1. Introduction
The trend in operational numerical weather prediction (NWP) since the implementation of the six-layer primitive equation model (Shuman and Hovermale 1968) has been toward improving horizontal resolution and the attendant model physics. In 1985, the forecast component of the National Centers for Environmental Prediction (NCEP, formerly NMC) Regional Analysis and Forecast System (RAFS; Hoke et al. 1989), the Nested Grid Model (NGM), was implemented, with an increase in horizontal resolution in interior domains to approximately 84 km from the 127-km resolution of the Limited-Area Fine Mesh Model (LFM). One benefit of such resolution increases was a significant improvement in forecasts of cyclogenesis (e.g., Sanders 1986a; Sanders 1987; Oravec and Grumm 1993). Further increases in resolution have come about with the recent introduction of the Eta and meso–Eta Models (Black 1994).
With this history of NWP model enhancement in mind, one might expect concomitant increases in forecast skill. Figure 1a shows the time history of the 500-hPa geopotential height skill score for 36-h predictions (adapted from Shuman 1989) for North America and adjacent waters for the period 1967–88 and the 24–36-h forecast skill with respect to climatology of the probability of measurable precipitation (POP) at approximately 100 U.S. stations (Carter and Polger 1986) over this same period. There is a strong upward trend in the 500-hPa forecast skill, amounting to an approximate 2% increase per year over the 22-yr span, affirming our expectations as far as advances in the prediction of the large-scale circulation pattern and attendant mass fields are concerned. However, POP skill shows a considerably slower growth over this same period (less than 0.7% per year), with no significant gains following the introduction of the higher-resolution NGM. Figure 1b shows National Weather Service (NWS) threat scores for day 1 (12–36 h) quantitative precipitation forecasts (QPF) for thresholds of 12.5 mm (0.5 in.), 25 mm (1.0 in.), and 50 mm (2.0 in.) for the period 1980–95 (adapted from Olson et al. 1995 with subsequent updates). These data indicate that the growth in QPF skill for this forecast range has been relatively slow in comparison to the rapid advances in circulation forecasts.
While numerous investigators have documented the difficulty of the precipitation forecast problem (e.g., Bosart 1983; Glahn 1985; Sanders 1986b; Junker et al. 1989; Olson et al. 1995), it was Anthes (1983) who suggested that a fundamental limit to significant improvements in regional predictions results from the increasingly “stochastic” nature of smaller-scale phenomena. It seems likely that much of the relatively slow advance in precipitation forecast skill to date can be attributed to the inability of the past and current suite of operational models to explicitly simulate mesoscale precipitation systems, a prominent example of which are the banded-type structures of winter cyclones (Moore and Blakley 1988; Shields et al. 1991; Ramamurthy et al. 1993). Roebber et al. (1994) provide an example of how the relative success in the prediction of cyclogenesis that has come about with improved model resolution does not necessarily translate immediately into improved QPF. In that sense, one can reasonably expect that the “stochastic limit” has not yet been reached and that further increases in resolution will lead to some additional gains in precipitation forecast skill. Recent experience with higher-resolution operational models suggests this to be the case. Mesinger (1996) compared QPF between the meso–Eta Model (with 48-km horizontal resolution) and the 80-km Eta and NGM and found that the higher-resolution model showed gains in threat score with respect to the coarser-resolution models. As operational models move into the nonhydrostatic mode and the explicit representation of mesoscale precipitation systems becomes possible, it will be necessary to gain some understanding of the limits to further progress that may be encountered.
This issue is explored in the present study through the use of regional analogs. By restricting the search to regional analogs (see section 2), we avoid the problem identified by Lorenz (1969): good hemispheric analogs do not occur within the time constraints imposed by available observational data. Since we are concerned with cyclone timescales (order 48 h), the rapid (within 4–5 days) degradation of good 500-hPa regional analogs discussed by Gutzler and Shukla (1984) will not unduly impact this study. Requiring analogy in the vertical structure of the mass field between two flows insures that the quasigeostrophic forcing of ascent (through the Laplacian of temperature advection and differential vorticity advection) is comparable, while circulation fidelity between those flows over a period of several days suggests that the history (and consequently the moisture content) of the air parcels is similar. It is important to note that this latter point is less robust, since moisture sources can be local (e.g., through evaporation of soil moisture). This issue will be addressed in section 3.
Thus, this approach allows us to study the sensitivity of the amount and distribution of precipitation to relatively small changes in the mass/circulation field by considering a range of cases with similar synoptic-scale structure. In this paper, we will study the correspondence in sensible weather between a base case, representing a very ordinary lee cyclogenesis event, and good time-coherent analogs, with special emphasis on a selected case. As such, the analysis should provide some insight into the problems governing typical precipitation forecast situations.
We shall address the regional analog selection procedure in section 2. Section 3 will present the analyses of the base case and the regional analogs and discuss the physical factors leading to precipitation differences. Figure 2 provides a locator map for places identified in the text. In a subsequent paper, we shall perform detailed diagnostics on mesoscale model simulations of the base case and a selected case in an effort to better quantify these processes. Section 4 provides a concluding discussion of our results.
2. Analog selection procedure
On 14 February 1992, lee cyclogenesis occurred in eastern Colorado; the cyclone subsequently tracked southeastward toward Kansas and Oklahoma (acquiring moisture in the southerly flow from the Gulf of Mexico) and then northeast toward the lower Great Lakes (Figs. 3a and 3c). Hereafter, we shall refer to this event as the base case. Precipitation analyses, based upon hourly data obtained from the National Climatic Data Center (NCDC) TD-3240 precipitation dataset, were constructed for this case and are shown in Figs. 4a,b. This dataset contains considerably more station observations than are available from the standard synoptic reporting network and thus allow better resolution of smaller-scale precipitation features.
Precipitation totals for this case were predominately in the 10–25-mm range (with some locally higher amounts) in the midwest, southeastern United States, and lower Great Lakes region during the 48 h of interest. Precipitation during the 24-h period from 0000 UTC 14–15 February was dominated by convection in southern Missouri and northeastern Arkansas (Fig. 4a). The next 24 h resulted in two main precipitation belts: a largely stable precipitation zone to the northeast of the cyclone center and a broad swath of convective precipitation along a line from Athens, Georgia (72311; AHN), to LaFayette, Louisiana (Fig. 4b). The base case is in no way exceptional in terms of cyclone development or in the severity of the weather associated with it. Rather, it corresponds to the climatological norm for winter/early spring in this region and as such represents a typical forecast situation. For example, the 20-yr climatology of Whittaker and Horn (1982) shows an average of 3.3 February–March cyclogenesis events per year in the region bounded by 35°–40°N and 100°–110°W.
The next step was to examine the candidate analogs for correspondence with the base case throughout the 48 h of primary interest (beginning at the initial time of 0000 UTC 14 February 1992). This correspondence was quantified using AC scores computed at both levels every 12 h in a 9 × 9 gridpoint region centered on and moving with the position of the base case surface cyclone. Since the objective of this exercise was to identify cases that had established synoptic-scale correspondence for an extended period (i.e., beyond 1 day), only cases that maintained AC scores of 0.5 or better for at least 36 h at either level were retained. This step led to a further reduction to a total of three cases that could be considered to be good regional, time-coherent analogs (these final cases were also subjectively judged to verify the results of the analysis). Each of the analogs so identified are listed in Table 1, while further discussion of these cases is presented in section 3.
3. Analysis of observations: Base case versus regional analogs
Figure 5 provides the time history of surface and 500-hPa AC for each of the identified regional analogs (section 2) relative to the base case. In all of these cases, the surface correspondence is good for at least 36 h; however, in only one case was there consistent correspondence at the surface and aloft throughout the full 48 h (20–22 February 1980). Consequently, when we turn to a more detailed analysis of the distinctions between the base case and the analogs below, we will focus our efforts on this particular case.
Figure 6 shows the composite sea level pressure and 500-hPa geopotential height fields for the analogs at 24 h into the 48-h period of correspondence with the base case (see Fig. 3a). The correspondence of the regional flow between the analog composite and the base case is striking and represents, from a forecast point of view, situations which might be considered to result in quite similar sensible weather across the midwestern United States at that time. Precipitation analyses were constructed for the composite from the NCDC TD-3240 hourly precipitation dataset for the period analogous to 0000 UTC 14–15 February 1992 (Fig. 4c). The precipitation differences between the base case (Fig. 4a) and the composite analog are extraordinary: essentially, no precipitation fell in the composites in association with the lee cyclogenesis. A similar analysis was constructed for the period corresponding to 0000 UTC 15–16 February 1992 and is shown in Fig. 4d. Here, we find that the composite precipitation pattern takes on a broad form reminiscent of that of the base case (Fig. 4b) with a swath of precipitation to the north of the cyclone track extending from Iowa to southern Ontario and a second, convective band across the southeastern United States. However, the details of the precipitation distributions vary considerably with somewhat less composite precipitation occurring in the north and substantially reduced precipitation in the southeast.
The individual analog of 16–18 February 1986 is consistent with the composite patterns shown in Figs. 6 and 4c–d. On the day prior to the 48-h period of analogy, a 500-hPa short wave was approaching the West Coast with an associated landfalling extratropical cyclone generating rain along the Pacific coast of the United States. At this time, conditions over the Midwest were clear and dry under anticyclonic 500-hPa flow and surface high pressure (in the base case at this time, large-scale conditions were similar except that an eastward propagating 500-hPa short wave was producing rain over the Midwest and the Gulf Coast). On the first day of circulation analogy, a lee cyclone had formed in 1986 as in 1992. However, the upper-level flow was somewhat stronger and more westerly with a southwest curvature in the low-level flow apparent; in the midwest, clouds developed, but no precipitation occurred (in the base case, a somewhat cleaner separation between the polar and subtropical jets was evident, and the flow at low levels was more southerly with rain developing in the midwest). During the second day of circulation analogy, the lee cyclone in both 1986 and 1992 passed south of Lake Michigan. In both cases, the flow was directly from the Gulf of Mexico, and stratiform precipitation fell over the lower Great Lakes with convection erupting in the southeast. However, during this time in 1986, the cyclone was becoming less organized and the precipitation was generally lighter than in 1992. The analog of 10–12 February 1975 followed a similar evolution to that of the 1986 case. As before, a 500-hPa short wave approached the Pacific coast and produced rain in that region as the cyclone made landfall. Again, the midwest was clear and dry rather than wet as in 1992. The short wave propagated across the Continental Divide and initiated a lee cyclone; the upper-level flow again showed a confluence between the polar and subtropical jet streams. In this case, the southwesterly flow was strong (500-hPa winds of 48 m s−1), and the cyclone followed a more southerly course to the mid-Atlantic states (note the rapid fall of surface AC in the 36–48-h time frame in Fig. 5). Precipitation was light in the Great Lakes region and the southeastern United States.
In order to present a physical basis for the distinctions between the base case and the analogs, we will present a detailed comparison of the base case with the analog of 20–22 February 1980. For this case, the best of the three analogs, we obtain anomaly correlations of 0.8–0.9 and root-mean-square differences of 2–4 hPa at sea level and 20–30 m at 500 hPa throughout the period of interest. As shown in Figs. 3b and 3d, the basic evolution of the synoptic pattern in the February 1980 case (hereafter referred to as the analog) is quite similar to that of the base case. Therefore, we shall divide the comparison of the base case with the analog into two sections based on two equal 24-h periods beginning at the initial time of the base case (0000 UTC 14 February 1992).
a. Period 1: 00–24 h
As shown in Figs. 4a and 4e, precipitation totals associated with the base case and the analog were quite different across the Midwest, despite relatively similar circulation patterns in the two events. This is suggestive that significant moisture differences may have been present during this period (0000 UTC 14–15 February 1992 and 1200 UTC 20–21 February 1980 for the base case and the analog, respectively). Careful inspection of the sea level pressure patterns shown in Figs. 3a and 3b reveals a slight southwest curvature of the isobars in 1980, suggesting air trajectories initially from the dry Mexican plateau rather than the Gulf of Mexico. However, differences in local sources of moisture (evaporation of soil moisture) and in the thermal structure and subsequent evaporation of water from the Gulf of Mexico might also play a role. Funk (1991) has noted that moisture availability, moisture convergence, and the strength of the low-level inflow are considerations that are emphasized by NCEP forecasters in determining QPF.
As a first step in investigating these issues, sea surface temperature (SST) analyses constructed from the Comprehensive Ocean–Atmosphere Data Set (COADS) were examined for February 1992 and February 1980 (not shown). These analyses indicated little difference in the surface thermal structure, with the exception of a narrow swath of 18°C SSTs along the northern Gulf Coast in 1992 (compared to 20°C in 1980). Since the cooler SSTs along the coast in 1992 would have tended to inhibit evaporation relative to 1980, this factor does not appear to have been a significant contributor to moisture differences in the two cases.
The first term on the right-hand side of (4) represents the vertical integral of the water vapor flux divergence (
Calculations using (4) were made for a total of nine triangles based on nine sounding stations (Fig. 9) covering an area approximately 7.6 × 105 km2 in the lower midwest and are shown in Table 2. Moisture convergence and strong moisture advection with some contribution from local evaporation led to rapid moistening of the atmosphere and the development of light regional precipitation between 0000 and 1200 UTC 14 February 1992. In the subsequent 12 h (1200–0000 UTC 14–15 February), moisture convergence and local evaporation increased, while moisture advection was maintained; since the soundings were already quite moist, much of this additional water substance was precipitated out. In contrast, during the first 12 h of the analog period (1200–0000 UTC 20–21 February 1980), all of these sources (moisture convergence, moisture advection, local evaporation) were weaker and thus led only to a slow moistening of the atmospheric column such that by the end of this time, precipitable water was still considerably less than in 1992 (Table 2 and Fig. 8); these conditions persisted over the following 12 h (0000–1200 UTC 21 February) and observed precipitation was scattered (Fig. 4e).
These results suggest in the base case a complex interplay between precipitation events prior to our period of interest (which appear to have acted as a seeding process generating a local source of moisture upwind of the affected region, leading to strong moisture advection during the first 12 h) and synoptic-scale dynamics, which resulted in the development of the flow from the Gulf of Mexico and convergence in a then moist environment during the latter 12 h. In the analog, moisture advection was much weaker during the entire 24-h period (no upwind seeding, and no subsequent development of flow from the gulf), and subsequent low-level convergence and associated ascent were taking place in regions of limited water vapor. Detailed calculations of surface evaporation and air parcel trajectories will be presented in the subsequent numerical modeling study referred to previously to further establish these conclusions.
Because the differences in the circulation between the base case and the analog are relatively subtle, it is important to consider precisely how these differences in the low-level flow arose. Figure 10 shows 3-h surface pressure tendency calculations early in the first 24-h period (0300 UTC 14 February 1992 in Fig. 10a and 1500 UTC 20 February 1980 in Fig. 10b), based upon hourly observations from the surface synoptic network (tendencies were calculated based upon a centered difference in time, using observations from the hour prior to and following the desired time). These analyses clearly indicate that a southerly geostrophic flow was becoming established at this time in 1992, with surface pressure falls (rises) centered to the northwest (east) of Little Rock, Arkansas (St. Louis, Missouri). In 1980, however, a west-to-southwest geostrophic flow was developing as a result of a band of surface pressure rises along a line from central Texas through southern Arkansas and pressure falls to the north in Kansas and Oklahoma. Since surface pressure changes can be expressed as the sum of the vertically integrated mass divergence and terrain effects (Palmén and Newton 1969, 134–135), we have evaluated the influence of upslope–downslope winds on the pressure tendency fields in these cases. This diagnostic was evaluated by considering the horizontal gradients of the terrain height interpolated to the surface reporting station location from the National Center for Atmospheric Research (NCAR) 10-min (approximately 19 km) global terrain dataset. In each case, terrain effects did not qualitatively modify the patterns shown in Fig. 10. From this result, we can conclude that the observed surface pressure tendencies were largely forced by column mass divergence (convergence) in the regions of pressure falls (rises).
Although column mass divergence cannot be quantitatively evaluated from direct observations, we can make some qualitative inferences on the processes responsible for the observed surface pressure tendencies. Figure 11 shows the 500-hPa geopotential height and absolute vorticity fields, along with the position of the subtropical jet (as evaluated at 200 hPa) for 0000 UTC 14 February 1992 (Fig. 11a) and 1200 UTC 20 February 1980 (Fig. 11b). Figure 12 shows the 250-hPa divergence computed from the NMC Northern Hemisphere 381-km octagonal gridded winds for these times. In 1992, with the ridge line extending from eastern Montana southeastward through western Arkansas, the resultant pattern of 500-hPa vorticity advection is consistent with the 250-hPa divergence (Fig. 12a) and the observed surface pressure tendencies 3 h later (Fig. 10a). The subtropical jet does not appear to have contributed significantly to the surface pressure falls in the region of interest, since the right entrance region of the jet streak and associated cross-stream divergent ageostrophic flow is positioned over Louisiana and the Gulf of Mexico, south and southwest of Arkansas (see also Fig. 13a below).
Although the flow in 1980 (Fig. 11b) is broadly similar to 1992, particularly with respect to the pattern of 500-hPa vorticity advection, differences in the subtropical jet are readily apparent. Elaboration of this observation is provided by vertical cross sections along a line from Lake Charles, Louisiana (72240; LCH) to Omaha, Nebraska (72553; OMA) (location indicated as the dashed line in Fig. 11a) at these times (Fig. 13a,b). In 1980 (Fig. 13b), the subtropical jet was lower, stronger (peak winds well in excess of 55 m s−1; dark shading in Fig. 11b), and exhibited greater tropospheric depth than the jet in 1992 (Fig. 13a). Furthermore, there is an indication of weaker midtropospheric static stability underneath the jet core in 1980 (note the relative spread of the 300- and 320-K isentropes); these factors taken together increase the likelihood of a more robust surface response to jet-level dynamics in 1980. Also of importance is the northward shift of the jet axis to northern Oklahoma and Arkansas (Fig. 11) in this case, thus positioning the regions of northeastern Texas and Louisiana beneath the anticyclonic shear side of the subtropical jet. The divergence pattern of Fig. 12b is the result of this combination of midtropospheric vorticity advection and jet-level dynamics. Downstream of the ridge axis, vorticity advections of either sign are small, and the resultant pattern of divergence extending from Louisiana to Mississippi reflects the thermally direct ageostrophic circulations associated with the right entrance region of a strong jet streak (Fig. 13b) propagating southeastward; the shifting of these circulations toward the warm side of the thermal ridge is predicted by the two-dimensional Sawyer–Eliassen equation (Keyser and Shapiro 1986). The 250-hPa convergence extending across Texas represents the combination of effects from the right exit region of the upstream jet streak and strong anticyclonic vorticity advection (AVA) along the backside of the short wave over Oklahoma. The thermally indirect ageostrophic circulation in the jet exit region (Fig. 13c), shifted toward the warm side of the thermal ridge, produces strong upper-level convergence across central and southeastern Texas, while the AVA produces convergence (in opposition to the jet entrance region circulation) farther upstream. Cyclonic vorticity advection along the jet stream axis counters the convergence produced in the jet exit region (Fig. 13c), producing a meridional gradient of surface pressure change with the strongest rises to the south of the jet stream axis. Of course, these two-dimensional interpretations are complicated in real flows by curvature effects. For example, the anticyclonic curvature of the flow depicted in Fig. 11b suggests that some reduction of the along-stream divergence (convergence) in the downstream (upstream) jet streak would occur (Keyser and Shapiro 1986; Loughe et al. 1995). Nonetheless, keeping in mind the evolution of this pattern over the next several hours, these features are consistent with the observed surface pressure tendencies.
Rawinsonde observations from Monett, Missouri (72349; UMN), for the base case and the analog (Fig. 15a–d) provide further clarification as to the means by which the observed precipitation differences were realized in the two cases. In the base case, a subsidence inversion was present at 0000 UTC 14 February (Fig. 15a), indicating air with a history of descent (5 mm precipitable water). However, as noted above, moisture convergence and strong moisture advection with additional contributions from local evaporation led to a rapid rise in moisture by 1200 UTC in advance of the warm front (25 mm precipitable water); penetration of the frontal surface by the rawinsonde is suggested by the moist inversion whose top is near 900 hPa (Fig. 15b). Also notable is the deep tropospheric nature of the moisture profile in Fig. 15b. The vertical profile of the time-averaged divergence, vertical motion and moisture budget terms near UMN during this 12-h period, obtained through analysis of rawinsonde triangles as described above, reveal deep ascent and significant moisture convergence and moisture advection extending to near the 600-hPa level (Fig 16). From 1100 to 1300 UTC, 19.1 mm of precipitation was measured at the Portable Automated Mesonet II station located near UMN, and NWS radar summaries (not shown) indicated an area of thunderstorms with tops near 290 hPa in this region. Analysis of the 1200 UTC sounding shows that parcels displaced from the top of the inversion would be unstable to upright convection. A warming and moistening of the 900-hPa level by only 1°C erodes the convective inhibition and yields convective available potential energy (CAPE) of 104 J kg−1, with maximum parcel levels above 300 hPa. Thus, warm and moist advection associated with frontal upglide embedded within a broad region of synoptic scale lift (resulting from a band of cyclonic differential vorticity advection ahead of the upper-level short-wave trough; see Figs. 3a, 11a, and 12a) was able to release the positive buoyant energy extant in the base case atmosphere. Vertical motions from synoptic-scale ascent and convection may also have acted to transport moisture to levels above 600 hPa, as suggested by a comparison between Figs. 15b and 16b.
In the analog case, conditions at UMN are quite similar to those of the base case at the initial time of 1200 UTC 20 February 1980 (Fig. 15c). The slight backing of the winds between 850–700 hPa in 1980 and consequent cold-air advection and quasigeostrophic subsidence is replaced by 0000 UTC 21 February (Fig. 15d) by warm-air advection. Triangle calculations of divergence and vertical motion show area-averaged ascent of nearly 0.8 Pa s−1 at this time (compared to about 1.1 Pa s−1 in the base case). However, as shown by the moisture budget analysis (Table 2), significant moisture differences between these cases are apparent at UMN and throughout this part of the Midwest (Figs. 8a–d), and no convective response to the existing synoptic forcing was possible. Thus, slight deviations in the low-level flow between the base case and the analog, the result of a complex interaction between the basic synoptic pattern and the subtropical jet present in both cases, led to substantial differences in the sensible weather (Figs. 4a and 4e). This result should send a strong cautionary signal to researchers and forecasters who base the acceptability of a set of forecasts solely on measures of the large-scale circulation such as the AC.
b. Period 2: 24–48 h
During this period, the regional distribution of precipitation was relatively similar in the base case and the analog, with a stable precipitation region evident northeast of the surface cyclone and a zone of convective precipitation in the southeastern states (Figs. 4b and 4f). However, particularly in the convective region, the details of the observed precipitation structures differ substantially. For example, in the base case, the heaviest precipitation was associated with a squall-line structure which developed into a line echo wave pattern (LEWP) in north-central Georgia during the 6-h period between 1300 and 1900 UTC. During this time, 25 mm of precipitation fell at AHN. In contrast, in the analog case, convective precipitation was less organized and generally occurred later in the period.
The origin of these differences is not obvious. For example, an analysis of precipitable water shows that the distribution of available moisture is quite comparable during this time (Figs. 8g,h), and the synoptic patterns remain similar (Figs. 3c,d). However, an examination of the rawinsonde observations at AHN (Fig. 15e–h) in these two cases is instructive. Casual inspection of the 1200 UTC 15 February 1992 AHN sounding (Fig. 15e) does not suggest that convection was imminent, given the subsidence inversion above 950 hPa. However, radar echoes (not shown) had been advancing toward AHN for several hours prior to this time and light (nonconvective) rainfall had just begun at the station in response to synoptic-scale ascent (triangle diagnostics indicate areally averaged ascent near AHN at this time of 1.8 Pa s−1). Under these conditions, it is reasonable to expect that some erosion of the subsidence inversion would result through evaporative cooling and moistening as the rain fell through the dry layer. The sounding taken 12 h later (0000 UTC 16 February 1992;Fig. 15f) is consistent with this hypothesis. In addition, cooling and drying at upper levels resulted from cold advection aloft during the period (as suggested by the backing winds in the 500–250-hPa layer at AHN at 0000 UTC). Modification of the 1200 UTC AHN sounding with these principles in mind easily leads to the production of CAPE, sufficient to allow parcel ascent to the observed levels of approximately 315 hPa. It is also worth noting that the sounding taken at 0000 UTC shows that convective potential remains (maximum parcel levels approach 250 hPa), although no precipitation was reported at AHN after this time.
Analysis of the rawinsonde observation taken at AHN at 0000 UTC 22 February 1980 (Fig. 15g) reveals that lifting of near-surface parcels would release CAPE and allow these parcels to attain heights above 220 hPa. Triangle diagnostics indicate areally averaged ascent near AHN at this time of 1.3 Pa s−1. However, despite this potential instability and the presence of lift, no precipitation was reported at this time or during the following 6 h, the period corresponding to the development of the LEWP and 25 mm of precipitation in the base case. Inspection of the sounding taken at AHN at 1200 UTC 22 February 1980 (Fig. 15h) reveals considerable potential for convection as temperatures have warmed at low levels and cooled aloft. A parcel lifted from the top of the inversion near 900 hPa would have CAPE of 2023 J kg−1 and would attain heights above 170 hPa. It was, in fact, after this time, in advance of the cold front, that more substantial precipitation fell in this case (but not in the base case). These results suggest that mesoscale triggering mechanisms may have been important in the development of the convection in each case. To consider this issue, maps of the 250-hPa and 850-hPa divergence for the base case and analog were constructed from the NMC Northern Hemisphere 381-km octagonal gridded winds for these times (Fig. 17). These data suggest synoptic-scale differences in the horizontal distribution, and magnitude of the divergence profiles (and vertical motions) in this region also existed; we cannot rule out the possibility that these differences were alone sufficient to produce convection in the one case but not the other. To more fully explore these issues, we will need to turn to some experiments with mesoscale numerical models in a subsequent paper.
4. Discussion
In this paper, we have posed the basic question: How sensitive is the precipitation associated with baroclinic winter storms to small variations in the synoptic-scale circulation? The unique approach employed to examine this question was to analyze observed data using ensembles of regional analogs. Our results have highlighted the fact that despite considerable synoptic-scale circulation analogy, observed differences in precipitation can be substantial in relatively ordinary cases of land-based cold-season cyclogenesis. We believe that these results should strongly qualify claims of increasing forecast skill, particularly when couched in forms that measure the skill of forecasts of the large-scale circulation (e.g., Kalnay et al. 1990).
In order to further understand how such differences can arise, we will pursue detailed comparisons of one of the regional analogs with the base case in a subsequent paper. However, it might be useful to generalize our results beyond this case and the set of three analogs studied here. For example, one approach would be to expand the search to include days in other cold-season months.
However, rather than perform such an analysis, we have elected to reverse the basic question: Given analogy in the observed distribution of precipitation, what are the resulting circulation structures? We have addressed this issue by examining daily precipitation totals (obtained from the NCDC TD-3220 daily weather observations dataset) for 14–15 February 1992 for 70 stations in the region bounded by 30°–45°N and 81°–100°W (Fig. 18). We then computed correlation coefficients between the observed precipitation on each of those days with daily precipitation measurements for each February day of the 25-yr period from 1965 to 89. A list of the dates, whose precipitation distribution can account for the greatest proportion of the precipitation variance that occurred on 14–15 February 1992 or in either 24-h period individually, is given in Table 3. Remarkably, the best precipitation analog (25–26 February 1985) can account for only roughly one-quarter of the total precipitation variance in the 2-day period; this is the result of poor precipitation correspondence in the first 24 h but fairly good analogy in the second 24 h of the 48-h period of comparison. It is important to note that the precipitation analogy in the second 24 h resulted from a much different circulation structure than occurred in the base case, specifically, a weak low pressure system which propagated along a frontal boundary extending from the Gulf of Mexico to North Carolina. Accordingly, the correlation for this period was weighted entirely by the precipitation that fell in the southeastern United States, since there was no precipitation at all over the lower Great Lakes (as occurred in the base case). There were many February cases in which it was possible to find relatively good analogs with the precipitation distribution that occurred in the second 24 h, but in each of these cases, the precipitation distribution was focused in the southeastern United States with little or no precipitation over the lower Great Lakes; furthermore, analogy with the first 24 h was utterly lacking in all of these cases.
The best analog to the first 24 h (14–15 February 1980) was able to account for only about one-third of the base case precipitation distribution variance during that 24-h period. This analogy, which occurred about one week prior to the 1980 regional analog that we have studied in some detail in this paper, resulted from lee trough formation to the east of the Rocky Mountains in zonal 500-hPa flow. The lee trough, in combination with a northwest to southeast ridge axis extending from Alberta, Canada to Georgia, set up a southerly geostrophic flow in the lowest levels of the atmosphere, resulting in good moisture flow from the Gulf of Mexico. Subsequently, a weak wave began to propagate northeastward toward the lower Great Lakes but quickly dissipated with no upper-level support (the vorticity advection in advance of a mobile short-wave trough, typical with such developments, was completely lacking in this case). Thus, there was no correspondence between the precipitation that fell the next day vis-a-vis the base case.
It was not possible to find any cases in which precipitation analogy extended over the full 48-h period of the base case. The best analog in that respect (28 February–1 March 1987) showed only weakly positive correlations with the base case in both periods. This event was organized by a high-amplitude 500-hPa trough associated with a lee cyclone that formed near the border between Colorado and New Mexico. This cyclone exhibited a more extensive circulation than occurred with the base case, with sea level central pressures near 990 hPa and cyclonic flow extending from Manitoba to northern Florida. Likewise, precipitation was more widespread in this case, with measurable precipitation falling across virtually all of the United States from the Plains states eastward.
Thus, we find that whether we seek analogs on the basis of similar synoptic-scale circulation structures or on the basis of the precipitation fields themselves, the resultant sensible weather is crucially dependent on the details. It is the physical mechanisms that govern these details (in both the base case and the 1980 circulation analog) that we will turn our attention to in a subsequent paper.
Acknowledgments
This work began while the lead author (PJR) was a postdoctoral research associate with L. Bosart at The University of Albany, State University of New York, supported in part by NSF Grant ATM 94-13012. This research has received additional support under NASA JOVE Grant NAG8-1205. Our interest in this work was sparked by the ongoing effort of many researchers to improve forecasts across a broad range of spatial and temporal scales and in particular how these efforts relate to the precipitation forecast problem. Our thanks are extended to the anonymous reviewers whose efforts led to a much expanded but improved paper.
REFERENCES
Anthes, R. A., 1983: Regional models of the atmosphere in middle latitudes. Mon. Wea. Rev.,111, 1306–1335.
Black, T. L., 1994: The new NMC mesoscale Eta Model: Description and forecast examples. Wea. Forecasting,9, 265–278.
Blanchard, B., M. T. McFarland, T. J. Schmugge, and E. Rhodes, 1981: Estimation of soil moisture with API algorithm and microwave emission. Water Resour. Bull.,17, 769–774.
Bosart, L. F., 1983: An update on trends in skill of daily forecasts of temperature and precipitation at the State University of New York at Albany. Bull. Amer. Meteor. Soc.,64, 346–354.
——, and F. Sanders, 1981: The Johnstown flood of July 1977: A long-lived convective system. J. Atmos. Sci.,38, 1616–1642.
Carter, G. M., and P. D. Polger, 1986: A 20 year summary of National Weather Service verification results for temperature and precipitation. NOAA Tech. Memo. NWS FCST-31, National Weather Service, NOAA/U.S. Department of Commerce, 50 pp.
Farrell, R. J., and T. N. Carlson, 1989: Evidence for the role of the lid and underrunning in an outbreak of tornadic thunderstorms. Mon. Wea. Rev.,117, 857–871.
Funk, T. W., 1991: Forecasting techniques utilized by the forecast branch of the National Meteorological Center during a major convective rainfall event. Wea. Forecasting,6, 548–564.
Glahn, H. R., 1985: Yes, precipitation forecasts have improved. Bull. Amer. Meteor. Soc.,66, 820–830.
Gutzler, D. S., and J. Shukla, 1984: Analogs in the wintertime 500-mb height field. J. Atmos. Sci.,41, 177–189.
Hao, W., and L. F. Bosart, 1987: A moisture budget analysis of the protracted heat wave in the southern plains during the summer of 1980. Wea. Forecasting,4, 269–288.
Hoke, J. E., N. A. Phillips, G. J. Dimego, J. J. Tucillo, and J. G. Sela, 1989: The regional analysis and forecast system of the National Meteorological Center. Wea. Forecasting,4, 323–334.
Hollingsworth, A., K. Arpe, M. Tiedtke, M. Capaldo, and H. Savijari, 1980: The performance of a medium range forecast model in winter—Impact of physical parameterizations. Mon. Wea. Rev.,108, 1736–1773.
Junker, N. W., J. E. Hoke, and R. H. Grumm, 1989: Performance of NMC’s regional models. Wea. Forecasting,4, 368–390.
Kalnay, E., M. Kanamitsu, and W. E. Baker, 1990: Global numerical weather prediction at the National Meteorological Center. Bull. Amer. Meteor. Soc.,71, 1410–1428.
Keyser, D., and M. A. Shapiro, 1986: A review of the structure and dynamics of upper-level frontal zones. Mon. Wea. Rev.,114, 452–499.
Lorenz, E. N., 1969: Atmospheric predictability as revealed by naturally occurring analogs. J. Atmos. Sci.,26, 636–646.
Loughe, A. F., C.-C. Lai, and D. Keyser, 1995: A technique for diagnosing three-dimensional ageostrophic circulations in baroclinic disturbances on limited-area domains. Mon. Wea. Rev.,123, 1476–1504.
Mass, C. F., H. J. Edmon, H. J. Friedman, N. R. Cheney, and E. E. Recker, 1987: The use of compact discs for storage of large meteorological and oceanographic data sets. Bull. Amer. Meteor. Soc.,68, 1556–1558.
Mesinger, F., 1996: Improvements in quantitative precipitation forecasts with the Eta regional model at the National Centers for Environmental Prediction: The 48-km upgrade. Bull. Amer. Meteor. Soc.,77, 2637–2649.
Moore, J. T., and P. D. Blakley, 1988: The role of frontogenetical forcing and conditional symmetric instability in the Midwest snowstorm of 30–31 January 1982. Mon. Wea. Rev.,116, 2155–2171.
O’Brien, J. J., 1970: Alternative solution to the classical vertical velocity problem. J. Appl. Meteor.,9, 197–203.
Olson, D. A., N. W. Junker, and B. Korty, 1995: Evaluation of 33 years of quantitative precipitation forecasting at NMC. Wea. Forecasting,10, 498–511.
Oravec, R. J., and R. H. Grumm, 1993: The prediction of rapidly deepening cyclones by NMC’s Nested Grid Model: Winter 1989–autumn 1991. Wea. Forecasting,8, 248–270.
Palmén, E., and C. W. Newton, 1969: Atmospheric Circulation Systems: Their Structure and Physical Interpretation. Academic Press, 603 pp.
Ramamurthy, M. K., R. M. Rauber, B. P. Collins, and N. K. Malhotra, 1993: A comparative study of large-amplitude gravity-wave events. Mon. Wea. Rev.,121, 2951–2974.
Roebber, P. J., J. R. Gyakum, and D. N. Trat, 1994: Coastal frontogenesis and precipitation during ERICA IOP #2. Wea. Forecasting,9, 21–44.
Sanders, F., 1986a: Explosive cyclogenesis over the west-central North Atlantic Ocean, 1981–84. Part II: Evaluation of the LFM model performance. Mon. Wea. Rev.,114, 2207–2218.
——, 1986b: Trends in skill of Boston forecasts made at MIT, 1966–84. Bull. Amer. Meteor. Soc.,67, 170–176.
——, 1987: Skill of NMC operational models in prediction of explosive cyclogenesis. Wea. Forecasting,2, 322–336.
Shields, M. T., R. M. Rauber, and M. K. Ramamurthy, 1991: Dynamical forcing and mesoscale organization of precipitation bands in a Midwest winter cyclonic storm. Mon. Wea. Rev.,119, 936–964.
Shuman, F. G., 1989: History of numerical weather prediction at the National Meteorological Center. Wea. Forecasting,4, 286–296.
——, and J. B. Hovermale, 1968: An operational six-layer primitive equation model. J. Appl. Meteor.,7, 525–547.
Sutcliffe, R. C., 1947: A contribution to the problem of development. Quart. J. Roy. Meteor. Soc.,73, 370–383.
Trenberth, K. E., 1978: On the interpretation of the diagnostic quasi-geostrophic omega equation. Mon. Wea. Rev.,106, 131–137.
Whittaker, L. M., and L. H. Horn, 1982: Atlas of Northern Hemisphere Extratropical Cyclone Activity, 1958–77. Department of Meteorology, University of Wisconsin—Madison, 65 pp.
(a) The 500-hPa geopotential height skill score for 36-h predictions for North America and adjacent waters (adapted from Shuman 1989) for the period 1967–88 (open circles) and the 24–36-h forecast skill with respect to climatology of the probability of measurable precipitation (POP) at approximately 100 U.S. stations for the same period (solid circles); (b) National Weather Service threat scores for day 1 QPF (12–36 h) for thresholds of 12.5 mm (0.5 in., open circles), 2.5 mm (1.0 in., open squares), and 50 mm (2.0 in., open triangles) for the period 1980–95 (adapted from Olson et al. 1995).
Citation: Monthly Weather Review 126, 2; 10.1175/1520-0493(1998)126<0437:TSOPTC>2.0.CO;2
Locator map for geographic references in the text.
Citation: Monthly Weather Review 126, 2; 10.1175/1520-0493(1998)126<0437:TSOPTC>2.0.CO;2
Sea level pressure (solid lines, 4-hPa interval), 500-hPa geopotential height (dashed lines, 6-dam interval), and track of the surface cyclone for the 48 h centered on the analysis time (arrow) for (a) 0000 UTC 15 February 1992; (b) for the time corresponding to (a) for the 1980 regional analog (1200 UTC 21 February 1980); (c) 1200 UTC 15 February 1992; and (d) for the time corresponding to (c) for the 1980 regional analog (0000 UTC 22 February 1980).
Citation: Monthly Weather Review 126, 2; 10.1175/1520-0493(1998)126<0437:TSOPTC>2.0.CO;2
Precipitation analyses, based upon the NCDC TD-3240 cooperative station precipitation dataset, for the 24-h period from (a) 0000 UTC 14–15 February 1992; (b) 0000 UTC 15–16 February 1992; (c) for the time corresponding to (a) for the composite of the three regional analogs identified in Table 1; (d) for the time corresponding to (b) for the composite of the three regional analogs identified in Table 1; (e) 1200 UTC 20–21 February 1980; and (f) 1200 UTC 21–22 February 1980. Contours are 1, 5, 10, 15, 20, 25, and 50 mm. The track of the surface cyclone is shown for the same period as the analysis (arrow).
Citation: Monthly Weather Review 126, 2; 10.1175/1520-0493(1998)126<0437:TSOPTC>2.0.CO;2
Time history of surface (filled symbols) and 500-hPa (open symbols) anomaly correlation scores for the three identified regional analogs, corresponding to the 48-h period of the base case (0000 UTC 14–16 February 1992). Circles correspond to 0000 UTC 10–12 February 1975, squares to 1200 UTC 20–22 February 1980, and triangles to 0000 UTC 16–18 February 1986.
Citation: Monthly Weather Review 126, 2; 10.1175/1520-0493(1998)126<0437:TSOPTC>2.0.CO;2
As in Fig. 3 but for the time corresponding to Fig. 3a for the composite fields of the three regional analogs identified in Table 1.
Citation: Monthly Weather Review 126, 2; 10.1175/1520-0493(1998)126<0437:TSOPTC>2.0.CO;2
Antecedent precipitation index (API) in millimeters for (a) 0000 UTC 14 Feb 1992 and (b) 1200 UTC 20 Feb 1980.
Citation: Monthly Weather Review 126, 2; 10.1175/1520-0493(1998)126<0437:TSOPTC>2.0.CO;2
Precipitable water and streamlines of the surface-to-500-hPa mean wind for (a) 0000 UTC 14 February 1992; (b) 1200 UTC 20 February 1980; (c) 1200 UTC 14 February 1992; (d) 0000 UTC 21 February 1980; (e) 0000 UTC 15 February 1992; (f) 1200 UTC 21 February 1980; (g) 1200 UTC 15 February 1992: and (h) 0000 UTC 22 February 1980. Contours are 10, 20, and 30 mm.
Citation: Monthly Weather Review 126, 2; 10.1175/1520-0493(1998)126<0437:TSOPTC>2.0.CO;2
(Continued)
Citation: Monthly Weather Review 126, 2; 10.1175/1520-0493(1998)126<0437:TSOPTC>2.0.CO;2
Distribution of rawinsonde stations comprising the network used to compute the moisture budget. The outline of the overall area (used to determine the values of Table 2 and representing 7.6 × 105 km2) is shown by the thick solid line, while the individual triangles within the area are shown by the thin solid lines connecting the sounding sites. The stations used were: Jackson (JAN; 72235); Lake Charles (LCH; 72240); Longview (GGG; 72247); Nashville (BNA; 72327); Little Rock (1M1; 72340); Monett (UMN; 72349); Amarillo (AMA; 72363); Topeka (TOP; 72456); and Peoria (PIA; 72532).
Citation: Monthly Weather Review 126, 2; 10.1175/1520-0493(1998)126<0437:TSOPTC>2.0.CO;2
Surface pressure tendencies (3 h) based on a centered time difference of hourly surface reports at (a) 0300 UTC 14 February 1992 and (b) 1500 UTC 20 February 1980. Dashed (solid) contours indicate pressure falls (rises), with a contour interval of 0.25 hPa (3 h)−1.
Citation: Monthly Weather Review 126, 2; 10.1175/1520-0493(1998)126<0437:TSOPTC>2.0.CO;2
The 500-hPa geopotential height (solid, 6-dam interval) and absolute vorticity (dashed, 2 × 10−5 s−1 interval) fields, along with the position of the subtropical jet (based upon 200-hPa data), where the arrow indicates the jet axis and the light (dark) shading denotes winds in excess of 50 m s−1 (55 m s−1) for (a) 0000 UTC 14 Feb 1992 and (b) 1200 UTC 20 Feb 1980. The bold dashed lines in (a) and (b) indicate the positions of the cross sections shown in Figs. 13a–c.
Citation: Monthly Weather Review 126, 2; 10.1175/1520-0493(1998)126<0437:TSOPTC>2.0.CO;2
The 250-hPa divergence fields based on the NMC Northern Hemisphere 381-km octagonal gridded winds for (a) 0000 UTC 14 Feb 1992 and (b) 1200 UTC 20 Feb 1980. The solid (dashed) contours (interval of 1 × 10−5 s−1) indicate divergence (convergence).
Citation: Monthly Weather Review 126, 2; 10.1175/1520-0493(1998)126<0437:TSOPTC>2.0.CO;2
Vertical cross sections through the subtropical jet. Analyzed are isotachs (solid lines, 5 m s−1 interval), isentropes (dashed lines, 10-K interval), and divergence (gray lines divergence, dashed gray lines convergence; interval of 1 × 10−5 s−1). The divergence values are obtained from the rawinsonde triangle method (see text for details) and interpolated to the location of the cross section from adjacent triangles. The position of the cross section is indicated by the bold–dashed line in Fig. 11a (Lake Charles/LCH to Longview/GGG to Little Rock/1M1 to Monett/UMN to Topeka/TOP to Omaha/OMA) for (a) and (b) and in Fig. 11b (Brownsville/BRO to Stephenville/SEP to Amarillo/AMA to Dodge City/DDC to North Platte/LBF) for (c). The analyses are for (a) 0000 UTC 14 February 1992 and (b)–(c) 1200 UTC 20 February 1980.
Citation: Monthly Weather Review 126, 2; 10.1175/1520-0493(1998)126<0437:TSOPTC>2.0.CO;2
Sutcliffe (1947)–Trenberth (1978) forcing of ascent for (a) 0000 UTC 14 February 1992 and (b) 1200 UTC 20 February 1980. Shown are 500–200-hPa thickness (solid lines, 6-dam interval) and the 300-hPa vorticity (dashed lines, 2 × 10−5 s−1 interval).
Citation: Monthly Weather Review 126, 2; 10.1175/1520-0493(1998)126<0437:TSOPTC>2.0.CO;2
Thermodynamic diagrams for (a) Monett, Missouri (UMN), at 0000 UTC 14 February 1992; (b) UMN at 1200 UTC 14 February 1992; (c) UMN at 1200 UTC 20 February 1980; (d) UMN at 0000 UTC 21 February 1980; (e) Athens, Georgia (AHN), at 1200 UTC 15 February 1992; (f) AHN at 0000 UTC 16 February 1992; (g) AHN at 0000 UTC 22 February 1980; and (h) AHN at 1200 UTC 22 February 1980. Shown are temperature (°C, solid bold), dewpoint (°C, dashed bold), and wind direction and speed (pennant—25 m s−1, long barb—5 m s−1, short barb—2.5 m s−1). In the diagram, the solid lines slanting toward the upper right are isotherms, the solid lines slanting toward the upper left are dry adiabats, dashed lines slanting toward the upper left are moist adiabats, and the dotted lines slanting toward the upper right are lines of constant saturation mixing ratio.
Citation: Monthly Weather Review 126, 2; 10.1175/1520-0493(1998)126<0437:TSOPTC>2.0.CO;2
(Continued)
Citation: Monthly Weather Review 126, 2; 10.1175/1520-0493(1998)126<0437:TSOPTC>2.0.CO;2
The vertical profile of the 12-h average (a) divergence (dashed; 10−5 s−1) and vertical motion (10−1 Pa s−1, solid) and (b) moisture budget terms (10−9 s−1) for three triangles (72363–72340–72349; 72340–72349–72532; 72349–72456–72532; see Fig. 9) near UMN for the period 0000–1200 UTC 14 February.
Citation: Monthly Weather Review 126, 2; 10.1175/1520-0493(1998)126<0437:TSOPTC>2.0.CO;2
Divergence fields, based on the NMC Northern Hemisphere 381-km octagonal gridded winds, for (a) 1200 UTC 15 February 1992 at 250 hPa; (b) 0000 UTC 22 February 1980 at 250 hPa; (c) 1200 UTC 15 February 1992 at 850 hPa; and (d) 0000 UTC 22 February 1980 at 850 hPa. The solid (dashed) contours (interval of 1 × 10−5 s−1) indicate divergence (convergence).
Citation: Monthly Weather Review 126, 2; 10.1175/1520-0493(1998)126<0437:TSOPTC>2.0.CO;2
Station locations of the NCDC TD-3220 daily weather observations used in the precipitation analog analysis.
Citation: Monthly Weather Review 126, 2; 10.1175/1520-0493(1998)126<0437:TSOPTC>2.0.CO;2
Table 1. Time-coherent regional analogs to the 14–16 February 1992 base case. The selection procedure is described in the text.
Table 2. Triangle moisture budget results for the first 24-h period of the base case (0000 UTC 14–15 Feb 1992) and the analog (1200 UTC 20–21 Feb 1980). The variables q, V, and W represent the specific humidity, wind vector, and precipitable water, respectively, and the overbar denotes an average over the area of all of the triangles (the 7.6 × 105 km2 region depicted in Fig. 9). All quantities are expressed in millimeters.
Table 3. A list of 48-h February periods for the years 1965–89, whose precipitation distribution can account for the greatest proportion of the precipitation variance that occurred on 14–15 February 1992 (denoted as %V, positive correlations only). The highest 24-h period correlation date is indicated for each 24-h period by the boldface type. Correlation for the first (second) 24 h is indicated as R1 (R2).
