Atmospheric Infrared Sounder (AIRS) sounding evaluation and analysis of the pre-convective environment


  • Danelle Botes,

    1. Atmospheric Science Department, University of Alabama in Huntsville, Huntsville, Alabama, USA
    Search for more papers by this author
  • John R. Mecikalski,

    Corresponding author
    1. Atmospheric Science Department, University of Alabama in Huntsville, Huntsville, Alabama, USA
      Corresponding author: J. R. Mecikalski, Atmospheric Science Department, University of Alabama in Huntsville, 320 Sparkman Dr., Huntsville, AL 35805-1912, USA. (
    Search for more papers by this author
  • Gary J. Jedlovec

    1. Global Hydrology and Climate Center, NASA Marshal Space Flight Center, Huntsville, Alabama, USA
    Search for more papers by this author

Corresponding author: J. R. Mecikalski, Atmospheric Science Department, University of Alabama in Huntsville, 320 Sparkman Dr., Huntsville, AL 35805-1912, USA. (


[1] The Atmospheric Infrared Sounder (AIRS) is a hyperspectral instrument onboard the National Aeronautics and Space Administration's (NASA) Earth Observing System (EOS) Aqua satellite. This study investigates the performance of AIRS soundings in characterizing the stability in the pre-convective environment of the southeastern United States. AIRS soundings are collocated with radiosonde observations within ±1 degree and 2 h of the Aqua overpass. For each case, the AIRS sounding with maximum PBest quality indicator (signifying the pressure level above which the sounding is of best quality) is chosen for analysis. Rapid Update Cycle soundings from 1800 UTC analyses are used to evaluate the results from AIRS. Precipitable water and stability indices including convective available potential energy, convective inhibition, Lifted Index, K-Index, and Total Totals are derived from all soundings. Results indicate that AIRS underestimates instability due to a dry bias at the surface and roughly 900 hPa. A simple method is presented for reconstructing a RAOB-like inversion (in terms of magnitude and altitude) within AIRS soundings, hence developing more representative RAOB-like soundings that can benefit the operational forecaster.

1. Introduction

[2] Soundings of atmospheric water vapor (WV) and temperature (T) retrieved from sensors onboard meteorological satellites have been providing valuable information for nearly three decades. Data void and inaccessible regions on Earth are provided coverage from satellites, subsequently improving numerical weather prediction (NWP) forecasts [Ohring, 1979; English et al., 2000], enhancing our understanding of atmospheric WV, stability and storm structures, and assisting in diagnosing regional climate. Another main benefit for using satellite remote sensing in meteorology is high spatial resolution (to ∼10 km), and sampling times (reaching ∼5 min). (Note that sun-synchronous satellites like AIRS have 12-hour repeat cycles for the latitudes considered in this study.) The vertical resolution of satellite-based soundings is gradually being improved as the amount of spectral information [or infrared (IR) channels] increases on a given sensor that measures atmospheric radiation, and as instrument quality and retrieval assumptions improve. Current hyperspectral sensors, like the Atmospheric Infrared Sounder (AIRS) onboard the National Aeronautics and Space Administration's (NASA) Earth Observing System (EOS) Aqua satellite, make use of thousands of IR channels that each represent a unique wavelength of radiation, in order to enhance the accuracy and vertical resolution of T and WV retrievals.

[3] The AIRS data set spans a period over nine years during which multiple versions of the retrieval algorithm have been released. Many validation studies have been performed to improve retrieval accuracy and cloud-clearing methods (to eliminate the effects of clouds in the IR field of view, FOV) so to enable high-quality retrievals in up to 90% cloudiness [Fetzer et al., 2003; Susskind et al., 2003]. The AIRS Science Team aims to provide retrieved fields that are accurate within 1 K in a 1 km layer for T, and 10% in a 2 km layer for relative humidity (RH) [Aumann et al., 2003; Divakarla et al., 2006; Tobin et al., 2006a]. These accuracies are not always achieved, and therefore appropriate quality indicators and error corrections need to be applied to AIRS fields prior to use [Susskind, 2006].

[4] Examining soundings of atmospheric T and RH is a valuable part of the nowcasting process for convective storms [Gates, 1961; Stull, 1991]. Sounding-derived indices that diagnose static stability have proven to be highly valuable when forecasting the occurrence of convective weather [Rasmussen, 2003]. These stability indices are routinely derived from radiosonde observations (RAOBs) and NWP model simulations [Ducrocq et al., 1998], however the literature is currently lacking in research on the use and evaluation of stability indices derived from AIRS soundings. This study aims to investigate the usefulness of AIRS T and RH information in assessing atmospheric stability. The primary science questions guiding this study are: (1) Do AIRS-derived stability indices properly characterize the environment ∼1–6 h in advance of thunderstorm formation? (2) How well do AIRS soundings capture low-level inversions, which provide critical nowcasting information on the initiation and intensity of the convective storms that form on a given day? (3) Which quality control parameters can be used to increase the usefulness of AIRS data?

[5] This paper is organized as follows: Section 2 provides a background, while section 3 overviews the data and methods used herein, as well as a description of the sounding data sets analyzed. Section 4 presents the main results, including a case study example, while section 5 summarizes the errors and uncertainties of the analysis, and highlights the main conclusions.

2. Background

2.1. Atmospheric Stability Estimation

[6] Atmospheric instability, coupled with the presence of a lifting mechanism and sufficient low-level WV, are ingredients required for the formation of thunderstorms [Johns and Doswell, 1992; Doswell et al., 1996]. In order to forecast the formation of cumulus clouds and deep convection in the pre-convective environment, it is necessary to accurately evaluate atmospheric stability throughout various layers. The turbulent boundary layer is often capped by a statically stable layer, and this “capping” inversion occurs naturally when warm and dry mid-tropospheric air is transported atop a more moist surface layer (e.g., the elevated mixed layer [Carlson et al., 1983]). This often occurs over the United States (U.S.) Great Plains and Midwest when moist Gulf of Mexico air is met by dry mid-level air from desert-like regions to the west.

[7] Several indices are commonly used to diagnose pre-convective atmospheric (in)stability that require T, WV and wind data from real-time or forecast soundings. These include convective available potential energy (CAPE), convective inhibition (CIN), the Lifted Index (LI) [Galway, 1956], the K-Index (KI) [George, 1960], Total Totals (TT) [Miller, 1967], and the Severe Weather Threat Index (SWEAT) [Peppier, 1988]. These indices have been in use for years, and are formulated to supply forecasters with a single measure for evaluating the potential for severe weather such as hail, tornadoes or damaging winds [Schaefer, 1986; Rasmussen and Blanchard, 1998; Rasmussen, 2003; Craven and Brooks, 2004; Wagner et al., 2008], and aid forecasters in developing situational awareness of pending severe storm events with lead times of >6 h [Petersen and Homan, 1989; Menzel et al., 1998; Feltz et al., 2003; Koenig and De Coning, 2009].

2.2. Satellite Soundings of the Atmosphere

[8] The usefulness of measuring atmospheric T and WV in order to diagnose the state of the atmosphere lead to the development of satellite-based sounding instruments over 40 years ago [Hilleary et al., 1966; James, 1967; Yates, 1970; Chesters et al., 1986], beginning with the U.S. Satellite Infrared Spectrometer (SIRS) and Infrared Interferometer Spectrometer (IRIS) onboard Nimbus 3, launched in 1969 [Wark and Hilleary, 1969; Hayden, 1971; Wick, 1971; Smith, 1972]. SIRS and IRIS were capable of remotely sensing atmospheric T structures by using statistical relationships and regression methods [Kaplan et al., 1977; Westwater and Strand, 1968; Smith et al., 1970; Crosby et al., 1973; Fritz, 1977]. More recent advances in processing satellite data for sounding production include hybrid methods that incorporate physical and statistical processes [Smith, 1970; Hillger and Vonder Haar, 1977, 1981].

[9] Satellite sensors can have broad viewing swaths stretching over 1000s of kilometers, with instantaneous FOVs (IFOVs) or “pixels” up to a few kilometers, thus having spatial resolutions unavailable on ground-based sounding instruments. Depending on the satellite orbit, i.e., polar-orbiting or geostationary, temporal resolutions range from twice-daily overpasses to ∼5 min viewing of the same scene. The vertical resolution of satellite sensors is dependent on the number of IR channels, spectral resolution, and the IR channels used within a retrieval algorithm, as well as scene T lapse rate, surface-atmosphere thermal contrast, instrument noise, and retrieval algorithm constraints, among other things. Present-day hyperspectral sensors operate with thousands of IR channels to sharpen weighting functions per channel. Even though progress has been made using hyperspectral sensors, the vertical resolution and accuracy obtainable from satellite sensors still cannot match those from direct RAOB measurements. Since profiles from satellite soundings appear smoothed out, therefore not accurately capturing low-level features like inversions, this introduces errors when these data are used to diagnose atmospheric structure.

2.3. Previous Inter-comparison Studies

[10] RAOBs are the most conventional source of truth data for satellite retrievals, although aircraft measurements can be used, as often collected through special field campaigns [Schwartz and Benjamin, 1995; Frehlich and Sharman, 2010]. Some comparison studies aim at validation of indirect measurements using RAOB data [Xuebao et al., 2005; Divakarla et al., 2006; Tobin et al., 2006a] or aircraft measurements [Gettelman et al., 2004; Tobin et al., 2006b], while others focus on mesoscale analyses in the pre-convective environment by using satellite-retrieved data. For the latter, static stability fields are usually derived from satellite soundings in cloud-free areas and then compared with similar RAOB analyses [Fetzer et al., 2006; Martinez et al., 2007], or level-dependent thermodynamic fields like T or dewpoint temperature (Td) are created by interpolation of several satellite soundings in advance of being compared to RAOB soundings [Hillger and Vonder Haar, 1981; Jedlovec, 1985; Mostek et al., 1986]. Studies show that satellite measurements are in fair agreement with RAOB data: T root mean square (RMS) differences are ≤2 K and RH differences are 10–20%. The above studies note that such results are to be expected when being mindful of the differences between data sets, including specific identification of significant T and moisture changes in RAOBs, and directly observed (RAOB) versus retrieved (e.g., AIRS) fields. Many previous inter-comparison studies also found that WV biases were present in certain atmosphere layers, with only slight retrieval degradation in surface layers.

[11] AIRS product validation studies have been performed [Fetzer et al., 2003; Gettelman et al., 2004; Xuebao et al., 2005; Divakarla et al., 2006; Taylor et al., 2008], with results meeting the AIRS design requirements. Bruce et al. [1977] found that RMS differences of ≤1 K need to be expected purely because of horizontal T variations over distances; RMS values may increase to 2 K as the distance between a balloon launch site and a satellite observation increases. Hagan et al. [2004] studied AIRS upper-level WV retrievals between 500 and 100 hPa in comparison to aircraft and RAOB data. The in situ measurements were taken within 2 h and 100 km of the AIRS observations and averaged to correspond to the AIRS layers. Results indicated uncertainty in the satellite retrievals of up to 25% for the collocated soundings, but AIRS captured WV variability near 150 hPa fairly well. Bedka et al. [2010], in a comparison between ground-based microwave (MW) radiometer and AIRS, showed that accuracy of precipitable water (PW) vapor was within 5% at three Atmospheric Radiation Measurement program climate sites for most conditions. Exceptions were: (1) very dry cases (PW < 1 cm) over the U.S. Southern Great Plains land site, both daytime and nighttime, where AIRS was 15–30% too moist, and (2) nighttime observations over land sites for PW > 1 cm, where AIRS was ∼10% too dry. The Bedka et al. [2010] study speculates that moist bias is related to surface emissivity (ε).

[12] Divakarla et al. [2006] compared AIRS, RAOBs, operational retrievals from Advanced TIROS Operational Vertical Sounder (ATOVS), and model forecast soundings over land and sea for cloud-cleared and initially clear sky cases. Data sampled within ∼3 h and 100 km from the RAOB site were retrieved, after which the ATOVS, RAOB and model data were interpolated to AIRS vertical levels. Clear-air AIRS soundings were more accurate and precise over the ocean than over land, and RMS differences between AIRS and RAOBs were close to those expected, i.e., 1 K in 1 km layers for T and better than 15% in 2 km layer for WV. RongHui et al. [2008] compared stability indices (CAPE, CIN, LI, KI and the Showalter Index) in regions of land-falling typhoons using European Center for Medium-Range Weather Forecasts (ECMWF) analyses, ATOVS products, cloud-cleared AIRS physical retrievals, and surface observations. In general, the location of significant stability index values was not in agreement between the AIRS and ECMWF data. The conclusions were that the vertical and horizontal resolutions of the different data sets have an influence on the calculated stability indices (yet interpolation would not improve results). As instability is present in cloudy or partly cloudy regions, cloudy satellite measurements are more meaningful than the clear-sky retrievals, hence cloud-clearing methods are of vital importance.

[13] Wu [2009] compared Level 2 AIRS T and RH soundings to dropsondes in the NASA African Monsoon Multidisciplinary Analyses (NAMMA) campaign to study the Saharan air layer. For collocation, all AIRS soundings within 0.5° of the dropsonde location were averaged together. AIRS data underestimated WV in the 1000–925 hPa surface layer, but positive bias values were present for the rest of the sounding. AIRS T was found to be in good agreement with NAMMA dropsondes as RMS errors ranged from 0.79 to 1.61°C. Fetzer et al. [2004] studied boundary layer T inversions over the subtropical northeastern Pacific Ocean, comparing cloud-free AIRS with ECMWF model soundings. AIRS inversion strength was characterized by the difference between the warmest T in the 925–700 hPa layer and the T at 1000 hPa. Values ≤ 6 K were observed, and all inversion tops were at or below 850 hPa. AIRS near-surface T was in good agreement with ECMWF at vertical scales <1 km, with differences of <2 K. Devasthale et al. [2010] quantified clear-sky T inversions in the Arctic Ocean using AIRS soundings. Inversions were detected by searching for an increase in the T from the surface to 400 hPa, enabling the inclusion of both surface-based and elevated inversions. Mean inversion strengths ranged from 2.5 to 3.9 K during summer, and 7.8–8.9 K during winter. Pavelsky et al. [2011] also studied inversion strength over polar oceans using AIRS; AIRS was found to underestimate inversion strength relative to the RAOB by an average of 2.05 K, which was due to a height-dependent bias in AIRS T retrieval.

3. Data and Methodology

3.1. Sounding Data Sets

3.1.1. AIRS Version 5 Level 2 Support Product (AIRX2SUP)

[14] AIRS is in sun-synchronous polar orbit ∼705 km above Earth, crossing the equator at ∼1:30 P.M. local time each day, on the NASA “Afternoon-Train” or A-Train. The A-Train passes over the continental U.S. (CONUS) in ascending orbit from east to west between 1700 and 2200 UTC. The Aqua satellite (launched May 2002) carries onboard six different instruments to sense atmospheric T and WV, and to examine Earth's radiation budget [e.g., Anderson et al., 2005; Waquet et al., 2009]. AIRS has a cross-track scanning spectrometer with 2378 IR channels between 3.74 and 4.61, 6.20–8.22 and 8.8–15.5 μm [Aumann et al., 2003], which is essential for atmospheric T and RH soundings. AIRS also has four visible (VIS) and near-IR (NIR) channels between 0.40 and 0.94 μm, which are mainly used for the detection of clouds in the IR FOV. The IR resolution obtainable from AIRS is 13.5 km in the horizontal and 1 km in the vertical, and the VIS/NIR spatial resolution is ∼2.3 km [Chahine et al., 2006].

[15] The Advanced Microwave Sounding Unit (AMSU) is comprised of two sub-instruments, AMSU-A and the Humidity Sounder for Brazil (HSB), that assist AIRS in obtaining atmospheric soundings in the presence of clouds. AMSU-A, mainly a T sounder, has 15 channels between 23 and 89 GHz that covers the strong O2 absorption band at 60 GHz (important for T sounding) and the weak WV absorption line at 23 GHz (for PW measurements [Chahine et al., 2006]). HSB is a humidity sounder with 5 channels between 89 and 183.3 GHz, that covers the strong WV band at 183 GHz. The AIRS/AMSU IFOVs are optimally aligned at the surface since the larger AMSU footprint covers a set of 9 AIRS footprints with a horizontal nadir resolution of 45 km, used to create a single cloud-cleared IR IFOV (see Figure 1) [Aumann et al., 2003]. The AIRS/AMSU sounding system (hereafter referred to as just AIRS) was designed to target issues such as cloud contamination in the IR FOV as the percentage of actual clear FOVs is small (10%) [Huang and Smith, 2004; Carrier et al., 2007]. MW sounders are ideal for sensing atmospheric T in and below clouds, and therefore AMSU data play an important role in the AIRS retrieval process. The cloud-clearing, as part of the retrieval process, is generated at an AMSU footprint center that overlaps 9 adjacent AIRS footprints [Susskind et al., 2003]. Each AIRS/AMSU co-registered footprint represents a single sounding, with ∼300 000 soundings retrieved by AIRS daily. AIRS is subsequently the first hyperspectral radiometer that produces T and RH soundings with near-radiosonde accuracy in partly cloudy conditions, by taking advantage of measured radiance at wavelengths between strong absorption lines (thereby sharpening the weighting functions).

Figure 1.

AIRS scanning configuration indicating pixel sizes and alignment of AIRS and AMSU instantaneous field of views [Aumann et al., 2003].

[16] The AIRS Level 2 product consists of T and WV soundings, among other parameters, retrieved by the AIRS system. The retrieval algorithm includes both physical and statistical methods and makes use of calibrated radiance measurements taken from the AIRS and AMSU instruments [Goldberg et al., 2003; Susskind et al., 2003]. The AIRX2SUP soundings consist of 100 fixed levels between 1013.9 and 0.016 hPa, which are the same pressure (p) levels used by the radiative transfer model. AIRX2SUP fields used here are: (1) geo-location fields, (2) standard p levels (hPa), (3) a first guess surface p, (4) retrieved air T (K) at each corresponding standard p level, (5) retrieved column WV density for each layer above the indicated p, used to calculate mixing ratio (q) and Td, (6) total PW vapor (kg m−2), and (7) mean topography (m) used to calculate p heights. Per-level T were used to construct T profiles, however q and Td are layer quantities, and thus are layer-means bounded at the bottom by the p that they are reported at. Only levels 1013.94–103.02 hPa (54 levels) were used. For each sounding, the first guess surface p, as derived from the Global Forecast System model, and data from a digital elevation model, were compared to the standard p levels in order to remove levels below topography.

[17] The AIRX2SUP product also contains quality indicators for all retrieved parameters, identifying the quantities that have passed through the entire retrieval system and meet the accuracy requirements set up by the AIRS Science Team. PBest is the p level above that the T and WV sounding are of best quality [Susskind, 2006]. This level was determined by analyzing the differences between error characteristics for each AIRS sounding and large-scale truth data, and then comparing these differences to a predetermined threshold value to test for retrieval failure. If the T value exceeded that threshold at three consecutive levels, the uppermost of the three levels became the level of PBest. When successful retrievals are made in cloudy conditions, it is believed that PBest can be an indicator of the level of the cloud tops. PBest has been used for AIRS data assimilation activities [Jedlovec et al., 2006; Zavodsky et al., 2007], and is employed here to identify the highest quality soundings. Data from the PBest level were used to compute the severe weather indices requiring surface-based information (T, Td) in this study.

3.1.2. Radiosonde Observations

[18] Archived RAOBs were obtained from the Earth System Research Laboratory Radiosonde Database, and have undergone gross error and hydrostatic consistency checks [Schwartz and Govett, 1992]. Only soundings from 1800 to 1900 UTC balloon launches were selected, nearest the time of Aqua overpasses in the Southeastern U.S. region, based on the daytime severe weather events analyzed for this study. A ∼30 min offset between RAOB launches and AIRS soundings should therefore occur for the cases analyzed. Significant and mandatory levels, ranging from the surface to 100 hPa for the selected stations, were used to calculate RH and q at each level.

3.1.3. Rapid Update Cycle (RUC) Model Data

[19] RUC is maintained by the National Center for Environmental Prediction [Benjamin et al., 2004]. Grid point soundings were used as an additional comparison data set to AIRS and RAOB soundings. For this study, 1800 UTC RUC analysis soundings were obtained from the National Operational Model Archive and Distribution System. The isobaric gridded data are 20 km in horizontal resolution on 37 levels at 25 hPa vertical resolution, from 1000 to 100 hPa. T, RH, CAPE, CIN, LI, and PW were retrieved from model output, and used to calculate q and Td at all grid points. RUC RH was defined with respect to saturation over water in the RUC isobaric fields and in the surface RH field, with the conversion to q and Td done following Bolton [1980]. CAPE and CIN are routinely calculated in the RUC by using the parcel lifted from the most buoyant level in the lowest 255 hPa, therefore these quantities can be regarded as “most unstable” CIN and CAPE. The RUC LI is routinely calculated using a surface-based parcel. However, in an effort to assure that AIRS CAPE, CIN and LI are computed similarly to that of RUC, RUC T and Td fields were found for the AIRS PBest level to determine parcel characteristics. PW is calculated by the integration of layer specific humidity and mass (surface to model top, 50 hPa).

3.2. Case Event Identification and Selection

[20] Figure 2 shows the study domain, along with the location of 14 upper air stations that were chosen because 1800–1900 UTC balloon launches were available (March–August, 2006–2010). Only warm season months were considered since low-level heating drives most thunderstorms during these times. As Aqua passes over CONUS in an easterly direction (over the study domain) from ∼1800–1930 UTC this provides soundings of environments ahead of mid- to late-afternoon convective storms, with the focus being on the Southeastern U.S. (versus the Great Plains) where overpass-to-RAOB launch time differences were the smallest between 1800 and 1900 UTC (as compared to regions further west). However, the selection of case events becomes severely limited by the availability of RAOBs simply because relatively few RAOBs are available at asynoptic times.

Figure 2.

The domain chosen for this study, which covers the southeast United States with coordinates −95° W, 25° N; −75° W, 40° N, as indicated by the black box. In the image, the locations of 14 upper air stations are indicated by the black stars and station identifiers. “Redst” refers to the Redstone Arsenal in Huntsville, Alabama.

[21] The initial RAOB data set consisted of over 500 soundings, but several constraints further reduced the number of usable cases: (1) The Aqua swath width of 1650 km caused data gaps, and continuous coverage between adjacent overpasses were not available. (2) The spatial resolution changes from swath-edge to swath-center, therefore AIRS soundings are spaced further apart when retrieved at the swath-edge. And, (3) if the upper air station happened to be located on the swath-edge, fewer AIRS soundings were available, especially in party cloudy conditions. It was also required that convective events occur within ∼111 km (1°) of an upper air station, within 6 h of an Aqua overpass, and before 2359 UTC. An additional constraining factor was cloud cover amount in the pre-convective environment at the Aqua overpass time. Only a small fraction of all global AIRS measurements are made in truly cloud-free environments, and the location of cloud-free FOVs can vary substantially. Although operational AIRS retrievals are rejected only when cloud fractions exceed 90%, the presence of clouds can reduce the value of PBest, thereby limiting the sounding to that part marked as having the best quality. Ideally, the level of PBest would be as close to the surface as possible since the AIRS soundings are being used to calculate stability indices that rely heavily on low-level data. Table 1 lists the collocated AIRS and RAOB data for all cases. All RUC soundings are valid at 1800 UTC. The final data set consists of 76 AIRS, RUC and RAOB sounding sets collected within <2 h of each other.

Table 1. List of Case Studies With Upper-Air Stations, Corresponding RAOB Launch Time, AIRS Sampling Time and AIRS Surface Class
DateStationStation IdentifierLaunch Time (UTC)AIRS Overpass (UTC)Surface Class
11 March 2006Springfield7244018001905Land
12 March 2006Springfield7244018001947Land
13 March 2006Jackson7223518001847Land
13 March 2006Birmingham7223018001847Land
30 March 2006Springfield7244018001935Land
20 April 2006Peachtree City7221518001811Land
21 April 2006Slidell7223318001853Land
1 May 2006Springfield7244018001935Land
3 May 2006Springfield7244018001923Land
9 May 2006Jackson7223518001841Land
10 May 2006Jackson7223518001929Land
10 May 2006Lake Charles7224018001929Land
20 May 2006Peachtree City7221518001823Land
24 May 2006Springfield7244018001941Land
25 May 2006Springfield7244018001847Land
19 July 2006Greensboro7231718001853Land
28 August 2006Shreveport7224818001941Land
29 August 2006Birmingham7223018001847Land
29 August 2006Jackson7223518001847Land
3 April 2007Little Rock7234018001841Land
3 April 2007Nashville7232718001841Land
3 April 2007Springfield7244018001841Land
25 April 2007Springfield7244018001941Land
6 May 2007Springfield7244018001923Land
12 June 2007Charleston7220818001805Land
19 July 2007Springfield7244018001959Land
16 August 2007Springfield7244018001847Land
24 August 2007Springfield7244018001935Land
3 March 2008Jackson7223518001935Land
3 March 2008Slidell7223318001935Coastline
4 March 2008Peachtree City7221518001841Land
7 March 2008Tampa7221018001911Ocean
15 March 2008Peachtree City7221518001823Land
15 March 2008Charleston7220818001823Ocean
18 March 2008Shreveport7224818001853Land
30 March 2008Lake Charles7224018001917Land
4 April 2008Lake Charles7224018001935Coastline
22 April 2008Little Rock7234018001923Land
25 April 2008Shreveport7224818001953Land
8 May 2008Peachtree City7221518001923Land
10 May 2008Peachtree City7221518001911Land
10 May 2008Little Rock7234018001911Land
11 May 2008Charleston7220818001817Ocean
22 May 2008Jackson7223519001935Land
23 June 2008Charleston7220818001759Ocean
18 August 2008Slidell7223318001847Coastline
18 August 2008Birmingham7223018001847Land
19 August 2008Jackson7223518001929Land
30 August 2008Tallahassee7221418001911Coastline
26 March 2009Lake Charles7224018001917Ocean
26 March 2009Shreveport7224819001917Land
27 March 2009Lake Charles7224018001959Coastline
2 April 2009Nashville7232718001923Land
2 April 2009Little Rock7234018001923Land
10 April 2009Peachtree City7221518001835Land
10 April 2009Birmingham7223018001835Land
10 April 2009Redstone ArsenalRedst17001835Land
13 April 2009Jackson7223518001905Land
29 April 2009Shreveport7224818001905Land
9 June 2009Springfield7244019001859Land
18 June 2009Peachtree City7221518001853Land
21 July 2009Shreveport7224819001935Land
27 August 2009Jackson7223518001953Land
2 April 2010Little Rock7234018001847Land
30 April 2010Little Rock7234018001911Land
1 May 2010Little Rock7234018001953Land
2 May 2010Birmingham7223018001859Land
20 May 2010Jackson7223518001847Land
19 June 2010Springfield7244018001859Land
28 June 2010Jackson7223518001853Land
29 June 2010Shreveport7224818001935Land
29 June 2010Jackson7223518001935Land
11 July 2010Springfield7244018001959Land
21 July 2010Springfield7244018001859Land
23 July 2010Tallahassee7221418001847Land
5 August 2010Redstone ArsenalRedst18001817Land

3.3. Sounding Comparisons

[22] To facilitate the comparison between data sets AIRS soundings and RUC analyses were spatially collocated with each RAOB sounding. A 2 × 2° “window” was centered over each upper air station to identify the AIRS and RUC soundings whose centers fell within roughly 1° (∼111 km) from the RAOB launch site. This is consistent with past studies [Kitchen, 1989; Divakarla et al., 2006; Sun et al., 2010]. As an example, Figure 3 shows the locations of AIRS soundings from the 1905 UTC Aqua overpass on 11 March 2006. Only those soundings with PBest values > 700 hPa are shown. A black symbol denotes the Springfield, Missouri upper air station (center of 2 × 2° window). The horizontal resolution of each data set allows up to 19 AIRS soundings (close to nadir), and more than 100 RUC soundings, to be sampled over the window. There will be fewer AIRS soundings in the window when the upper air station is located close to a swath-edge. Thus, depending on the cloud fraction and location of the station with regards to the satellite swath, there may be far less than 19 AIRS soundings within the window, and hence the amount of usable AIRS soundings varied for each case study.

Figure 3.

The PBest quality indicator values for AIRS soundings sampled over the Springfield, Missouri upper air station (indicated by the black cross inside the box) on 11 March 2006 at 1905 UTC. The AIRS sounding with the maximum PBest value (PBestmax) within the 2 × 2° box centered on the upper air station, is chosen for analysis.

[23] Initially, the average of all AIRS and RUC soundings was calculated within the window and compared to the RAOBs. However, this permitted the inclusion of bad quality AIRS data, especially in partly cloudy regions. In order to optimally collocate the AIRS and RAOB soundings, the PBest variation within the window was analyzed, and Figure 4 shows this for all 76 case studies (maximum, minimum, mean). The PBest value for some AIRS soundings was zero, indicating that no part of the sounding is useful (only soundings with PBest > 0 were included in the display window). Also shown in Figure 4 are soundings retrieved over coastline versus ocean.

Figure 4.

The variation of the level-dependent PBest quality indicators for each of the 76 cases used in this study, along with the maximum, minimum and mean PBest values. The PBest values for AIRS soundings taken over oceans and those over coastline are indicated by open symbols. Standard error bars are shown in red.

[24] Figure 4 shows a fairly large variation of PBest values, with means of 900–500 hPa. For many cases, the variation of PBest within the window is small, with mean values > 820 hPa. However, other cases show a significantly larger spread of PBest values, therefore exemplifying larger error bars and low mean values. The cases with large spread suggest cloud contamination, since the window contains both good and very bad quality soundings. As a result, the one associated with the maximum PBest (PBestmax) is chosen to reduce the impact of cloud contamination. For some cases, more than one sounding had the same PBestmax, and then the sounding in closest proximity to the RAOB was selected. Some soundings showed anomalously large (>10°C) differences in surface T compared to other soundings in the same window, and these were removed despite having high PBestmax values. (It is not uncommon for erroneous retrievals to make it through the system without bad quality flags associated with them.) For the RUC soundings, the one closest to the upper air station was selected for comparison.

[25] After collocating the AIRS and RUC soundings with the RAOB soundings, all sounding sets were interpolated to common p levels for the calculation of statistical accuracy measures. Since the thermodynamic state of the lower troposphere plays a significant role in forecasting or defining convective weather, all AIRS, RUC and RAOB levels below 500 hPa were interpolated to 5 hPa increments, with levels above 500 hPa interpolated to 10 hPa increments up to 110 hPa. The 5 hPa increment allowed for comparison to more detailed boundary layer structure and significant level data from the RAOB observations despite being at a resolution above what AIRS is capable of resolving (∼1 to 3 km). Level-by-level calculations then compared T, Td and q from AIRS, RUC and RAOB, and the RMS errors were calculated as the square root of the mean squared error (MSE) [Wilks, 1995, pp. 278–280]. The RMS error will be a positive number, where zero indicates a perfect forecast. The mean error (ME), most commonly referred to as the bias, was also calculated for each level of data [Wilks, 1995, pp. 279–280]. A positive (negative) bias indicates that the forecast variable, in this case the AIRS variable, is on average too high (low). For the first 13 levels of the interpolated p array, there were <76 soundings used to calculate RMS errors and bias. However, starting at 930 hPa upward there were 76 soundings available for the calculation of statistics at each level.

3.4. Inversion Detection

[26] RAOB soundings from each case, in their original format and without any interpolated levels, were analyzed in order to detect positive T changes between 1.5 K and 6 K with a thickness less than 1 km (i.e., surface-based superadiabatic layers were eliminated). The inversion height in the RAOB sounding was defined as the first p level above the surface where T increased toward the next level (below 500 hPa). These criteria were chosen in order to permit analysis of capping inversion strength. The same criteria could not be used to detect inversions in AIRS soundings since no positive T gradients (from the surface upwards) were found in any of the 76 cases. As an alternative, AIRS soundings were analyzed to locate a decrease in the environmental lapse rate, indicating the presence of a stable layer. The center of the p layer associated with the minimum AIRS lapse rate below 500 hPa was taken as the inversion height. Both RAOB and AIRS inversions were then used to calculate the bias in AIRS inversion heights.

4. Results

4.1. Sounding Error and Bias Characteristics

[27] Since all 76 sets of AIRS, RUC and RAOB soundings do not start at the same p levels, the amount of soundings used to calculate RMS differences and biases at each level are not equal. At 990 and 995 hPa (the highest surface p in the interpolated data set) only 3 soundings from each data set are available, while between 42 and 75 soundings are available for calculations from 985 to 935 hPa, respectively. (Despite there only being 3 soundings in the lowest layer, these results are shown but not discussed, and no statistical significance is given to them (since the sample size is so small).) At 930 hPa, all 76 soundings were used for statistical calculations at each level. RMS differences and biases were calculated for T, Td and q.

[28] As AIRS usually cannot resolve individual RAOB/model sounding levels, the RMS differences and biases from several sounding levels were summed to create a layer average. Table 2 shows these layer averages AIRS T, Td and q. The column on the far right indicates how many sounding levels were summed. The data in Table 2 should be compared with Figures 5a–5d, 6a and 6b. Figure 5a shows RMS differences in AIRS T (solid lines) that are within 2 K for most of the RMS profile and within 2.5 K below 900 hPa. This agrees with the magnitude of observation errors in RAOB data of 1–2 K from Connell and Miller [1995], yet not those of ≤0.4 K as recently reported by Steinbrecht et al. [2008]. The largest differences can be seen below 900 (to 990) hPa where there is a sudden decrease in error at the surface, and also above 200 hPa. AIRS measurements are slightly too warm at 900 hPa. T inversions are often found in the abovementioned layers, while low-level clouds are frequently seen around 900 hPa, indicating that this is one source of error for AIRS measurements at those levels. (AIRS T soundings also suffer from accuracy errors associated with poor knowledge of surface ε, making it difficult to quantify the radiance contributions from the surface.)

Table 2. Layer Averages of RMS Differences and Bias Between AIRS and RAOB Soundings for All Case Studiesa
Layer (hPa)T RMS (K)T Bias (K)Samples
  • a

    The sample size used to calculate the average for each layer is shown in the right-most column. Here, temperature (T), dewpoint temperature (Td) and mixing ratio (q) are shown. See text for discussion.

Layer (hPa)Td RMS (K)Td Bias (K)Samples
Layer (hPa)q RMS (g kg−1)q Bias (g kg−1)Samples
Figure 5.

Profiles of: (a) temperature and (b) dewpoint temperature RMS differences (K), and (c) temperature and (d) dewpoint temperature bias (K) from 76 AIRS (solid lines) and RUC (dashed lines) soundings.

Figure 6.

Profiles of: (a) mixing ratio RMS differences (g kg−1) and (b) bias (g kg−1) from 76 sets of AIRS (solid lines) and RUC (dashed lines) soundings.

[29] Larger errors exist in the AIRS Td (solid lines in Figure 5b), especially near 400 hPa where RMS differences peak at 11 K. Below 900 hPa, RMS differences are 2–5 K, and there is a rapid increase in error from the surface through a shallow layer up to 850 hPa (not seen in Table 2 since RMS differences are averaged over the surface–900 hPa layer). There is a pronounced increase in q RMS error up to 850 hPa, where the highest AIRS-RAOB differences were 3 g kg−1 (Figure 6a). This level corresponds well with the average PBestmax level from all cases used, and it is often the height of low-level capping inversions. Above 850 hPa, RMS differences decrease asymptotically to zero at 110 hPa, which is because the absolute magnitude of q aloft is extremely small. The relatively small errors in low-level WV displayed in the AIRS data will slightly affect the accuracy of the derived stability indices.

[30] Bias profiles for AIRS T, Td and q are shown in Figures 5c, 5d and Figure 6b. Figure 5c shows a bias near −2 K in AIRS T directly above the surface. This corresponds well to the level of a rapid increase in AIRS T RMS. A smaller warm bias is visible below 850 hPa, and from 750 to 300 hPa. In between these layers there is a very small cold bias in AIRS T. AIRS measurements are biased dry below 750 hPa (solid line in Figure 5d), up to 3 K in Td, interpreted as a dry bias since it is associated with an underestimation in WV; the same behavior can be seen in the bias profile of AIRS q (solid line in Figure 6b). Above 700 hPa, a moist bias ≤4 K is shown in the profile. In Figure 6b, a similar dry bias (to >2.1 g kg−1) can be seen in q below 700 hPa, as well as a moist bias above. The strong 850 hPa dry bias does not seem to be related to a T bias at the same level.

[31] Figure 5a also shows RMS differences for RUC T (dashed lines). RMS differences below 900 hPa are only slightly larger than the accuracy of RAOB data that is ∼1 K [Connell and Miller, 1995]. However, larger errors can be seen from 900 hPa upwards. RMS differences >5 K are seen above 200 hPa, suggesting that the RUC model does not simulate upper-tropospheric T very well. The RMS differences for RUC Td and q shown in Figures 5b and 6a are generally larger than for AIRS, except below 900 hPa where Td RMS differences decrease to <1 K and q to <1 g kg−1. Larger Td errors up to 15 K are seen around 400 hPa. There is a rapid increase in WV error from the surface through a shallow layer just above it, which is evident in the Td and q RMS profiles from RUC. Thompson et al. [2003], evaluating environments of supercell thunderstorms using RUC proximity soundings, showed that Td errors are largest between 400 and 100 hPa, corresponding well to the results here.

[32] The RUC T bias profile in Figure 5c is also reversed from the AIRS bias profile, except at the surface (and directly above it) where a cold T bias of 2 K is found compared to the AIRS soundings. A cold bias of <1 K is seen up to 850 hPa, and somewhat larger cold bias is seen from 700 to 300 hPa. Figure 5d shows that the RUC model overestimated Td below 900 hPa by 1.5 K. A higher dry bias is seen from 900 to 600 hPa, possibly due to the inability of the RUC model's convective parameterization scheme to accurately simulate the presence of WV at this level [Buizza et al., 1999]. Similar to AIRS, the RUC seems to have overestimated mid-level WV above 600 hPa given a moist bias up to 6 K (Figure 5d). Figure 6b shows a similar moist bias (>1 g kg−1) below 900 hPa, but a much smaller surface moist bias (Figure 5d). A dry bias (<1 g kg−1) is seen near 700 hPa, corresponding to a Td bias. These results are consistent with Thompson et al. [2003], indicating RUC T and q have slight cold and wet biases below 850–900 hPa.

4.2. Error and Bias in Atmospheric Stability

[33] The 76 storm cases used were divided into six different categories based on the RAOB CAPE, which is referred to as the “true CAPE.” By doing this, the performance of AIRS stability indices as a function of the CAPE environment in which the thunderstorms occurred can be assessed. It is realized that the small number of cases (76) does not justify a rigorous statistical analysis, and therefore having six divisions of CAPE is toward showing a continuum of behavior as the comparisons are made (between RUC, AIRS and RAOB). In a simple experiment, the results were reduced to three CAPE categories, yet as the gradation of results from low to high “true CAPE” over six CAPE categories is relatively continuous, the choice was made to retain six CAPE divisions. The true CAPE (J kg−1) categories used are (the bracketed number indicates the amount of cases within a category): (1) <600 [10]; (2) 600–1200 [11]; (3) 1200–1800 [14]; (4) 1800–2400 [16]; (5) 2400–3000 [11]; (6) >3000 [14]. Stability indices and PW were calculated for all soundings, and then the RMS differences and biases were calculated for all events within each category.

[34] Figures 7a7f shows bar histograms of RMS differences for the stability indices and PW, for each CAPE category (CAPE categories are labeled from lowest to highest). Figure 7a shows that errors in AIRS CAPE increase as the true CAPE increases. Even in the lower CAPE categories (≤1200 J kg−1), RMS differences are 500–1000 J kg−1. These differences are much lower for the RUC soundings, although 500 J kg−1 is also exceeded for all categories, possibly due to AIRS having larger T and Td RMS at the surface compared to RUC. For CIN derived from the RUC soundings in these first three categories (Figure 7b), RMS differences are larger than for AIRS. There does not seem to be any systematic relation between the error in CIN between AIRS and RUC soundings, and the value of the true CAPE. RUC CIN and CAPE appear to be most accurate for category 3, true CAPE from 1200 to 1800 J kg−1. The PW uncertainty estimates shown in Figure 7c indicate RMS differences in AIRS of 4–7 mm, similar to those from RUC. There is no distinct relation between the PW error and the true CAPE value.

Figure 7.

RMS differences for stability indices (a) true CAPE, (b) CIN, (c) precipitable water, (d) LI, (e) KI, and (f) Total Totals (TT) derived from 76 AIRS (light gray) and RUC (dark gray) soundings. Results are categorized into six categories according to the CAPE values derived from RAOB soundings. See text for CAPE categories.

[35] Figures 7d7f show RMS differences for the LI, KI and TT derived from AIRS and RUC soundings. Errors up to 6 K are present in the LI derived from AIRS, about twice that compared to RUC. The LI values range from +10 to −8 K, and therefore the errors shown in Figure 7d are substantial. Since the LI is dependent on a surface-based air parcel, or some parcel from the lowest 300 hPa, errors in the surface T and Td will affect the accuracy of the LI. RMS differences for the RUC T and Td are lower than for AIRS, therefore the errors in the LI are reduced as well. For the KI however, the errors in the RUC data are larger. The same is seen for TT, which is closely related to the KI, due to the biases seen in Figures 5c and 5d. There does not seem to be any systematic relation between the magnitude of errors for LI, KI or TT, and true CAPE.

[36] Figures 8a8f show bias values for all the stability indices and PW. The bias calculation for CIN included negative values which refers to AIRS data that were on average too low, i.e., too negative. Figure 8a and 8b also shows that AIRS CAPE and CIN are mostly underestimated, with large negative biases seen. An overestimation in CIN derived from AIRS soundings is shown for the first two categories. The amount of underestimation in AIRS CAPE and CIN increases with true CAPE (due to the underestimation in T and Td from AIRS), which is similar for RUC. The RUC also depicts a mostly negative bias in CAPE and CIN, although the biases are lower than for AIRS. The PW bias (Figure 8c) is reversed in sign for AIRS and RUC. Similar to the moist bias shown for RUC Td and q, a moist bias is also seen for RUC PW. For most of the categories, AIRS underestimates PW (as also seen in Figure 6b for q).

Figure 8.

Bias for stability indices (a) true CAPE, (b) CIN, (c) precipitable water, (d) LI, (e) KI, and (f) Total Totals (TT) as derived from 76 AIRS (light gray) and RUC (dark gray) soundings. Results are categorized into six categories according to the true CAPE values derived from the RAOB soundings. See text for CAPE categories.

[37] Figures 8d8f show bias for LI, KI and TT for AIRS and RUC. For AIRS, positive biases are shown in the lowest true CAPE bin, with negative biases otherwise (again due to the underestimation in T and Td). Note that the bias values for the LI were multiplied by −1 in order to be consistent with the rest of the stability indices, for which the term “underestimation” refers to magnitudes that are on average lower than the RAOB data. A warm bias in AIRS T at 500 hPa (Figure 5c) might also produce an underestimation of AIRS LI. For the RUC LI, a moderate amount of overestimation occurs, most likely due to the positive bias in RUC low-level WV affecting the parcel lifted to 500 hPa in the LI computation. Large negative and positive biases are seen in RUC KI and TT, randomly varying with true CAPE, as expected from Figures 5c and 5d.

[38] Figure 9 shows a scatterplot of RAOB CAPE along with the difference between the AIRS and RAOB CAPEs. The symbols are colored according to PBestmax for the AIRS soundings. The diagonal line that is formed by the dots shows that in most cases, RAOB CAPE is greater than AIRS CAPE. PBestmax values for the 76 soundings were all >827 hPa. Soundings taken over ocean (black open triangle) or coastline (black open circle) are shown for completeness. Assuming the RAOB CAPE is the truth, all symbols in Figure 9 would be located along the black horizontal line if the AIRS and RAOB true CAPE were equal. However, large differences between AIRS and RAOB true CAPE are seen, mainly when the true CAPE is large. This agrees with the RMS differences shown in Figure 7a. The accuracy of the AIRS CAPE does not seem to be related to PBestmax.

Figure 9.

Scatterplot of RAOB CAPE and differences between AIRS and RAOB CAPE for all 76 soundings. Symbols are color-coded according to the PBestmax quality indicator associated with each AIRS sounding. Soundings classified over ocean (open triangle) and coastline (open circle) are indicated as well.

[39] These analyses show that large errors exist in both the AIRS and RUC stability indices and in PW. Relatively small positive biases in most of the stability indices, including PW, are evident in AIRS for true CAPE values < 600 J kg−1 and from 600 to 1200 J kg−1. Therefore, AIRS performed the best in environments of CAPE < 1200 J kg−1. T and Td from AIRS is fairly degraded near the surface, with the sources of degradation (at and directly above the surface) including: (1) surface contamination in the radiance measurements due to heterogeneity of the land surface and variation in spectral ε, (2) uncertainties in the surface p due to uneven topography, (3) contamination by low-level clouds that are warmer than the temperature from radiances below clouds, and (4) amplification of instrumental noise due to cloud-clearing methods.

4.3. Inversion Height Bias

[40] T inversions below 500 hPa were detected in RAOB soundings by locating positive T gradients >1.5 K and <6 K, with a thickness <1 km. In contrast, AIRS soundings were analyzed in order to locate a decrease in the environmental lapse rate, since the criteria for the detection of RAOB inversions could not be applied. Here, the center of the p layer associated with the minimum AIRS lapse rate below 500 hPa was defined as the inversion height. For the RAOB, the inversion height is the first p level where T increased toward the next level. Out of 76 cases, 19 low-level inversions were detected in both of the RAOB and AIRS soundings.

[41] Figure 10 shows a portion of a collocated AIRS and RAOB profile from 3 March 2008, in which an inversion of >3°C can be seen at 843 hPa. The AIRS does not indicate this T increase, but a change in T gradient is visible between the upper and lower boundaries marked as “A” and “B,” respectively. Several AIRS soundings within the 2 × 2° window that were not chosen for analysis, depict a stronger decrease in lapse rate, yet are not necessarily associated with PBestmax within the display window. The PBestmax for AIRS in Figure 10 is 853 hPa. The bias in AIRS inversion height was calculated by taking the difference between the mean for 19 AIRS and RAOB soundings. The mean p level of the inversion height in AIRS is 854 hPa, similar to Fetzer et al. [2004], despite the fact that our analysis is on land soundings only, i.e., all inversion tops were at or below 850 hPa. The average RAOB inversion height is 815 hPa. For all 19 soundings a bias of +39 hPa was found, indicating AIRS inversion heights are on average too low (yet within the expected ability of AIRS to resolve these inversion features). This calculation was repeated using the top of the RAOB inversion as the “inversion height,” with a +56 hPa bias then found in the AIRS inversions, indicating that the lapse rate minimum is even lower in altitude.

Figure 10.

Snippet of the AIRS (solid line) and RAOB (dashed line) temperature profiles from 3 March 2008. The figure indicates the RAOB inversion and the upper and lower boundaries associated with the change in AIRS lapse rate. The horizontal line between “A” and “B” represents the center of the inversion. See text for discussion.

[42] Last, to demonstrate a simple procedure for developing more representative RAOB-like soundings, an effort was made to further quantify AIRS inversion magnitudes. This exercise is toward increasing the operational use of AIRS soundings when characterizing the pre-convective environment. Inflection points similar to A and B in Figure 10 were computed along the T curve for 17 of the 19 AIRS soundings (as two T profiles did not exhibit any clear inflection point). The inflection points A (“Inversion Base (m); see Table 3) and B (“Inversion Top” (m)) represent the bottom and top locations of a change in slope rate in the T profile. The difference in T between inflection points A and B is indicative of the AIRS “Inversion Magnitude” (°C; Table 3), with the inversion p assigned to the mid-way point between A and B (horizontal line in Figure 10; which differs from taking the altitude of the minimum AIRS lapse rate as the “Inversion Height”). Table 3 shows these results, demonstrating this method for reconstructing a RAOB-like inversion (magnitude and altitude) within AIRS soundings.

Table 3. Subjectively Determined Inversion Base, Top and Height for 17 of 19 AIRS Soundings Exhibiting a Defined Decrease in Lapse Rate Consistent With the Presence of a Well-Defined Inversion in RAOB Observationsa
DateInversion Base (m)Inversion Top (m)Inversion Height (m)Inversion Magnitude (°C)
  • a

    Referring to Figure 13, the “Inversion Base” (m) is level “B,” the “Inversion Top” (m) is level “A,” and the “Inversion Height” (m) is here taken as the mid-way point between levels “A” and “B” (which differs from simply taking the altitude of the minimum AIRS lapse rate as the Inversion Height, as discussed in Section 4d). The “Inversion Magnitude (°C) is defined as the temperature difference between the Inversion Top and Inversion Base.

11 March 20061163244018017.1
30 March 20061703195918311.2
20 April 20061755227420153.0
03 May 20061472198417283.5
09 May 2006688250015946.6
10 May 20061454224718515.6
03 April 20071232226617496.2
03 April 2007
03 March 2008944247617106.6
15 March 2008891215615246.1
04 April 20081201197115864.2
10 May 2008
11 May 2008900192114114.0
26 March 20092458117518177.3
27 March 200961913759970.8
13 April 20091424270120637.2
30 April 20101636215718972.5
01 May 2010627190312654.3
02 May 20101472252419983.9

4.4. Case Study: 10 May 2008

[43] On 10 May 2008, severe thunderstorms occurred over Arkansas with multiple reports of hail, damaging winds and tornadoes. At 0600 UTC, a surface trough was situated east of the Rocky Mountains with a synoptic-scale low pressure system over Texas, New Mexico, Colorado, Oklahoma and Kansas, and an east-west orientated cold front stretching from Colorado to California (12 h prior to Figure 11). By 1400 UTC, convective cells had started to develop ahead of the warm front across the border between Oklahoma and Arkansas where Td values were near 15°C. Storms continued to develop over northwestern Arkansas and southwestern Missouri as the system moved toward the northeast. An 1800 UTC RAOB launch was performed at the North Little Rock, Arkansas upper air station, where overcast conditions were reported at that time. At 1800 UTC, scattered convective storms were located over southwestern Missouri and northern Arkansas, while more storms developed over eastern Oklahoma in a region of increasing Td. Aqua passed over this region at 1911 UTC 10 May. Figure 11, a composite Geostationary Operational Environmental Satellite (GOES)–surface–NEXRAD radar map at 1900 UTC, shows locations of storms over Missouri and northern Arkansas. Scattered, new convective development over eastern Oklahoma evolved into a significant storm that passed over North Little Rock at ∼2300 UTC, with a second line of storms passing at 0200 UTC on 11 May.

Figure 11.

CONUS satellite-surface map showing IR GOES satellite data, NEXRAD precipitation radar imagery, surface pressure contours (thin blue lines) and surface observations at 1900 UTC on 10 May 2008. The location of cold fronts (blue), warm fronts (red), occluded fronts (purple) and stationary fronts (blue and red) are indicated as well (taken from

[44] Figure 12 shows a MODerate resolution Imaging Spectroradiometer (MODIS) satellite image overlaid by PBest values > 700 hPa. Soundings with PBest from 750 to 700 hPa are indicated by dark blue symbols (not shown in color bar). The chosen AIRS sounding has a PBestmax value of 853 hPa, and was sampled over an area with broken cloud cover. Figure 13 shows a skew-T diagram of the non-interpolated profiles from the AIRS, RUC and RAOB soundings on 10 May 2008. Except for a slight moistening near 950 hPa, there is very little vertical structure in the AIRS profiles since the strong low-level inversion shown in the RAOB profile goes undetected by AIRS. The AIRS sounding also indicates a warmer surface than is evident from the RAOB sounding. As seen in Figures 11 and 12, a fair amount of cloud cover was present over Arkansas at the time of the Aqua overpass. Therefore, the PBestmax of 853 hPa was likely assigned to this sounding due to the presence of low clouds. The AIRS sounding is 1 h 11 min after the RAOB, when a warm front over southern Arkansas was leading to increased surface T from warm air advection.

Figure 12.

PBest quality indicators associated with individual AIRS soundings sampled over North Little Rock, Arkansas upper air station (black cross) at 1911 UTC on 10 May 2008. PBest values (hPa) are indicated by the color bar and overlay a MODIS satellite image. The dark blue symbols indicate PBest pressures less than 750 hPa.

Figure 13.

Skew-T diagram with non-interpolated AIRS (red), RUC (blue) and RAOB (green) soundings for 10 May 2008. Solid lines denote temperature profiles and dashed lines denote dewpoint temperature profiles. The RAOB was launched from North Little Rock, Arkansas. Sampling times for AIRS, RUC and RAOB are 1911, 1800 and 1800 UTC, respectively.

[45] The 1800 UTC RAOB from North Little Rock did pass through a low-level (960–890 hPa) cloud layer, as verified by an 1800 UTC surface observation of overcast skies. Also in Figure 13 are two instances of rapid and unrealistic cooling in the RAOB Td profile, specifically where superadiabatic lapse rate occur at 890 and 410 hPa (i.e., the wet-bulb effect [Hodge, 1956]). Rapid cooling is also evident in the RUC profile just below 800 hPa, where a layer of very dry air extending from ∼850 hPa upwards. In general, the gross features in the RUC and RAOB Td profiles are in fair agreement, and the RUC T profile is also in agreement with the RAOB, except that the RUC profile is much smoother from 600 hPa upwards and does not exhibit the larger changes visible in the RAOB T profile. The dry adiabatic lapse rate evident in the RAOB T profile between 750 and 600 hPa can also be seen in the RUC sounding.

[46] Stability indices and PW for the 10 May 2008 case for the AIRS, RUC and RAOB soundings are in Table 4, showing large discrepancies. There is little agreement in CAPE between the three data sets, and once again the RUC sounding displays the highest values of CAPE. The RUC surface T is 20°C with a Td of 17°C, showing a cool and moist surface layer that has not been subjected to daytime heating or warm air advection. The RAOB sounding displays a fairly low amount of CAPE, since the most unstable parcel selected for the CAPE calculation has a T = 15°C, located just above the low-level cloud layer. The AIRS sounding displays a warm surface layer with a surface Td of 19°C and CAPE of 1838 J kg−1, that was calculated using the most unstable parcel originating just above the surface with a T = 25°C. CIN values also show little agreement, however the AIRS has the lowest CIN due to low-level instability portrayed by a fairly moist surface and a high T lapse rate. For the RAOB and RUC, CIN values are much higher, indicating a moderate capping inversion.

Table 4. Stability Indices and Precipitable Water (PW) Values for AIRS, RUC and RAOB Soundings for the 10 May 2008 Case Studya
SourceCAPE (J kg−1)CIN (J kg−1)LI (K)KI (K)TT (K)PW (mm)
  • a

    See text for description.


[47] LI values derived from the AIRS and RUC soundings are in poor agreement with those from the RAOB, indicating that no parcel instability exists at low levels. This is mainly due to the presence of low clouds and cool surface T as depicted in the RAOB. In contrast, AIRS shows much warmer T below 925 hPa without the stable layer near 800 hPa, resulting in a large amount of buoyant instability. The RUC LI shows the highest LI for all three data sets, even though it only depicts a surface T of 20°C and a dry and stable layer around 800 hPa.

[48] The RAOB KI and TT are much lower than those from AIRS and RUC, due to the dry/stable layer starting just below 850 hPa, as well as a very high 700 hPa Td depression. The KI and TT derived from AIRS and RUC show better agreement. The AIRS KI is the highest, due to a larger amount of 700 hPa WV, while the RUC TT is the highest due to a larger amount of 850 hPa WV. The AIRS PW is the largest, showing a moderate amount of tropospheric WV as AIRS also has the highest surface T and Td that affects the total-column WV. The RAOB and RUC PW depict lower amounts of total PW, mainly due to lower surface Td.

[49] These results demonstrate problems associated with horizontal inhomogeneity due to cloud contamination of RAOBs. The RAOB sounding (that depicts more stable conditions) is in poor agreement with AIRS and RUC, which instead indicate moderate amounts of instability and the potential for severe thunderstorms. In this case, the 1800 UTC RAOB sounding is not representative of the pre-convective environment, highlighting the value of AIRS.

5. Summary and Conclusions

5.1. Sources of Analysis Discrepancies

[50] Fetzer et al. [2003] present a discussion of the error analysis associated with remotely sensed soundings, and specifically on AIRS validation studies. The influences on AIRS data for resolving fine-scale atmospheric structures are T lapse rate, surface-atmospheric thermal contrast, the number of and specific IR channels used in the retrieval algorithm (i.e., spectral resolution), instrument noise, retrieval algorithm constraints, and the broadness of weighting functions [Maddy and Barnet, 2008]. As a result, overly smooth AIRS vertical profiles are derived, and low-level features are undetected even though strict tests are applied in lower tropospheric T retrievals [Susskind, 2006]. These uncertainties are reflected in the error estimates that accompany retrieved AIRS products. Sounding quality control, among others, is eventually dependent on the use of these error estimates, especially for the determination of PBest.

[51] There are several limiting factors that may have introduced errors into the analyses presented here. First, cloud contamination in the IR pixel was unavoidable in some cases due to the presence of broken cloud cover at the time of the Aqua overpass, leading to cold T biases in AIRS retrievals. Second, the case study selection process was made difficult due to the limited amount of 1800 UTC RAOBs, the presence of broken cloud cover at the time of an Aqua overpass, and the limited number of times that an Aqua overpass occurred within 2 h of a RAOB launch (an optimal sampling difference is <30 min). Third, Sun et al. [2010] studied the expected errors when collocation comparisons are imperfect; standard deviation errors for T are 0.35–0.42 K, and for RH are 3.1–3.3%, for collocation mismatches ≤3 h and 100 km. Our collocation strategies used may therefore have introduced errors. Fourth, previous studies have delineated the performance of AIRS soundings over land and over ocean [Fetzer et al., 2003; Divakarla et al., 2006], however this study did not examine the performance of AIRS over oceans even though 10 of the 76 soundings were retrieved over ocean or coastline. Last, the definition of a T inversion used for this study will differ from those used by others, specifically because only the layer at which the inversion begins, and not the inversion strength, was determined.

5.2. Main Conclusions

[52] The main conclusions drawn from this study are the following:

[53] 1. AIRS T and Td profiles are smooth compared to collocated RAOB profiles, however there is fairly good agreement between AIRS and RAOB T, demonstrating the AIRS instrument's ability to measure atmospheric T with good accuracy in the pre-convective environment.

[54] 2. Height-dependent errors and bias exist in AIRS, of which the most pronounced are found in the Td and q profiles. In the q profiles, a 2.25 g kg−1 dry bias is seen near the surface, and a 1.8 g kg−1 dry bias from 850 to 900 hPa.

[55] 3. AIRS T indicates a RMS difference of 1.5 K at 850 hPa. The heights of the T discrepancies correspond with the heights of the largest WV discrepancies, namely above 900 hPa.

[56] 4. The errors and biases in low-level AIRS T and WV observations had a pronounced effect on the stability indices and PW derived from AIRS. The near-surface and ∼850 hPa dry biases resulted in an underestimation of the instability, as well as in PW. However, as atmospheric instability is also affected by the mid-tropospheric T, the slight warm bias at 500 hPa seen in AIRS will also contribute to the underestimation of instability.

[57] 5. The RMS error in AIRS CAPE increases with increasing true CAPE. Similarly, the negative bias in CAPE and positive bias in LI derived from AIRS also increases with increasing true CAPE values (up to 3000 J kg−1). The lowest bias for PW, and in four out of the five stability indices, was seen for cases with true CAPE 600-1200 J kg−1. Therefore, the AIRS soundings agree better with truth/RAOB data in environments with low to moderate CAPE. This is most likely related to the more accurate AIRS T lapse rate of the mid-troposphere in such environments. (See Divakarla et al. [2006] for discussion of height dependence on AIRS T soundings.) If the AIRS instrument is unable to sense cooler mid-levels, larger discrepancies between AIRS and the RAOB data will occur in large-CAPE environments.

[58] 6. PBest can vary significantly for AIRS soundings within a 2 × 2° display window. It is suggested that further work be done to explore the relation between PBest and cloud fraction, as well as the height of low-level clouds. PBestmax values do not appear correlated to the difference magnitude between AIRS and RAOB CAPE, however this does not eliminate the possibility that PBestmax can be used as an indicator of cloud height.

[59] 7. RUC seems to overestimate WV, especially in mid-levels above 600 hPa with Td biases up to 6 K. Biases in q however are mostly small, except below 900 hPa, which is opposite that of AIRS. An insignificant negative T bias is seen in RUC above 600 hPa (≤1 K).

[60] 8. RUC performs similarly to AIRS for CIN, PW, KI and TT, especially in CAPE categories 3–5, whereas biases in RUC LI and CAPE are overall smaller compared to AIRS.

[61] For conclusions 1–3, and as noted above, cloud contamination may be a significant contributing factor by producing either moist or dry sampling biases depending on the cloud type in the IR FOV, which in turn would affect the quality of AIRS-derived stability indices. The pre-convective environment is usually not cloud free given the high likelihood of sub-pixel scale (500 m to ∼2 km wide) cumulus clouds before convective storms initiate. The heights of the largest biases in Td and q correspond well with the heights of low-level clouds and T inversions, suggesting issues associated with the broadening of weighting functions and/or some amount of cloud contamination around 850 hPa. These are speculative statements, all requiring follow-up analysis in order to quantify, and a suggested direction for this research.

[62] In the 76 soundings, 19 included a robust T inversion below 500 hPa, as seen in the corresponding RAOBs. As distinct increases in T with height are not seen in AIRS, the minimum in T lapse rate is regarded as the location of the AIRS inversion. On average, the minimum lapse rate can be found at ∼854 hPa, while the average of the lowest height of RAOB inversions was 815 hPa, giving an inversion height bias of +39 hPa. According to our definition of an AIRS inversion, it is concluded that they are on average located too close to the surface. A simple, subjective method (Figure 10 and Table 3) is presented for reconstructing a RAOB-like inversion (in terms of magnitude and altitude) within AIRS soundings, hence developing more representative RAOB-like soundings that can help benefit the operational forecaster.


[63] This study was supported by National Aeronautics and Space Administration (NASA) grant NAG5-12536. The authors wish to thank three anonymous reviewers for substantially improving the quality of this manuscript.