Abstract

Data from six urban areas in a nationwide network of sites within the surface roughness layer are examined. It is found that the average velocity variances in time, derived by averaging the conventional variances from a network of n stations, are nearly equal to the velocity variances in space, derived as the variances among the n average velocities. This similarity is modified during sunlit hours, when convection appears to elevate the former. The data show little dependence of the ratio of these two variances on wind speed. It is concluded that the average state of the surface roughness layer in urban and suburban areas like those considered here tends toward an approximate equality of these two measures of variance, much as has been observed elsewhere for the case of forests.

1. Introduction

The surface roughness layer (SRL) is the lowest part of the turbulent atmosphere, where flow and fluxes depend on the presence and distribution of obstacles and other roughness elements. Classical micrometeorology focuses on levels that are sufficiently above the surface that flow and fluxes are not influenced by individual roughness elements. Rather, spatial homogeneity of the surface is assumed. The height of partitioning between the SRL and the micrometeorological “constant flux layer” is historically taken to be about d + 10z0, where d is the displacement height and z0 is the roughness length over an area that is assumed to be homogeneous. This corresponds (usually) to about 2 times the average height of the surface roughness elements. In urban areas, the SRL is of special interest because it is within this layer that people live and work, and where they are most susceptible to exposure to any dispersing material.

Dispersion within the urban SRL remains a topic of considerable uncertainty. The importance of relevant wind information is universally acknowledged, but acquisition of such data and the form they should take remain issues of debate. Discussion typically centers on consideration of whether any single near-origin location can provide data suitable for describing downwind dispersion. A recent analysis of data obtained in Washington, D.C. (Hicks et al. 2012), shows that there is benefit in considering results from a network of surrounding locations, should such a network be available. This was demonstrated by showing similarity between data derived by combining results from such nearby SRL network stations with observations made above the tops of major buildings in the Washington metropolitan area. The SRL data were from the “WeatherBug” network, operated by Earth Networks, Inc., a commercial provider of meteorological information for educational and other purposes. These data are obtained by using mechanical anemometers that are situated a few meters above the roof edges of buildings. To facilitate examination of the ability to derive meaningful and spatially representative data from such SRL instruments near some point of special interest, specific locations were selected for testing, these stations being part of the DCNet research network of the Air Resources Laboratory of the U.S. National Oceanic and Atmospheric Administration (NOAA). DCNet operates sonic anemometers that are placed 10 m above the roofs of selected buildings, most of which are in the Washington area (see online at http://dcnet.atdd.noaa.gov/).

Starting with the obvious facts that any single in situ anemometer can only report the variance in the wind field at its own location (a variance in time) but that dispersion across an area will be subject to variations in space as well as in time, an intriguing consideration is of the distinction between velocity variances (or their associated standard deviations) in time and in space . In simple concept, any detailed analysis of this matter should make use of a dataset of measurements made in the same way, simultaneously, and at a frequency sufficient to resolve the velocity variances, across a sufficiently dense spatial network. What constitutes “sufficient” would require site-dependent consideration. Given such a body of data, analysis could then explore the variances 1) across the whole body of data, 2) as would be observed at any single time across the spatial array, and 3) as yielded by analysis of conventional time series of observations at single locations.

In the earlier analysis of Washington urban data, and were found to be approximately equal. The present intent is to test the generality of this result. WeatherBug SRL data from six suburban areas in the eastern United States will be used: Washington; Chicago, Illinois; Boston, Massachusetts; New Orleans, Louisiana; Philadelphia, Pennsylvania; and New York, New York. The analysis presented here is not meant to answer all of the questions that the comments above invite but instead is meant to show how data from networks inside the SRL can be used to help to address some of the issues involved. It is recognized that the effects of differences among locations will become smaller as the height of observation is increased relative to the height of surface features, that is, as the measurements become more indicative of flow above the SRL.

Conventional “flat earth” micrometeorology relates to smoother and more uniform (in space and in time) circumstances. The emphasis that follows is on extending this conventional thinking to those parts of the atmosphere beneath the height of surface structures, where urban and suburban populations are mostly exposed. The intent is to take some first steps toward providing an alternative (or additional) source of readily available data for use in driving models, especially dispersion models, such that the spatial complexity of the area is taken into account. It is acknowledged that the problem of urban complexity has been addressed in considerable detail elsewhere (e.g., Grimmond and Oke 1999), resulting in the development of a building-effects parameter λ based on building and street dimensions. The present analysis stops short of a consideration of the relevance of λ for the urban areas now considered and seeks instead to explore the characteristics of meteorological data that are directly affected by such building effects.

2. Data sources and analysis

The data used here are from the vicinities of Boston, New York, Philadelphia, Washington, Chicago, and New Orleans. Data for a 1-yr period starting in September of 2006 are used. In all cases, the selection of a central point as a focus for the current investigation has been based on a number of considerations, all intended to correspond to a possible situation involving a release of some hazardous material in a location that would cause a large number of casualties: suburban rail yards in Chicago and Philadelphia, sports stadiums in New Orleans and Boston, the Pentagon in the Washington urban area, and a high-density residential area of New York City. This is in contrast to the earlier analysis, in which areas were centered on central Washington locations, where elevated sonic anemometers could be used to provide reference data. In the case presented here, the focal locations are intended to be no more than examples of where future attention might need to be directed. These locations are not sites of any existing meteorological network. It is the data from the WeatherBug network in the vicinity of these central locations that will be used here. All of the present data were obtained using mechanical anemometers, mainly vane-oriented helicoid propellers (see online at http://weather.weatherbug.com/).

Standard WeatherBug data archiving is of the usual meteorological state variables: wind speed, wind direction, temperature, humidity, pressure, and rainfall. For this analysis, a special archive was constructed that contains turbulence data as well as mean condition information, archived over 15-min periods (this being the shortest sampling duration compatible with standard micrometeorological requirements). Outputs from the anemometers were digitized at 0.5 Hz. In their vane-oriented application, the distance constants of the helicoid propeller systems were on the order of 0.7 m. Note that the specified starting speed of the anemometers is 1 m s−1, and hence datasets with an average wind speed below this value have been excluded from the current analysis. There has been no data selection or omission on the basis of siting.

Figure 1 shows (as circles) the locations of SRL sites used in the current analysis, for each of the six areas now examined. To ensure a reasonable number of reporting stations within each area of interest, a circular area of 7-km radius around each central location has been used here. In each panel in Fig. 1, a line indicates the radius (7 km) of the circular area within which stations have been identified. The number of SRL sites varies considerably: 19 for Boston (central location at 42.346°N, 71.096°W), 11 for Chicago (41.766°N, 87.779°W), 13 for Philadelphia (39.934°N, 75.205°W), 33 for Washington (38.869°N, 77.059°W), 23 for New York (40.758°N, 73.903°W), and 23 for New Orleans (29.951°N, 90.081°W). The New Orleans dataset reflects the rebuilding after Hurricane Katrina—the number of contributing stations increased during the reporting period. The analysis that follows is confined to locations in the eastern part of the continental United States. Cities farther to the west are of interest, but the number of sites reporting during the current time window is inadequate.

Fig. 1.

The distributions of surface roughness layer sites (circles) used in this analysis, all located within a radius of 7 km (as indicated by the red lines) from selected central locations (squares): A is Boston, B is Washington, C is New York, D is Chicago, E is Philadelphia, and F is New Orleans. Maps were prepared using Google Earth.

Fig. 1.

The distributions of surface roughness layer sites (circles) used in this analysis, all located within a radius of 7 km (as indicated by the red lines) from selected central locations (squares): A is Boston, B is Washington, C is New York, D is Chicago, E is Philadelphia, and F is New Orleans. Maps were prepared using Google Earth.

The focus on sampling from sites distributed in space imposes a limitation on the potential utility of the results, since (except in unusual situations) the characteristics of the flow and of turbulence will vary from place to place. For local dispersion applications, there is no option but to depend on 1) local meteorological observations (which would likely be rare), 2) the outputs of sufficiently comprehensive numerical models (which could be slow in coming), or 3) the educated interpretations of on-scene observations by informed responders. The last of these options seems most likely to be the basis for a response. It is the behavior of dispersing materials beyond the region of influence of local structures that is of main interest here.

From the wind speed u and wind direction θ data, reported to an on-site data analysis system at 0.5 Hz, values of the U (east–west) and V (north–south) instantaneous wind components were derived. For each site, averages and standard deviations were computed by the on-site system, yielding 15-min summaries that were then transmitted to a central location for archiving and subsequent analysis. Note that this process obviates the difficulties associated with analyzing wind direction data that pass through 360°. The data for 0.5 Hz were not saved. The following variables were then derived:

  1. and , the time (15 min) averages of the wind components for each of the n reporting sites, as reported by the individual sites,

  2. σ2(U) and σ2(V), the corresponding velocity variances over the same 15-min sampling periods, for each of the n sites and as reported by them,

  3. σ2() and σ2(), the variances across space of the archived 15-min-average velocity components defined in point 1 above, and

  4. σ2(U)〉 and 〈σ2(V)〉, the spatial averages of the site-specific velocity component variances, as derived from point 2 above.

From these core variables, the variance in space was computed as , the variance among the n averages of the U wind component, plus the corresponding variance of the V wind component. As a measure of the prevailing average variance in time, was derived as the average of the n evaluations of the U-component variance, plus the average of the n evaluations of the V-component variance:

 
formula
 
formula

In brief, corresponds to the average of the conventional velocity variances observed by the network of n stations and corresponds to the variance of the average velocities derived over 15-min intervals across the same network. Here, we refer to these as the variances in time and in space, respectively, while admitting that there are other ways in which such distinctions might be addressed.

In the case now addressed, these variances are constructed from data that are not “well positioned” from the perspective of classical models, since they are often affected by nearby obstructions. The resulting variations, however, are among those that affect the exposure of surface populations. The study here is intended to relate to the characteristics of the atmosphere as these affect people and not to the predictions or requirements of any particular simulation.

3. Discussion

Figure 2 shows results for a location in the Queens area of New York City, obtained for six months: October 2006, December 2006, February 2007, April 2007, June 2007, and August 2007. Data are shown as hourly averages, using data from the entire month, and are constructed from the statistical results derived from the basic 15-min datasets so as to simplify the plots and impose some smoothing. In Fig. 2 and in many of the figures to follow, a diurnal effect is obvious. In general, the variances increase quickly after sunrise, peak soon after noon, and decrease as night approaches. For four of the six months, and appear to be similar. For all except the June case, the variances are exceedingly close during the night; for June, the space variance appears to exceed the time variance during the night. For the daytime, the variance in time exceeds (or is equal to) that in space for all except the June case. In all cases the differences are small, however, and the main conclusion to be drawn is that the two estimates of the variance are very close.

Fig. 2.

Average wind variances over 1-month periods as functions of the local time of day. Times signs indicate σ2(t)—the velocity variance, averaged over the n stations found in an area of with 7-km radius, centered on a residential area of Queens, in New York City. Plus signs indicate σ2(s)—the variance in space computed using the average (15 min) velocities for each station. Averages are constructed over complete months (October 2006, December 2006, February 2007, April 2007, June 2007, and August 2007). Datasets with incomplete records are excluded, as are all cases with average wind speeds below 1 m s−1.

Fig. 2.

Average wind variances over 1-month periods as functions of the local time of day. Times signs indicate σ2(t)—the velocity variance, averaged over the n stations found in an area of with 7-km radius, centered on a residential area of Queens, in New York City. Plus signs indicate σ2(s)—the variance in space computed using the average (15 min) velocities for each station. Averages are constructed over complete months (October 2006, December 2006, February 2007, April 2007, June 2007, and August 2007). Datasets with incomplete records are excluded, as are all cases with average wind speeds below 1 m s−1.

The results for Boston (Fig. 3) are similar, although for this case it is the February and April nighttime data that show exceeding . Again, the obvious conclusion is that the diurnal cycles of the two variance quantities are about the same, although with tending to be greater than near noon.

Fig. 3.

As in Fig. 2, but for the 7-km circle surrounding Fenway Park in Boston.

Fig. 3.

As in Fig. 2, but for the 7-km circle surrounding Fenway Park in Boston.

All of the results shown here are doubtlessly influenced by the omission of all datasets for which the 15-min average wind speed was below 1 m s−1. To test the influence of this constraint, Fig. 4 also shows data from the Boston case, but for wind speeds above 2 m s−1. The overall conclusions are the same, although clearly the average variances are considerably greater in Fig. 4 than in Fig. 3. The 1 m s−1 starting speed of the anemometers prohibits reliable examination of very low wind speed data.

Fig. 4.

A test of the susceptibility of these plots of variance to the omission of light winds—all cases with average wind speeds below 2 m s−1 have been excluded. Boston (Fenway Park) data are used here, so that a comparison with Fig. 3 is valid.

Fig. 4.

A test of the susceptibility of these plots of variance to the omission of light winds—all cases with average wind speeds below 2 m s−1 have been excluded. Boston (Fenway Park) data are used here, so that a comparison with Fig. 3 is valid.

Four other cases are illustrated in Figs. 58, all for wind speeds that were greater than 1 m s−1. Figure 5, for Chicago, shows exceeding more often than for the previous two cases, although the data show the same approximate equality for the nighttime hours (the main exception being December). Figures 68 (New Orleans, Philadelphia, and Washington, respectively) differ substantially in that almost always exceeds . In all cases, however, the values of and are closest during nighttime.

Fig. 5.

As in Figs. 2 and 3, but for the 7-km circle around the Chicago rail yards, immediately south of Chicago Midway International Airport.

Fig. 5.

As in Figs. 2 and 3, but for the 7-km circle around the Chicago rail yards, immediately south of Chicago Midway International Airport.

Fig. 6.

As in Figs. 2 and 3, but for the 7-km circular area centered on the New Orleans Superdome.

Fig. 6.

As in Figs. 2 and 3, but for the 7-km circular area centered on the New Orleans Superdome.

Fig. 7.

As in Figs. 2 and 3, but for the area around the Philadelphia rail yards adjacent to the Schuylkill and Delaware Rivers.

Fig. 7.

As in Figs. 2 and 3, but for the area around the Philadelphia rail yards adjacent to the Schuylkill and Delaware Rivers.

Fig. 8.

As in Figs. 2 and 3, but for the 7-km area around the Pentagon, in the urban area of Washington.

Fig. 8.

As in Figs. 2 and 3, but for the 7-km area around the Pentagon, in the urban area of Washington.

Changes corresponding to seasonality are clearly evident in all of these figures, with the warmer months showing lower values of the variances than the colder months. It is not immediately clear why there are some cases in which and differ substantially throughout the day (as in several of the months illustrated in Figs. 68). In each of these cases, the area under scrutiny is dominated by residential buildings, and it seems possible that the result is a function of this difference from the closer surroundings of city centers as correspond to Figs. 2 and 3. Figure 5, for the southwestern suburbs of Chicago, appears to represent an intermediate case.

The daily variations evident in the diagrams are not closely linked to changes in wind speed, as can be seen in Fig. 9, which shows the wind speed dependence of the velocity standard deviation ratio Rst {}, for the case of Boston. There are no situations for which a dependence on wind speed is apparent, nor is there for any other of the areas considered. The performance limitations of the anemometers limit consideration of lighter winds than are depicted, but there appears to be no reason to expect a strong departure from the trends evident in Fig. 9—as wind speeds decrease, there is a consistent increase in the scatter associated with the prevailing value of Rst but there is no detectable trend in the average. Rst is clearly associated with the cycle of insolation, however. In each of Figs. 28, triangular marks along the abscissa show the times of sunrise, solar noon, and sunset. If conventional atmospheric stability was a controlling feature, then a role of wind speed should be apparent in the way that and are apportioned during that part of the day for which convection was active (and hence for a part of the time represented in Fig. 9). On first principles, it is expected that convection would increase the variance at individual locations and would, therefore, cause an increase in . The average wind fields would not be affected as much, and hence Rst should decrease as convection sets in. This is indeed found to correspond to the observations reported here, as is evident in Figs. 10 and 11, where the ratio Rst is plotted against time of day for the Philadelphia and Chicago cases, respectively. Inspection of the individual values of Rst from these sites shows that the standard deviations are similar to the linear averages, and hence the distributions are closer to lognormal than to normal. For this reason, the values plotted are computed geometrically and not linearly. This, in itself, appears to be a further indication that the wind regime measured by the SRL instrumentation is largely chaotic. The datasets from all of the considered sites behave similarly.

Fig. 9.

The dependence of the ratio of velocity standard deviations (space over time; Rst) on wind speed, using all available data for the Boston case regardless of the time of day.

Fig. 9.

The dependence of the ratio of velocity standard deviations (space over time; Rst) on wind speed, using all available data for the Boston case regardless of the time of day.

Fig. 10.

The variation of the velocity standard deviation ratio Rst on the time of day for the Philadelphia case.

Fig. 10.

The variation of the velocity standard deviation ratio Rst on the time of day for the Philadelphia case.

Fig. 11.

As in Fig. 10, but for Chicago.

Fig. 11.

As in Fig. 10, but for Chicago.

4. Further comments and conclusions

Studies of the lower atmosphere have been dominated by the analysis of detailed observations made at single, specially selected locations, most often situated so as to satisfy the demands of numerical models and above the height of direct influence by surface roughness elements (Oke 2004). One of the purposes in the study presented here is to move beyond the single-location constraint, in a way that might allow for the fact that areal dispersion in the surface roughness layer is influenced by more than just the statistics of the velocity vector at some single location. In the case of forecasting dispersion from a street-level release, initial diffusion takes place in a layer very close to the surface, below the tops of buildings or other surface components. Downstream dispersion involves velocity fields that are at heights sufficiently above the surface that details of the roughness layer lose importance (in “skimming flow”). In the former case, it is the local distribution of surface obstacles and the velocity fields caused by them that are of importance. In the latter, it is the areal characteristics of the surface such as its roughness length and displacement height. In both cases, the prevailing wind speed and convection are critical drivers of the relevant turbulence regimes (Roth 2000; Britter and Hanna 2003).

This analysis addresses an intermediate case. It confirms the following results:

  1. If the effects of convection are disregarded, then the street-level data collected in urban and suburban areas tend to display variances in space that are about the same as those in time. Average values for Rst are 1.00 ± 0.04 (New York—Queens), 1.02 ± 0.03 (Boston), 0.96 ± 0.03 (Chicago), and 0.87 ± 0.04 (New Orleans).

  2. Convection appears to increase the variance in time at specific locations more than the variance of the average wind components. That is, the onset of convection causes σ2(t) to increase more than σ2(s).

  3. The similarity between velocity variances in space and in time found here parallels the findings of Katul et al. (2004) for a forest canopy and those of Hanna and Zhou (2009) for a short-term experiment in New York City.

The overall situation invites more (and independent) attention. It is possible that these findings are little more than a consequence of taking different samples of the same randomly affected atmosphere; it appears clear, however, that the ratio Rst should decrease with increasing height, from a value (as found here) near unity within the roughness layer to a much smaller value at a level well above it. In the urban and suburban contexts of interest here, it seems likely that the value of unity should be a good first estimate, provided allowance is made for the (presently presumed) role of convection.

Most models used to predict dispersion within the urban roughness layer are based on understanding of averaged behaviors, as are addressed here. The analysis presented here shows, however, that the statistical distributions around these averages are such that the applicability of any prediction to any particular event is questionable. There appears to be no option but to provide guidance on a probabilistic basis, by propagating both averages and distributions through the computational systems.

The analysis presented here demonstrates that there can be value in data collected at sites that do not satisfy the siting criteria accepted by most contemporary researchers. There is need for a different view of urban data. Rather than dismiss data because they fail to satisfy some expectations that are based on criteria imposed (albeit with good intentions and sound reasoning) by modelers, among others, there is a growing need to look at such data with the intent to extract from them as much meaning as is possible. This is exacerbated by the flood of data following recent advances in data acquisition technology. There is an obvious corollary that appears applicable to investigative experiments: any deployment of instruments that is based on “representativeness” and on the predictions of models will tend to confirm the predictions of those models. A more rewarding approach might well be to deploy instruments preferentially at locations that are not representative, so that some first steps might be taken toward understanding actual exposure regimes better. Even a dense network of such sites, however, will not resolve such well-known and significant features of dispersion within the urban canopy as the channeling of pollutants along streets or their being waylaid in courtyards and short alleys. Investigating such matters will still require intensive studies with dedicated instrumentation that has been located with stricter, probably model-related, exposure criteria.

Note that this analysis is of averages of many occurrences and that the atmosphere might rarely conform to these averages. In common with similar micrometeorological and other atmospheric results, the conclusions should not be considered as appropriate for any specific case but rather as descriptive of ensembles of similar cases. Last, note that in this analysis there has been no site selection on the basis of instrument exposure or surroundings.

Acknowledgments

The data used in this analysis were provided under a memorandum of understanding between Earth Networks, Inc. (formerly AWS Convergence Technologies, Inc.), and the U.S. National Oceanic and Atmospheric Administration. The data were archived by the NOAA Atmospheric Turbulence and Diffusion Division (of the Air Resources Laboratory), located in Oak Ridge, Tennessee. Comments by anonymous reviewers contributed substantially to this presentation, and it is hoped that related discussion might continue. The work was supported by both Earth Networks, Inc., and the Air Resources Laboratory of NOAA. The principal author serves as a consultant to Earth Networks, Inc.

REFERENCES

REFERENCES
Britter
,
R. E.
, and
S. R.
Hanna
,
2003
:
Flow and dispersion in urban areas
.
Annu. Rev. Fluid Mech.
,
35
,
469
496
.
Grimmond
,
C. S. B.
, and
T. R.
Oke
,
1999
:
Aerodynamic properties of urban areas derived from analysis of surface form
.
J. Appl. Meteor.
,
38
,
1262
1292
.
Hanna
,
S. R.
, and
Y.
Zhou
,
2009
:
Space and time variations in turbulence during the Manhattan Midtown 2005 field experiment
.
J. Appl. Meteor. Climatol.
,
48
,
2295
2304
.
Hicks
,
B. B.
,
W. J.
Callahan
,
W. R.
Pendergrass
III
,
R. J.
Dobosy
, and
E.
Novakovskaia
,
2012
:
Urban turbulence in space and time
.
J. Appl. Meteor. Climatol.
,
51
,
205
218
.
Katul
,
G.
,
D.
Cava
,
D.
Poggi
,
J.
Albertson
, and
L.
Mahrt
,
2004
. Stationarity, homogeneity, and ergodicity in canopy turbulence. Handbook of Micrometeorology, X. Lee, W. Massman, and B. Lee, Eds., Atmospheric and Oceanographic Sciences Library, Springer, 161–180.
Oke
,
T. R.
,
2004
. Initial guidance to obtain representative meteorological observations at urban sites. World Meteorological Organization Rep. 81, Technical Doc. WMO/TD 1250, 47 pp.
Roth
,
M.
,
2000
:
Review of atmospheric turbulence over cities
.
Quart. J. Roy. Meteor. Soc.
,
126
,
941
990
.