Geophys. Res
,
100
,
11593
11610
.
McMillin
,
L.
, and
D. S.
Crosby
,
2000
:
A study of the solar radiation effect on the 4.3 micron channels of the HIRS.
J. Appl. Meteor
,
39
,
1791
1796
.
McPeters
,
R. D.
, and
G. J.
Labow
,
1996
:
An assessment of the accuracy of 14.5 years of Nimbus-7 TOMS Version 7 ozone data by comparison with the Dobson network.
Geophys. Res. Lett
,
23
,
3695
3698
.
Mo
,
T.
,
1995
:
A study of microwave sounding unit on the NOAA-12 satellite.
IEEE Trans. Geosci. Remote Sens
,
33
,
1141
1152
.
Nakazawa
,
T.
,
K.
Miyashita
,
S.
Aoki
, and
M.
Tanaka
,
1991
:
Temporal and spatial variations of upper tropospheric and lower stratospheric carbon dioxide.
Tellus
,
43B
,
106
117
.
Nash
,
J.
, and
F. J.
Schmidlin
,
1987
:
WMO international radiosonde comparison: Final report.
WMO Instruments and Observing Methods, S. Houvila, Ed., Vol. 30, WMO, 103 pp
.
Neuendorffer
,
A. C.
,
1996
:
Ozone monitoring TIROS-N operational vertical sounders.
J. Geophys. Res
,
101
,
18807
18828
.
Novelli
,
P. C.
,
L. P.
Steele
, and
P. P.
Tans
,
1992
:
Mixing ratios of carbon monoxide in the troposphere.
J. Geophys. Res
,
97
,
20731
20750
.
Novelli
,
P. C.
,
K. A.
Masarie
, and
P. M.
Lang
,
1998
:
Distributions and recent changes of carbon monoxide in the lower troposphere.
J. Geophys. Res
,
103
,
19015
19033
.
Prabhakara
,
C.
, and
R.
Iacovazzi Jr.
,
1999
:
Comments on “Analysis of the merging procedure for the MSU daily temperature time series.”.
J. Climate
,
12
,
3331
3334
.
Prabhakara
,
C.
,
R.
Iacovazzi Jr.
,
J. M.
Yoo
, and
G.
Dalu
,
1998
:
Global warming deduced from MSU.
Geophys. Res. Lett
,
25
,
1927
1930
.
Prinn
,
R. G.
,
R. F.
Weiss
,
P. J.
Fraser
,
P. G.
Simmonds
,
F. N.
Alyea
, and
D. M.
Cunnold
,
1998
:
The ALE/GAGE/AGAGE database.
DOE-CDIAC World Data Center, Dataset No. DB-1001
.
Sarmiento
,
J. L.
,
1993
:
Atmospheric CO2 stalled.
Nature
,
365
,
697
698
.
Scott
,
N. A.
, and
A.
Chédin
,
1981
:
A fast line-by-line method for atmospheric absorption computations: The Automatized Atmospheric Absorption Atlas.
J. Appl. Meteor
,
20
,
801
812
.
Scott
,
N. A.
, and
and Coauthors
,
1999
:
Characteristics of the TOVS pathfinder path-B dataset.
Bull. Amer. Meteor. Soc
,
80
,
2679
2701
.
Schauffler
,
S. M.
, and
J. S.
Daniel
,
1994
:
On the effects of stratospheric circulation changes on trace gas trends.
J. Geophys. Res
,
99
,
25747
25754
.
Schneider
,
H. R.
,
D. B.
Jones
,
C. S.
Wofsy
, and
M. B.
McElroy
,
2000
:
Analysis of residual mean transport in the stratosphere. Part II: Distribution of CO2 and mean age.
J. Geophys. Res
,
105
,
20013
20024
.
Smith
,
W. L.
,
H. M.
Woolf
,
C. M.
Hayden
,
D. Q.
Wark
, and
L. M.
McMillin
,
1979
:
The TIROS-N operational vertical sounder.
Bull. Amer. Meteor. Soc
,
60
,
1177
1187
.
Uddstrom
,
M. J.
, and
L. M.
McMillin
,
1994
:
System noise in the NESDIS TOVS forward Model. Part I: Specification.
J. Appl. Meteor
,
33
,
919
938
.
Uddstrom
,
M. J.
, and
A. V.
Korpela
,
1999
:
The DSD5 Archive: An archive of collocated radiosonde and TOVS data.
NIWA Tech. Memo. WLG99/51, 65 pp
.
Wofsy
,
S. C.
,
K. A.
Boering
,
B. C.
Daube Jr.
, and
M. B.
McElroy
,
1994
:
Vertical transport rates in the stratosphere in 1993 from observations of CO2, N2O, and CH4.
Geophys. Res. Lett
,
21
,
2571
2574
.
World Meteorological Organization
,
1999
:
Scientific assessment of ozone depletion: 1998.
WMO Rep. 44, Global Ozone Research Monitoring Project, Geneva, Switzerland, 55 pp
.

Footnotes

Corresponding author address: Dr. Alain Chédin, Laboratoire de Météorologie Dynamique, Ecole Polytechnique, Palaiseau Cedex 91128, France. Email: chedin@araf1.polytechnique.fr