1. Introduction

The High Resolution Infrared Radiation Sounder (HIRS) has been on board the National Oceanic and Atmospheric Administration (NOAA) polar-orbiting satellite series (hereafter abbreviated as N#, where # is the satellite number) for nearly 30 yr. The global measurements provide a valuable dataset to assess long-term variability of the atmospheric properties. For example, Wu et al. (1993) analyzed the three water vapor channels of HIRS and obtained statistics of the measurements from 1981 to 1988. Using a longer time series, McCarthy and Toumi (2004) focused on the upper-troposphere relative humidity derived from HIRS channel 12, analyzed its interannual variability, and had extensive discussion on the relationship with the El Niño–Southern Oscillation events. Wylie et al. (2005) extended the statistics from an earlier study of high clouds (Wylie and Menzel 1999) and derived trends in global cloud cover for both high clouds and all cloud types based on two decades of HIRS data. Recently, Liu et al. (2006) used HIRS measurements to investigate the temperature inversion strength in the Arctic from 1980 to 1996. The datasets of many of these long-term trend studies originated from the operational level-1b data, which were calibrated independently for each satellite. Because of the independent calibration that is based on each individual HIRS channel spectral response function (SRF) along with other factors, biases exist from satellite to satellite. These intersatellite biases have become a common source of uncertainty faced by long-term trend studies.

Efforts have been made in the past to adjust these HIRS intersatellite biases. Bates et al. (1996) intercalibrated 13 yr (1981–93) of HIRS with an empirical method. In the analysis, the data were binned on 2.5° × 2.5° grids for 5-day averaged data. A 13-yr mean for each location based on each pentad and each satellite was calculated. The anomaly relative to this mean was then computed for the dataset. The anomalies were compared empirically between two satellites for the same target assuming that the statistical distributions of anomalies were the same. Based on the comparison, the satellites were adjusted to a base satellite (N7) to produce a self-consistent dataset of global observations. This intersatellite calibrated dataset facilitated studies of the variability and trends of the upper-tropospheric humidity (Geer et al. 1999; Bates and Jackson 2001; Bates et al. 2001). Other studies on intersatellite calibration were also carried out on HIRS-derived outgoing longwave radiation to produce a long-term climate dataset (Lee et al. 2007).

Cao et al. (2005) used simultaneous nadir overpass (SNO) observations to intercompare radiances measured by HIRS on board N15, N16, and N17. The SNO observations were collocated at the satellite nadir within a few seconds. The method was developed to quantify the observed radiance differences measured by HIRS on different satellites with little ambiguity. Using SNO intercomparisons for channels sensing stratosphere, the study found that the seasonal bias variation in the stratosphere channels is highly correlated with the lapse rate factor. The study also provided extensive discussion on possible causes for HIRS radiance biases.

A similar method was applied to another sounder on board the NOAA series of polar satellites, the Microwave Sounding Unit (MSU), to obtain intersatellite calibrated data. Zou et al. (2006) developed a calibration algorithm consisting of a dominant linear response of the MSU raw counts to the Earth view radiance plus a smaller quadratic term. An intersatellite calibrated MSU pentad dataset from N10 to N14 was obtained based on the SNO analysis. The work also showed the application of intersatellite calibrated data in examining the anomaly trend of MSU channel-2 brightness temperatures over the oceans. Subsequently, Iacovazzi and Cao (2007) applied a similar SNO analysis to quantify the brightness temperature biases of the Advanced Microwave Sounding Unit-A instruments between the Earth Observing System Aqua and NOAA satellites N15, N16, and N18. Wang et al. (2007) used SNO observations from the Atmospheric Infrared Sounder on Aqua to access the radiance accuracy of N16 HIRS. These studies showed numerous applications of the SNO method.

Further examination of HIRS SNO radiance datasets developed by Cao et al. (2005) reveals that for some of the channels, the intersatellite biases vary significantly with the observed scene radiance. The present study analyzes the HIRS longwave channels (channels 1–12) and focuses primarily on the radiance-dependent channels. Most of the examples are based on N16 and N17 comparisons, but discussion of other satellites is included as well.

2. HIRS channel characteristics

The HIRS is one of the primary instruments on board the NOAA polar-orbiting satellite series since late 1978. The sounder provides routine measurement of the atmosphere on a global scale. It has 20 spectral channels, 12 of which are longwave channels. Among the longwave channels, channels 1–7 are located in the 15-μm carbon dioxide absorption band. The channel frequencies are selected to sound different levels of the atmosphere from the surface to the stratosphere. Channel 8 is a window channel. Channel 9 is in the ozone band. Channel 10 is in the water vapor continuum band. Channels 11 and 12 are the water vapor rotational–vibrational channels. Channels 13–19 center at wavelengths ranging around 4 μm (shortwave window and carbon dioxide absorption). Channel 20 is a visible channel.

The center wavenumbers of the HIRS longwave channels from N6 to N17 are provided in Table 1. The channel characteristics usually differ slightly from satellite to satellite. However, significant changes to a few channels have also occurred. For example, the HIRS/2 design was used in the N6–N14 satellites, and HIRS/3 was used in N15–N17. When the instrument was switched to the HIRS/3 design, the center wavenumber of channel 12 changed from near 1480 to near 1530 cm⁻¹. Channel 10 also had two significantly different sets of wavenumbers in the satellite series (near 1220 cm⁻¹ for N6–N10 and N12, and near 800 cm⁻¹ for N11 and N14–N17).

Figure 1 shows the transmission spectrum in the infrared field, where the HIRS sounding channels are located. The transmission values are calculated by using a moderate-resolution radiative transfer model (MODTRAN4) for a typical late-spring, midlatitude, cloud-free atmospheric profile. A detailed description of the MODTRAN4 model can be found in Acharya et al. (1999) and Berk et al. (2003).

The HIRS are discrete stepping, line-scan instruments. The optical field of view is 1.3° in the longwave band, which encompasses an area of ∼20 km at nadir on the earth. The calibration of individual HIRS infrared channels is based on a view of a warm target mounted to the instrument base and a view of the space every 40 scan lines. Details of the HIRS instruments are provided by the NOAA polar orbiter data user’s guide (Kidwell 1998) and the NOAA KLM (N15, N16, and N17) user’s guide (Robel 2006).

3. Dataset

The data analyzed in this study are derived from the Environmental Services Data and Information Management project (Cao et al. 2005). The SNOs occur when two satellites cross each other. These satellite intersections are found in the regions +70° to +80°, and −70° to −80° latitude zones once every 8–9 days. The data processing procedure has been described in detail by Cao et al. (2005). In summary, the radiance data for the matching orbits were obtained based on the orbital prediction of the SNO between pairs of satellites using orbital perturbation models and associated two-line elements (Cao et al. 2004b). The matching nadir pairs are found by searching the pixels with a nadir distance less than 20 km and a time difference less than 30 s. A spatial subset of HIRS radiance data is extracted within 30 scan lines around the SNO pixel. A pixel-by-pixel match between the two matching subsets is performed and optimization is done based on the radiance correlation to reduce geolocation errors between the datasets. The intersatellite biases are then calculated from the pixels in the nadir window consisting of 10 cross-track pixels and 11 scan lines around the SNO pixel. The subset of this nadir window acts as a factor to reduce the effect of pixel radiance variations, sensor modulation transfer function differences, and minor mismatch in pixel collocation.

Each collocated HIRS SNO file contains pairs of SNO observations obtained from either north or south latitude zones for each channel with timestamps of year and date. Figure 2 shows scatterplots of the SNO brightness temperatures for channel 8 between N16 and N17. Though the events of satellite intersection occurred between 70° and 80° latitude zones, the temperature ranges of the observation are considerably large. As displayed in Fig. 2, temperatures for channel 8 span from 210 to 280 K.

4. Analysis of intersatellite biases

a. Intersatellite biases between N16 and N17

Intersatellite biases are the results of a combination of factors, which may include changes in channel SRFs, calibration biases, etc. For the tropospheric channels, Cao et al. (2005) showed that the intersatellite radiance bias can be as large as 1.12 mW (m² sr cm⁻¹)⁻¹ over the Arctic region (e.g., channel 7 between N16 and N17), which is close to 1 K in brightness temperature. Among a number of sources that may have contributed to the bias, a large effect is likely from different SRFs between the satellites. Because several of the tropospheric sounding channels are located at the sharp slope of the transmission line, they are especially sensitive to even a small shift in wavenumbers. As an example, Fig. 3 shows the normalized response functions for two of these sounding channels, channels 4 and 7, for N16 and N17. The values in the plot are obtained from the KLM user’s guide (Robel 2006). Channel 4 centers at 701 cm⁻¹ for N16, but shifts to 703 cm⁻¹ for N17. Channel 7’s center shifts from 749 cm⁻¹ for N16 to 747 cm⁻¹ for N17. Although these shifts are small, the effects on the channel measurements are not negligible.

Many past studies examined the mean biases of HIRS channels. As the sounder measurements span a considerable range of radiance, the intersatellite biases can change at different scene radiance values. To examine the biases in more detail, analyses are carried out based on the variation of intersatellite biases across the observed scene radiance range. The results for channels 2–12 between N16 and N17 are displayed in Fig. 4. The biases are computed by subtracting the matched-up radiances for N17 from those for N16. In the figure, each plot represents the averaged intersatellite bias for the scene radiance centered at the indicated values. The averages are carried out for every 10 mW (m² sr cm⁻⁴)⁻¹ for channels 2–10, every 2 mW (m² sr cm⁻⁴)⁻¹ for channel 11, and every 1 mW (m² sr cm⁻⁴)⁻¹ for channel 12.

Figure 4 shows that there are significant variations of the bias values across scene radiance ranges for some of the channels, among which the most notable is channel 7. The channel-7 bias value changes from 0.2 mW (m² sr cm⁻⁴)⁻¹ at the low scene radiance to as large as 1.6 mW (m² sr cm⁻⁴)⁻¹ at high radiance. This scene radiance dependency feature is an important factor to consider when performing intersatellite calibration. For instance, if only the mean or median of the bias value for channel 7 is used in an intersatellite calibration, it can lead to more than 0.5 mW (m² sr cm⁻⁴)⁻¹ of false trends in warm and cold regions. Several other channels (channels 2–4) also have significant variations of more than 0.4 mW (m² sr cm⁻⁴)⁻¹ across scene radiances.

Among these channels, channels 4 and 7 show the largest intersatellite biases; however, their signs are opposite. This is mostly due to the opposite directions of their channel SRF shifts. As shown in Fig. 3, the channel-4 normalized response function shifts from a lower to higher wavenumber, and thus moves to a higher transmitting spectrum (see Fig. 1). Even though in the SNO dataset the two satellites are observing the same scene, they actually measure different layers of the atmosphere. For channel 4, with the SRF for N17 moving to a higher transmitting wavenumber, the observation from N17 measures a lower layer than that from N16, and thus a larger radiance value compared to N16 (thus negative bias from N16 to N17). For channel 7, the response function shifts to a lower wavenumber from N16 to N17, thus the opposite effect dominates and positive biases are produced.

The bias pattern between two satellites with respect to scene radiance largely depends on the effect of the two different response functions. Theoretically, the bias pattern can be computed by using a radiative transfer model. To obtain a physical illustration of how intersatellite bias varies with radiance, we used the radiative transfer for Television and Infrared Observation Satellite (TIROS) Operational Vertical Sounder (TOVS) (RTTOV-8) (Saunders et al. 1999) to compute the channel brightness temperature biases between N16 and N17. This model is widely used in the Advanced TIROS Operational Vertical Sounder (ATOVS) community. In the model, the integration over the channel response function is simulated directly using parameters derived from a line-by-line model. Given an atmospheric profile of temperature, water vapor, ozone, and carbon dioxide together with satellite zenith angle and surface temperature, pressure, and surface emissivity, RTTOV computes the top of atmosphere radiances in each of the channels. The RTTOV was one of the participating models in the intercomparison of HIRS and AMSU channel radiance computation (Garand et al. 2001). It was shown that the RTTOV simulated radiance biases for HIRS channels are generally small among the models examined. For the model computation of HIRS intersatellite biases, approximately 500 atmospheric profiles from the European Centre for Medium-Range Weather Forecasts (ECMWF) reanalysis were used as input (Chevallier 2001). These profiles were selected from 70° to 80°S and 70° to 80°N latitudes of the reanalysis data.

Simulations of HIRS channel radiances are computed for N16 and N17, and the differences are obtained. Figure 5 displays the bias patterns of channels 4 and 7 based on RTTOV-8 simulations. In general, we do not expect the model simulations to exactly match the SNO observation for a number of reasons. The response functions in the model dataset are prelaunch laboratory measurements. They are often slightly different from the postlaunch response functions of the satellite instruments in operation. Other factors contributing to the differences include possible model bias and different atmospheric constituent amounts between those in the model database and those in the actual atmosphere. However, for the channels having large biases, the use of a radiative transfer model is expected to show the same signs (positive or negative) of the biases and the general temperature-dependent features.

Figure 5 shows that for channel 4, negative biases are obtained from the model simulations, which agree with what is observed from SNO. The largest bias of about −0.7 mW (m² sr cm⁻⁴)⁻¹ is found at the scene radiance of 30 mW (m² sr cm⁻⁴)⁻¹, and gradually smaller negative biases are found as the scene radiance becomes larger. The SNO plot for channel 4 in Fig. 4 is very similar to this bias variation pattern for the radiance observations in the 30–60 mW (m² sr cm⁻⁴)⁻¹ range. No observation from SNO at lower scene radiance was obtained to compare with the model simulation. For channel 7, the bias variation trend obtained from RTTOV-8 is also similar to that from the SNO plots. Both show increases of channel-7 biases with radiance from values close to zero to values larger than 1.5 mW (m² sr cm⁻⁴)⁻¹.

To obtain a perspective of the intersatellite biases in brightness temperatures, the observed noise equivalent brightness temperature (neΔT) for channels 2–12 between N16 and N17 is computed and plotted in Fig. 6. The neΔT is defined as

View Expanded

where ΔR represents the radiance difference for each channel as shown in Fig. 4, T represents the brightness temperature, and B is the Plank function.

Figure 6 shows that for the channels having large variations of biases with scene radiance, the observed neΔT values also vary significantly with brightness temperature. The largest neΔT variation is found in channel 7, with values changing from 0.2 K at low brightness temperatures to 1.2 K at high brightness temperatures. This illustrates larger uncertainty of intersatellite calibration at larger brightness temperature values for this channel. In general, for the channels having significant radiance biases (as shown in Fig. 4), large values of neΔT are found, which indicates that large biases are likely seen in the derived brightness temperatures for these channels.

For the window channel (channel 8), the bias values are generally close to zero, and the bias variation with respect to temperature is small. The channel is located in the relatively flat transmission spectrum, and thus is relatively insensitive to the slight spectrum changes between satellites N16 and N17.

The biases between N16 and N17 are also relatively small [less than ±0.4 mW (m² sr cm⁻⁴)⁻¹] for several other channels, including channels 6, 9, 10, 11, and 12. However, larger biases are found for these channels from other satellite pairs that have larger spectral changes. These are discussed in the following section.

5. Summary and discussion

Data from the HIRS SNO obtained from satellite intersection observations are examined. Though the observations are from high latitudes, the SNO channel observations span a considerably large temperature range. The analysis indicates that a shift in the center wavenumber from one satellite to another can lead to a significant difference in the radiance values observed between two satellites. The shifting of the center wavenumber results in different weighting functions with the peak energy contribution from different altitudes. Therefore, the bias between two HIRS instruments represents the difference in measurements of two slightly different vertical layers. The direction of the center wavenumber shift in combination with the change in the SRF (e.g., the shape of SRF and bandwidth) plays an important role in determining the sign (positive or negative) of the bias.

The intersatellite biases for a number of HIRS channels are found varying significantly with the scene radiance. Many of these channels with radiance-dependent biases are the tropospheric sensing channels centered along the sharp slope of the transmission line. The biases for these channels may be small and close to zero at one end of the scene radiance ranges, but can be as large as over 1 mW (m² sr cm⁻⁴)⁻¹ at the other end. For example, for channel 7 between N16 and N17, the intersatellite bias changes from a value of 0.2 mW (m² sr cm⁻⁴)⁻¹ at the scene radiance of 40 mW (m² sr cm⁻⁴)⁻¹ to 1.5 mW (m² sr cm⁻⁴)⁻¹ at the radiance of 90 mW (m² sr cm⁻⁴)⁻¹. This variation of the bias is equivalent to approximately 1 K in brightness temperature. If such a large variation is not properly considered in an intersatellite calibration, it can lead to false long-term trends, especially for warm and cold seasons.

The intersatellite biases of the window channel (channel 8) are relatively small compared to other channels for most of the satellite pairs, because of the flatter transmission line for channel 8’s spectrum and thus the channel’s relative insensitivity to the slight spectrum change from satellite to satellite. The variations of bias across scene radiance ranges are also relatively small.

Channel 10 underwent large spectral changes in several of the HIRS instruments. Channel 12 also changed to a different spectrum when the new generation of the NOAA polar series started with N15. When these changes occurred, large biases were carried into satellite observations.

It should be noted that the SNO data from the satellite intersections are obtained only over the high latitudes. Though the year-round temperature span of the observations is relatively large, the SNO matchups do not cover the high temperatures typically observed over the tropical regions. For these regions, additional data or methods are required for intersatellite calibration. However, the results in this study show enough of a temperature range to determine whether the bias is radiance dependent.

The agreement between forward radiative transfer modeled biases and observed biases indicates that most of the bias can be explained by physical differences in the SRFs. With accurate measurements of SRFs, a radiative transfer model can be a useful tool to estimate intersatellite biases, especially for satellite pairs with short (or without) overlap.

The analysis in this study derived a radiance-dependent HIRS intersatellite bias dataset for each pair of the satellites in an effort to provide a more accurate method for HIRS intersatellite calibration. In addition to the intercalibration within the HIRS instruments, selected HIRS channel measurements have been used to calibrate other satellite instruments such as the geostationary satellite measurements (Breon et al. 2000; Knapp 2008). Often a single polar-orbiting satellite is used as a reference (e.g., Gunshor et al. 2004). With the availability of the intersatellite calibrated HIRS data, there is a potential to extend the time series of intersatellite calibrated geostationary satellite data significantly.

Combining all HIRS instruments, the dataset represents the longest atmospheric sounding measurements from satellite observations. This provides an important database to derive long-term time series of many environmental variables on a global scale, such as atmospheric temperature, water vapor, and cloud properties. A well-calibrated adjustment to the measurements of different satellites is an essential step in reducing bias in long-term trend studies.