1. Introduction
The Advanced Microwave Sounding Unit-A (AMSU-A) is a cross-track scanning total power microwave radiometer that measures layer temperatures from the surface to the top of the stratosphere. The first AMSU-A instrument, on board the National Oceanic and Atmospheric Administration-15 (NOAA-15) satellite, was launched in May 1998. Seven additional AMSU-A radiometers were launched to date. Table 1 summarizes its channel characteristics and specifications (Goodrum et al. 2010). AMSU-A consists of three antenna systems: A1–1 hosts channels 6, 7, and 9–15; A1–2 provides channels 3–5, and 8; A2 provides channels 1 and 2. Each antenna system is designed to have a nominal field of view (FOV) of 3.3° and scans the earth at 30 view angles, with a spatial resolution of ~50 km at nadir. AMSU-A is a direct successor of the Microwave Sounding Unit (MSU; 1978–2006, four channels) and was initially designed for numerical weather forecasting and temperature soundings. However, like MSU, AMSU-A data have been widely used for atmospheric temperature climate trend studies due to its near-global coverage under all weather conditions except precipitation (Christy et al. 2003; Mears and Wentz 2009a; Mo 2009; Zou and Wang 2009, 2011).
AMSU-A channel characteristics and specifications. Polarization: horizontal (H) and vertical (V).
The measurement capability of AMSU-A significantly surpasses those of MSU. AMSU-A provides 11 channels (4–14) of high vertical resolution layer atmospheric temperature measurements, utilizing the intensity of oxygen emissions near the 50–60-GHz absorption band. The weighting functions for channels 4–14 cover from the lower troposphere to the top of the stratosphere (see Fig. 1). Yet, the climate community has only utilized limited AMSU-A data for atmospheric temperature trend analysis so far. To date, most studies focused on channels 5, 7, and 9, which are equivalent to MSU channels 2 [midtropospheric temperature (TMT)], 3 [upper-tropospheric temperature (TUT)], and 4 [lower-stratospheric temperature (TLS)], respectively. Two research groups, including the Remote Sensing Systems (RSS) and the NOAA/National Environmental Satellite, Data, and Information Service/Center for Satellite Applications and Research (STAR), have merged the three AMSU-A channels to MSU measurements and generated 30+ years of atmospheric temperature thematic climate data records (TCDRs) (Mears and Wentz 2009a; Zou and Wang 2011). Another group, the University of Alabama at Huntsville (UAH), only used two of the three channels (Christy et al. 2003). Channels 10–14 provide valuable middle to top stratospheric temperature satellite observations after the Stratospheric Sounding Unit (SSU, 1978–2006), yet measurements from these channels as well as those from channels 4, 6, and 8 have not been extensively analyzed by the climate community. Mo (2009) provides the only global temperature anomaly trend analysis using all AMSU-A channels; however, the study was limited to NOAA-15 and issues caused by channel failures, sensor degradation, and orbital drift were not fully addressed. It is necessary to compare NOAA-15 measurements with data from other AMSU-A instruments to ensure continuity and reliability of AMSU-A long-term time series. Zou and Wang (2011) recently recalibrated and reprocessed AMSU-A atmospheric temperature channels; it is a great opportunity to apply these recalibrated radiances in climate change investigations.
Weighting functions of AMSU-A atmospheric temperature channel (4–14) at near nadir (1.67°, solid lines) and limb scan positions (48.33°, dashed lines).
Citation: Journal of Atmospheric and Oceanic Technology 31, 4; 10.1175/JTECH-D-13-00134.1
The purpose of this study is to generate multisatellite merged, monthly averaged, global, consistent AMSU-A-only atmospheric temperature TCDRs (channels 4–14, 1998–2011) for long-term climate trend studies. This study is unique in three aspects. First, it generated consistent high vertical resolution atmospheric temperature TCDRs using all atmospheric channels. Second, it extensively analyzed and merged observations from more AMSU-A instruments than previous studies. Third, new methodologies were developed to address issues particular to generating 11 channels of AMSU-A TCDRs. The paper is organized as follows. AMSU-A level 1c radiance datasets and premerging adjustments required for generating the TCDRs are presented in section 2. Section 3 is dedicated to explaining premerging adjustment methodologies as well as quality control and merging techniques. Major trend features in the TCDRs will be presented in section 4. A summary will be given in section 5.
2. AMSU-A level 1c radiance datasets and adjustments required for TCDR development
AMSU-A instruments have been flown on eight polar-orbiting satellites: NOAA-15–NOAA-19, the Meteorological Operation-A (MetOp-A) and MetOp-B, and the National Aeronautics and Space Administration (NASA) Earth Observing System (EOS) Aqua. Two AMSU-A level 1c radiance datasets are available to date. The first dataset is NOAA operational (OPR) calibrated level 1c radiances (Goodrum et al. 2010). The second dataset is the STAR-recalibrated and -reprocessed AMSU-A level 1c radiances based on new calibration coefficients derived by an Integrated Microwave Intercalibration Approach (IMICA), formerly referred to as the simultaneous nadir overpass (SNO) method (Zou et al. 2006; Zou and Wang 2011). Compared to the OPR calibration, the IMICA addresses a variety of calibration biases and resulted in more consistent multisatellite radiance observations: 1) new constant offsets and nonlinearity coefficients for channels 4–14 were estimated using SNO matchups and global mean temperatures analysis to minimize sun-heating-induced instrument temperature variability in radiances and scene temperature dependency in biases; 2) time-dependent bias drifts in NOAA-16 and MetOp-A were corrected using time-dependent level 1c calibration coefficients; and 3) calibration nonlinearity drift was corrected and frequency shift was found in NOAA-15 channel 6. Figure 2 compares monthly averaged global land and/or ocean mean intersatellite difference time series for the two level 1c radiance datasets using representative channels. We used averaged absolute bias (b), standard deviation (sd), and absolute linear trend (trd) of the global mean intersatellite difference times series for all pairs considered in this study to represent how well observations from different satellites agree with each other. The purpose of using absolute bias (and trend) is to avoid biases (and trends) of individual pairs with opposite signs to cancel each other. Generally speaking, the smaller the three statistics, the more consistent are the observations from different satellites. Our statistical analysis indicated that averaged absolute biases, standard deviations, and absolute trends of the IMICA-recalibrated global mean intersatellite difference time series are generally much smaller than the OPR-calibrated radiances (thorough discussions of Fig. 2 will be given later in this section). Moreover, NOAA-15 (1998–) and NOAA-16 (2001–) provide the two longest observations among all AMSU-A data; unfortunately, the OPR-calibrated NOAA-15 channel 6 and most NOAA-16 channels suffer time-dependent calibration drifts and cannot be used for generating TCDRs (Zou and Wang 2011). On the other hand, the IMICA-recalibrated channels successfully corrected the calibration drifts and are suitable for developing consistent TCDR products.
Global mean intersatellite differences time series for the IMICA-recalibrated (nongray lines) and the NOAA OPR-calibrated (gray lines in the background) level 1c radiances: (a) representative channels/areas from the first group and (b) channels/areas from the second group. The y axis labels show global mean temperatures derived from all satellites and their −0.5 K and +0.5 K offsets. The terms b, sd, and trd represent the averaged absolute bias, standard deviation, and absolute linear trend, respectively, of the global mean intersatellite difference times series for all satellite pairs considered in this study.
Citation: Journal of Atmospheric and Oceanic Technology 31, 4; 10.1175/JTECH-D-13-00134.1
Because of the advantages of IMICA-recalibrated level 1c radiances, they were used in this study to construct the AMSU-A-only TCDRs. Specifically, we incorporated IMICA-recalibrated atmospheric temperature channels from NOAA-15 through NOAA-18, MetOp-A, and Aqua. Table 2 summarizes the time periods and instrument/channel status of all AMSU-A data used. The time period for channels 4–13 is from November 1998 to December 2011. For channel 14, the time period is from January 2001 to December 2011 because NOAA-15 channel 14 failed in October 2000, before the launch of NOAA-16; therefore, it cannot be merged with other satellites. NOAA-19 has not been recalibrated. MetOp-B was launched in September 2012.
AMSU-A data periods and instrument/channel status.
In addition to correcting calibration errors, there is also a need to adjust AMSU-A observations to correspond to consistent channel central frequencies, a uniform observation time, and an identical sensor incidence angle before merging data from multiple satellites to generate homogeneous atmospheric temperature TCDRs (Christy et al. 2003; Mears and Wentz 2009a; Zou and Wang 2009; Wang et al. 2012). To better elucidate the required adjustments to level 1c radiances, we divided the 11 IMICA-recalibrated channels into two groups based on their distinct features of intersatellite bias. The first group consists of channels 4 and 5 over ocean areas (CH_4_5_OCEAN) and channels 6–12 globally. The second group consists of channels 4 and 5 over land areas (CH_4_5_LAND) and channels 13 and 14 globally (CH_13_14). The intersatellite bias features of the two groups are compared in Fig. 2. Channels 7–11 in the first group were not plotted because their intersatellite bias features are similar to CH_4_5_OCEAN and channel 12.
In the first group channels/areas (see Fig. 2a), observations from different satellites agree well with each other after the IMICA recalibration, with standard deviations of intersatellite difference time series less than 0.036 K. The averaged absolute biases are typically less than 0.1 K, except for channel 6. The large bias in channel 6 between NOAA-15 and NOAA-18 was due to NOAA-15 frequency shift from its prelaunch central frequency measurement (Zou and Wang 2011). The intersatellite differences from the frequency shift were not corrected in the IMICA-recalibrated level 1c radiances. Therefore, it must be corrected in this study before NOAA-15 channel 6 observations can be merged with other satellites. For channel 12, Metop-A has a ~0.25-K cold bias relative to other satellites after the IMICA recalibration. However, the bias was mostly constant. Constant bias does not affect trend and can be easily removed using a constant bias correction method (see section 3e).
For the second group channels/areas (see Fig. 2b), global mean intersatellite difference time series analysis indicated that standard deviations are equal or larger than 0.080 K and averaged absolute biases are larger than 0.1 K even after the IMICA recalibration. The larger intersatellite biases are mainly caused by diurnal drift effect. Satellites carrying AMSU-A instruments were launched at different local equator crossing times (LECT; see Fig. 3). NOAA-15, NOAA-17, and MetOp-A are morning satellites; NOAA-16, NOAA-18, and Aqua are afternoon satellites. Moreover, the LECT for NOAA-15 and NOAA-16 have drifted several hours since their launches, with NOAA-15 drifted from 0730 (1998) to 0440 (2011) and NOAA-16 drifted from 1400 (2001) to 2000 (2011). Global intersatellite bias analysis by Zou and Wang (2011) and in this study indicated that the diurnal effect can be ignored for the first group channels/areas. Applying diurnal corrections to these channel/areas does not reduce intersatellite biases and neither improves climate data records. On the contrary, the diurnal drift effects are big enough in the second group channels/areas to cause spurious trends if not corrected. Therefore, observations from the second group channels must be adjusted to uniform observation time before satellite merging.
AMSU-A LECT (1998–2011).
Citation: Journal of Atmospheric and Oceanic Technology 31, 4; 10.1175/JTECH-D-13-00134.1
In addition to the frequency shift and the diurnal drift effects, sensor incident angle effect in the level 1c radiances also needs to be corrected. AMSU-A is a cross-scan radiometer, with its scan angles varying from 1.67° to 48.33°. Sensor incidence angles vary with scan angles and cause changes in the length of optical path between Earth and satellites and therefore changes in channel weighting functions (see Fig. 1). For the atmospheric temperature channels, the sensor incidence angle effect can be as large as 15 K for the limb scan positions (Goldberg et al. 2001) and introduces large sampling noise in the datasets if not corrected (Christy et al. 2000; Mears and Wentz 2009a). Moreover, the altitudes of NOAA-15, -16, and -18 dropped several kilometers since their launches. Satellite orbital decay (loss of altitude) causes sensor incidence angle drifts over time. In addition, satellite roll and pitch angle variations over time also cause changes in sensor incidence angle. Satellite altitude and attitude drifts, if large enough, can cause spurious trend in the long-term time series (Mears and Wentz 2009a).
In this study, we used time series of off-nadir biases, the global mean temperature difference between a particular scan position and nadir (averages of scans 15 and 16), to represent sensor incidence angle effect. Off-nadir bias statistics for representative channels (5 and 8 from antenna system A1–2, 9 and 12 from A1–1) in NOAA-15 AMSU-A are shown in gray in Fig. 4. Each point summarizes off-nadir biases for one scan position, with y values representing the mean of off-nadir bias time series. The size of each point is proportional to the standard deviation of the off-nadir bias time series for this scan. Averaged off-nadir biases for the 16 near-nadir scans (8–23) are also shown. There are two obvious features in Fig. 4: 1) A1–1 and A1–2 channels show distinct off-nadir biases and 2) channels 9 and 12 on A1–1 have larger off-nadir bias variations in near-limb scans. Off-nadir biases for other channels/satellites are similar to those in Fig. 4.
Off-nadir bias statistics for representative channels on NOAA-15 AMSU-A for before (gray) and after (black) the correction of the sensor incidence angle effect. Each point summarizes the off-nadir bias time series for one scan, with y values representing the mean bias. The size of each point is proportional to the standard deviation of the off-nadir bias time series for this scan.
Citation: Journal of Atmospheric and Oceanic Technology 31, 4; 10.1175/JTECH-D-13-00134.1
3. Methods for generating AMSU-A atmospheric temperature TCDRs
Our procedure for construction of multisatellite merged, monthly global gridded AMSU-A-only TCDRs consists of the following steps: 1) adjusting all observations to a consistent sensor incidence angle, 2) correcting the NOAA-15 channel 6 frequency shift, 3) adjusting channels 4 and 5 and 13 and 14 observations to local solar noon time, 4) generating monthly global gridded individual satellite time series and quality control, and 5) merging multisatellite observations to generate the AMSU-A atmospheric temperature TCDRs. Figure 5 shows the flowchart of the procedure. Each step will be discussed in detail in the following subsections.
Flowchart of the procedure for generating AMSU-A-only TCDRs.
Citation: Journal of Atmospheric and Oceanic Technology 31, 4; 10.1175/JTECH-D-13-00134.1
a. Correction of sensor incidence angle effect
Different techniques have been developed to address the sensor incidence angle effect in MSU and AMSU-A observations (Goldberg et al. 2001; Christy et al. 2003; Zou et al. 2006; Mears and Wentz 2009a; Zou and Wang 2009). In this study, sensor incidence angle effect due to the sensor scan angle and satellite altitude variations were corrected using correction terms derived from radiative transfer simulations. AMSU-A satellite attitudes decay were very small and do not drift over time (Dr. H. Meng, NOAA/NESDIS/STAR, 2011, personal communication); therefore, they were not considered at the current stage. For each observation, the sensor incidence angle effect (
The NOAA/Joint Center for Satellite Data Assimilation Community Radiative Transfer Model (CRTM) (Han et al. 2006) was used to simulate all brightness temperatures for AMSU-A. Surface skin temperatures and atmospheric profiles required for the CRTM simulation were obtained from the NASA Modern-Era Retrospective Analysis for Research and Applications (MERRA) reanalysis (Rienecker et al. 2011). In the MERRA reanalysis, surface skin temperatures are hourly data with ⅔° longitude and ½° latitude spatial resolution; atmospheric temperature, moisture, and ozone profiles are 3-hourly data at 42 pressure levels with a grid resolution of 1.25° longitude × 1.25° latitude. All CRTM input parameters were interpolated to the geolocation and time of satellite observations. The sensor incidence angle correction term as well as the NOAA-15 channel 6 frequency shift term presented in the next subsection is related to the change of weighting functions. Wang et al. (2012) indicated that correction terms derived from CRTM simulations were not sensitive to the errors in the atmospheric profile itself, since these errors were cancelled out with the differences between simulations.
Figure 4 compares off-nadir biases for before and after the correction of sensor incidence angle effect. Mean off-nadir biases were less than 1 K for all cases after the correction, significantly reduced compared to the nearly 15-K off-nadir biases before the correction. Moreover, temperature variations within individual scans, indicated by the size of each point, were also depressed. Nevertheless, residual biases still exist, especially in near-limb scans. The cause of the residual biases was uncertain; it may be due to instrument antenna patterns, instrument asymmetry, ground-source radio frequency interference, and uncertainties in CRTM input parameters. Similar constant residual off-nadir biases were reported by Mears and Wentz (2009a). In this study, we excluded data with large residual off-nadir biases and used the 16 near-nadir scans (scan positions 8–23) only for constructing the AMSU-A-only TCDRs. The residual off-nadir biases for these 16 near-nadir scans were very small (on the order of 0.1 K), and they were further removed by subtracting constant offsets derived using the individual satellite global land and ocean mean temperatures. The purpose of this additional adjustment is to make the global temperatures at each scan have the same mean values as that at nadir. Long-term time series analysis indicated that the correction of AMSU-A sensor incidence angle effect changes the absolute value of the global mean temperatures only but has a negligible impact on biases, standard deviations, and trends of global mean intersatellite difference time series. In other words, it reduced sampling noises only but would not affect layer temperature trends in the final TCDR products.
The method used in this study adjusts the sensor incidence angle effect in each channel independently. It is different from the Goldberg et al. (2001)’s method used in the STAR-merged MSU–AMSU-A time series (Zou and Wang 2009, 2011), which adjusts a target channel using adjacent channels whose weighting functions peak above and/or below it. The performances of the two methods are comparable in the absence of malfunctioning channels. However, multiple AMSU-A channels failed during satellite operations (see Table 2 for the onset dates for each channel failures). In this situation, the Goldberg et al. (2001) algorithm was inapplicable. Specific channels with such problems include NOAA-16 and Aqua channel 5 (due to channel 4 failures), MetOp-A channels 6 and 8 (due to channel 7 failure), and NOAA-15 channels 12 (due to channel 11 failure) and 13 (due to channel 14 failure). The sensor angle effect correction based on CRTM simulations do not depend on neighbor channels and thus can provide adequate adjustments in all situations. This maximizes utilizations of AMSU-A observations.
b. Correction of frequency shift for NOAA-15 AMSU-A channel 6
The IMICA-recalibrated NOAA-15 channel 6 was corrected to the prelaunch-determined central frequency before it was merged to the same channel from other instruments. The frequency shift correction term (
Figure 6 illustrates the global mean time series of the NOAA-15 AMSU-A channel 6 frequency shift correction term. Figure 7 shows the global map of the multiyear averaged adjustments. The adjustments are ~0.6 K over the polar regions, consistent with the magnitude of the frequency shift effect determined during the IMICA recalibration process (Zou and Wang 2011). For midlatitude and tropical regions, the adjustments are ~0.8 and ~1.4 K, respectively. After the correction, NOAA-15 channel 6 becomes more consistent with NOAA-18, with the standard deviation and trend of intersatellite difference time series reduced from 0.056 to 0.051 K and from 0.042 to 0.027 K decade−1, respectively.
Global mean time series for the NOAA-15 AMSU-A channel 6 frequency shift correction term.
Citation: Journal of Atmospheric and Oceanic Technology 31, 4; 10.1175/JTECH-D-13-00134.1
Global map of the multiyear averaged adjustments applied to NOAA-15 AMSU-A channel 6 by the frequency shift correction.
Citation: Journal of Atmospheric and Oceanic Technology 31, 4; 10.1175/JTECH-D-13-00134.1
c. Correction of diurnal drift effect
Diurnal drift effects in CH_4_5_LAND and CH_13_14 as stated in section 2 were corrected using diurnal anomaly climatologies from two difference sources. For CH_4_5_LAND, the diurnal drift effect is mainly due to diurnal variations of surface and atmospheric temperatures over land areas. The RSS group generated a 5-yr averaged monthly diurnal anomaly climatology for AMSU-A channel 5 using the National Center for Atmospheric Research (NCAR) Community Climate Model (CCM) (Mears et al. 2002, 2003). In this climatology, diurnal anomalies are functions of sensor scan angle, Earth location, month, and local solar hour. For each observation, the diurnal correction term (
As in Fig. 2b, but for after the corrections of the sensor incidence angle and diurnal drift effects. Time series for after the correction of the sensor incidence angle effect but before the correction of the diurnal drift effect are shown in the background (gray lines).
Citation: Journal of Atmospheric and Oceanic Technology 31, 4; 10.1175/JTECH-D-13-00134.1
Multiyear averaged spatial patterns of intersatellite biases for before and after the correction of the diurnal drift effect.
Citation: Journal of Atmospheric and Oceanic Technology 31, 4; 10.1175/JTECH-D-13-00134.1
The diurnal effect in CH_13_14 is caused by strong semidiurnal tides in the stratosphere (Seidel et al. 2005; Zou and Wang 2011; Wang et al. 2012). The land–sea contrast in CH_13_14 is not prominent, different from channels 4 and 5. Two new multiyear averaged diurnal/semidiurnal anomaly climatologies for CH_13_14 were developed using CRTM simulations with MERRA 3-hourly, monthly averaged atmospheric temperature, moisture, and ozone profiles as inputs. Figure 10 shows CRTM-simulated diurnal/semidiurnal anomaly climatologies near the equator (1.25°N, 1.25°E) for January and July. The magnitudes of CRTM-simulated solar diurnal and semidiurnal tides are ~1.8–2.0 K for channel 14 (2 hPa) and ~1.0–1.8 K for channel 13 (5 hPa) at this location. For January, the magnitudes of diurnal/semidiurnal cycles are similar at the two channels. For July, channel 14 exhibits larger diurnal/semidiurnal cycles than channel 13. It should be noted that the diurnal mode was cancelled out by averaging the ascending and descending orbits so the diurnal effects came from the semidiurnal mode, which has a magnitude much smaller than the diurnal mode.
CRTM-simulated monthly diurnal anomaly climatology at 1.25°N × 1.25°E for January and July: (a) channel 13 and (b) channel 14.
Citation: Journal of Atmospheric and Oceanic Technology 31, 4; 10.1175/JTECH-D-13-00134.1
CH_13_14 observations were adjusted to local solar noon time using two CRTM-simulated diurnal anomaly climatologies. By comparing intersatellite difference time series (the bottom two sets of time series in Fig. 8) and multiyear averaged intersatellite difference spatial patterns (the bottom four images in Fig. 9) for before and after the correction, the CRTM-simulated diurnal anomaly climatologies appeared to significantly reduce the diurnal drift effect in CH_13_14, with averaged intersatellite biases dropping from 0.143 to 0.115 K for channel 13 and from 0.194 to 0.066 K for channel 14; the standard deviation of intersatellite difference time series dropped from 0.080 to 0.049 K for channel 13 and from 0.093 to 0.043 K for channel 14; and the averaged trend of intersatellite difference time series reduced from 0.191 to 0.079 K decade−1 for channel 13 and from 0.322 to 0.090 K decade−1 for channel 14.
Zou and Wang (2011) suggested that channels 6–10 do not need a diurnal drift adjustment due to small intersatellite biases for these channels. Also shown in section 2, intersatellite difference time series has averaged absolute bias less than 0.1 K and standard deviations less than 0.036 K for channels 11 and 12, indicating a small diurnal drift effect for these two channels. Therefore, no diurnal drift correction was applied for channels 6–12 in this study.
d. Generating individual satellite time series and quality control
After the corrections of sensor incidence angle, frequency shift, and diurnal drift effects, all AMSU-A IMICA-recalibrated level 1c radiances were adjusted to nadir, local solar noon time, and prelaunch-specified channel frequencies. The next step is to remove erroneous or low-quality observations and generate global gridded monthly averaged individual satellite time series.
Two levels of quality control were employed. First, we used pixel-level quality control to identify and remove various bad or low-quality observations in level 1c radiances: 1) precipitation-contaminated pixels were identified and excluded using Grody et al.’s (2000) precipitation screening scheme. The screening was only applied to channels 4–6, and approximately 6%–7% of observations were removed. 2) Also excluded were anomalous observations with temperatures exceeding minimum/maximum values (see Table 3) that were statistically determined using global mean temperature time series from all satellites. Anomalous observations may be due to instrument malfunctions or failures of precipitation screening. After the pixel-level quality control, AMSU-A observations were averaged over grid cells to generate gridded (2.5° × 2.5°), monthly averaged individual satellite layer temperature time series. Warm target temperature time series for each antenna system are also generated for multisatellite merging purpose.
Minimum and maximum temperature threshold values used in AMSU-A TCDR pixel-level quality control.
Second, grid-level quality control was applied to identify and remove abnormal months in the individual satellite temperature time series. Abnormal months were determined by two means: 1) total number of pixels used for generating a monthly averaged grid is less than 60% of the theoretical maximum number of pixels, which can be calculated using the total scan time in a month (8 s per scan line for AMSU-A); and 2) visually examining images of monthly gridded individual satellite temperatures and intersatellite temperature differences as well as global mean intersatellite temperature difference time series. As result, the following months of the gridded temperatures were removed: 1) NOAA-15 at October 1998, NOAA-15 channel 11 at April 2002, NOAA-16 channel 4 after January 2007, NOAA-17 at June 2002 and October 2003, NOAA-18 at August 2005, MetOp-A channel 7 after August 2008, Aqua before August 2002 for all channels, and Aqua channel 4 after August 2007. Generally, only few missing orbits were found in the AMSU-A observations. The above-mentioned months were usually associated with the beginning or ending of an instrument, subinstrument, or channel’s life time, and therefore they were affected by undersampling or severely degraded observations. 2) NOAA-15 channel 6 data after 2009 were not used. The STAR IMICA-derived calibration coefficients were estimated using SNO matchups and global mean temperatures from November 1998 to December 2009. The nonlinearity of this channel changed over time; we found existing coefficients were less effective after 2009. 3) Aqua months at October 2003 and November 2003 were removed because of large sampling errors due to missing data. 4) NOAA-16 channels 9–14 at November 2002 were excluded due to anomalies caused by its phase lock loop oscillator manipulations (NOAA 2013). 5) NOAA-17 channel 14 months from November 2002 to January 2003 were also not used due to unexplained large intersatellite biases compared to other satellites.
e. Generating multisatellite-merged AMSU-A-only TCDRs
Next, we merged individual satellite time series to generate multisatellite-merged AMSU-A-only atmospheric temperature TCDRs. Intersatellite biases were further reduced during the merging process through minimizing biases due to the residual warm target effect and Earth location–dependent constant biases.
Christy et al. (1998, 2000) first discovered the warm target effect—that is, strong correlations between intersatellite difference time series and warm target temperature time series in certain MSU satellite pairs. Similar correlations also exist in AMSU-A data. Discussions of the underlying cause of the warm target effect—that is, inaccurate calibration nonlinearity in level 1c calibration—are given in detail in Zou et al. (2006). Two methods were developed to correct the warm target effect: 1) the Christy method, which estimates empirical correction terms (target factors) using intersatellite difference time series and warm target temperature time series (Christy et al. 2000); and 2) the SNO calibration method (a major component in the IMICA), derives new nonlinear calibration coefficients using SNO matchups and global mean temperatures (Zou et al. 2006; Zou and Wang 2011). Zou and Wang (2010) found that a combination of the two methods can optimally remove the warm target effect. In this study, the warm target effect was first reduced by the SNO calibration through using the STAR IMICA-recalibrated AMSU-A level 1c radiances. The residual warm target effect caused by imperfect nonlinear calibration was further addressed using the Christy method. Monthly averaged global mean (ocean only for channels 4 and 5 due to large residual intersatellite biases over land, globally for other channels) scene temperatures and their corresponding warm target temperature anomaly time series were used to construct a series of regression equations for overlapping periods; target factors were obtained by solving the equations. Our results indicated that the residual warm target effects are very small in channel 14; therefore, no Christy bias correction was applied to this channel. In addition, the Christy correction does not affect Aqua channels because its warm target temperatures are well controlled and do not vary much over time (Zou and Wang 2011). Target factors for the majority of instruments/channels are less than 0.005, indicating that the SNO calibration performs well in reducing the warm target effect in level 1c radiances. NOAA-17 channels and some channels of MetOp-A have larger target factors (~0.02). The nonlinear calibration coefficients for NOAA-17 and MetOp-A may be less accurate compared to other AMSU-A instruments because the SNO matchups used for intersatellite calibration of the two instruments were very short, 14 months for NOAA-17 and 2 years (2007–09) for MetOp-A (Zou and Wang 2011). MetOp-A requires a reintercalibration in the future to derive more accurate calibration coefficients using longer SNO matchup time series. In this study, because of the complementary effects between the Christy correction and the SNO method, the two methods together can effectively minimize the warm target effect in all AMSU-A observations.
After correction of the residual warm target effect, the next step is to remove Earth location–dependent constant intersatellite biases. Mears and Wentz (2009a) employed a zonal mean constant bias correction scheme, which calculates latitude zonal mean biases by averaging grid cells over a 2.5° band and removes the biases sequentially. Zou et al. (2009) and Zou and Wang (2010) suggested a gridcell-dependent constant bias correction scheme, that is, estimating constant biases at each grid cell. For channels 4 and 5, in which observations over land have larger errors than ocean due to imperfect corrections of diurnal drift effect, the Zou et al. (2009) and Zou and Wang (2010) methods can prevent errors over land from contaminating ocean. The disadvantage of this scheme is that sampling errors may cause inaccurate intersatellite biases estimation for some grid cells. In this study, we adopt a new method that combines the advantages of the two schemes for channels 4 and 5. The new method calculates 2.5° latitudinal band zonal mean constant biases separately over land and ocean so that the errors over land do not affect ocean. For channels 6–14, which do not show distinct land–ocean bias patterns, the zonal mean constant bias correction (Mears and Wentz 2009a) was used.
Figure 11 shows the global mean intersatellite difference time series and corresponding statistics for representative channels after the corrections of the residual warm target effect and Earth location–dependent constant biases. The Christy correction slightly reduced the standard deviations of the intersatellite difference time series in some channels, such as channels 4, 6, and 13. The Christy correction and Earth location–dependent constant bias correction together reduced the averaged intersatellite biases to less than 0.004 K for all channels. After the two corrections, different satellites agree well with each other during their overlapping periods.
As in Fig. 2, but for after the corrections of the residual warm target effect and Earth location–dependent constant biases. Time series for before the two corrections were plotted in the background (gray lines).
Citation: Journal of Atmospheric and Oceanic Technology 31, 4; 10.1175/JTECH-D-13-00134.1
Finally, 11 channels of multisatellite-merged monthly averaged atmospheric temperature TCDRs were generated by averaging gridded individual satellite time series from all satellites. The mean annual cycle was calculated using data from the entire study period and the anomaly time series were derived by subtracting the mean annual cycle from the merged temperature time series.
4. Major trend features in the AMSU-A-only TCDRs
This study focused primarily on the methodologies for constructing the AMSU-A only atmospheric temperature TCDRs. The 13-yr AMSU-A-only TCDRs developed in this study are relatively short for some climate signals to be statistically robust and thus are sensitive to additional observations. However, the observations are long enough for certain features to be suggestive and warrant further investigations. Figure 12 shows monthly averaged global (−82.5° to 82.5°) mean temperature anomaly time series and trends. Individual satellite anomaly time series after all corrections were also plotted using distinct colors to illustrate the temporal coverage of each satellite. The 2002/03 and 2009/10 El Niño events with moderate magnitude can be observed in all tropospheric channels. The channel 4 (lower troposphere) global mean anomaly time series shows a warming trend of 0.120 K decade−1 with a 95% confidence interval of ±0.080 K decade−1. The warming was more prominent over ocean areas (0.131 K decade−1), where the diurnal drift effect is ignorable. Zou and Wang (2010) suggested that trends over land and oceans should be comparable, assuming the atmosphere is well mixed. The fact that trends over land were less warming than oceans for channel 4 suggests there is still room to improve the diurnal adjustment for this channel. Indeed, intersatellite biases over land after the diurnal correction were still larger than oceans (see Figs. 9 and 11), although much smaller than those before the diurnal adjustment.
AMSU-A-only TCDR global mean layer temperature anomaly time series and trends (gray dashed vertical lines show the 2002/03 and 2009/10 El Niño events).
Citation: Journal of Atmospheric and Oceanic Technology 31, 4; 10.1175/JTECH-D-13-00134.1
In contrast to channel 4, channels 5–7 show small warming trends ranging from 0.025 to 0.055 K decade−1 with large uncertainties. In the tropopause and stratosphere, channels 8–13 show cooling trends from −0.023 to −0.657 K decade−1, indicating a systematic increase of cooling trends as the center of channel weighting functions move higher. Channel 14 also shows a large cooling trend from 2001 to 2011 (−0.721 K decade−1). The 95% confidence intervals indicated that the cooling trends for channels 10–14 are robust. Note that channel 14 was not corrected for the Zeeman splitting effect. The Zeeman effect caused a spatial bias pattern proportional to Earth’s magnetic field if not corrected (Han et al. 2006). This dominant pattern is similar for different satellites. Unfortunately, time variations of Earth’s magnetic field were not included in CRTM for adjustment, so it is unclear how the Zeeman effect may affect the channel 14 trends.
Global spatial trend patterns derived from the AMSU-A-only TCDRs are illustrated in Fig. 13. The most striking feature are the cooling trends over the midlatitudes, tropics, and the Antarctic regions in the upper stratosphere (channels 12–14), contrasted by the persistent warming trends over Canada. Lower to midstratospheric channels (9–11) show cooling trends in the tropics and the polar regions but warming trends over midlatitude areas. Over the tropics, cooling trends in the stratospheric channels gradually changed to warming trends in the tropospheric channels. For channels 4 and 5, the polar regions are dominated by warming trends, while both the United States and the eastern China show cooling trends.
Spatial trend patterns derived from AMSU-A-only TCDRs (1998–2011 for channels 4–13, 2001–11 for channel 14).
Citation: Journal of Atmospheric and Oceanic Technology 31, 4; 10.1175/JTECH-D-13-00134.1
We also studied the contributions of major bias-correction procedures to AMSU-A trends by comparing global mean temperature anomaly trends derived from this study and those from Mo (2009) for the time period of 1998–2007. Several factors may cause trend differences between the two studies: 1) the correction of the warm target effect, which was applied in this study through the use of the IMICA-recalibrated radiances (see section 2) and the Christy correction method (see section 3e)—no warm target effect was corrected in Mo (2009); 2) the correction of the diurnal drift effect—no diurnal drift correction was applied in Mo’s (2009) data processing procedure; 3) the length of data—Mo (2009) used 6 months of extra data (May 1998–October 1998) from the NOAA-15 sensor evaluation period. Based on Fig. 2 in Mo (2009), the extra months have positive anomalies in channels 4–8. Our experiments indicated that these positive anomalies in the beginning of Mo’s (2009) anomaly time series reduced temperature trends by ~0.1–0.13 K decade−1. For channels 9–14, the impacts of these extra months on the trends are limited. 4) The time-dependent constant bias and nonlinearity for the IMICA-recalibrated NOAA-15 channel 6. Except for the above-mentioned four factors, our results indicated that other bias-correction procedures do not have significant impacts on the trend differences between the two studies. Moreover, our quality control results indicated NOAA-15 AMSU-A observations have a high data integrity with few missing orbits during its channels’ normal operation periods. Though Mo (2009) used daily instead of monthly averaged global mean temperatures for trend analysis, the daily averaged time series usually have larger noises but do not alter trends significantly.
Table 4 compares global mean temperature anomaly trends derived from this study and those from Mo (2009). The trend differences as well as the impacts of the warm target effect and the diurnal drift effect corrections on channel trends were also listed in the table for analysis purpose. Generally speaking, both studies show warming in the troposphere and cooling in the stratosphere, except for channel 6 in Mo (2009). Compared to Mo (2009), global mean trends derived from this study vary more systematically as the peaking of channel weighting functions moves higher. For channel 4, the trends derived by the two studies are very close. However, our detailed analysis indicated that the correction of the diurnal drift effect has a −0.079 K decade−1 impact on the channel 4 trend (~32%). In addition, the impact of the warm target effect correction is −0.049 K decade−1. The impacts of the two corrections together are very close to the effect of extra data in Mo (2009) (−0.130 K decade−1). In other words, the channel 4 trend derived by Mo (2009) would be ~0.130 K decade−1 warmer than this study if the extra data were not used. For channel 5, the trend difference is ~42%, mainly due to the correction of the diurnal drift effect applied in this study (−0.042 K decade−1) and the extra data in Mo (2009). The influence of the warm target effect correction on this channel is negligible. Our channel 6 trend (0.055 K decade−1) differs significantly from the −0.546 K decade−1 cooling trend in Mo (2009). It was shown that operational-calibrated channel 6 data had inaccurate calibration nonlinearity and channel central frequency (Zou and Wang 2011); thus, the trend in Mo (2009), which was based on operational calibration, was unreliable. These problems were successfully addressed by the IMICA intercalibration. Hence, the channel 6 trend derived in this study is more reliable and more consistent with other channels. Similar to channel 5, the 53% and 106% of trend differences in channels 7 and 8 may be mainly due to the correction of the warm target effect and the extra data used by Mo (2009). For channels 9, 10, and 12, all the potential factors have relatively small impacts on anomaly trends. As a result, the trends from the two studies generally agree well with each other. For channel 13, the trend difference between the two studies is mostly due to the correction of the diurnal/semidiurnal drift effect applied in this study because the other two factors have a limited influence on trend for this channel. The time series for channels 11 and 14 from Mo (2009) are too short for climate studies because the two channels on NOAA-15 failed in April 2002 and October 2000, respectively. On the other hand, we generated consistent long-term TCDRs for channels 11 and 14 by incorporating observations from the other five AMSU-A instruments.
Comparison of global mean temperature anomaly trends between this study and Mo (2009) (K decade−1).
AMSU-A channels 4, 5, 7, and 9 were equivalent channels for the merged MSU–AMSU-A temperature of lower troposphere (TLT), TMT, TUT, and TLS (Christy et al. 2003; Mears and Wentz 2009a,b; Zou and Wang 2011), although their weighting function peaks are slightly different. The TLT times series, the counterpart of channel 4 TCDR, is constructed using near-limb observations from MSU channel 2 and AMSU-A channel 5 observations (Christy et al. 2003; Mears and Wentz 2009b). Figure 14 compares our channels 4, 5, 7, and 9 TCDRs with the latest MSU–AMSU-A merged time series developed by the UAH (version 5.5), RSS (version 3.3), and STAR (version 2.0) at global and tropical regions. The STAR version 2.0 time series did not include merged MSU–AMSU TLT product—the AMSU-A-only channel 4 data developed in this study serve as the STAR TLT product; the UAH group does not support the TUT product. The global mean temperature anomaly trend of the channel 4 TCDR is 0.042 K decade−1 cooler than the coincident UAH TLT time series but 0.061 K decade−1 warmer than the coincident RSS TLT time series during the same time period. Similar differences were also observed in the tropical region, which implies a ~0.1-K trend difference between the UAH and RSS TLT products. The MSU–AMSU-A merged TMT time series are all warmer than the channel 5 TCDR, with trend differences of 0.045, 0.022, 0.059 K decade−1 for the UAH, RSS, and STAR groups, respectively. Similar large differences were also observed over the tropics. The trend difference between the channel 5 TCDR and the STAR TMT was mainly due to the STAR group’s MSU and AMSU-A merging method. For channel 7, the trend from this study is consistent with the STAR TUT time series, but it differs from the RSS time series. Interestingly, large disagreements were also observed among the MSU–AMSU-A merged TLT, TMT, and TUT products developed by different groups during the 1998–2011 period. On the contrary, the TLS products from the three groups are generally consistent with each other; meanwhile, they also agree better with the AMSU-A channel 9 TCDR. Several factors may cause the observed inconsistency among the AMSU-A TCDRs and the MSU/AMSU-A merged time series: 1) the central frequency differences between the MSU and AMSU-A equivalent channels; 2) calibration schemes; 3) the correction of the diurnal drift effect; 4) differences in quality control and data merging procedures; and 5) errors in near-limb observations from MSU channel 2 and AMSU-A channel 5 (for TLT only). While the AMSU-A-only TCDRs provide a potential new means to reconcile the differences among MSU–AMSU-A merged atmospheric temperature time series, more efforts are required to pinpoint the underlying causes of the discrepancies in the future.
Comparing AMSU-A channels 4, 5, 7, and 9 TCDRs with the TLT, TMT, TUT, and TLS time series developed by the UAH, RSS, and STAR groups: global: −70° to 82.5° for channel 4 and −82.5° to 82.5° for other channels (solid curve), and tropical: −20° to 20° (dashed curve).
Citation: Journal of Atmospheric and Oceanic Technology 31, 4; 10.1175/JTECH-D-13-00134.1
5. Summary
In this study, we focused on methodologies for the construction of multisatellite merged, monthly averaged, global gridded AMSU-A-only atmospheric temperature TCDRs (11 channels, 1998–2011) using the IMICA-recalibrated AMSU-A level 1c radiances. The sensor incidence angle effect in all channels and the frequency shift in NOAA-15 channel 6 were corrected using CRTM simulations. The diurnal drift effect in channels 4 and 5 was corrected using the RSS diurnal anomaly climatologies; the diurnal drift effects in CH_13_14 were corrected using diurnal/semidiurnal anomaly climatologies derived from MERRA monthly diurnal atmospheric profiles. Six satellite merged AMSU-A-only TCDRs were developed after the corrections of the residual warm target effect and zonal mean constant biases. Eleven channels of AMSU-A-only TCDRs from the lower troposphere to the top of stratosphere were generated for atmospheric temperature trend studies. The well-merged AMSU-A-only TCDRs can also be used to generate merged MSU–AMSU-A and SSU–AMSU-A upper-air temperature time series.
Acknowledgments
The study was supported by NOAA Grant NESDISNESDISPO20092001589 (SDS0915). The authors thank Dr. Yong Chen from the NOAA CRTM team for providing shifted NOAA-15 channel 6 spectral coefficients and Dr. Huan Meng for kindly providing representative AMSU-A attitude datasets. The opinions contained in this paper are those of the authors and should not be construed as an official NOAA or U.S. government position, policy, or decision.
REFERENCES
Christy, J. R., Spencer R. W. , and Lobl E. S. , 1998: Analysis of the merging procedure for the MSU daily temperature time series. J. Climate, 11, 2016–2041, doi:10.1175/1520-0442-11.8.2016.
Christy, J. R., Spencer R. W. , and Braswell W. D. , 2000: MSU tropospheric temperatures: Dataset construction and radiosonde comparisons. J. Atmos. Oceanic Technol., 17, 1153–1170, doi:10.1175/1520-0426(2000)017<1153:MTTDCA>2.0.CO;2.
Christy, J. R., Spencer R. W. , Norris W. B. , Braswell W. D. , and Parker D. E. , 2003: Error estimates of version 5.0 of MSU–AMSU bulk atmospheric temperatures. J. Atmos. Oceanic Technol., 20, 613–629, doi:10.1175/1520-0426(2003)20<613:EEOVOM>2.0.CO;2.
Goldberg, M. D., Crosby D. S. , and Zhou L. , 2001: The limb adjustment of AMSU-A observations: Methodology and validation. J. Appl. Meteor., 40, 70–83, doi:10.1175/1520-0450(2001)040<0070:TLAOAA>2.0.CO;2.
Goodrum, G., and Coauthors, cited 2010: NOAA KLM user’s guide. [Available online at http://www.ncdc.noaa.gov/oa/pod-guide/ncdc/docs/klm/cover.htm.]
Grody, N. C., Weng F. , and Ferraro R. R. , 2000: Application of AMSU for obtaining hydrological parameters. Microwave Radiometry and Remote Sensing of the Earth’s Surface and Atmosphere, P. Pampaloni and S. Paloscia, Eds., VNU Science Press, 339–351.
Han, Y., van Delst P. , Liu Q. , Weng F. , Yan B. , Treadon R. , and Derber J. , 2006: JCSDA Community Radiative Transfer Model (CRTM)—Version 1. NOAA Tech. Rep. NESDIS 122, 40 pp. [Available online at ftp://ftp.emc.ncep.noaa.gov/jcsda/CRTM/CRTM_v1-NOAA_Tech_Report_NESDIS122.pdf.]
Mears, C. A., Schabel M. C. , Wentz F. J. , Santer B. D. , and Govindasamy B. , 2002: Correcting the MSU middle tropospheric temperature for diurnal drifts. 2002 IEEE International Geoscience and Remote Sensing Symposium, Vol. III, IEEE, 1839–1841, doi:10.1109/IGARSS.2002.1026272.
Mears, C. A., Schabel M. C. , and Wentz F. J. , 2003: A reanalysis of the MSU channel 2 tropospheric temperature record. J. Climate, 16, 3650–3664, doi:10.1175/1520-0442(2003)016<3650:AROTMC>2.0.CO;2.
Mears, C. A., and Wentz F. J. , 2009a: Construction of the Remote Sensing Systems V3.2 atmospheric temperature records from the MSU and AMSU microwave sounders. J. Atmos. Oceanic Technol., 26, 1040–1056, doi:10.1175/2008JTECHA1176.1.
Mears, C. A., and Wentz F. J. , 2009b: Construction of the RSS V3.2 lower-tropospheric temperature dataset from the MSU and AMSU microwave sounders. J. Atmos. Oceanic Technol., 26, 1493–1509, doi:10.1175/2009JTECHA1237.1.
Mo, T., 2009: A study of the NOAA-15 AMSU-A brightness temperatures from 1998 through 2007. J. Geophys. Res., 114, D11110, doi:10.1029/2008JD011267.
NOAA, cited 2013: POES status. NOAA/NESDIS/Office of Satellite Operations. [Available online at www.oso.noaa.gov/poesstatus/.]
Rienecker, M. M., and Coauthors, 2011: MERRA: NASA’s Modern-Era Retrospective Analysis for Research and Applications. J. Climate, 24, 3624–3648, doi:10.1175/JCLI-D-11-00015.1.
Seidel, D. J., Free M. , and Wang J. , 2005: Diurnal cycle of upper-air temperature estimated from radiosondes. J. Geophys. Res., 110, D09102, doi:10.1029/2004JD005526.
Wang, L., Zou C.-Z. , and Qian H. , 2012: Construction of stratospheric temperature data records from stratospheric sounding units. J. Climate, 25, 2931–2946, doi:10.1175/JCLI-D-11-00350.1.
Zou, C.-Z., and Wang W. , 2009: Diurnal drift correction in the NESDIS/STAR MSU/AMSU atmospheric temperature climate data record. Atmospheric and Environmental Remote Sensing Data Processing and Utilization V: Readiness for GEOSS III, C.-Z. Zou and W. Wang, Eds., International Society for Optical Engineering (SPIE Proceedings, Vol. 7456), 74560G, doi:10.1117/12.824459.
Zou, C.-Z., and Wang W. , 2010: Stability of the MSU-derived atmospheric temperature trend. J. Atmos. Oceanic Technol., 27, 1960–1971, doi:10.1175/2009JTECHA1333.1.
Zou, C.-Z., and Wang W. , 2011: Intersatellite calibration of AMSU-A observations for weather and climate applications. J. Geophys. Res., 116, D23113, doi:10.1029/2011JD016205.
Zou, C.-Z., Goldberg M. D. , Cheng Z. , Grody N. C. , Sullivan J. T. , Cao C. , and Tarpley D. , 2006: Recalibration of microwave sounding unit for climate studies using simultaneous nadir overpasses. J. Geophys. Res., 111, D19114, doi:10.1029/2005JD006798.
Zou, C.-Z., Gao M. , and Goldberg M. D. , 2009: Error structure and atmospheric temperature trends in observations from the Microwave Sounding Unit. J. Climate, 22, 1661–1681, doi:10.1175/2008JCLI2233.1.
