TMCnet News

Low-Level Jets in the North American Regional Reanalysis (NARR): A Comparison with Rawinsonde Observations [Journal of Applied Meteorology and Climatology]
[September 19, 2014]

Low-Level Jets in the North American Regional Reanalysis (NARR): A Comparison with Rawinsonde Observations [Journal of Applied Meteorology and Climatology]


(Journal of Applied Meteorology and Climatology Via Acquire Media NewsEdge) ABSTRACT Climatological analyses of low-level jets (LLJs) can be negatively influenced by the coarse spatial and temporal resolution and frequent changes in observing and archiving protocols of rawinsonde observations (raobs). The introduction of reanalysis datasets, such as the North American Regional Reanalysis (NARR), provides new resources for climatological research with finer spatial and temporal resolution and potentially fewer inhomogeneities. To assess the compatibility of LLJ characteristics identified from NARR wind profiles with those obtained from raob profiles, LLJs were extracted using standard jet definitions from NARR and raobs at12 locations in the central United States for four representative years that reflect different rawinsonde protocols. LLJ characteristics (e.g., between-station differences in relative frequency, diurnal fluctuations, and mean speed and elevation) are generally consistent, although absolute frequencies are smaller for NARR relative to raobs at most stations. LLJs are concurrently identified in the NARR and raob wind profiles on less than 60% of the observation times with LLJ activity. Variations are seen between analysis years and locations. Of particular note is the substantial increase in LLJ frequency seen in raobs since the introduction of the Radiosonde Replacement System, which has led to a greater discrepancy in jet frequency between the NARR and raob datasets. The analyses suggest that NARR is a viable additional resource for climatological analyses of LLJs. Many of the findings are likely applicable for other fine-resolution reanalysis datasets, although differences between reanalyses require that each be carefully evaluated before its use in climatological analyses of wind maxima.



1. Introduction Low-level jets (LLJs) have a substantial influence on local and regional climate. Climatological studies of LLJs have frequently used vertical wind profiles at in- dividual rawinsonde stations to identify wind maxima (e.g., Bonner 1968; Mitchell et al. 1995; Arritt et al. 1997; Whiteman et al. 1997; Walters et al. 2008). The coarse temporal (2 times per day) and spatial (often greater than 300 km) resolutions of routine rawinsonde observations (raobs), as well as changes over time in rawinsonde sta- tion locations, rawinsonde types, observational protocols, and archiving procedures (Bosart 1990; Schwartz and Doswell 1991; Winkler 2004), often limit the climato- logical applications of these observations, however. In addition, Elliott and Gaffen (1991) and Winkler (2004) warn potential users that quality-control procedures ap- plied to rawinsonde archives likely have not removed all erroneous or questionable observations. Other sources of upper-air wind observations, such as the National Oceanic and Atmospheric Administration (NOAA) National Profiler Network, have proven also to be problematic for climatological analyses because of the limited spatial extent of observations and/or the short record length, as well as contamination by migrating birds (Mitchell et al. 1995; Arritt et al. 1997; Daniel et al. 1999).

Several recent studies of LLJs (e.g., Muñoz et al. 2008; Weaver et al. 2009; Weaver and Nigam 2011) have instead utilized wind fields from the North American Regional Reanalysis (NARR), which is a long-term, dynamically consistent, high-resolution climate dataset that is available from 1979 to the present for the North American domain (Mesinger et al. 2006). NARR has a finer spatial (32 km) and temporal (3 hourly) resolution than do rawinsonde observations and may also be affected to a lesser degree by data-quality issues, because NARR is an assimilated product that incorporates first-guess model fields with observations from dropsondes, pilot balloons, commercial aircraft, satellites, and radar wind profilers in addition to rawinsonde observations (Shafran et al. 2004).


Previous comparisons of NARR wind fields with ob- servational datasets and/or other reanalysis fields are relatively limited, with most comparisons focusing on winds at a single elevation or pressure level. For exam- ple, NARR 10-m near-surface wind speeds over south- ern Greenland were found to slightly underestimate wind speeds measured at a data buoy, with a least squares error of approximately 2 m s21 (Moore et al. 2008). Similarly, Mo et al. (2005, their Fig. 14a) show weaker early-morning wind speeds in NARR when compared with profiler wind speeds as analyzed by Higgins et al. (1997, their Fig. 5) for a location near Oklahoma City. On the other hand, Li et al. (2010), who compared mean daily 80-m NARR wind speeds with observed winds at five rawinsonde stations in the Great Lakes region, re- ported a slight bias (from 20.64 to 10.59 m s21) in wind speed, with the sign of the bias varying across locations. Zonally averaged NARR wind speeds at the 925-hPa level over the Caribbean Sea were found to be weaker than wind fields from the 40-yr European Centre for Medium-Range Weather Forecasts reanalysis (ERA-40; Muñoz et al. 2008). The only earlier analysis that focused explicitly on the vertical wind profile is that of Mesinger et al. (2006), who compared average wind profiles for North America in January and July obtained from NARR, raobs, and the National Centers for Environ- mental Prediction-U.S. Department of Energy second Atmospheric Model Intercomparison Project (AMIP-II) Global Reanalysis (GR2) and found that NARR better approximated the raob wind profiles than did GR2.

The dearth of comparisons of the vertical profile of NARR winds with the profiles obtained from other wind sources is a concern, because many climatological studies have identified LLJs on the basis of vertical wind shear in addition to the maximum wind speed at the jet ''nose.'' For example, Bonner (1968), in his classic climatological analysis of LLJs, defined a jet as a speed maximum of at least 12 m s21 with a decrease of at least 6 m s21 above the jet nose. Numerous authors (e.g., Mitchell et al. 1995; Arritt et al. 1997; Whiteman et al. 1997; Anderson and Arritt 2001) have utilized similar criteria, whereas other studies also required a decrease in wind speed below the jet nose (e.g., Andreas et al. 2000; Bantaetal.2002; Walters and Winkler 2001; Walters et al. 2008). [See Table 1 in Walters et al. (2008) for a summary of com- monly used jet definitions.] Thus, in addition to differ- ences in wind speed, differences in wind shear resulting from the vertical resolution of NARR wind profiles in the lower troposphere also may affect the frequency at which LLJs are identified.

In light of the growing use of reanalysis datasets such as NARR for climatological research, this study com- pares LLJs identified from raob vertical wind profiles with those identified from NARR vertical wind profiles for 12 locations in the central United States, a region of frequent jet occurrence (Bonner 1968; Helfand and Schubert 1995; Arritt et al. 1997; Higgins et al. 1997; Carbone and Tuttle 2008; Walters et al. 2008). In par- ticular, this study investigates 1) whether the charac- teristics (e.g., frequency, strength, and diurnal variation) of LLJs as inferred from NARR are generally in agreement with those inferred from rawinsonde obser- vations, 2) whether the agreement between LLJs iden- tified from raobs and NARR varies geographically, and 3) whether changes in the vertical resolution of raobs over the last few decades have had an impact on the characteristics of LLJs identified from raob and NARR wind profiles. NARR was the reanalysis dataset selected for this analysis given its wide use, but the findings have implications for other reanalysis datasets with similar vertical resolutions. This analysis should not be consid- ered a ''verification'' of NARR, because raob observa- tions are ingested into NARR and hence the two datasets are not independent. Instead, we investigate the impli- cations of using NARR rather than raobs for climato- logical analyses of LLJs.

2. Data and methods a. Study period and study locations Four analysis years (1980, 1990, 2000, and 2010) were selected to represent ''eras'' in the rawinsonde record with different rawinsonde types, reporting and coding practices, and/or archiving procedures. Soundings for 1980 represent the period during which the coding of the Global Telecommunication System (GTS) message for international distribution of rawinsonde observations was performed manually, the soundings for 1990 repre- sent the Automatic Radiotheodolite System (MiniART) period with automated generation of GTS messages, the soundings for 2000 are from the replacement Micro- computer Automatic Radiotheodolite (MicroART) sys- tem, and the soundings for 2010 are part of the recently installed Radiosonde Replacement System (RRS). Note, however, that for any segment of the rawinsonde record, stations were not necessarily using rawinsondes from the same manufacturer or using identical ground equipment. Individual years separated in time were employed rather than a sequential time series because transitions to a new system with accompanying changes in observing and coding practices usually occur over an extended (several year) period, with the transition at different times de- pending on the station. Each year was defined as ex- tending from December through November (e.g., 1980: 1 December 1979-30 November 1980) to allow for com- parisons on a seasonal basis.

The LLJ comparisons were conducted for 12 locations (Bismarck, North Dakota; Green Bay, Wisconsin; North Platte, Nebraska; Topeka, Kansas; Springfield, Missouri; Amarillo, Texas; Norman, Oklahoma; Midland, Texas; Fort Worth, Texas; Del Rio, Texas; Brownsville, Texas; and Lake Charles, Louisiana) within and surrounding the well-documented region of frequent LLJ activity in the central United States (Bonner 1968; Mitchell et al. 1995; Walters et al. 2008)(Fig. 1). The constraint that all stations must have made the transition to the same observing and coding practices for each of the four analysis years (i.e., 1980, 1990, 2000, and 2010) limited the number of sta- tions included in the analysis relative to all available rawinsonde locations.

b. Wind profiles Twice-daily (0000 and 1200 UTC) raobs were obtained from the archive maintained by NOAA's Earth System Research Laboratory [ESRL, formerly the Forecast Systems Laboratory (FSL)]. The ESRL archive was uti- lized because it had undergone preprocessing of ''raw'' rawinsonde observations that included hydrostatic and gross error checks. The rawinsonde observations for 1980 and 1990 were extracted from the NOAA/FSL Radiosonde Data Archive on CD-ROM, whereas the data for 2000 and 2010 were downloaded from the NOAA/ESRL Radio- sonde Database (see online at http://www.esrl.noaa.gov/ raobs/). Observations for December of 1979 were miss- ing from the ESRL record for Norman; thus, compari- sons for this location are based on a slightly shorter period of record. The ''merged'' rawinsonde profiles, for which the different components of the GTS message (e.g., ''TTAA,'' ''TTBB,'' and ''PPBB'') had been com- bined and sorted by the pressure level of the mea- surement, were employed in the analyses below. [See Schwartz and Govett (1992) for more details on the ESRL archive.] The Grid Analysis and Display System (GrADS) soft- ware (see online at http://iges.org/grads/) was used to extract NARR vertical wind profiles for the latitude/longitude coordinates corresponding to the rawinsonde sites. GrADS applies a bilinear interpolation to obtain estimated values for a station using the four native grid points that are nearest to the station location. The vertical wind profiles were augmented with the NARR 10-m winds, because near-surface wind information is included in the archived raob profiles. The raob and NARR wind profiles were considered to represent the same geographic location. Although the spatial drift of a weather balloon over its ascent may exceed 200 km (McGrath et al. 2006), LLJs by definition occur in the lower troposphere where the drift is relatively minimal.

In making comparisons of LLJs detected from NARR with those detected from raobs, one needs to be mindful of differences in vertical resolution, under the assump- tion that having a larger number of vertical levels with wind information increases the likelihood of detecting an LLJ. Whereas the vertical resolution for the NARR archive is fixed (25-hPa intervals below 700 hPa and 50-hPa intervals above 700 hPa), the vertical resolution of a raob sounding varies from one observation to another and with the different segments of the rawinsonde re- cord. The latter is seen in the average number of wind data levels at or below the 550-hPa level for the 12 study locations during the four analysis years (Fig. 2). The average number of raob wind levels was greatest in 1990 or 2000, depending on the station, and subsequently decreased in 2010 to a number more closely comparable to the number of data levels (i.e., fixed pressure levels plus 10-m wind) in NARR. The merged raob soundings in the ESRL archive include a code for each data level, with ''4'' indicating a standard pressure level (currently 1000, 925, 850, 700, and 500 hPa in the lower and mid- troposphere), ''5'' indicating a ''significant'' level usually defined on the basis of temperature and/or humidity lapse rates but also for regular 50-hPa intervals in the 1980 and 1990 soundings, and ''6'' indicating a wind-only data level. The changes with time of the different codes provide some insight on the temporal variation in the total number of wind levels. For example, relative to 1980, the number of significant levels in 1990 that also included wind information increased and in addition wind-only data levels were included in the sounding. The decrease in wind data levels at Green Bay and Midland in 2000 corresponds to the significant levels in the archived sounding no longer including wind in- formation, whereas at the other stations the number of significant levels with wind information is similar to that in 1990. At all stations, the frequency of wind-only levels increased from 1990 to 2000. During 2010, wind in- formation was no longer included for significant levels and instead was reported only for standard pressure levels and wind-only levels.

c. Comparison procedures As a first step for comparing the raob and NARR wind profiles, the overall agreement of the lower- and mid- tropospheric wind speeds was evaluated for each obser- vation time, regardless of the presence of an LLJ. Raob wind speeds for 2000 and 2010 were linearly interpolated to regular 50-hPa levels to make them compatible with the archived levels in the raobs for 1980 and 1990. Dif- ferences in wind speed between the two data sources were calculated for each 50-hPa pressure level and ob- servation time, and box plots were created to display the differences for 0000 and 1200 UTC separately.

Next, the characteristics (frequency, speed, and ele- vation) of LLJs identified in the NARR and raob wind speed profiles were compared. The following criteria were used to extract LLJs for both datasets: 1) the jet nose (i.e., the elevation of the maximum wind speed) had to be located at or below 700 hPa, 2) wind speeds at the jet nose were $ 12 m s21, 3) the wind speed above the jet nose decreased by $6ms21 to the next minimum or to 550 hPa (whichever was lower), and 4) a wind speed decrease of $6ms21 was found below the jet nose. The thresholds for wind speed and the wind shear above the jet nose correspond to those initially employed by Bonner (1968). Although Bonner did not consider wind shear below the jet nose, this criterion was applied in several recent studies (e.g., Andreas et al. 2000; Banta et al. 2002; Walters and Winkler 2001; Walters et al. 2008)andis included here to facilitate comparison across the multiple LLJ definitions that have been employed in climatological analyses. The maximum elevation of the jet nose (700 hPa) follows that of Walters and Winkler (2001) and Walters et al. (2008) and recognizes that not all LLJs are bound- ary layer features but rather some are synoptically driven with often slantwise airflow (e.g., Uccellini and Johnson 1979). Similar to Bonner (1968), LLJs were further classified into intensity categories, with jets in category 1, category 2, category 3, and category 4 having jet speeds $12 m s21 (vertical speed shear $ 6ms21), $16 m s21 (vertical shear $ 8ms21), $20 m s21 (vertical shear $ 10 m s21), and $24 m s21 (vertical shear $ 12 m s21), respectively. LLJs were classified as southerly (northerly) if they originated between 1138 and 2478 (2938 and 678). All raob-derived jet frequencies were expressed as a percentage of available raobs for that location and time, given that the number of missing observations varied across stations and analysis years (6.7% of raobs were missing when summed across all stations and the four analysis years). NARR-derived jet frequencies, on the other hand, were calculated as a percentage of all observation times. Then, t tests for equality of the mean (assuming unequal variance) were conducted to assess the statistical significance of differences between the two data sources in jet speed and elevation.

3. Results a. Wind speed deviations A comparison of NARR and raob wind speeds at 50-hPa intervals in the lower and midtroposphere indicates considerable variability, with NARR wind speeds sub- stantially (up to 4-6 m s21) weaker than raob speeds at some observation times and stronger at other times. Large variability in the speed differences is present for all pressure levels, study locations, and years and for both 0000 and 1200 UTC.

Vertical profiles of the median value and range of the speed deviations at 1200 UTC are shown in Fig. 3 for the four analysis years combined. (Note that for some lo- cations, especially Amarillo, Fort Worth, and Green Bay, the small range of the wind speed deviations for the pressure level closest to the surface is a reflection of a smaller number of observations because the station pressure is frequently lower than this pressure level.) The raob wind speeds at the eastern stations (i.e., Green Bay, Springfield, Norman, Fort Worth) tend to be somewhat stronger than the NARR speeds at 900 hPa, as compared with the surrounding pressure levels, as seen by relative large positive median deviations. At Amarillo and Midland, small positive deviations are seen at 850 hPa relative to the surrounding pressure levels. For all these stations, the level with the largest positive median deviation is located approximately 50-100 hPa above the surface. Elsewhere, the median deviations are similar for all pressure levels, with the exception of Del Rio for which the negative median deviation at 900 hPa suggests somewhat larger wind speeds for NARR relative to raobs. At 0000 UTC, the median speed deviations at all stations hover near zero for all pressure levels, except for Amarillo and Brownsville (Fig. 4). At Amarillo, a small positive median deviation is again seen at 850 hPa, whereas at Brownsville the sign of the median speed deviation varies by level, with a negative value (NARR . raob) at 950 hPa and a pos- itive value (NARR , raob) at 800 hPa.

b. LLJ frequency LLJ frequencies, summed across all wind directions and the two observation times (0000 and 1200 UTC), are displayed by station in Fig. 5. In general, the jet frequen- cies obtained from the two datasets display a similar geographic pattern, with higher frequencies identified from both the NARR and raob wind profiles at stations in the central plains and southern Texas (e.g., Topeka, Amarillo, Norman, Fort Worth, and Brownsville) relative to elsewhere. The magnitude differs between the two datasets, however, with considerably smaller LLJ fre- quencies identified from the NARR wind profiles than the raob profiles. Exceptions, where the frequency of LLJs in the NARR profiles is higher than that for the raob profiles, are found at Brownsville during all anal- ysis years and at Del Rio in 2000. Another interesting feature is the larger frequency of LLJs in the raob wind profiles in 2010 than in the other analysis years, seen at all stations except Norman and Lake Charles. In contrast, NARR-derived LLJ frequencies increased by a much smaller amount, if at all, in 2010 relative to earlier years. When only stronger (category 2 or greater) LLJs are considered (not shown), NARR also underestimated LLJ frequency when compared with raobs, and a similar increase in jet frequency is seen in 2010 for the rawin- sonde profiles.

To investigate whether the increase in 2010 of raob- detected LLJs may be related to the RRS introduction, LLJs were extracted from raobs and NARR wind pro- files for all years from 2000 (the representative year for the MicroArt era) to 2010, and the annual LLJ fre- quency, regardless of jet direction or time of observation, was calculated (Fig. 6). At all study locations, an increase in raob-derived LLJ frequencies coincides roughly with the introduction of RRS, although the magnitude of the increase varies among locations. This result suggests that changes in protocols for identifying wind levels for in- clusion in the GTS message influence LLJ detection from raob profiles. The small increase or lack of increase in jet frequencies in the NARR profiles from 2000 to 2010 implies that the first-guess fields and the other data sourcesincorporatedintoNARRhelptominimizethe impact of inhomogeneities in the raob time series. The smaller number of LLJs detected from the raob profiles in 2010 at Norman and, to a lesser extent, at Lake Charles is intriguing, especially because this decrease occurs several years after the introduction of the RRS. An explanation for these differences is not obvious.

Differences in jet frequency between the two datasets were also analyzed by jet direction. For southerly LLJs (S-LLJs), large positive (raobs . NARR) differences are found at stations in the central plains (e.g., Fort Worth, Norman, Topeka), with smaller differences ob- served at surrounding stations, especially the northern stations (Fig. 7). This geographic variation is less pro- nounced in 2010, when, as compared with other years, smaller differences between the two datasets are seen in the central plains, especially at Norman and Fort Worth, but comparatively larger differences are found at most other stations. The negative (NARR . raob) difference at Norman in 2010 agrees with the lower frequency of raob-identified LLJs during that year (see discussion above). In addition, S-LLJ frequency is generally larger for NARR than for raobs at the southernmost stations. Negative deviations are found at Brownsville for all four study years but with a particularly large deviation in 2000, at North Platte in 1980, and at Del Rio and Lake Charles in 2000. For the majority of stations, northerly LLJs (N-LLJs) are somewhat less frequent in the NARR than in the raob wind profiles during all four years, with slightly larger differences observed in 2010. The modest differences in N-LLJs between the two datasets, as compared with the larger differences observed for S-LLJs, reflect the relatively infrequent occurrence of N-LLJs at the study locations. Note that NARR un- derestimates the raob-derived frequency of N-LLJs at Brownsville, in contrast to the large overestimation seen for S-LLJs. Another interesting feature is the modest overestimation of N-LLJs in NARR at Lake Charles during three of the four analysis years.

For S-LLJs only, we also investigated whether differ- ences between the two data sources varied by time of day (Fig. 8). A similar analysis was not completed for N-LLJs because of the relatively infrequent occurrence of jets from this direction. At 0000 UTC, S-LLJ frequencies de- rived from NARR are smaller than those from raobs, with greater differences (especially at Del Rio) observed in 2010 than in the other analysis years. An exception to this general pattern is the large negative (NARR . raobs) deviation in jet frequency in every year except 2010 at Brownsville. At 1200 UTC, the largest positive (raobs . NARR) deviations in S-LLJ frequency are found at the stations in northern Texas and Oklahoma, except in 2010 when smaller differences are observed at Amarillo and a positive deviation is observed at Norman. At Brownsville, much smaller differences are seen between NARR- and raob-derived S-LLJ frequencies at 1200 UTC than at 0000 UTC. This agrees with the somewhat stronger (weaker) NARR low-level wind speeds at 0000 UTC (1200 UTC) in the mean wind profiles for this location as shown above (Figs. 3 and 4). In contrast, the greater frequency of S-LLJs in the NARR wind profiles at Norman during 2010 is only evident at 1200 UTC.

Numerous studies have shown that S-LLJs in the study area display a distinct diurnal cycle with higher frequencies in the early-morning hours (e.g., Bonner 1968; Arritt et al. 1997; Whiteman et al. 1997; Walters et al. 2008). To evaluate whether the jets extracted from the NARR wind profiles display this known climato- logical characteristic, the relative frequency of S-LLJs at 1200 UTC was calculated as a percentage of the fre- quency of LLJs at 0000 and 1200 UTC combined (Fig. 9). As for raobs, S-LLJs detected from the NARR wind profiles are more frequent at 1200 UTC than at 0000 UTC, although NARR depicts a somewhat stronger diurnal cycle at most stations. Exceptions occur at Brownsville and Lake Charles where a weaker diur- nal cycle is seen, in line with the negative deviation (NARR . raobs) in S-LLJ frequency at 0000 UTC for these two stations. Diurnal variations in N-LLJ frequency are also similar for the two datasets, with both datasets indicating a greater proportion of N-LLJs occurring at 1200 UTC than 0000 UTC.

To better understand the differences in jet frequency observed above, NARR and raob wind profiles with the same date and time stamp were compared. For this portion of the analysis, only those observation times were included for which rawinsonde soundings were available. Overall, LLJs were concurrently identified in both the NARR and raob wind profiles anywhere from 33% to 57% of the observation times when an LLJ was present in either one or both of the datasets. Although some differences are evident among the four analysis years (not shown), larger variations in the frequency of concurrent LLJs exist by jet direction, time of day, and station (Fig. 10). In general, S-LLJs at 1200 UTC are more likely to occur concurrently in NARR and raob wind profiles, particularly for the study locations in the central and southern plains, whereas for N-LLJs and for S-LLJs at 0000 UTC, a jet is more often evident in the raob wind profile but not in the corresponding NARR wind profile. Much less often, an LLJ is identified in the NARR wind profile but not in the raob profile. Excep- tions occur at Brownsville, and to a lesser extent at Lake Charles, where, as noted above, negative (NARR . raobs) deviations in S-LLJ frequency at 0000 UTC are observed.

Because typical definitions of LLJs employ threshold values for both wind speed and wind shear, we also assessed the ''limiting factor'' preventing a NARR wind profile from being classified as a jet when an LLJ was identified in the concurrent raob profile (Fig. 11). For S-LLJs at 0000 UTC, NARR low-level wind speeds exceeded 12 m s21 on more than 50% of the times when an S-LLJ was identified in the raob profile, implying that wind shear was the factor most often preventing a NARR profile from being classified as an S-LLJ. This pattern is seen at all stations. For S-LLJs at 1200 UTC, wind speed was typically the limiting factor at Bismarck, North Platte, Midland, Fort Worth, Del Rio, and Brownsville, whereas the lack of sufficient wind shear was the constraining factor at the remaining stations. At almost all locations, wind shear was most often the constraining factor for N-LLJs.

c. LLJ average speed and intensity category In general, the average speed of LLJs identified from NARR is weaker than that of raob-identified LLJs, al- though the differences are often not significant. For S-LLJs (Table 1), statistically significant (at the 95% confidence level) differences in mean speed are found in most analysis years only at Amarillo and Midland, and only for the 1200 UTC observation time. Elsewhere, with few exceptions, nonsignificant differences in S-LLJ mean speed are observed for both 0000 and 1200 UTC and for all four years. Because of the larger sample size, more stations have significant differences when S-LLJ mean wind speed is calculated across the four analysis years, although the number is still modest, with significant differences seen at four stations (North Platte, Springfield, Amarillo, and Midland) at 1200 UTC and two stations (Bismarck and Brownsville) at 0000 UTC. The wind speed analysis for N-LLJs was not separated by time of day, because of the relatively small number of jets from this direction. When N-LLJs from both observation times are collectively considered, significant differences are few for any specific analysis year, but seven of the locations (Topeka, Amarillo, Norman, Fort Worth, Del Rio, Brownsville, and Lake Charles) display significant differences between NARR and raobs when N-LLJ wind speed is calculated across all four years (Table 2).

Using concurrent jets only, we also investigated the agreement in jet intensity category between the two datasets. A one-by-one comparison of concurrent S-LLJs, when summed across the four analysis years, suggests that at most stations the majority of the raob- and NARR-identified jets fall within the same intensity cat- egory (Fig. 12), although differences are seen by station and by time of day. For example, at 1200 UTC, a larger proportion of concurrent S-LLJs fall within the same in- tensity category at Midland, Del Rio, Brownsville, Fort Worth, and Lake Charles relative to more northern sta- tions. When the intensity category differs, the S-LLJ in the NARR wind profile is usually weaker than the con- current jet identified from raobs. An exception occurs for 0000 UTC S-LLJs at Brownsville, where a large number of NARR-identified LLJs have a higher intensity cate- gory than the corresponding raob jet. For N-LLJs, when summed across all analysis years and the two observation times, the agreement in intensity category is generally greater for the easternmost locations than for those lo- cated farther west.

d. LLJ average elevation In general, the elevations of S-LLJs identified from the NARR wind profiles are lower (i.e., closer to the surface) than those identified from raobs, although most differences are not statistically significant. For S-LLJs at the 1200 UTC observation time, stations with significant differences for two or more of the analysis years include Bismarck, Amarillo, Norman, and Brownsville, although at Amarillo the NARR-identified jets are on average located at a higher elevation than those in the raob wind profiles (Table 3). At 0000 UTC, significant differences in S-LLJ elevation for more than one analysis year are found only at Brownsville, where NARR S-LLJs are lo- cated on average over 300 m closer to the surface than raob S-LLJs. Similarly, most differences in N-LLJ ele- vation are not significant (Table 4). Significant differ- ences for two or more of the analysis years are only found at North Platte, Norman, and Lake Charles; for all three locations, the NARR-identified LLJs are located closer to the surface. The number of locations with significant elevation differences is largest in 2010 when NARR N-LLJs at four locations are located closer to the ground relative to the raob N-LLJs.

4. Discussion An initial assumption was that changes with time in the vertical resolution of raob wind profiles, in contrast to the fixed vertical resolution of NARR profiles, would have a substantial influence on the agreement between the two data sources with respect to LLJ frequency and characteristics. In particular, we assumed that LLJs would be detected in raobs more frequently during those ''eras'' of the record with a larger number of wind measurement levels, leading to a greater discrepancy in jet frequency between raobs and NARR. These as- sumptions were not confirmed by the analyses. Instead, the frequency of LLJs in the raob wind profiles, and the discrepancy with the NARR frequencies, was largest in 2010, the representative year for the RRS era, even though the number of levels with wind information in the NARR and raob wind profiles was similar. These findings suggest that it is not so much the vertical reso- lution but rather the protocols by which wind levels are selected for inclusion in the GTS message that influence the detection of LLJs in raobs and, as a consequence, the agreement between NARR and raob jet frequency. When wind-only levels were introduced into the GTS message with the MiniART and MicroART systems, the wind information was reported in the United States at fixed 305-m (1000 ft) intervals. Wind levels in the GTS message for the RRS are selected in an iterative process that identifies the levels of greatest departure from the linearly interpolated wind speed over increasingly small segments of the sounding (National Weather Service 2014), however, similar to how thermodynamically sig- nificant levels are selected. The RRS procedures appear to better identify fluctuations with height in the vertical profiles of wind speed, leading to higher LLJ frequencies.

Other examples of an increase in LLJ frequency de- spite a decrease in the number of vertical levels with wind information in the rawinsonde soundings occur at Green Bay and Midland in 2000, the representative year for the MicroART era. Significant levels (i.e., those levels coded as a ''5'' in the archived raobs) for this analysis year no longer included wind speed and di- rection, resulting in a decrease in the number of raob levels with wind information by close to one-half of those in 1990 (the representative year for the MiniART era). Wind information continued to be reported for significant levels at the other study locations, yet the discrepancy in the frequency of raob- and NARR-identified LLJs at Green Bay and Midland in 2000 is as large as or larger than that at the other locations. One might infer that in 2000 these two stations were already using procedures similar to those of the RRS era to identify wind-only levels, but this explanation is incomplete because a sub- stantial increase in LLJ frequency is seen at both stations after the introduction of the RRS. Observation and re- porting procedures at individual raob locations are unfortunately not well documented (M. Govett 2012, personal communication).

Differences in LLJ speed between the two datasets can also affect LLJ detection, especially because widely used jet definitions, including the one used here, employ fixed speed thresholds. We initially assumed that NARR wind speeds would on average be weaker than raob speeds and less likely to meet the criteria for LLJs be- cause the NARR wind speeds are areal averages for grid cells, which would likely smooth local speed maxima, and because additional smoothing was applied during the interpolation of the NARR gridpoint values to the rawinsonde locations. The difference in wind speed for paired observation times, regardless of whether an LLJ was present, provided some initial credence to this contention. For several stations, NARR wind speeds at 1200 UTC were weaker than raob speeds for elevations at which low-level jet profile signatures are normally found (i.e., 50-100 hPa above the surface). The time of day (i.e., 1200 UTC) and the relatively low elevation of these deviations suggest that NARR may not fully cap- ture the supergeostrophic winds commonly seen above the boundary layer in the central plains at night and in the early morning (e.g., Blackadar 1957; Bonner et al. 1968; Bonner and Paegle 1970), although a more detailed anal- ysis is necessary to confirm this conjecture. Further cre- dence that weaker low-level winds in the NARR profiles contribute to an underestimation of jet frequency at 1200 UTC is provided by the analysis of the ''limiting factors'' of the jet definition. At 1200 UTC, the maximum speeds for stations in the central and southern plains in the NARR wind profile did not meet the critical threshold (12 m s21) for the majority of times when an S-LLJ was observed in the raob wind profile but not in the NARR profile. On the other hand, the NARR jet frequencies were also smaller than raob frequencies at 0000 UTC, even though the mean speed deviations were close to zero for all pressure levels and most locations. At this observation time, it appears that the coarser vertical resolution, and hence smoother wind profiles, contribute to the lower jet frequency, with the NARR profiles often not meeting the wind shear criteria of the jet definition.

The analyses presented here also highlight several ''peculiarities'' at individual locations. In 2010 (the RRS representative year), NARR substantially overestimated (relative to raobs) S-LLJ frequency at Norman for 1200 UTC but not for 0000 UTC or for N-LLJs. This dif- ference was not evident at surrounding stations or during the other analysis years. A related inconsistency at Norman is the large decrease in LLJ frequency in the raob wind profiles from 2009 to 2010, which is also not seen at other study locations. An explanation for these differences is not obvious, and investigation of potential causes is handi- capped by the lack of detailed information for individual stations in the metadata for the rawinsonde network. NARR-derived LLJ frequencies were also larger than raob frequencies at Brownsville, but for this location the largest deviations were observed for S-LLJs at 0000 UTC and primarily for the manual MiniART and MicroART eras. A possible explanation is differences between NARR and raobs in the detection of sea- breeze circulation at this coastal location. In support of this contention, the majority of 0000 UTC S-LLJs at Brownsville, whether identified from the NARR or from raob wind profiles, had onshore wind directions between 1608 and 1658 (not shown). It may be that the strength of the sea-breeze circulation was overestimated by NARR in comparison with raobs, or, as an alternative, that the fixed wind-only levels in the early raob record were insufficient to detect sea-breeze circulation, in contrast to the wind- only levels in the more recent RRS profiles that are se- lected using a more sophisticated algorithm.

LLJs were concurrently observed on only about 30%- 60% of the observation times that a jet was identified in either the raob or NARR datasets. The relatively low frequency of concurrent LLJs is not surprising given that the jet definition used here is composed of fixed thresh- olds for the speed at the jet nose and the amount of shear above and below the jet nose. Small departures from these thresholds can result in a wind profile not being classified as having an LLJ. It is encouraging that the proportion of concurrent jets was greatest for the more frequent 1200 UTC S-LLJs. On the other hand, the NARR wind profiles often did not capture the less fre- quent N-LLJs seen in the raob wind profiles. The fact of the relatively infrequent concurrent LLJs in the NARR and raob datasets imposes some limitations on the use of NARR for LLJ case-study analysis and for utilizing NARR as ''truth'' for evaluating simulations from nu- merical models such as regional climate models.

A surprising finding was the between-station differ- ences in the agreement of NARR and raobs for several of the LLJ characteristics. The variations between sta- tions were often considerably larger than the differences between the rawinsonde eras. It is difficult to tease out without considerable additional analysis whether these between-station differences primarily reflect differences in observing and reporting protocols among the study locations, geographic variations in the ability of the NARR Eta Model to simulate lower-tropospheric air- flow, or some combination of both factors. These between- station differences, in addition to the differences between rawinsonde eras, complicate any attempts to modify clas- sical jet definitions to accommodate differences between NARR and raob wind profiles.

Although our findings indicate that, with the excep- tion of Brownsville and Norman, LLJ frequency derived from the NARR wind profiles is smaller than that de- rived from the raob profiles, we feel that, when inter- preted cautiously and used judiciously, there remains considerable potential for NARR-based climatological analyses of LLJs. Relative (rather than absolute) jet frequencies for different stations and time periods were generally similar for NARR and raobs. Also, the dif- ferences between the two datasets in the diurnal varia- tions, speed, and elevation of LLJs were modest. Both datasets displayed a marked diurnal cycle in jet fre- quency with more LLJs at 1200 than at 0000 UTC. Furthermore, differences in LLJ speed and elevation between the two datasets were insignificant for most sta- tions and rawinsonde eras. A potential advantage of using NARR for LLJ climatological studies is that the jet frequencies appear to be less influenced by changes in observing and archiving practices, such as the RRS in- troduction, when compared with raobs. This provides an opportunity to investigate temporal trends in LLJ fre- quency, the study of which is currently hampered by inhomogeneities in the raob time series.

Application of the results of this analysis to other fine- resolution reanalysis datasets is limited by the differ- ences in the data-assimilation systems and underlying forecast models for the different reanalyses. Our results indicate that careful evaluation is needed before utiliz- ing a reanalysis dataset for climatological studies of low- level wind maxima. Reanalysis-based climatological analyses of LLJs need to be interpreted cautiously, just as the interpretation of raob-based jet climatological analyses needs to consider the many inhomogeneities in the original time series.

5. Conclusions The recent development of reanalysis datasets with increasingly finer temporal and spatial resolution has made new data sources available for climatological analyses. Long-term studies of low-level jets are par- ticularly in need of alternative data sources, because the rawinsonde record is complicated by multiple changes in observing and archiving protocols that have introduced inhomogeneities into the time series of vertical wind observations. Furthermore, LLJ climatological analyses are limited by the coarse spatial and temporal resolution of raobs. With these concerns in mind, this study in- vestigated whether the North American Regional Re- analysis is a viable additional resource for studying the climatological characteristics of LLJs.

The comparison of NARR-identified LLJs with raob- identified LLJs for four years (1980, 1990, 2000, and 2010) that represent so-called eras with different ra- winsonde observing and reporting protocols and at 12 stations (Bismarck, Green Bay, North Platte, Topeka, Springfield, Amarillo, Norman, Midland, Fort Worth, Del Rio, Brownsville, and Lake Charles) across the central United States showed that NARR captures the general climatological features of LLJs, including dif- ferences among locations in relative LLJ frequency and diurnal variations in jet frequency. In addition, differ- ences in jet speed and elevation between NARR and raobs were insignificant for many locations. LLJ fre- quency was lower for NARR than for raobs at most stations, however. Also, on fewer than 60% of the times when an LLJ was observed in one of the datasets was an LLJ found concurrently in the other dataset, although the agreement was greater for southerly jets than for northerly jets. Therefore, it is advisable that NARR wind fields be used cautiously for LLJ case-study anal- yses or validation of numerical model simulations of wind maxima.

In addition, some intriguing differences were seen by analysis year and location. In particular, LLJs were identified in the raob profiles considerably more fre- quently after the introduction of the Radiosonde Re- placement System than when the earlier rawinsonde systems were in effect. A similar increase in jet fre- quency was not seen in the NARR wind profiles, sug- gesting that NARR is the more appropriate of the two datasets to assess trends in LLJ frequency. Between- station differences arise because of limitations of either NARR or raobs. For example, the lower S-LLJ fre- quencies for NARR at 1200 UTC in the central and southern plans suggests that NARR imperfectly simu- lates supergeostrophic airflow in this area, whereas the large discrepancy between NARR and raobs in terms of jet frequency in 2010 at Norman is likely due to the anomalously small number of LLJs observed in the raob wind profiles for this particular year. Further work is needed to explain the large frequency of S-LLJs in the NARR 0000 UTC wind profiles at Brownsville.

Our analyses suggest that, depending on the applica- tion, NARR is a viable additional resource for clima- tological analyses of low-level wind maxima. A careful evaluation with rawinsonde observations is a necessary initial step, however, and the interpretation of the en- suing jet climatological analysis must consider temporal and spatial variations in the agreement of jet charac- teristics as identified from NARR and raob wind pro- files. We expect many of our findings to also be applicable to other fine-resolution reanalysis datasets, although differences between reanalyses in terms of their data- assimilation systems and underlying forecast models re- quire that they be individually evaluated before their use for climatological investigations of wind maxima.

Acknowledgments. We thank the anonymous re- viewers for their helpful suggestions and comments. We also thank Xindi Bian for his help with extracting the low-level jets from the North American Regional Reanalysis. This research was supported by the National Science Foundation under Grants BCS-0924768 and BCS-0924816. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the National Science Foundation.

REFERENCES Anderson, C., and R. Arritt, 2001: Representation of summertime low-level jets in the central United States by the NCEP- NCAR reanalysis. J. Climate, 14, 234-247, doi:10.1175/ 1520-0442(2001)014,0234:ROSLLJ.2.0.CO;2.

Andreas, E. L, K. J. Claffy, and A. P. Makshtas, 2000: Low-level at- mospheric jets and inversions over the western Weddell Sea. Bound.-Layer Meteor., 97, 459-486, doi:10.1023/A:1002793831076.

Arritt, R., T. Rink, M. Segal, D. Todey, and C. Clark, 1997: The Great Plains low-level jet during the warm season of 1993. Mon. Wea. Rev., 125, 2176-2192, doi:10.1175/1520-0493(1997)125,2176: TGPLLJ.2.0.CO;2.

Banta, R., R. Newsom, J. Lundquist, and Y. Pichugina, 2002: Nocturnal low-level jet characteristics over Kansas during CASES-99. Bound.-Layer Meteor., 105, 221-252, doi:10.1023/ A:1019992330866.

Blackadar, A. K., 1957: Boundary layer wind maxima and their significance for the growth of nocturnal inversions. Bull. Amer. Meteor. Soc., 38, 283-290.

Bonner, W. D., 1968: Climatology of the low-level jet. Mon. Wea. Rev., 96, 833-850, doi:10.1175/1520-0493(1968)096,0833: COTLLJ.2.0.CO;2.

_____, and J. Paegle, 1970: Diurnal variations in boundary layer winds over the south-central United States in summer. Mon. Wea. Rev., 98, 735-744, doi:10.1175/1520-0493(1970)098,0735: DVIBLW.2.3.CO;2.

_____, S. Esbensen, and R. Greenberg, 1968: Kinematics of the low-level jet. J. Appl. Meteor., 7, 339-347, doi:10.1175/ 1520-0450(1968)007,0339:KOTLLJ.2.0.CO;2.

Bosart, L. F., 1990: Degradation of the North American radio- sonde network. Wea. Forecasting, 5, 527-528, doi:10.1175/ 1520-0434(1990)005,0527:DOTNAR.2.0.CO;2.

Carbone, R. E., and J. D. Tuttle, 2008: Rainfall occurrence in the U.S. warm season: The diurnal cycle. J. Climate, 21, 4132- 4146, doi:10.1175/2008JCLI2275.1.

Daniel, C. J., R. W. Arritt, and C. J. Anderson, 1999: Accuracy of 404-MHz radar profilers for detection of low-level jets over the central United States. J. Appl. Meteor., 38, 1391-1396, doi:10.1175/1520-0450(1999)038,1391:AOMRPF.2.0.CO;2.

Elliott, W. P., and D. J. Gaffen, 1991: On the utility of radiosonde humidity archives for climate studies. Bull. Amer. Meteor. Soc., 72, 1507-1520, doi:10.1175/1520-0477(1991)072,1507: OTUORH.2.0.CO;2.

Helfand, H., and S. Schubert, 1995: Climatology of the simu- lated Great Plains low-level jet and its contribution to the continental moisture budget of the United States. J. Cli- mate, 8, 784-806, doi:10.1175/1520-0442(1995)008,0784: COTSGP.2.0.CO;2.

Higgins, R., Y. Yao, E. Yarosh, J. Janowiak, and K. Mo, 1997: In- fluence of the Great Plains low-level jet on summertime pre- cipitation and moisture transport over the central United States. J. Climate, 10, 481-507, doi:10.1175/1520-0442(1997)010,0481: IOTGPL.2.0.CO;2.

Li, X., S. Zhong, X. Bian, and W. E. Heilman, 2010: Climate and climate variability of the wind power resources in the Great Lakes region of the United States. J. Geophys. Res., 115, D18107, doi:10.1029/2009JD013415.

McGrath, R., T. Semmler, C. Sweeney, and S. Wang, 2006: Impact on balloon drift errors in radiosonde data on climate statistics. J. Climate, 19, 3430-3442, doi:10.1175/ JCLI3804.1.

Mesinger, F., G. DiMego, E. Kalnay, and K. Mitchell, 2006: North American Regional Reanalysis. Bull. Amer. Meteor. Soc., 87, 343-360, doi:10.1175/BAMS-87-3-343.

Mitchell, M., R. Arritt, and K. Labas, 1995: A climatology of the warm season Great Plains low-level jet using wind profiler observations. Wea. Forecasting, 10, 576-591, doi:10.1175/ 1520-0434(1995)010,0576:ACOTWS.2.0.CO;2.

Mo, K., M. Chelliah, M. Carrera, and R. Higgins, 2005: Atmo- spheric moisture transport over the United States and Mexico as evaluated in the NCEP regional reanalysis. J. Hydrome- teor., 6, 710-728, doi:10.1175/JHM452.1.

Moore, G. W. K., R. S. Pickart, and I. A. Renfrew, 2008: Buoy observations from the windiest location in the World Ocean, Cape Farewell, Greenland. Geophys. Res. Lett., 35, L18802, doi:10.1029/2008GL034845.

Muñoz, E., A. Busalacchi, S. Nigam, and A. Ruiz-Barradas, 2008: Winter and summer structure of the Caribbean low-level jet. J. Climate, 21, 1260-1276, doi:10.1175/2007JCLI1855.1.

National Weather Service, cited 2014: Radiosonde Replacement Sys- tem WMO levels generation for coded messages. NOAA/NWS Tech Doc., 5 pp. [Available online at http://www.ua.nws.noaa.gov/ Documents/RRS-Coded-Messages.pdf.] Schwartz, B. E., and C. A. Doswell III, 1991: North American ra- winsonde observations: Problems, concerns, and a call to ac- tion. Bull. Amer. Meteor. Soc., 72, 1885-1886, doi:10.1175/ 1520-0477(1991)072,1885:NAROPC.2.0.CO;2.

_____, and M. Govett, 1992: A hydrostatically consistent North Ameri- can radiosonde database at the Forecast Systems Laboratory, 1946 to present. NOAA Tech. Memo. ERL FSL-4, 81 pp. [Available online at http://www.esrl.noaa.gov/raobs/intl/radiosonde.pdf.] Shafran, P. C., J. Woollen, W. Ebisuzaki, W. Shi, Y. Fan, R. W. Grumbine, and M. Fennessy, 2004: Observational data used for assimilation in the NCEP North American Regional Reanalysis. Preprints, 14th Conf. on Applied Climatology, Seattle, WA, Amer. Meteor. Soc., 1.4. [Available online at https://ams.confex.com/ams/pdfpapers/71689.pdf.] Uccellini, L. W., and D. R. Johnson, 1979: The coupling of upper and lower tropospheric jet streaks and implications for the development of severe convective storms. Mon. Wea. Rev., 107, 682-703, doi:10.1175/1520-0493(1979)107,0682: TCOUAL.2.0.CO;2.

Walters, C. K., and J. A. Winkler, 2001: Airflow configurations of warm season southerly low-level wind maxima in the Great Plains. Part I: Spatial and temporal characteristics and re- lationship to convection. Wea. Forecasting, 16, 513-530, doi:10.1175/1520-0434(2001)016,0513:ACOWSS.2.0.CO;2.

_____, _____, R. P. Shadbolt, J. van Ravensway, and G. D. Bierly, 2008: A long-term climatology of southerly and northerly low- level jets for the central United States. Ann. Assoc. Amer. Geogr., 98, 521-552, doi:10.1080/00045600802046387.

Weaver, S. J., and S. Nigam, 2011: Recurrent supersynoptic evo- lution of the Great Plains low-level jet. Bull. Amer. Meteor. Soc., 24, 575-582.

_____, S. Schubert, and H. Wang, 2009: Warm season variations in the low-level circulation and precipitation over the central United States in observations, AMIP simulations, and ideal- ized SST experiments. J. Climate, 22, 5401-5420, doi:10.1175/ 2009JCLI2984.1.

Whiteman, C. D., X. Bian, and S. Zhong, 1997: Low-level jet cli- matology from enhanced rawinsonde observations at a site in the southern Great Plains. J. Appl. Meteor., 36, 1363-1376, doi:10.1175/1520-0450(1997)036,1363:LLJCFE.2.0.CO;2.

Winkler, J. A., 2004: The impact of technology upon in situ at- mospheric observations and climate science. Geography and Technology, S. D. Brunn, S. L. Cutter, and J. W. Harrington Jr., Eds., Springer, 461-490.

CLAUDIA K. WALTERS Department of Social Sciences, University of Michigan-Dearborn, Dearborn, Michigan JULIE A. WINKLER Department of Geography, Michigan State University, East Lansing, Michigan SARA HUSSEINI AND RYAN KEELING Department of Social Sciences, University of Michigan-Dearborn, Dearborn, Michigan JOVANKA NIKOLIC AND SHIYUAN ZHONG Department of Geography, Michigan State University, East Lansing, Michigan (Manuscript received 7 November 2013, in final form 24 April 2014) Corresponding author address: Claudia K. Walters, Dept. of Social Sciences, University of Michigan-Dearborn, 4901 Ever- green Rd., Dearborn, MI 48128.

E-mail: [email protected] (c) 2014 American Meteorological Society

[ Back To TMCnet.com's Homepage ]