Automated aircraft water vapor reports, provided by 148 aircraft worldwide through the WMO’s AMDAR program, are at least as accurate as rawinsonde observations and have greater influence on 1–2 day NWP forecasts than all other in-situ moisture data over the United States
Currently, 148 aircraft-based Water Vapor Sensing Systems (WVSS) deliver observations operation-ally across the globe daily, principally across the United States. Although previous studies [summarized by Petersen (2016)] have demonstrated the value of aircraft wind and temperature reports obtained through the Aircraft Meteorological Data Relay (AMDAR) program (WMO 2003a) to both regional and global numerical weather prediction (NWP) scales, the quality and significance of these newly available moisture observations are less well recognized. Because moisture often varies at smaller scales than wind and temperature, temporally and spatially frequent moisture observations can have exceptionally large impacts on forecasts of disruptive weather events and could help offset the dwindling number of upper-air moisture observations available globally. Forecasts of disruptive weather events made at National Weather Service (NWS) field offices and participating airlines demonstrate the benefits of these newly available observations, particularly profiles made during aircraft ascents and descents.
A series of assessments are presented here comparing data from WVSS sensors installed on Boeing 757 aircraft operated by United Parcel Service (UPS) with collocated conventional observations. The results support the hypothesis that the aircraft moisture measurements can help fill data voids over land between rawinsonde observations (herein called raobs), both in time and space. Information derived regarding spatial and temporal moisture variability observed between neighboring aircraft could be important in optimizing the use of these observations in future regional- and storm-scale forecast systems. A series of case studies and a review of initial NWP studies performed elsewhere are also presented to demonstrate the importance of these high temporal and spatial resolution data on short-range forecasts over the United States.
BACKGROUND.
Various studies over the past decade have shown that, in addition to temperature and wind observations, detailed measurements of the vertical, horizontal, and temporal atmospheric moisture and related moisture flux structures are necessary to improve forecasts of location, intensity, and timing of precipitation events, including the onset and strength of convective storms (e.g., Hartung et al. 2011; Otkin et al. 2011). These events can have major impacts on public safety and economic efficiency. To meet this need, Fleming (1996) established the WVSS project with the goal of developing a moisture sensor appropriate for use on commercial aircraft that could offer a cost-effective means of supplementing other in situ observations, both temporally and spatially.
As an independent component of the WVSS development process, the University of Wisconsin Cooperative Institute for Meteorological Satellite Studies (UW-CIMSS) has conducted a series of independent evaluations of the systems over a 10-yr period to determine the accuracy of the aircraft humidity observations relative to temporally and spatially collocated raobs. WVSS data for these tests were obtained from UPS Boeing 757 aircraft landing at and departing initially from Louisville, Kentucky, and later Rockford, Illinois, for multiple 2-week-long episodes. Because 60%–80% of the WVSS equipped planes typically land or take off daily from these major UPS operational hubs, these locations allowed close comparison with raobs at the airports, without the logistical complications inherent in launching balloons in congested air traffic areas near more major airports.
The first-generation system (WVSS-I) used raob-like sensors to measure relative humidity (RH). Tests of this system, however, showed that many of the observations were unacceptably affected by a combination of factors, including decreased accuracy over time due to the collection of contaminants on the sensors, the inability to respond quickly to rapid changes in moisture, and excessive relative humidity biases and measurement errors below the reliable minimum threshold of the sensor caused by pressure-induced heating on aircraft traveling at high Mach numbers (Fleming et al. 2002).
A major redesign effort provided the basis for the current WVSS instrument, which can also be referred to as WVSS-II. First available in late 2004, it is applicable to all aircraft sizes and speeds and does not require frequent recalibration (Fleming and May 2006).
The reengineered WVSS system measures specific humidity (SH) using a laser-diode approach that senses the number of water vapor molecules moving past the sensor in a specific volume of air. WVSS observations are made independent of temperature and aircraft speed and are available across a wide range of values, from below 50 to over 40,000 ppmv (parts per million by volume, corresponding to approximately 0.03–24.9 g kg–1, respectively), and with an advertised accuracy of ±50 ppmv or ±5%, whichever is greater (Spectra Sensors 2012). The four times per second internal sampling rate (corresponding to one observation every 60–70 m at cruise levels) and 2.3-s data output rate provide vertical moisture profiles comparable in quality and resolution to rawinsonde reports. Observations are made during all phases of flight (ascent, cruise, and descent) and are attached to independently measured temperature and wind data to form a single AMDAR report for transmission to the ground. [See Petersen (2016) for details about AMDAR spatial and temporal reporting frequencies.]
Chamber tests were performed by Deutscher Wetterdienst (DWD) to document the accuracy of the WVSS system compared with research-grade moisture sensors in a controlled environment (Hoff 2009). These tests showed that the laser-diode system exceeded accuracy specification across a wide range of observing environments (Fig. 1), with the only substantial errors of the WVSS observations occurring at SH values less than 0.04 g kg–1. DWD concluded that WVSS data should be valuable in applications where SH remains above this threshold. Climatologically, these include midlatitude regions from the ground to at least 400 hPa during winter and to 200 hPa during summer.
Test results of the 2006 version of WVSS against calibrated reference systems performed in a climate chamber at the DWD Meteorological Observatory Lindenberg. [See Hoff (2009) for details.]
Citation: Bulletin of the American Meteorological Society 97, 11; 10.1175/BAMS-D-14-00211.1
Installation of the new WVSS sensors on UPS aircraft began in 2005. Initial aircraft-to-raob evaluations of this system (Petersen et al. 2005, 2006a,b) again revealed a number of performance problems, including 1) flaws in data-encoding procedures (whereby reports of SH greater than 10 g kg–1 had a precision of only 1 g kg–1), 2) water vapor contamination within the laser-diode sensing chamber that produced moist biases and thereby prevented the systems from observing very small amounts of atmospheric moisture (consistent with the findings from the DWD chamber tests), and 3) engineering problems caused by temperature-sensitive electronic components that produced large irregular biases and other errors.
After final engineering and communications modifications were completed (Helms et al. 2009; NOAA 2009), new WVSS hardware and software were installed on 25 UPS aircraft beginning in mid-2009. Results described here used data obtained only from these revised sensors. Similar sensors have since been installed operationally on approximately 115 Southwest Airlines Boeing 737 aircraft, as well as for evaluation on 9 aircraft flying in Europe. A map of typical daily WVSS ascent/descent coverage currently available over the United States is shown in Fig. 2.
Sample distribution of WVSS observations made during takeoff and/or landing between the surface and approximately 3,000 m (10,000 ft) from 1–7 October 2016. (Courtesy of NOAA/ESRL.)
Citation: Bulletin of the American Meteorological Society 97, 11; 10.1175/BAMS-D-14-00211.1
Prior to final intercomparisons of the reengineered WVSS systems with raobs, 224 WVSS observations obtained from aircraft on the ground were compared with contemporaneous aviation routine weather report (METAR) surface observations at 43 airports across the contiguous United States (CONUS) and southern Canada (Petersen et al. 2009). These comparisons were conducted primarily at night when the local effect of thermal mixing and consequent moisture variability should be minimal, although it should be noted that the METAR temperature and moisture are made 2 m above a grassy surface while the aircraft WVSS data were obtained at a higher height above concrete. The findings provide a benchmark for assessing instrument accuracy without the influences of aircraft motion. Results in Fig. 3 show good agreement across a range of SH values between 2 and 20 g kg–1, with standard deviation (std dev) agreements between independent moisture observations within 0.42 g kg–1 for SH and a negligible moist bias of approximately 0.01 g kg–1. This equates to an RH std dev of 3.53%. [For all RH comparisons shown in this paper, temperatures from the appraisal standard were used in the WVSS RH calculations, thus avoiding the effects of known AMDAR temperature biases described by Zhu et al. (2015) and discussed in more detail later.]
Comparison of 225 AMDAR SH observations (g kg−1) with surface METAR reports made at 43 sites across the United States and southern Canada for the period from 5 to 17 Sep 2009. Linear regression fit statistics shown in lower right.
Citation: Bulletin of the American Meteorological Society 97, 11; 10.1175/BAMS-D-14-00211.1
WVSS EVALUATION DESIGN AND RAWINSONDE COLOCATION RESULTS.
Comparisons of WVSS reports with collocated raobs were obtained during 16 evenings over three separate observing periods in fall 2009, spring 2010, and summer 2010 at the Rockford, Illinois, airport (RFD). These tests were designed to assess the general performance of the systems and to provide statistical evidence of the accuracy of the reengineered WVSS system. Of the 25 WVSS equipped UPS Boeing 757 aircraft, 17 were available for use in the statistical evaluations discussed later.
All raobs were made using Vaisala RS92-SGP instruments with Humicap moisture sensors, which employ twin heated thin-film capacitors to provide RH reports. The moisture observations have advertised reproducibility (std dev differences between twin soundings) of 2% and a total uncertainty in a sounding for reports with T > –60°C of 5% (2-sigma confidence level of cumulative effects, including repeatability, long-term stability, measurement conditions, and measurement electronics, as well as dynamic effects including response time). All of the raobs used in this test were acquired within 60 days of the experiments to minimize the effects of instrument aging on observation accuracy. [See Miller et al. (1999), WMO (2011), and Vaisala (2015) for more details on specific instrument details and performance.]
Raobs were launched adjacent to the RFD runway at approximately 3-hourly intervals. The first launches were made immediately before a period when a group of WVSS equipped aircraft landed and the last immediately after the final aircraft departed, with intermediate launches made during the short period between the last landing and first departure.
DIRECT INTERCOMPARISONS OF AMDAR-WVSS AND RAOB OBSERVATIONS.
Before statistical analyses were performed, individual WVSS systems were compared with nearby raobs. As in all results presented here, the raob and aircraft data were vertically interpolated using linear–log pressure methods to common pressure levels based on the nominal reporting frequency of the instantaneous aircraft ascent and descent reports available for this study (10-hPa intervals from the surface to 850 hPa, and 25-hPa intervals from there upward).
Figure 4 shows differences between an individual WVSS equipped aircraft and raobs throughout the spring 2010 data collection period. The AMDAR temperature (T) reports show a persistent warm bias at all but the lowest levels, although descending data (dashed lines) have slightly smaller biases. The spread of the T difference profiles is indicative of random variations at all levels. These results are consistent with previous studies by Ballish and Kumar (2008) and Zhu et al. (2015).
Differences between ascent (solid) and descent (dashed) observations from WVSS equipped aircraft (tail 1370) and collocated raobs taken within ±1 h between 27 Apr and 10 May 2010 at Rockford, Illinois. (left) T (°C) and (right) SH (g kg−1). Different colors distinguish between individual WVSS reports made throughout the period.
Citation: Bulletin of the American Meteorological Society 97, 11; 10.1175/BAMS-D-14-00211.1
By contrast, the WVSS SH reports show little bias and very small random differences in the lowest 200 hPa of the profiles. Larger random differences are seen immediately above 800 hPa in several ascents and above 650 hPa in both some ascent and descent reports, in part a reflection of the greater distance between higher-altitude WVSS reports and the validating raobs at these levels.
Comparisons between aircraft and individual raobs provide another subjective means of evaluating observations from individual WVSS equipped aircraft. Differences between WVSS and raobs for two evenings with different synoptic conditions and seasons are shown in Figs. 5a and 5b. In both cases, the majority of both ascent (solid) and descent (dashed) profiles showed good agreement with raobs taken immediately prior to and following the aircraft reports, including strong gradients across moisture inversions. This was especially true in the lowest 300 hPa of the soundings where all aircraft were flying essentially along the same paths and close to the raob launch site. The agreement between the successive, independent WVSS profiles provides confidence in the accuracy of and consistency between the individual WVSS observations. Aloft, the WVSS and raob differences show greater variability, part of which may be related to small-scale moisture variations and/or clouds in the area along the aircraft and/or the raob path and the increased separation between the paired observations at higher levels.
(a) Differences between ascending WVSS SH (g kg−1) observations and raobs taken around 0827 UTC on 30 Apr 2010 at Rockford, Illinois. Hourly rate of change observed between bounding raobs shaded. Colors indicate different aircraft providing reports, with arrival/departure times and encoded tail numbers noted in upper right. (b) As in (a), but for descending observations taken around 0310 UTC on 24 Aug 2010. (c) Plots of collocated WVSS SH (g kg−1) observations (colored) and raobs (black) taken around 0310 UTC on 24 Aug 2010 at Rockford, Illinois. Changes observed between bounding raobs shaded.
Citation: Bulletin of the American Meteorological Society 97, 11; 10.1175/BAMS-D-14-00211.1
Ascent WVSS moisture observations (Fig. 5a) predominantly agree within ±0.5 g kg–1 and fall within the hourly variations (shaded) between the temporally bounding raobs. Ascent reports generally dominated the aircraft datasets and typically show little bias.
Consistency between descent reports (Fig. 5b) was also very good, as illustrated by the fit of the WVSS data between the bounding raobs across much of the strong low-level moisture inversion in this summer case (see Fig. 5c). Although the WVSS and raob moisture profiles agree very closely, the slight misalignment of observing levels between the WVSS observations and the validating raobs at the time, location, and vertical displacement of the strong inversion led to a large plus/minus difference couplet between 850 and 950 hPa in Fig. 5b. Large individual differences like these, especially in moist environments, can have a large impact on the overall statistical evaluation presented later. It should be noted that overestimates of SH in WVSS descent data near the melting level were noted in some of the fall 2009 data, but were not present in later observations. Aircraft temperature profiles, however, continued to have notable warm biases, especially during aircraft ascent (not shown).
STATISTICAL ASSESSMENT OF COLLOCATED AMDAR AND RAOB PROFILES.
The analysis presented here used all available data from 15 UPS aircraft and excluded two early systems that experienced mechanical failures. WVSS observations were compared both with the single closest-time raobs and with bounding raobs interpolated linearly at each level to the aircraft observation time. Because the time interpolation process removed a portion of the random differences noted between the two datasets (11% for SH and lesser amounts for T and wind), only time-interpolated results are shown here. Colocations were made using all raobs within 60 min and 50 km of the WVSS reports, with raob locations adjusted for balloon drift.
Because the objective of these evaluations was to determine the quality of the WVSS data, in situ observations in which the effect of small-scale, local atmospheric variability was minimal were sought for comparison. Based on the analyses of individual cases described above, additional criteria were applied to restrict evaluations to cases with relatively small changes between successive raobs, both in time and between vertical levels. Temporal differences in RH between sequential raobs were limited to 7% and vertical differences between adjacent vertical levels in individual profiles to less than 10%. These restrictions diminish both the effects of scattered clouds that could have been along the raob trajectory and of vertical motion in shallow banks of moisture and fronts occurring between the 3-hourly raob launches. A minimum of 10 observation matchups was required at every level before statistics were calculated. In total, 1,175 collocations were used in the assessments—711 during aircraft ascent and 464 during descent.
Figure 6 compares individual WVSS reports with the collocated raobs for all levels, all observation periods, and both ascents and descents. Several characteristics of the WVSS data become readily apparent that are consistent with the individual profiles shown in Fig. 5. Overall, the more numerous ascent data show fewer outliers than the descent reports. The fits of the ascent data are extremely good for the middle- and upper-moisture range (>6 g kg–1), although there are a number of outliers from two specific aircraft at larger time differences in the 9–12 g kg–1 validation range. These outliers include the small-scale effects of the strengthening inversion for the August case in Fig. 5. It is noteworthy that the highest SH reports (generally obtained near the surface) agreed very closely both with the validation raobs and with other WVSS observations made at the time, as will be discussed later. Overall, the WVSS dataset matchups reveal a systematic difference (bias) of 0.15 g kg–1 and a random difference (std dev) of 0.62 g kg–1.
Scatterplot of 1,175 collocated WVSS and time-interpolated raob SH reports for all levels for all intercomparison periods at Rockford, Illinois. Markers color coded for 15 individual aircraft providing WVSS reports using encoded tail numbers at left. Marker size indicates time spread between WVSS and raobs, with larger markers indicate matchups that are closer in time. Ascents presented as solid squares, descents by open squares. Intercomparison and least-squared fit statistics included in lower right.
Citation: Bulletin of the American Meteorological Society 97, 11; 10.1175/BAMS-D-14-00211.1
SH data from descending aircraft showed a slightly smaller bias (+0.02 g kg–1) and larger random difference (std dev of 0.69 g kg–1) when compared with +0.20 g kg–1 bias and 0.54 g kg–1 std dev for ascending reports. Flights that descended into moist conditions showed slightly larger systematic moist biases relative to the collocated raobs, while the ascending flights showed smaller random differences (std devs). The fact that reports from multiple aircraft showed these same behaviors implies both good consistency between the aircraft reports and the possibility that the raobs taken at approximately 3-hourly intervals may not have fully captured the small-scale moisture structures observed by the WVSS aircraft at intervening times.
To determine whether WVSS data showed any degradation as a function of percent saturation, WVSS and raob SH intercomparison statistics were collected in 5% bins across the full range of observed raob RH values (Fig. 7). The results display only small variations in SH fit statistics, especially for RH values less than 85%. During ascent, WVSS biases remain small and positive (moist) except near saturation, where the biases became slightly negative (dry), while during descent, neutral to dry biases are more prevalent across a larger span of higher-raob RH values, with negative biases noted at RHs greater than 80%. Random differences in the ascent data remain fairly consistent for environments below 50% RH, with std dev near 0.5 g kg–1, but approach 1.1 g kg–1 in environments between 50% and 85% RH, and then return to 0.5 g kg–1 in the highest RH ranges. The larger std dev observed at higher RHs during descent might have been due in part to possible hysteresis in the instruments when descending through cloudy/moisture layers.
Comparison of systematic (bias) and random (std dev) differences of all WVSS SH (g kg−1) observations with time-interpolated raobs during the full intercomparison periods at 5% RH intervals during ascent (blue) and descent (green). Histogram at bottom shows number of observations in each RH interval.
Citation: Bulletin of the American Meteorological Society 97, 11; 10.1175/BAMS-D-14-00211.1
Vertical profiles of the WVSS and raob SH statistics are shown in Fig. 8. The best agreements appear in the lowest 50 hPa, with random differences (std dev) on the order of 0.5 g kg–1 and biases varying between -0.3 and +0.3 g kg–1. At all other levels, biases remain chiefly between -0.1 and 0.4 g kg–1. This systematic behavior can readily be removed and modified over time by monitoring instrument performance against a calibrated standard on a regular basis, such as comparing the expanded number of WVSS observations to coincident METARs and/or nearby operational raob profiles.
(left) Plots of SH comparison statistics between collocated observations from all WVSS and time-interpolated raobs taken for all 2009/10 intercomparison periods at Rockford, Illinois (bias, g kg−1, red; RMS, g kg−1, black; std dev, g kg−1, blue). Fit statistics included in lower right indicate a bias of 0.15 g kg−1 and std dev of 0.62 g kg−1 over the full assessment period using 1,175 data matches. Hatching indicates RMS of change between successive raobs throughout test period, normalized to 3-hourly rates. (right) Number of observations intercomparisons used (black), mean distance between reports (km, red), and mean time difference between reports (min, blue).
Citation: Bulletin of the American Meteorological Society 97, 11; 10.1175/BAMS-D-14-00211.1
Above 950 hPa, random differences range between 0.4 and 1.0 g kg–1. As shown on the right panel of Fig. 8, the increases above 850 hPa occur as the space and time separation between observations increases. Above 600 hPa, the decrease in random differences is due in part to decreases in magnitude of SH at these levels (not shown). Overall, the majority of the random differences in the WVSS-to-raob matchups fall within the shaded 3-hourly root-mean-square (RMS) changes observed between successive raobs throughout the full test period, indicating that WVSS observations are consistent with moisture variations detected between raob launches. Based on these results, WVSS performance appears to be well within the current WMO standards for both global and mesoscale weather forecasting applications (WMO 2003b, II-1–II-2).
Further partitioning of these profile statistics by ascent and descent (Fig. 9) shows that the time differences between ascents and raobs were less than those for descents and that the distance between the observations made during aircraft descent (which are generally made into the wind and therefore toward the raob ascent path) was generally less than during aircraft ascent. The number of upper-altitude reports available for comparison was also lower in the descent dataset owing to the more gradual (and therefore lengthier) descent paths taken by aircraft during landing. Descent data showed slightly smaller random differences when compared with raobs in the lowest 70 hPa, with larger differences aloft. Biases during descent were negligible, while during ascent averaged less than 0.2 g kg–1. Overall, random differences during descent of 0.67 g kg–1 were slightly larger than the 0.58 g kg–1 std dev noted during ascent.
As in Fig. 8 (left), but for (a) ascent and (b) descent aircraft data. Fit statistics included in lower right of each panel indicate a bias of 0.18 g kg−1 and std dev of 0.61 g kg−1 for 711 reports made during ascent and a bias of 0.1 g kg−1 and a std dev of 0.68 g kg−1 for reports made during descent.
Citation: Bulletin of the American Meteorological Society 97, 11; 10.1175/BAMS-D-14-00211.1
Investigation of other AMDAR observations show that, as in past tests, T observations (Fig. 10a) show positive biases, even for observations that were separated by less than 10 km. It should be noted that the aircraft T observing system is independent of the WVSS hardware. Although the bias is small near the surface, it increases rapidly to nearly 0.5°C between 925 and 850 hPa. Above that level, the bias decreases by half but then increases again near 600 hPa. Overall, the profiles collected during the three test periods showed a warm bias of approximately 0.36°C and are consistent with, but slightly larger than, results from the same aircraft during earlier WVSS tests. [Further discussion of possible sources of this bias, including possible hysteresis effects, is available in Zhu et al. (2015).] By contrast, random variability between the aircraft and raob T observations remained fairly uniform, averaging slightly below 0.84°C across all levels.
As in Fig. 8 (left), but for T (°C); T bias of 0.36°C and std dev of 0.84°C using 1,175 observation matches. (b) As in (a), but for RH calculated from aircraft SH and raob T; RH bias of 1.8% and std dev of 10.7%.
Citation: Bulletin of the American Meteorological Society 97, 11; 10.1175/BAMS-D-14-00211.1
When the independent aircraft T and SH observations are combined to derive RH, the warm bias in the aircraft T compensates for the small moist bias in the aircraft SH data, producing an RH difference profile with a bias of less than 0.5% throughout the lower troposphere (not shown). These results, however, do not accurately reflect the quality of the WVSS data alone. Because aircraft T biases may misrepresent WVSS moisture report RH accuracy in a variety of applications, including NWP quality control systems, it is recommended that the WVSS SH observations be used in their native form whenever possible.
To depict WVSS performance solely in terms of RH, the effects of the aircraft T biases were removed by using raob T in both the raob and WVSS RH calculations (Fig. 10b), similar to what was done previously using METAR data. Overall, RH biases are approximately 1.8% too moist and std devs are approximately 10.6%. Although the random differences between WVSS and raob observations in the lowest 200 hPa were generally between 4% and 6%, RH std dev values in colder environments above 700 hPa (where saturation SH is lower) reached as high as 15%.
STATISTICAL ASSESSMENT AMONG AMDAR MOISTURE OBSERVATIONS FROM MULTIPLE AIRCRAFT.
Another measure of the robustness of the WVSS observations was obtained by intercomparing reports made between pairs of WVSS aircraft within specific time, height, and spatial intervals. This approach provides information about the consistency between WVSS observations that is independent of possible errors in raob moisture reports. Although overall WVSS system biases cannot be determined this way, the results provide additional information about atmospheric temporal and spatial variability that could be important in determining how best to use moisture data in future storm-scale forecast systems. For this exercise, all WVSS observations that were included in the raob intercomparisons during 2009/10 evaluation periods were used. Observations from different aircraft that fell vertically within 50 m of each other were paired, without applying any vertical interpolation. The nearly 4,000 data pairs found at and below 5-km altitude were then sorted twice, first into four 15-min time bins for separations times up to 1 h and then into four 15-km distance bins for separation lengths out to 60 km.
Figure 11 shows that the RMS differences between WVSS observations made within 60 km of each other improves from about 0.42 g kg–1 for observations with 45–60-min separation times to 0.18 g kg–1 for time separations less than 15 min. Likewise, the differences between all observations made within 1 h of each other decreases from about 0.47 to 0.22 g kg–1 as separation distance between aircraft shortens from 45–60 to 0–15 km. In both cases, the variability between observations at the shortest ranges is less than half that at the largest intervals, owing in part to variations in atmospheric representativeness over the different time and length scales. Projecting these results to simulate perfect collocations (zero distance and time separations), the observation fits would be within 0.17 g kg–1. These differences are several times smaller than those between WVSS observations and individual raobs using the same time separation intervals, suggesting that errors in the raob moisture measurements may be contributing a larger component to the raob-to-WVSS comparison statistics than the WVSS observation errors.
Solid lines show variability (RMS) between SH observations among pairs of nearby WVSS observations made by 15 UPS Boeing 757 aircraft sorted by separation distance (red) and time difference (blue) intervals for all levels from the surface to 5 km from six weeks in three seasons of 2009/10. Vertical axis shows RMS differences in g kg−1 and horizontal axis shows 15-km wide distance and 15-min long time collocation bins. Thin lines indicate linear fits to observed differences, including projections to fits for perfect collocations. Dashed green line shows differences between WVSS and raob SH observations as a function of same time separations.
Citation: Bulletin of the American Meteorological Society 97, 11; 10.1175/BAMS-D-14-00211.1
These results, combined with the previously discussed raob intercomparisons, confirm both that WVSS observations meet international measurement quality criteria and suggest that they could be considered as a potential new observational comparison standard for a variety of meteorological and climate applications. Further comparisons among operational WVSS observations could provide a readily available means of understanding moisture variability in different locations, seasons, and weather regimes that should be helpful in improving the utility of these high-resolution point observations in future data assimilation systems at many scales.
IMPACTS OF AIRCRAFT MOISTURE OBSERVATIONS ON LOCAL FORECASTING.
An original justification for the participation of UPS in evaluations of the WVSS systems was the need to provide airline forecasters with observations needed for improving predictions of a variety of weather events that have negative impacts on the overnight operations of the airline. Use of the asynoptic aircraft moisture profiles (defined as observations taken at irregular periods throughout the day/night rather than at prespecified times) has since spread to a growing number of forecast offices (Baker et al. 2011). The following short examples illustrate how local forecasters across the United States incorporate WVSS-enhanced AMDAR observations (including temperature, winds, and moisture—denoted here as FullAMDAR) into their daily operations as a means of monitoring and improving upon standard observations and NWP guidance.
PRECIPITATION TYPE FORECASTS.
This case from the afternoon and evening of 21 December 2013 illustrates how WVSS observations at 2000 UTC were used to improve the local forecast of sleet for the evening before the busy Christmas travel weekend. Meteorologists at the National Weather Service Chicago office use WVSS data to support aviation forecasts for O’Hare (ORD) and Midway (MDW) airports. The data are especially important here, as the nearest twice-daily raob soundings are more than 200 km distant.
NWP guidance from 1200 UTC showed a storm system approaching from the plains was bringing in mild air around 1 km above the ground, while temperatures at the surface remained near freezing, a condition that can be favorable for formation of sleet and impact airport and flight operations if sufficient moisture is also present. As such, the forecast problem was one of determining if the model guidance was providing accurate depictions of the low-level temperature/moisture structures in the Chicago area. The succession of FullAMDAR soundings throughout the day showed progressively warmer temperatures than expected, with a 2000 UTC FullAMDAR descent sounding into MDW (Fig. 12) reporting temperatures at 850 hPa (1.5 km) near +5°C. Not only were these reports warmer than the models’ forecast, but also sufficiently warm to make sleet unlikely.1
Vertical profiles of temperature, wind, and moisture data from AMDAR sounding including WVSS reports from aircraft arriving at MDW at 2000 UTC 21 Dec 2013 showing warm layer between 800 and 875 hPa with a shallow subfreezing layer below. Plot from NOAA/ESRL shows temperature and derived dewpoint profiles on a skew T–logp diagram in large panel, a wind hodograph in the upper left, and a variety of stability parameters on the upper right of the diagram and wind plots by level on the right. All winds presented in knots (1 kt = 0.5144 m s−1).
Citation: Bulletin of the American Meteorological Society 97, 11; 10.1175/BAMS-D-14-00211.1
Based on these observations, NWS meteorologists correctly modified their 2213 UTC aviation forecast discussion, mentioning the importance of the aircraft soundings in determining the precipitation type of the winter storm. The appended text stated “Precipitation was moving into the Chicago area this afternoon…mainly in the form of rain/freezing rain” and included “Recent aircraft soundings from Chicago-Midway, IL (MDW) and Rockford, IL (RFD) depict warming aloft…with temperatures ranging from around +3°C at RFD to +4/5°C over MDW. This is warmer than most of the model guidance has depicted…which indicates a reduced potential for sleet.”
CEILING AND VISIBILITY FORECASTS.
Moisture availability and stratification are important factors for forecasting the formation, maintenance, and dissipation of low cloud layers. Cloud layer forecasts of 2,000 ft (610 m) or lower are important to aviation as they require some users to file for alternate airports and carry the additional fuel needed to fly to that airport.
This case illustrates how frequent WVSS soundings from Houston Hobby Airport (HOU) allowed meteorologists to monitor changes in atmospheric water vapor structures over time and update/improve local forecasts.
The proximity of Houston to moisture from the Gulf of Mexico results in frequent low cloudiness that can affect air traffic into this busy hub airport by reducing ceiling and visibility below critical operational minima. On the evening of 21 May 2014, a problem facing the night shift forecaster was to determine if low-level moisture would be as pronounced as during the past few evenings (not shown), as was indicated by NWP guidance (not shown).
Several FullAMDAR soundings were available throughout the afternoon and evening (e.g., Fig. 13). WVSS humidity profiles indicated that moisture between 0.5 and 1 km above ground level was both less abundant than the past few evenings and also smaller than indicated by the most recent numerical guidance. The persistent lack of lower-level moisture in the FullAMDAR soundings provided sufficient information to allow aviation forecasters to modify the 0618 UTC 22 May 2014 aviation forecast discussion to state “VFR will slowly turn to MVFR conditions tonight into tomorrow morning. AMDAR soundings show slightly less moisture this evening than at this time yesterday. GFS and NAM forecast soundings both support IFR and MVFR conditions tomorrow morning, but given AMDAR soundings and some possible dry air working its way around the ridge, [we] think clouds will remain more scattered. TAF sites will likely go between VFR and MVFR though the night until the sun rises, which will help the CIGS rise.”
As in Fig. 12, but from aircraft departing HOU at 2315 UTC 21 May 2014.
Citation: Bulletin of the American Meteorological Society 97, 11; 10.1175/BAMS-D-14-00211.1
The forecast allowed the airport to operate without concern of delays for either commercial or general aviation. Surface observations (Fig. 14) verify the changes made to the forecast, with clouds remaining few or scattered below 2,500 ft (762 m) for all but a single hour throughout the night and busy morning flight period.
Surface METAR reports from HOU from the morning hours of 22 May 2014. Low cloud observations highlighted.
Citation: Bulletin of the American Meteorological Society 97, 11; 10.1175/BAMS-D-14-00211.1
FOG FORECASTING AND NWP VALIDATION.
For decades, forecasters have used the process described by Petterssen (1956) for forecasting the onset, intensity, and dissipation of fog based on a vertical structure in which moisture increases with height below a near-surface inversion, with light winds and clear skies. In many situations, however, the needed vertical profiles of temperature and moisture are not available owing to the lack of local and/or real-time raobs. AMDAR observations including moisture profiles have helped fill this data gap at many sites across the country, including differentiating between significantly different meteorological conditions that can exist between nearby airports (e.g., see Rahn and Mitchell 2016).
For example, on 29 March 2005, UPS meteorologists forecasting for operations at their important Louisville airport hub (SDF) were faced with the problem of determining whether the NWP guidance showing dry advection near the surface until about 0400 UTC, followed by decoupling and possible local fog formation after 0700 UTC, would be applicable at the airport or need to be modified. A series of aircraft descent profiles were investigated (Fig. 15). The first descent report at 0425 UTC (dark blue profile in Fig. 15) confirmed the model forecasts of a sharp decrease in moisture with height. Furthermore, a 3°C inversion was capped by winds of nearly 10 m s–1, indicating favorable conditions for downward mixing of drier air. Based on these reports, forecasters expected patchy ground fog across the city, but not at SDF airport itself, owing to an increased heat island effect there. Subsequent aircraft soundings (later reports in Fig. 15) corroborated the initial profile and confirmed the diagnosis of further localized erosion of the low-level moist layer.
As in Fig. 12, but from five aircraft arriving at SDF between 0425 and 0649 UTC 29 Mar 2005. Times of aircraft descents for different aircraft noted in lower left.
Citation: Bulletin of the American Meteorological Society 97, 11; 10.1175/BAMS-D-14-00211.1
This case illustrates the important role that FullAMDAR observations can play in local forecasts. At Louisville Bowman Field (LOU) visibility was reduced to 0.4 km by ground fog later in the evening, extending in patches into low-lying locations across the city and verifying the NWP forecast. However, only 8 km away, the visibility at SDF never dropped below 4 km, as indicated by the FullAMDAR reports that supplemented the NWP guidance there.
SEVERE THUNDERSTORM FORECASTS.
Severe thunderstorm forecasters are often hampered by inadequate knowledge of the temperature, wind field, and moisture above the ground. This information is so important that special raobs can be launched in the United States around 1800 UTC to provide forecasters with added data for the afternoon and evening forecasts. These special launches are costly and are normally reserved for days when organized severe thunderstorms are expected. Forecasters across the United States have found FullAMDAR profiles to be a useful supplemental data source, both for determining the likelihood and strength of severe thunderstorms as well as differentiating between severe and nonsevere events.
Such a situation occurred during the midday hours of 29 May 2014. Forecasters in Tulsa, Oklahoma, were faced with determining the severity of thunderstorms expected that afternoon. A series of FullAMDAR soundings were available throughout the day. The combination of temperature, WVSS moisture, and wind information contained in these profiles (e.g., Fig. 16) showed not only a persistent lack of suitable instability and a capping inversion but also insufficient wind shear to support severe thunderstorm development. Using this information, the area forecast discussion was modified to indicate that, although scattered thunderstorms were likely to occur, the chance of severe thunderstorms was very low, stating “The 1450 UTC AMDAR sounding is not overly favorable for convection…despite being virtually uncapped. Lapse rates from 700 hPa on up are either moist adiabatic or less…with a unidirectional northeast flow thru the column at less than 20 kts. Nevertheless…with insolation ongoing, there will be scattered showers and a few storms, but instability will be limited by the poor lapse rates aloft and there will be virtually no storm relative flow. Thus…severe weather is not expected this afternoon.” No severe weather was reported.
As in Fig. 12, but from aircraft departing Tulsa International Airport at 1449 UTC 29 May 2014.
Citation: Bulletin of the American Meteorological Society 97, 11; 10.1175/BAMS-D-14-00211.1
IMPACTS OF WVSS ON NWP MODELS.
The recent expansion in the number of AMDAR aircraft providing WVSS observations has allowed several NWP centers adequate data to begin conducting impact tests using these spatially and temporally denser moisture observations. Although these tests have been much less extensive than those for AMDAR temperature and wind observations [see Petersen (2016) for discussion and further references], all tests show positive impacts on short-range forecasts over the United States. Several efforts using different global forecast models are noteworthy.
In cooperation with the National Centers for Environmental Prediction (NCEP), Hoover et al. (2016) tested methods for incorporating WVSS SH observations in the Global Data Assimilation System (GDAS) and Global Forecast System (GFS) using data for both warm (April–May 2014) and cold seasons (December 2014–January 2015). Results demonstrated positive impact on warm season 12–36-h precipitation forecasts and improved observation-minus-background (6-h forecast) values of SH as well. Comparisons with total precipitable water measurements at surface global positioning system (GPS) sites showed warm season forecast improvements out to 72 h (statistically significant between 0–36 and 60–66 h) and cold-season improvements through 24 h (significant at 0 and 6 h). Aircraft moisture observations were included in the operational GDAS in mid-2016.
To provide more specific insight into the value of assimilated AMDAR moisture observations in operational models, an evaluation of the impact of WVSS observations in the Navy Global Environmental Model (NAVGEM; Hogan et al. 2014) was conducted using forecast sensitivity observation impact (FSOI) methods (Langland and Baker 2004). FSOI values are computed for each observation as part of the operational run of NAVGEM. When summed for all observations and variables over a specified period, they provide a measure of the relative importance of that instrument type in reducing the 24-h forecast error, measured by the total moist energy error norm.
Figure 17 (left) portrays relative FSOI values for in situ moisture observations over the CONUS, calculated as the percentage of the total reduction in the 24-h error of all humidity observations and presented from June through October 2015, as well as for shorter summer and fall subperiods. Among conventional humidity observations, the WVSS reports provide a reduction in 24-h forecast error, which is approximately 50% greater than when using only raob humidity data, because they are both more numerous than raob and surface observations and better distributed in time and space compared to the twice-daily raobs. The relative importance of WVSS is slightly greater in warm season than in fall, with greater surface land impacts in fall. The 15,400 daily WVSS observations used from ascents had nearly 50% more impact than the 10,760 daily descent reports. However, when considered in terms of relative impacts per observation, individual ascent and descent reports had nearly equal impact.
Relative contribution of different in situ moisture data sources over the CONUS in the NAVGEM v1.3 during a 5-month period from Jun to Oct 2015 as determined by percentage of total FSOI attributable to each data source. Results calculated for a latitude–longitude domain extending 25°–50°N, 125°–60°W. (left) Impacts of WVSS, raobs, and all surface observations for full period and two seasonal subperiods, along with relative impact per individual observation between WVSS ascent and descent profile reports for full period. (right) Relative FSOI for WVSS and raobs profile reports for 100-hPa thick layers of the troposphere.
Citation: Bulletin of the American Meteorological Society 97, 11; 10.1175/BAMS-D-14-00211.1
If viewed in profile form (Fig. 17, right), WVSS observations have greater impact than raobs at most levels of the troposphere below 400 hPa, with surface observations dominating the lowest layer and raobs leading only in the 600–699-hPa layer. The prominence of WVSS reports below this layer is likely related to the increased vertical observing frequency of the aircraft reports in this area [see Petersen (2016) for details], while the availability of multiple reports throughout the day at multiple locations as aircraft move away from airports and limited raob sites may increase the importance of WVSS data above 600 hPa.
These results demonstrate that aircraft-based WVSS observations have already become a primary source of upper-air moisture observations over the United States for short-range NWP. Increased impacts are anticipated as the network of WVSS equipped aircraft expands further. On the global scale, however, results (not shown) indicated that balloonborne observations will continue to be the primary in situ source of moisture over many parts of the globe until a broader international WVSS network becomes available elsewhere, followed in importance by surface observations.
SUMMARY AND RECOMMENDATION.
Tests conducted by numerous NWP centers over the past decade have concentrated on assessing the impact of AMDAR temperature and wind observations on the skill of regional and global NWP systems. Results show that aircraft data taken en route and during ascent/descent provide important information for improving forecasts, both for individual events and for long-term performance. Those tests, however, had not taken into account the additional improvements that can occur from the new WVSS moisture-observing systems being deployed as part of the global AMDAR enhancement effort.
Evaluations of these new moisture-observing systems being deployed on U.S. aircraft show that WVSS observations
provide excellent quality horizontally and vertically, even across sharp inversions;
agree with collocated raobs to within 0.6 g kg–1, with minimal biases (approximately 0.15 g kg–1); and
display consistency between observations from different aircraft of at least 0.2 g kg–1 (RMS), indicating that WVSS observations perform as well as high-quality raobs.
The volume of WVSS data available over the CONUS has recently grown to a level that can support data impact tests in NWP models. Initial results have shown short-range forecast impacts larger than from any other moisture observations, including twice-daily raobs. Humidity forecast improvements like these are essential to enhance prediction of both the timing and location of precipitation events.
Although the improvements attributable to WVSS observations have been concentrated in areas of highest data availability, similar advancements are expected in other areas as the spatial and temporal coverage of the reports increases globally. This will be particularly important both in areas where the continuation of upper-air observing programs are under budgetary threat and in forecast situations where additional observations are needed to fill the time and space gaps between once- or twice-daily raob launches.
Existing AMDAR observations are extremely cost effective, currently contributing only about 0.25% of the expense of the global observing system (Eyre and Reid 2014) with temperature/moisture/wind profiles typically costing less than 5% of a full raob launch. Although these data do not meet all balloonborne observing requirements (in particular, data in and above the stratosphere are needed for both weather and climate purposes), the availability of high-quality tropospheric profiles over land at space and time resolutions not affordable using conventional observing systems offers a unique opportunity for improving weather forecasts across the globe, including terminal and weather hazard forecasts benefiting airlines. As such, programs should be supported which both increase the number of aircraft providing humidity information and expand the AMDAR observing network into areas not currently covered adequately.
ACKNOWLEDGMENTS
This paper was commissioned by the World Meteorological Organization, Geneva, Switzerland. Particular thanks go to Dean Lockett and Frank Grooters in supporting these papers. The contributions and efforts of Sarah Bedka in the early stages of the WVSS evaluations were invaluable, as were those of Erik Olson and the multiple other CIMSS personnel involved in the field tests in Rockford, Illinois, and Louisville, Kentucky. The efforts of David Helms of NOAA were instrumental in the formative and testing stages of the WVSS program and deserve to be recognized, as do the reviewers for their helpful suggestions. Finally, the roles of Kenneth Macleod, Charles Sprinkle, and Jeff Stickland in developing the AMDAR program through WMO need to be acknowledged.
REFERENCES
Baker, R., R. Curtis, D. Helms, A. Homans, and B. Ford, 2011: Studies of the effectiveness of the Water Vapor Sensing System, WVSS, in supporting airline operations and improving air traffic capacity. Second Aviation, Range and Aerospace Meteorology Special Symp. on Weather-Air Traffic Management Integration, Seattle, WA, Amer. Meteor. Soc., 335. [Available online at https://ams.confex.com/ams/91Annual/webprogram/Paper181457.html.]
Ballish, B. A., and V. K. Kumar, 2008: Systematic differences in aircraft and rawinsonde temperatures—Implications for NWP and climate studies. Bull. Amer. Meteor. Soc., 89, 1689–1707, doi:10.1175/2008BAMS2332.1.
Eyre, J., and R. Reid, 2014: Cost-benefit studies of observing systems. Met Office Forecasting Research Tech. Rep. 593, 11 pp.
Fleming, R. J., 1996: The use of commercial aircraft as platforms for environmental measurements. Bull. Amer. Meteor. Soc., 77, 2229–2242, doi:10.1175/1520-0477(1996)077<2229:TUOCAA>2.0.CO;2.
Fleming, R. J., and R. D. May, 2006: The 2nd generation water vapor sensing system and benefits of its use on commercial aircraft for air carriers and society. SpectraSensors, 16 pp. [Available online at www.eol.ucar.edu/system/files/spectrasensors.pdf.]
Fleming, R. J., D. R. Gallant, W. F. Feltz, J. G. Meitin, W. R. Moninger, S. F. Williams, and R. T. Baker, 2002: Water vapor profiles from commercial aircraft. UCAR Rep., 37 pp. [Available online at http://mailman.eol.ucar.edu/system/files/watervapor2.pdf.]
Hartung, D. C., J. A. Otkin, R. A. Petersen, D. D. Turner, and W. F. Feltz, 2011: Assimilation of surface-based boundary layer profiler observations during a cool-season weather event using an Observing System Simulation Experiment. Part II: Forecast assessment. Mon. Wea. Rev., 139, 2327–2346, doi:10.1175/2011MWR3623.1.
Helms, D., K. Johnston, G. Sanger, B. Taubvurtzel, R. Petersen, A. Homans, and A. Hoff, 2009: Testing and deployment of the Water Vapor Sensing System II. 25th Conf. on International Interactive Information and Processing Systems (IIPS) for Meteorology, Oceanography, and Hydrology, Phoenix, AZ, Amer. Meteor. Soc., 4A.5. [Available online at https://ams.confex.com/ams/89annual/techprogram/paper_149759.htm.]
Hoff, A., 2009: WVSS assessment at the DWD. Deutsche WetterDienst, 13 pp. [Available online at http://amdar.noaa.gov/docs/WVSS-II_Assessment_DWD.pdf.]
Hogan, T. F., and Coauthors, 2014: The Navy Global Environmental Model. Oceanography, 27, 116–125, doi:10.5670/oceanog.2014.73.
Hoover, B. T., D. A. Santek, A.-S. Daloz, Y. Zhong, R. Dworak, R. A. Petersen, and A. Collard, 2016: Forecast impact of assimilating aircraft WVSS-II water vapor mixing ratio observations in the Global Data Assimilation System (GDAS). UW SSEC Publication 16.02.H1/Project Rep., 38 pp. [Available online at http://library.ssec.wisc.edu/research_Resources/publications/pdfs/SSECPUBS/SSEC_Publication_No_16_02_H1.pdf.]
Langland, R. H., and N. Baker, 2004: Estimation of observation impact using the NRL atmospheric variational data assimilation adjoint system. Tellus, 56A, 189–201, doi:10.1111/j.1600-0870.2004.00056.x.
Miller, E. R., J. Wang, and H. L. Cole, 1999: Correction for dry bias in Vaisala radiosonde RH data. Proc. Ninth ARM Science Team Meeting, San Antonio, TX, Atmospheric Radiation Measurement Climate Research Facility, 9 pp. [Available online at www.arm.gov/publications/proceedings/conf09/extended_abs/miller_er.pdf.]
NOAA, 2009: Retest and evaluation report for the SpectraSensors Water Vapor Sensing System II (WVSS). NOAA/NWS Rep., 18 pp. [Available online at http://amdar.noaa.gov/docs/WVSS300_test_results.pdf.]
Otkin, J. A., D. C. Hartung, D. D. Turner, R. A. Petersen, W. F. Feltz, and E. Janzon, 2011: Assimilation of surface-based boundary layer profiler observations during a cool-season weather event using an Observing System Simulation Experiment. Part I: Analysis impact. Mon. Wea. Rev., 139, 2309–2326, doi:10.1175/2011MWR3622.1.
Petersen, R. A., 2016: On the impact and future benefits of AMDAR observations in operational forecasts—Part I: A review of the impact of automated aircraft wind and temperature reports. Bull. Amer. Meteor. Soc., 97, 585–602, doi:10.1175/BAMS-D-14-00055.1.
Petersen, R. A., W. Feltz, and S. Bedka, 2005: Final report of results of the June 2005 WVSS – Rawinsonde Intercomparison Study. University of Wisconsin–Madison, Cooperative Institute for Meteorological Satellite Studies Rep., 16 pp. [Available online at http://amdar.noaa.gov/docs/UW_WVSSII_JUN05_Assessment_Final_Report.pdf.]
Petersen, R. A., W. Feltz, E. Olson, and S. Bedka, 2006a: Evaluation of the WVSS-II moisture sensor using co-located in-situ and remotely sensed observations. 10th Symp. on Integrated Observing and Assimilation Systems for the Atmosphere, Oceans, and Land Surface (IOAS-AOLS), Atlanta, GA, Amer. Meteor. Soc., P2.6. [Available online at https://ams.confex.com/ams/Annual2006/techprogram/paper_102987.htm.]
Petersen, R. A., W. Feltz, E. Olson, and S. Bedka, 2006b: Final report: Results of the November 2006 WVSS—Rawinsonde Intercomparison Study. University of Wisconsin–Madison, Cooperative Institute for Meteorological Satellite Studies Rep., 42 pp. [Available online at http://amdar.noaa.gov/docs/UW_WVSS-II_Nov2006_Assessment_FINAL.pdf.]
Petersen, R. A., S. Bedka, W. Feltz, E. Olson, and D. Helms, 2009: WVSS-II Moisture Observations: A low-cost tool for validating and monitoring asynoptic satellite data. Meteorological Satellite Conf., Bath, United Kingdom, EUMETSAT, 10 pp. [Available online at www.eumetsat.int/website/wcm/idc/idcplg?IdcService=GET_FILE&dDocName=PDF_CONF_P55_S1_05_PETERSEN_V&RevisionSelectionMethod=LatestReleased&Rendition=Web.]
Petterssen, S., 1956: Weather Analysis and Forecasting. Vol. 2, McGraw-Hill, 266 pp.
Rahn, D. A., and C. J. Mitchell, 2016: Diurnal climatology of the boundary layer in Southern California using AMDAR temperature and wind profiles. J. Appl. Meteor. Climatol., 55, 1123–1137, doi:10.1175/JAMC-D-15-0234.1.
Spectra Sensors, 2012: Atmospheric Water Vapor Sensing System (WVSS) technical specifications. Spectra Sensors, 4 pp. [Available online at www.spectrasensors.com/files/525/.]
Vaisala, 2015: Vaisala Radiosonde RS92-SGP. Vaisala, 2 pp. [Available online at www.vaisala.com/Vaisala%20Documents/Brochures%20and%20Datasheets/RS92SGP-Datasheet-B210358EN-F-LOW.pdf.]
WMO, 2003a: Aircraft Meteorological Data Relay (AMDAR) reference manual. World Meteorological Organization WMO-958, 84 pp. [Available online at www.wmo.int/pages/prog/www/GOS/ABO/AMDAR/publications/AMDAR_Reference_Manual_2003.pdf.]
WMO, 2003b: Manual on the Global Observing System. World Meteorological Organization WMO-544, 50 pp. [Available online at www.wmo.int/pages/prog/www/OSY/Manual/WMO544.pdf.]
WMO, 2011: WMO intercomparison of high quality radiosonde systems. World Meteorological Organization WMO/TD-1580, 238 pp. [Available online at www.wmo.int/pages/prog/www/IMOP/publications/IOM-107_Yangjiang.pdf.]
Zhu, Y., J. Derber, R. J. Purser, J. Whiting, and B. A. Ballish, 2015: Variational correction of aircraft temperature bias in the NCEP’s GSA analysis system. Mon. Wea. Rev., 143, 3774–3803, doi:10.1175/MWR-D-14-00235.1.
This is significant when sufficient moisture is also present in that sleet often requires aircraft deicing, which is costly and causes delays.
