Journal of Geophysical Research: Atmospheres

Understanding intersatellite biases of microwave humidity sounders using global simultaneous nadir overpasses



[1] Simultaneous nadir overpasses (SNOs) of polar-orbiting satellites are most frequent in polar areas but can occur at any latitude when the equatorial crossing times of the satellites become close owing to orbital drift. We use global SNOs of polar orbiting satellites to evaluate the intercalibration of microwave humidity sounders from the more frequent high-latitude SNOs. We have found based on sensitivity analyses that optimal distance and time thresholds for defining collocations are pixel centers less than 5 km apart and time differences less than 300 s. These stringent collocation criteria reduce the impact of highly variable surface or atmospheric conditions on the estimated biases. Uncertainties in the estimated biases are dominated by the combined radiometric noise of the instrument pair. The effects of frequency changes between different versions of the humidity sounders depend on the amount of water vapor in the atmosphere. There are significant scene radiance and thus latitude dependencies in the estimated biases and this has to taken into account while intercalibrating microwave humidity sounders. Therefore the results obtained using polar SNOs will not be representative for moist regions, necessitating the use of global collocations for reliable intercalibration.

1. Introduction

[2] Tropospheric humidity measurements from microwave humidity sounders such as Advanced Microwave Sounding Unit-B (AMSU-B) [Saunders et al., 1995] and Microwave Humidity Sounder (MHS) [Bonsignori, 2007] have been proven to have significant impact on the skill of numerical weather prediction [Andersson et al., 2007]. This is primarily due to their ability to measure humidity under all-sky conditions compared to clear-only sampling by infrared sounders [e.g.,John et al., 2011]. Recently, some attempts have also been made to use microwave humidity sounding data for climate applications [e.g., Xavier et al., 2010; Eymard et al., 2010; Buehler et al., 2008]. However, although microwave temperature sounding data have been intercalibrated and extensively used for climate studies [Thorne et al., 2010], this has not yet been done for the humidity sensors. The main reason for this is the short span of the data, primarily since late 1998; although Special Sensor Microwave Humidity Sounder (SSM/T-2) data began in 1994, these early measurements were not widely used except for research [e.g.,Miao et al., 2001; Selbach et al., 2003; Sohn et al., 2003; Chung et al., 2011]. The error characteristics of SSM/T-2 radiances data are not fully understood, and careful validation is essential before they can be used to assess, in particular, long-term trends in upper tropospheric water vapor which is an important climate variable, yet poorly simulated by current climate models [e.g.,John and Soden, 2007].

[3] Unfortunately, there is a lack of stable and reliable ground-based or in situ reference measurements of atmospheric humidity to intercalibrate satellite instruments [Seidel et al., 2009]. Cao et al. [2004, 2005]have developed a method to find simultaneous nadir overpasses (SNOs) of polar orbiting satellite pairs and use them for intercalibration. There are regular near-polar SNOs and during an SNO, similar instruments on the different satellite platforms measure radiation emitted from the same area of Earth and/or its atmosphere at the same time. Therefore any difference in the radiance measured by the satellites can be used to intercalibrate the measurements. This is being developed in support of the Global Space-Based Inter-Calibration System (GSICS) initiative to provide climate quality satellite datasets [Goldberg et al., 2011].

[4] SNO data have been proven useful for intercalibration of instruments such as HIRS and MSU/AMSU-A [e.g.,Zou et al., 2006; Wang et al., 2007; Cao et al., 2005; Iacovazzi and Cao, 2007; Shi et al., 2008]. However, Iacovazzi and Cao [2008] showed that for those channels which are sensitive to the Earth's surface, there are large uncertainties in the estimated intersatellite bias due to surface inhomogeneity which arises mainly from variable surface emissivity of SNO scenes at subpixel scales.

[5] The concerns expressed by Iacovazzi and Cao [2008] can be put in the context of microwave humidity sounders as follows. The peak emission for a sounding channel occurs at an atmospheric level for which the optical depth, integrated from the top of the atmosphere, becomes approximately one [e.g., Petty, 2006]. Therefore depending upon the amount of water vapor in the atmosphere, the peak emission levels of humidity sounding channels move up and down, in contrast to temperature sounding channels which use the absorption of well mixed gases such as oxygen or carbon dioxide. Thus the sounding height of a humidity channel is at its maximum in a wet tropical atmosphere and becomes lower as the satellite moves toward higher latitudes. Figure 1 shows how the total opacity, which is the vertically integrated absorption coefficient, varies as a function of the amount of water vapor in the atmosphere. For the dry atmospheres sampled by SNOs which normally occur between 70° and 80° latitudes [Cao et al., 2004], the opacity is of order one even for the channel closest to the 183.31 GHz water vapor line center. Analysis of ERA-Interim [Dee et al., 2011] four-times daily precipitable water vapor data for 2010-01 and 2010-07 showed that more than 50% of the values are below 3 mm at latitudes 70–80° except for the Arctic region in summer. This is consistent with the results ofMelsheimer and Heygster [2008]. So for all channels on microwave humidity sounders, there is a significant contribution from the Antarctic surface and the Arctic surface in winter, and the radiation which reaches the satellite is then determined substantially by the surface skin temperature and the surface emissivity: the atmospheric contribution is relatively small. High-latitude surfaces are highly inhomogeneous, consisting of land, water, ice, or snow whose emissivities are significantly different [Weng et al., 2001; Weng and Yan, 2004]. Land surfaces have an emissivity close to 0.95 (note that surfaces with snow or sand have lower emissivity at these frequencies); ocean emissivity varies considerably depending on oceanic characteristics including surface roughness which is influenced by overlying atmospheric conditions; and snow and sea-ice emissivity also varies considerably [Mathew et al., 2008]. Therefore measures are necessary to reduce the noise related to surface inhomogeneity.

Figure 1.

Total opacities for seven profiles with 0.2, 0.5, 1.0, 3.0, 10.0, 37.5, and 66.5 mm precipitable water vapor. These profiles are taken from the Chevallier et al. [2006]data set. The 37.5 mm corresponds to the median and 66.5 mm corresponds to the 95th percentile. Shaded regions represent the passband positions of AMSU-B channels. The channel numbers are printed below the passbands. Note that for MHS, Channel 2 is centered at 157 GHz (instead of at 150 GHz for AMSU-B) and Channel 5 has only one passband at 190.31 GHz. Figure adapted fromJohn and Buehler [2004].

[6] Furthermore, near-polar SNOs only sample brightness temperatures which are not representative of lower latitudes. Owing to nonlinearity in the calibration, error in warm target measurements, and obstructed space view, intersatellite biases can vary with scene radiance [e.g.,Zou et al., 2006]. Therefore there are several reasons why near-polar SNOs are inadequate for intercalibrating the microwave humidity sounders.

[7] Owing to atmospheric drag, the Earth's nonsphericity, and gravitational pull from celestial bodies, the orbit of a polar orbiting satellite drifts and its local equator crossing time changes. When the equator crossing times of a pair of satellites become nearly the same, SNOs can occur at all latitudes for a short period, typically 1 or 2 months. We use these SNOs at all latitudes to estimate the adequacy of polar SNOs to intercalibrate microwave humidity sounders.

[8] Section 2 gives a short technical description of the humidity sounders and their channel characteristics. Section 3revisits a recent comparison between simulated AMSU-B and MHS to show that global SNOs are necessary for reliable intercalibration.Section 4 describes the data and methods of analysis. Section 5 presents the results and section 6 provides a summary and conclusions.

2. Functional Description of Microwave Humidity Sounders

[9] The Advanced Microwave Sounding Unit-B (AMSU-B) and the Microwave Humidity Sounder (MHS) are five-channel microwave radiometers. They are designed to measure the radiation emitted from the Earth's surface and atmosphere in order to estimate global fields of tropospheric humidity. The microwave absorption characteristics of the atmosphere are shown inFigure 1 and the instrument specifications are given in Table 1. AMSU-B is onboard NOAA-15 (N15), N16, and N17 and MHS is onboard N18, N19, and MetOpA (MA).

Table 1. Channel Characteristics of the Instrumentsa
ChannelfC (GHz)Δf (GHz)PassbandsNEΔT (K)Beam Width (deg)Polarization
  • a

    Here fC is the central frequency of the channel (taken from Kleespies and Watts [2007]), Δf is the passband width, NEΔT is the noise equivalent temperature from the first flight models (NOAA KLM User's guide [Goodrum et al., 2007]). Nominal polarizations are for nadir view only and rotate with view angle.


[10] Channels 1 and 2 at 89 GHz and 150 GHz (at 157 GHz for MHS), enable deeper penetration through the atmosphere to the Earth's surface. Channels 3–5 are located in the strongly opaque water vapor absorption line at 183.31 GHz and provide information on the atmospheric humidity at different levels. The passbands of Channels 3, 4, and 5 are at 183.31 ± 1.00 GHz, 183.31 ± 3.00 GHz, and 183.31 ± 7.00 GHz (only at 183.31 + 7.00 GHz for MHS), respectively. The passbands of the channels are also shown in Figure 1. Note that the five channels on AMSU-B are formally numbered as channels 16–20 (channels 1–15 belong to AMSU-A which is a temperature sounding instrument), but in this article we call them channels 1–5 to be consistent with MHS channel numbers.

[11] At each channel frequency, the antenna beamwidth is a constant 1.1 degrees (full width at half maximum). Ninety contiguous cells are sampled on the Earth's surface, with each scan covering ±49.5 degrees on each side of the subsatellite point. These scan patterns and geometric resolution translate to a 16.3 km diameter cell at nadir at a nominal altitude of ∼833 km. Each channel is also sensitive to radiation of a particular polarization as defined in Table 1, the direction of which rotates with scan angle. The differences in the polarization for channels 3 and 4 on MHS compared to ASMU-B will only manifest themselves for very dry atmospheres (or high topography) where these channels become sensitive to surface radiation such as over Antarctica.

3. Comparison of Simulated AMSU-B and MHS Measurements

[12] Kleespies and Watts [2007]compared the brightness temperatures simulated for MHS and AMSU-B using the 48 profiles ofStrow et al. [2003]. Significant differences were found only for channels 2 and 5 and in both cases mean MHS brightness temperatures were colder than those of AMSU-B. We revisit the study to investigate the dependence of bias between the two instruments on surface and atmospheric conditions, enabling us to interpret the results of our SNO analysis for these channels.

[13] Figure 2shows simulated brightness temperature differences between MHS and AMSU-B using 5000 diverse profiles, sampled from ECMWF forecasts to span the natural variability of the real atmosphere [Chevallier et al., 2006]. The simulations used a line-by-line radiative transfer model [Buehler et al., 2005a] that was already used in a number of intercomparison studies [Buehler et al., 2004; John and Buehler, 2005; Moradi et al., 2010]. Surface emissivity at these frequencies varies considerably with surface type, with higher emissivity for land (∼0.95) and lower emissivity for ocean (∼0.6). Therefore in the simulations we used an emissivity of 0.95 for land profiles and 0.6 for ocean profiles. For profiles from a model grid box which has both land and sea we calculated the emissivity based on land cover linearly varying between 0.6 and 0.95. Results are shown only for channels 2 and 5 because for the other three channels the differences are negligibly small. The differences are shown as functions of brightness temperature (Figure 2, right) and precipitable water vapor (PWV; Figure 2, left) which is the vertically integrated water vapor density.

Figure 2.

(left) Simulated brightness temperature differences between MHS and AMSU-B as a function of precipitable water vapor for (top) Channel 2 and (bottom) Channel 5 using a diverse atmospheric profile data set compiled from ECMWF forecasts [Chevallier et al., 2006]. Profiles are separated for ocean and land. Ocean emissivity is 0.6, land emissivity is 0.95, and emissivity of mixed grid point profiles varies linearly between 0.6 and 0.95. (right) Also shown is simulated brightness temperature differences as a function of mean brightness temperatures of AMSU-B and MHS. Note the simulations are only for nadir.

[14] Figure 2(top) shows the differences for channel 2 which is at 150 GHz on AMSU-B but at 157 GHz on MHS. Channel 2 is a sounding channel in a humid atmosphere but with a surface contribution which increases as atmospheric moisture decreases. With a very moist atmosphere, the surface has little effect and the brightness temperature of MHS is lower than that of AMSU-B because the atmosphere is slightly more opaque at 157 GHz than at 150 GHz (seeFigure 1), raising the sounding altitude slightly. With a less moist atmosphere, the higher atmospheric opacity at 157 GHz than at 150 GHz makes the radiometrically cold surface have less influence on MHS than on AMSU-B, leading to higher brightness temperature for MHS. This is especially true for the ocean (blue symbols inFigure 2) because of its lower emissivity. When there is very little water vapor, the difference is close to zero because both instruments sample the surface which has a similar emissivity at 150 and 157 GHz. It is clear from Figure 2 that the differences can have a wide range from −2 to 7 K depending on atmospheric and surface conditions (Kleespies and Watts [2007]reported −1.54 ± 2.03 K bias) and thus it is not straightforward to combine AMSU-B and MHS channel 2 radiances by adding an offset to one of the measurements.

[15] Figure 2(bottom) shows the brightness temperature difference for channel 5. As for channel 2, there is little difference when the atmosphere is almost free of water vapor and it starts increasing with water vapor. When precipitable water is around 3 mm the trade off between surface and sounding channel effects come into play and the difference starts to decrease. When there is about 15 mm of precipitable water the channel becomes a sounding channel and insensitive to the surface. MHS channel 5 is measuring colder radiances compared to the AMSU-B one, by about −0.6 K irrespective of surface type in an atmosphere with 20 mm or more precipitable water vapor.

[16] Figure 2(right) show simulated brightness temperature differences for channels 2 and 5 as a function of average scene brightness temperatures of AMSU-B and MHS. Transition from surface to sounding channel is clearly seen for sea points due to radiometrically colder surface which amplifies the water vapor signal from the atmosphere. Although it is possible to combine AMSU-B and MHS data for channel 5 [Kleespies and Watts, 2007] by adding a global offset to account for the frequency changes, the systematic major variations in bias between a dry polar atmosphere and a moist lower-latitude atmosphere lead us to conclude that biases from these channels estimated from polar SNOs cannot represent humid lower latitudes.

4. Collocation and Analysis Methods

[17] We used the collocation method based on Holl et al. [2010] (some details given in Appendix A). Sensitivity of distance and time thresholds for selecting collocations to the uncertainty in the estimated bias is shown in Figure 3. Consequently, to overcome spatial inhomogeneity we used only those pixel pairs whose centers are closer than 5 km, which is less than one third of the 16.3 km pixel diameter at nadir. We discard any measurements with time differences exceeding 300 s, to avoid changes in scene properties such as clouds. We used only four pixels each on either side of nadir, to avoid errors arising from limb effects and scan asymmetry. This also minimizes the impact of polarization differences. Collocations over both land and ocean are used throughout this study.

Figure 3.

Sensitivity test to select distance and time threshold for collocations. Standard deviation of brightness temperature difference in Kelvin for each grid box is shown. Distance grid is equally spaced with 1 km distance, but time grid has variable width. We have randomly selected 120 points from each grid box to calculate statistics. Grid boxes in white have too few collocations to make statistics. Collocations are taken from 70–80° latitudes in both hemispheres of MA–N17 collocations.

[18] Using those pixel pairs which satisfied these stringent criteria, we first calculated differences in brightness temperatures and then derived the mean difference or bias ( inline image) and the standard deviation of the differences image The standard deviation of collocated brightness temperatures (or in other words, SNO variability) has mainly two sources [Iacovazzi and Cao, 2008]: one is the combined radiometric noise (NEΔT, which is the smallest change in input brightness temperature that can be detected in the system output (i.e., calibrated brightness temperatures, including contributions from calibration noise)) of the two instruments and the other is the scene (spatial and temporal) inhomogeneity. In order to have robust statistics, we collected data for a month to calculate inline image and image This is an advance over previous studies which consider individual SNO events which have fewer pixel pairs for computing statistics. We also calculate standard errors of mean values, namely image divided by the square root of the number of collocations.

[19] Clouds affect these channels [Sreerekha et al., 2008], but we have not screened for them. This is mainly because in polar conditions it is difficult to differentiate between clouds and the surface. Owing to our stringent spatiotemporal collocation criteria, we assume that measurements from both instruments are affected by clouds in a similar way.

5. Results

5.1. Selection of SNOs

[20] Figure 4 shows the equator crossing times of the ascending nodes [Ignatov et al., 2004] of NOAA and MetOp polar-orbiting satellites. The orbital parameters of these satellites are designed so that equator crossing time will drift away from local noon because if a satellite crosses the equator at noon, this can affect the functioning of both the satellite and the instruments on board owing to different solar illumination. The drift creates the possibility that satellite pairs will have similar equator crossing times for short periods. During these time periods SNOs can occur globally. We have discovered that in recent years there have been SNOs at all latitudes and this is to our knowledge the first study to exploit this.

Figure 4.

Equator crossing times of the ascending nodes of NOAA/MetOpA polar orbiting satellites for the ATOVS time period. Drifting of the orbits can be seen. MetOpA is maintained in a stable orbit.

[21] We have identified 4 months of data with sufficient number of collocations satisfying our stringent criteria (Δx less than 5 km and Δt less than 300 s) at all latitudes. They are 2008-08 for the N16-N15 pair, 2009-04 and 2009-05 for the MA-N17, and 2009-09 for the N19-N18. We assign newer platforms NOAA-16, MetOpA, and NOAA-19 as primary satellites and NOAA-15, NOAA-17, and NOAA-18 as secondary satellites. Bias is calculated as primary satellite minus secondary satellite.

[22] We partitioned the collocations into 18 10° latitude bins. Figure 5(top) shows the latitudinal distribution of the number of collocations. The number of collocations varies for each satellite pair, with the N19-N18 pair having the most, about 50,000–100,000, collocations in each latitude bin. Most of the bins have 5000 or more collocations which are enough collocations to compute robust statistics.

Figure 5.

Number of collocations in 10° latitude bins. Each collocation satisfies stringent spatiotemporal criteria (Δx < 5 km and Δt < 300 sec). Black circles show collocations of NOAA-15 and NOAA-16 during 2008-08, green and blue circles show collocations of MetOpA and NOAA-17 during 2009-04 and 2009-05, respectively, and red circles show the collocations of NOAA-19 and NOAA-18 during 2009-09. Note the logarithmic y-axis scale. Map plots show geographical distribution of SNOs.

5.2. Interpretation of SNOs

[23] Biases are expected to vary with scene-radiance (section 3), so estimates of biases derived from SNOs at all latitudes will be particularly valuable if they vary systematically with latitude. On the other hand, the estimates will be less useful if they are noisy. So before presenting our main results we consider these two aspects.

5.2.1. Meridional Distribution of Brightness Temperature

[24] In order to interpret the meridional distribution of biases in the measurements, we need to know the meridional distribution of the brightness temperatures. Figure 6 shows the mean and standard deviation of brightness temperatures for each latitude bin for each channel. A common feature is the very low brightness temperatures south of 70°S. Variability is greater over these southern polar regions because the heterogeneous surface conditions show through the very dry atmosphere. Channel 1 is a surface channel at all latitudes: as seen in Figure 1, the total opacity is less than one even for the very wet profile. Accordingly it also shows low brightness temperatures for the midlatitude southern hemisphere and for the Arctic, as does channel 2 for the same reason. This might be associated with less landmass in these latitudes and lower ocean emissivity. Because of its sensitivity to the surface, channel 1 also has high variability.

Figure 6.

Mean and standard deviation of brightness temperature for latitude bins. Only those collocations with center of pixels less than 5 km apart and measurement time difference less than 300 s are used to compute statistics. The width of latitude bins is 10 degrees. First row: NOAA-19 and NOAA-18, 2009-09. Second and third rows: MetOpA and NOAA-17 during 2009-05 and 2009-04, respectively. Fourth row: NOAA-15 and NOAA-16, 2008-08.

5.2.2. Uncertainties in SNO Method

[25] An important source of uncertainty for the SNO method is the radiometric noise of the instruments [Iacovazzi and Cao, 2008]. This is normally expressed as noise equivalent brightness temperature (NEΔT): the first flight model values taken from Goodrum et al. [2007] for each channel are given in Table 1. NEΔT is time varying, it generally increases as the instrument starts to degrade. It can also increase due to a change in the operating conditions of the satellite and due to radio frequency interference (RFI) from nearby transmitters or other instruments.

[26] Mean NEΔT values for the analysis time period for all the channels are given in Table 2. NEΔT were determined from the warm target views during the analysis time period. Note that for N19 channel 3 the noise was about 2.5 K for the first half of September 2009 but more than 7 K for the second half of the month. Note the performance of this channel became better by the beginning of 2011.

Table 2. Mean NEΔT for the Analysis Time Period, Determined From the Warm Target Views for All Satellitesa
  • a

    Units are in Kelvin.


[27] Scene inhomogeneity is another source of uncertainty in the SNO method owing to spatial and temporal mismatches in collocated pixels. Figure 7 shows standard deviations of brightness temperature differences reflecting these uncertainties. Horizontal lines indicate the combined instrument noise based on values given in Table 2. Channel 1 shows standard deviations from 1 to 2 K which are higher than the combined instrument noise. For channel 2, N16–N15 (both having AMSU-B) and N19–N18 (both having MHS) show standard deviations from 1 to 1.5 K, approximately consistent with the specified NEΔT of these instruments. The MA–N17 pair (MA has MHS and N17 has AMSU-B) has higher standard deviation which can be explained by the differences between 150 and 157 GHz emissions for very different surface emissivities (land and sea) north of 40°S (see discussions insection 3).

Figure 7.

Standard deviation of the brightness temperature differences image Black circles show NOAA-16–NOAA-15 during 2008-08, green and blue circles show MetOpA-NOAA-17 during 2009-04 and 2009-05, respectively, and the red circles show NOAA19–NOAA18 during 2009-09. Horizontal lines indicate the combined instrument noise based on values given inTable 2.

[28] Channel 3 shows different standard deviations for different satellite pairs. If NEΔT accords with prelaunch specifications (Table 1), we would expect the effective variability associated with NEΔT to be equivalent to 1.5 K for the AMSU-B–AMSU-B combination, 1.18 K for the AMSU-B–MHS combination ( inline image= 1.18 K), and 0.7 K for the MHS–MHS combination. The MA-N17 pair comparing AMSU-B and MHS shows SNO variability close to instrument specification; even at high latitudes, where the surface is highly variable, standard deviations remain small. Thus there is very little contribution from scene inhomogeneity, given our stringent collocation criteria. The other two satellite pairs show significantly higher variability than expected from prelaunch instrument specifications. N19–N18 has the highest value of about 9 K, owing to known instrument problems causing exceptionally high noise in channel 3 on N19 at the time of the comparison. N16–N15 has about 4 K standard deviation which is also much higher than the specified noise of the instruments. Our analysis (Table 2) indicates that for both N16 and N15 the NEΔT have increased due to instrument problems.

[29] Channels 4 and 5 also show standard deviations consistent with NEΔT of the instruments, except for the N16–N15 pair which shows inflated standard deviations owing to instrument degradation (Table 2). The standard deviations are invariant with latitude for these channels as well which leads to the conclusion that there is little influence by scene inhomogeneity in the estimated bias, given our stringent collocation criteria.

5.3. Meridional Distribution of Bias

[30] Figure 8 shows the bias and its standard error for all latitude bins and for all satellite pairs. Intersatellite bias is time varying, so the results shown here represent only the time period analysed.

Figure 8.

Black circles and vertical bars show bias ( inline image) for latitude bins and its standard error estimated using SNOs. First row: NOAA-19–NOAA-18 during 2009-09. Second and third row: MetOpA–NOAA-17 during 2009-05 and 2009-04, respectively. Fourth row: NOAA-16–NOAA-15 for 2008-08. Red circles show bias estimated from zonal mean brightness temperatures (seesection 5.3.6 for details). Note that some of the red circles are out of the plot range.

5.3.1. Channel 1 (89 GHz)

[31] Channel 1 shows very small intersatellite biases. N16 is about 0.15 K warmer than N15 for most latitude bins. MA measurements are also warmer than N17 measurements by about 0.2 K but with a few outliers. Though N19–N18 bias is small there is a latitude dependence, with negative bias for high latitudes and positive bias for low latitudes.

5.3.2. Channel 2 (150/157 GHz)

[32] N16-N15 bias is very stable across all latitudes at about 0.5 K except for the two southernmost latitude bins. N19–N18 bias is clearly latitude-dependent, being as low as −0.3 K at high southern latitudes and 0.1 K at low latitudes, with a pattern similar to channel 1. The MA-N17 pair (AMSU-B and MHS combination) shows large biases, up to 4 K and high variability in bias with latitude, as expected from our analysis of simulated brightness temperatures for this channel insection 3. Biases are consistent for April and May 2009.

5.3.3. Channel 3 (183.31 ± 1.00 GHz)

[33] N16-N15 shows the largest biases, ranging from 1 to 2 K with a latitude dependence: 1.8 K bias near the South Pole which decreases to 1 K near the North Pole. N19–N18 has biases ranging from zero to 0.4 K with no obvious latitude dependence, but with large standard error. Note that N19 had exceptionally high noise for this channel. MA-N17 biases vary between −0.15 K to 0.4 K, being positive at high latitudes and near-zero or negative at low latitudes.

5.3.4. Channel 4 (183.31 ± 3.00 GHz)

[34] N16-N15 has significant intersatellite bias which varies linearly from 4 K at the South Pole to near zero at the North Pole. N19–N18 biases are close to 0.2 K at low latitudes and the biases vary considerably with latitude. MA–N17 also show latitude dependence with high latitudes showing positive biases up to 0.5 K whereas at low latitudes the biases are very slightly negative.

5.3.5. Channel 5 (183.31 ± 7.00/+7.00 GHz)

[35] N16–N15 biases also vary linearly with latitude for channel 5, from −3 K near the South Pole to −5.5 K near the North Pole. N19–N18 biases are close to zero except for the two southernmost latitude bins where the bias is close to −0.3 K. MA-N17 has the AMSU-B–MHS combination and this channel on MHS has only the upper sideband, so larger biases are expected. The biases show strong latitude dependence: positive at higher latitudes and near-zero or negative at lower latitudes.

[36] It is interesting to note that bias patterns are broadly similar for channels 3, 4, and 5 for AMSU-B and for channels 3 and 4 for MHS. This similarity might be manifested by common local oscillator and mixer used by these channels.

5.3.6. Consistency Check on Estimated Bias

[37] As an independent estimate to check the bias obtained from SNO method, biases were also calculated using zonal averaged brightness temperatures. This method was already used by Shi and Bates [2011]for infrared channels. We used only near-nadir brightness temperatures to avoid scan bias/limb effect. This method works well when the sampling times of two satellites are similar which is the case of our analysis. If sampling times were different, differences would arise from the diurnal cycles of humidity and temperature [Zou et al., 2006]. Biases are calculated for 18 latitude bins as in our SNO analysis. Red circles in Figure 8 show the estimated biases using this method. This method works well in general except for channel 1 due to very small biases for this channel. Thus it is confirmed that the latitude dependence of bias estimated using SNO method is correct and polar SNOs alone cannot be used to estimate intersatellite biases. Though the zonal average brightness temperature method is found to be useful for the analyzed time period, it is not certain whether it will work for other time periods due to differences in temporal sampling of the satellites and we are currently investigating this.

5.4. Dependence of Bias on Scene Radiance

[38] Intersatellite bias can vary with scene radiance, TB [e.g., Shi et al., 2008]. Therefore in Figure 9 we show biases and their standard errors estimated from global SNOs as a function of TB of each satellite. We did not average TBs across a satellite pair because the dependence of bias on TB can vary between satellites. Blue circles in Figure 9 show dependence of bias as a function of TB measured by primary satellites (N16, MA, and N19) and green circles do likewise but with reference to secondary satellites (N15, N17, and N18). We have collated biases into 10 K brightness temperature bins and then computed their mean and standard error for all bins with 100 or more data values.

Figure 9.

Blue circles show bias ( inline image) as a function of brightness temperature of primary satellites and the green circles show bias as a function of brightness temperature of secondary satellites and the vertical bars show standard errors of the biases. Primary satellites are N16, MA, and N19. Secondary satellites are N15, N17, and N18. First row: NOAA-19–NOAA-18 during 2009-09. Second and third row: MetOpA–NOAA-17 during 2009-05 and 2009-04, respectively. Fourth row: NOAA-16–NOAA-15 for 2008-08.

5.4.1. Channel 1

[39] N16–N15 bias tends to decrease with increasing scene radiance for N15 but not for N16. MA–N17 biases also tend to decrease with N17 TB but increase with MA TBs. N19–N18 shows a strong increase in bias with both TB.

5.4.2. Channel 2

[40] For both N16–N15 and N19–N18 biases generally increase with TBof the primary satellites. MA-N17 has peak biases at a TB of about 250 K for either satellite which is exactly what is expected based on our discussion in section 3 using simulated radiances (see Figure 2).

5.4.3. Channel 3

[41] For channel 3, N19–N18 bias varies with N19 TBs, from about −20 K at 160 K to 45 K at 310 K. This clearly indicates the instrument problem for this channel. This apparent large bias can be explained by the large noise of N19. The range of brightness temperature is larger for N19 due to higher noise. This in turn will lead to a negative bias for colder N19 TB bins and to a positive bias for warmer N19 TBbins. N16-N15 bias also shows dependence on TBs of both satellites. The bias starts increasing with N16 TBs and then stays constant from 170 K to 230 K and then dips before increasing again. Note that TBs below 230 K are mostly from the two southernmost bins where the channel is a window channel. The bias varies from about −2 K to 7 or 8 K with N16 TBs above 230 K. Bias seems to decrease with N15 TBs (3 K to −1 K), so the overall bias is reduced due to contrasting bias dependence on N15 and N16 TBs. MA–N17 biases generally decrease with both Tbs.

5.4.4. Channel 4

[42] Channel 4 bias patterns are similar to those of channel 3 for MA–N17. N19–N18 bias increases with N19 TBs but does not show any clear relationship with N18 TBs. N16–N15 bias stays constant with both TBs below 240 K and then starts to decrease up to 260 K. The bias then starts to increase with N16 TBs but continues to decrease with N15 TBs.

5.4.5. Channel 5

[43] Channel 5 on N19–N18 shows a strong increase in bias with both TBs, −0.5 K at 160 K and increasing to 0.1 K at 300 K. MA–N17 pair shows larger biases due to the frequency difference as discussed in section 3 with biases increasing with TBs and then starts decreasing when the channel becomes a sounding channel. N16–N15 biases show a very strong dependence on N15 TBs, −6 K at 250 K and 0.5 K at 300 K.

5.4.6. Explaining Latitude Dependence of Bias

[44] The latitude dependence of intersatellite biases can be explained by their dependence on scene radiance. For example, the N19–N18 pair shows rather monotonically increasing bias with increasing TBs for all channels. This leads to similar meridional distribution of TBs and bias, except for channel 3 due to the large noise of N19. Another example is channels 3–5 of MA–N17 which show decreasing bias with increasing TBs, and thus shows opposite meridional patterns for bias and TBs.

6. Summary and Conclusions

[45] Cao et al. [2004, 2005] have shown that the simultaneous nadir overpass (SNO) method is useful for intercalibrating satellite instruments. Nevertheless, Iacovazzi and Cao [2008] expressed concerns over using SNOs for surface sensitive channels. Owing to their normal occurrence in the polar regions, SNOs have potential problems for their use in intercalibrating microwave humidity sounding channels which are surface sensitive under dry polar atmospheric conditions. But as a result of orbital drift, SNOs can occur globally for a short period of time for polar orbiting satellite pairs when their local equator crossing times become close. We used these global SNOs to evaluate intercalibration using only polar SNOs.

[46] There are three satellite pairs with global SNOs for microwave humidity sounders. They are NOAA-16(N16)–NOAA15(N15) during 2008-08, MetOpA(MA)–NOAA-17(N17) during 2009-04 and 2009-05, and NOAA-19(N19)–NOAA-18(N18) during 2009-09. N15, N16, and N17 have the Advanced Microwave Sounding Unit-B (AMSU-B) and N18, N19, and MA have the Microwave Humidity Sounder (MHS). We have shown using simulated brightness temperatures that the differences for these channels between AMSU-B and MHS are dependent on the amount of water vapor in the atmosphere and on the scene radiance. The differences for channel 2 ranges between −2 and 7 K and for channel 5 from −1 to 3 K, but for other channels the differences are negligible.

[47] The method used to obtain collocations is based on Holl et al. [2010]. We used only those collocations with spatial differences less than 5 km and temporal differences less than 300 s, based on sensitivity analyses, to avoid uncertainties due to scene inhomogeneities. We then partitioned the collocated measurements into 18 10° bins. All channels show a large range (∼100 K) in brightness temperature across the latitudes with coldest brightness temperature near the South Pole and the warmest in the tropics. The main source of uncertainty in the SNO method is the combined radiometric noise of the instrument pair. The standard deviations of brightness temperature differences are invariant with latitude indicating that scene inhomogeneities play only a minimal role, given our stringent collocation criteria. The sounding channels (channels 3–5) show different values of standard deviations across the satellite pairs which is consistent with their radiometric noise. For example, the largest standard deviation of about 9 K is shown by N19–N18 pair for channel 3 owing to the anomalously large noise of N19.

[48] Channel 1 generally shows small intersatellite biases and less latitude dependence compared to other channels. Channel 2 has higher bias (up to 3.5 K) for the AMSU-B–MHS combination which is consistent with the results based on simulated radiances shown insection 3. N19–N18 shows a strong latitude dependence for biases in channel 2. N16–N15 shows the largest biases for channels 3, 4, and 5 and also shows a strong latitude dependence. We have validated the biases estimated from global SNOs by biases estimated from zonal mean near-nadir brightness temperatures. We suggest that it is not appropriate to use SNOs over a restricted latitude range to intercalibrate humidity sounders. The reason for the latitude dependence of biases primarily originates from their dependence on scene radiance which themselves have a latitude dependence. It was shown that the dependence of biases on one satellite could be different from another. Channel 3 of N16–N15 shows this behavior with biases increasing with N16 brightness temperatures and decreasing with N15 brightness temperatures. We suggest that it is important to take into account the dependence of biases on scene radiance during the intercalibration procedure.

[49] It has to be kept in mind that the present study explores the global SNOs which are available only for a short time during the life of satellites and thus cannot be used to estimate temporal evolution of bias [e.g., Zou et al., 2006]. Another method for intercalibration that is being developed is monitoring of satellite radiometer biases using NWP fields (R. W. Saunders et al., Monitoring satellite radiometer biases using NWP fields, manuscript in preparation, 2012) which allows global sampling during the entire life time of the satellites. Our plan is to combine the SNO method with the NWP method to intercalibrate microwave humidity sounders.

[50] This work is being done as part of a project to homogenize radiances measured by microwave humidity sounders. The next step is to include SSM/T-2 data as well in our analyses. Intersatellite biases are generally estimated using only near-nadir radiances. To apply these bias estimates to measurements at other viewing angles requires that there are no scan-dependent biases, but in reality there are scan-dependent biases [e.g.,Buehler et al., 2005b]. These asymmetries will also be estimated as part of this project.

Appendix A:: Collocation Methodology

[51] We used the collocation code from Holl et al. [2010]. This code is designed for any pair of satellite sensors and not specifically designed for the study performed here. However, the selection of data fulfilling spatial criteria was modified to improve performance and is different from the algorithm described by Holl et al. [2010]. The first steps are the same: orbits with temporal overlap are located, and then the segments within those pairs of orbits that have a time overlap (plus or minus the maximum time for a collocation) are located. This is described in detail by Holl et al. [2010]. In this study, all individual measurements are binned according to their latitude/longitude values, resulting in two “gridded” datasets for this orbital segment. For all grid cells where one sensor has measurements and the other sensor has measurements in the same or a nearby grid cell, all time differences between measurements from the one and the other sensor are calculated. Here, “nearby” is a function of cell size and maximum collocation distance. The number of neighboring cells to explore for collocations is chosen such that no collocations can be missed. For example, if the maximum collocation distance is 15 km and cells are 1° × 1°, a cell at 85°N is only 9.7 km wide (note that most satellites do not reach so close to the pole). In order not to miss any collocations, this means that measurements from a cell centering at (0°, 85°N) are compared to all measurements in the 15 cells spanning from (2°W, 86°N) to (2°E, 84°N). However, measurements from a cell at (0°, 45°N), where the cell is 79 km wide, need to be compared only within the nine cells spanning from (1°W, 46°N) to (1°E, 44°N). If this spatial criterion is also met, the collocation is selected for further processing.


[52] We thank three anonymous reviewers, Richard Allan, and John Eyre, Nigel Atkinson, Marc Shroeder, and Joerg Shulz for valuable comments. Viju John and David Parker were supported by the Joint DECC/Defra Met Office Hadley Centre Climate Programme (GA01101) and Viju John was also supported by the UK JWCRP. This work contributes to COST Action ES604–Water Vapor in the Climate System (WaVaCS). Thanks to Lisa Neclos of the NOAA CLASS for AMSU-B and MHS Level-1b data, EUMETSAT NWP-SAF for the AAPP software to process the data, and ARTS community for their radiative transfer model.