Temperature and wind data from a rural micronet and nearby site of the Oklahoma Mesonet are analyzed to study the frequency, strength, and formation processes of cold-pool events in a region with gentle terrain. Spatial analyses were performed for a 2-yr-long temperature record from 26 temperature/humidity surface stations, deployed across a 120 m × 320 m micronet located in a region of gently sloped terrain with maximum elevation changes of ∼25 m. Cold pools frequently formed at the base of a gentle slope in a small depression of only ∼6-m depth that is also sheltered by trees. The strength of each cold-pool event was classified according to a cold-pool index based on average nocturnal temperature perturbations within the cold-pool region. Wind data collected with sonic anemometers on a 15-m-tall tower at the micronet for a period of three months (spring 2005) suggest that flow sheltering by vegetation plays an important role in the cold-pool formation. The wind data also show signatures of katabatic flow for about 50% of the strong cold-pool events. However, a heat budget analysis for these nights suggested that the katabatic flows were associated with warm-air advection along the slope and that if katabatic jets had penetrated the cold pool, they would have produced substantial warming in the region of the cold pool. Since such warming was not observed, it is concluded that the katabatic jets did not actually penetrate the cold pool but likely flowed over it. An analysis of Richardson numbers demonstrates that cold-pool formation frequently occurs under strongly stable conditions that tend to suppress vertical turbulent mixing in the surface layer. Observations that significant temperature changes can occur even with elevation changes on the order of 6 m have important implications in agriculture as well as in data assimilation.
The spatial variability of nighttime (NT) temperatures and correlation between low temperatures and topographic depressions has been widely discussed in the literature. As stated in Geiger’s (1965) review of early studies, names such as “cold island,” “frost hole,” “cold air pool,” or “frost hollow,” are often used to describe the phenomenon of cold-air buildup in low-lying areas. Since then, the variability of near-surface temperatures in terrain of different complexity has been investigated in a number of studies (Mahrt and Heald 1983; Tabony 1985; Thompson 1986; LeMone et al. 2003; Mahrt 2006), and the role of trees and forests on nocturnal cold-pool (CP) formation has also been discussed (Gustavsson et al. 1998; Blennow 1998; Karlsson 2000). Measurement campaigns were further conducted to study nocturnal temperature distributions and drainage flows in basins and limestone sinkholes in mountainous terrain (e.g., Whiteman et al. 2004; Clements et al. 2003) as well as in shallow gullies (Mahrt et al. 2001; Soler et al. 2002). One common finding of these studies is that large spatial variations in near-surface temperatures develop shortly after sunset under clear-sky conditions with weak synoptic forcings. However, there does not appear to be a unique physical–dynamical process producing the CP in those studies. In particular, the role of a katabatic flow in CP development is not always clear-cut, and it might actually have a negative feedback that stops CP intensification in some cases.
Acevedo and Fitzjarrald (2001) concluded that the rates of change of mean quantities during the early evening transition (EET)—the first two hours after sunset—are “many times larger than they are subsequently until sunrise.” In that study, the EET is described as the period in which the surface energy budget changes rapidly and a shallow stable boundary layer develops. Effects of surface heterogeneities on meteorological quantities can become more pronounced as a result of the limited turbulent mixing and related small eddy sizes within this stable surface layer. As turbulent mixing decreases, the surface layer becomes decoupled from the boundary layer and the air confined within the shallow surface layer cools very rapidly. Turbulence decays faster in valleys, sinkholes, gullies, other small local depressions, or regions sheltered by vegetation, whereas it typically persists longer on hill tops and other well-exposed areas. Higher wind speeds also make the EET less pronounced because increased turbulent kinetic energy prevents the surface layer from decoupling from the boundary layer. André and Mahrt (1982) concluded that “turbulent heat flux divergence and clear-air radiative cooling contribute about equally to the development of nocturnal surface inversions.” André and Mahrt (1982) also found that the inversion layer continues to grow because of radiative cooling effects but its lowest part, which is turbulent and strongly stably stratified, remains rather shallow throughout the night.
Whiteman (2000) illustrates that CP formation in valleys also happens because the cooling rate is directly proportional to the valley’s volume, which is always less than the volume of an air column over flat terrain with a surface area identical to the valley’s top cross section. Valley shape also plays a role, with convex sidewalls tending to further enhance CP formation compared to concave or V-shaped valleys. Whiteman et al. (2004) studied the role of reduced sky-view factors in valleys and presented an idealized three-layer model of surface cooling in which some of the downward radiation received at the sinkhole’s basin floor does not originate from the atmosphere but from the sinkhole sidewalls. The model reproduced the observed cooling rates in sinkholes of different sizes and sky-view factors well, and Whiteman et al. (2004) concluded that “sky-view factor is indeed the most important topographic parameter controlling basin cooling.” During the course of the night, the basin floor sees an increasingly cooler atmosphere. Thus, the air continues cooling throughout the night, though at a much slower rate after the initial rapid temperature drop during the EET.
Another factor often used to explain CP formation is the drainage of cold air along the slopes of valleys or basin walls. For example, Geiger (1965) discusses that cold pools in low-lying areas continue to grow throughout the night as a layer of cold air forms because of radiative cooling along slopes and drains downslope. More recently, LeMone et al. (2003) also attribute the low-altitude cold air observed during the 1997 Cooperative Atmospheric Surface Exchange Study (CASES-97) field campaign in a watershed area in Kansas to radiative cooling and pooling of drainage flows that form during nights with clear skies and weak winds. Sun et al. (2003), who studied the heat balance in the nocturnal boundary layer during the follow-up study, CASES-99, also associated near-surface temperature variations with transport of cold air despite an observed uphill flow. Mahrt et al. (2001) and Soler et al. (2002) confirmed that drainage flows occurred in a shallow gully within the CASES-99 watershed area during nights with large variations of near-surface temperatures. These flows were very shallow (a few meters depth), and the downslope flow speed was less than 1 m s−1. They also noted that gully flows developed for temperature differences larger than 6°C between the gully basin and highest elevation (elevation difference ∼8 m); but that temperature inversions of the same magnitude did not automatically coincide with gully flow. These findings suggest that drainage flows may not play a major role in those CP formations.
Fiebrich and Crawford (2001) and Hunt et al. (2007) further highlighted that microscale variations of nighttime temperatures can be significant even in regions with very small changes in terrain elevation and without significant obstructions. Hunt et al. (2007) documented CP development without organized drainage flows and stated that “the most likely cause for the observed conditions was because of in situ cooling.” Thompson (1986) argued that the widely used explanation of CP formation by drainage flows is not always supported by experimental findings and suggested that the air in hollows or valleys is colder as a result of the diminishing turbulent heat transfer in sheltered locations, not because of cold-air drainage. Clements et al. (2003) further concluded that their measurements in the Peter Sinks sinkhole in Utah did not agree with the conceptual model of CP formation in valleys and basins, which assumes that cold pools deepen because of cold-air drainage. They observed only very weak (∼0.2 m s−1) and shallow (∼1 m) downslope flows and thus attributed only a minor role in CP formation to the downslope flows. Turbulent heat fluxes tended to cease ∼1.5 h after sunset, and the continued cooling throughout the night was explained by radiative flux divergence.
Thompson’s (1986) conclusion that the reduced or even absent turbulent heat transfer in sheltered locations is likely a major factor in the cooling also agrees with findings that CP formation is more prevalent in wooded areas in which flow sheltering by vegetation becomes important (Gustavsson et al. 1998; Blennow 1998; Karlsson 2000). The atmosphere in dense tree canopies is characterized by lower wind speeds and reduced mechanical mixing. Accordingly, the flow inside the canopy often becomes decoupled from the flow above the canopy, and downward momentum and heat transfer inside the canopy is very low (Kaimal and Finnigan 1994). However, the reduced sky view inside a canopy can prevent excessive cooling and lead to higher minimum temperatures than over open ground (Kalma et al. 1986). The observation that clearings within forests are prone for cold hollows (Gustavsson et al. 1998; Karlsson 2000) suggests that these regions may be subject to reduced mechanical mixing but increased sky view.
The foregoing discussion illustrates that although CP formation in complex terrain has been widely investigated, basic questions about the formation mechanisms, such as the role of drainage flow and the competing roles of flow sheltering and radiation trapping in forested areas, still exist. Previous studies have also mostly focused on horizontal temperature variability over a network of surface stations deployed over an area on the order of tens of kilometers (Mahrt and Heald 1983; Tabony 1985; LeMone et al. 2003; Gustavsson et al. 1998; Blennow 1998; Karlsson 2000), or over deep valleys and on steep and/or larger-scale slopes (see Mahrt et al. 2001 and references therein), or in enclosed basins in complex terrain (e.g., Whiteman et al. 2004; Clements et al. 2003) with relatively short measurement periods covering only a few weeks or months.
The present study focuses on the development of strong temperature gradients in a very shallow, partially sheltered, local depression at the base of a small hill (20-m elevation gain) bordering a lake in central Oklahoma. A dense network of surface stations, the Lake Thunderbird Micronet, was deployed from the hilltop down to the lake. Additionally, a 15-m-tall instrumented meteorological tower was situated in the center of the Micronet. A 2-yr-long temperature record of high spatial resolution was analyzed along with wind data for a 3-month period. Our study provides a detailed look at the microclimate of a region typical of many agricultural and recreational landscapes in the Midwest of the United States and elsewhere. On the one hand, our site is characterized by fairly complex small-scale surface features (planar hill slope, sheltered depression, forest, lake shore) but on the other hand, our study area is relatively unaffected by larger-scale terrain effects. Despite their ubiquity, there are very few detailed investigations of such settings. Perhaps the most similar kind of study was the shallow drainage flow study of Mahrt et al. (2001). However, even though the horizontal and vertical length scales of the Micronet site are similar to those considered in that study, the site topography and surrounding terrain are distinctly different: Mahrt et al. (2001) studied drainage flows in a shallow V-shaped gully that was also influenced by larger-scale (10 km) drainage flows.
Perhaps the most striking microclimatological feature at the Micronet is the relatively frequent development of strong nocturnal cold pools in the local, shallow depression at the base of the hill. We investigated the climatology—frequency, strength, and temporal evolution of the CP events—for different seasons. Although seasonal effects exist, CP events are not limited to the winter months. The wind data were used to investigate the near-surface conditions along the slope during CP events.
As mentioned earlier, the role of katabatic flow in CP production remains controversial. This controversy may be, in part, due to misconceptions about the nature of katabatic flows. As described in Shapiro and Fedorovich (2008), when a sloping surface is cooled, a temperature difference is setup between the air in the surface layer and the environmental air at the same elevation. The buoyancy force associated with this temperature deficit projects in the downslope direction and induces a downslope flow (katabatic flow). However, this process does not require that the along-slope temperature gradient be positive or negative and thus both along-slope warm- and cold-air advection are consistent with katabatic flows (Shapiro and Fedorovich 2007). On the basis of our findings, katabatic flows do develop over the gentle Micronet slope, but the CP events cannot be explained by the pooling of cold air that flows downhill, because in all cases the along-slope temperature gradient points uphill. In fact, energy budget calculations based on our data indicated that if the katabatic jet had penetrated the CP region, the warm-air advection would have produced very substantial warming in the CP region. Since such warming is not observed, we speculate that the relatively warm air (with respect to the CP) in the katabatic jet flows over the CP. Accordingly, in situ cooling forced by radiative heat losses and diminishing turbulent heat transfer in sheltered regions are likely the main causes for CP development at the Micronet.
2. Research methods
a. Description of experiment site
The data analyzed in this study were primarily collected at the Lake Thunderbird Micronet, which is located 15 km northeast of the National Weather Center (NWC) in Norman (NRMN), Oklahoma. The general terrain in the study area is characterized by mostly north–south-oriented undulating ridges, with elevations ranging between about 1050 and 1150 ft MSL (Fig. 1a). The micronet itself has 28 stations spaced roughly 30 m apart and covers an area of approximately 120 m × 320 m. It was established as an outdoor laboratory for environmental research and student training (Shapiro et al. 2009).
Measurement sites stretch from the shoreline of Lake Thunderbird up to a small hill, with the highest measurement sites being located roughly 300 m from the shoreline and 20 m above the water level of the lake (Fig. 1). The main topographic feature is a gentle slope downward from north to south, with the ridge top being slightly north (outside of) the micronet. The highest point of the micronet is in the middle of its northern border. The mean elevation gradient from south to north is ∼65 m km−1 or roughly 4°, with a smaller downward slope from west (site 19) to east (site 21) in the southernmost 100 m of the micronet (gradient ∼58 m km−1). The southeast part of the micronet (stations 20–23) is relatively flat and represents a local minimum in elevation, as the terrain slopes down into this area from the west, north, and east. Figure 2a shows the portion of the micronet south of station 8. Stations 22–23 are located in the wooded area near the lakeshore, while stations 16–18 and 20–21 are located in the clearing just north of this wooded area (see also Fig. 1b). It can be clearly seen how the terrain slopes into this region from the west and from the north. Station 27 (Fig. 2b) on the southwestern edge of the micronet is closest to the lake. More details about the micronet are given in Shapiro et al. (2009).
b. Instrumentation and data collected
Each micronet surface station is equipped with a HOBO H8 Pro RH/Temperature logger placed inside the radiation shields and mounted on wooden posts 1 m above ground (Figs. 2b and 2c). The operating range of the temperature sensor is specified as −30° to +50°C, the accuracy as ±0.2°C, the resolution as 0.02°C, and the response time in still air as 34 min. For our studies, we analyzed temperature data recorded every five minutes at 26 surface stations (of the original 28 sites, the logger at site 10 was defective and a micrometeorological tower replaced station 8 in 2004). The study period was from 1 January 2005 to 31 December 2006, although data are missing for the periods 25–31 January in both 2005 and 2006. Before their original deployment in 2002, the HOBO H8 Pro RH/Temperature loggers were tested in the laboratory, with all sensors collecting data for about one week while being exposed to the same temperatures. These initial tests revealed no drift errors for any of the sensors. The sensors then remained out in the field and were only briefly checked and cleaned during each of the downloading visits every 6–10 weeks. An assessment of the sensors’ performance after being out in the field for such a long time is included in section 2c.
The micronet data are compared with temperature data measured at 1.5-m height at the NRMN Oklahoma Mesonet site, 17 km west of the micronet. The layout and instrumentation used at Oklahoma Mesonet sites are described in McPherson et al. (2007). Wind speed and wind direction data, measured at NRMN, are also used in the data analysis. To investigate the role of atmospheric stability in CP formation, the gradient Richardson number (Ri) at NRMN is estimated using (Stull 2000)
where g is the acceleration due to gravity, Γd ≅ 0.01°C m−1 is the dry adiabatic lapse rate, and T9m and T1.5m are the air temperatures measured at 9 and 1.5 m above ground, respectively. The wind velocity components u2m, u10m, υ2m, and υ10m at the 2- and 10-m levels are computed using the wind speed at the respective level and the wind direction at 10 m. The height differences between the NRMN measurement levels are ΔzT = 7.5 m for air temperature and Δzu = 8.0 m for wind speed.
From 12 March to 31 May 2005, wind and temperature data were also collected at an instrumented 15-m-tall micrometeorological tower that replaced station 8 (Fig. 2d). The base of the tower is about 170 m north and 12 m above stations 20–22, which are used to identify the peak (negative) temperature perturbations of the CP. Five RM Young 81000 sonic anemometers were mounted at the tower at 1.5, 3, 6, 10, and 15 m, respectively, and time series were sampled with a frequency of 5 Hz. For each of these levels [hereinafter (micronet tower level) MTL1–MTL5], 5-min statistics of the wind data were calculated and included in the CP analysis.
Our analysis focuses primarily on nocturnal temperature variations beginning with the onset of the early evening transition. Therefore, we have chosen 0200 to 0755 UTC (2000–0155 LST) as a daily study period, which will hereinafter be referred to as NT hours.
c. Data analysis
A variety of quality-assurance (QA) checks have been applied to the daily records of the 5-min data from the 26 surface stations operating in 2005 and 2006. First, daily records for sensors recording any 5-min temperature value outside of the sensors operating range, that is, below −30° or above 50°C, were flagged (QA criterion 1). Only 25 daily site records violated these checks, that is, less than 0.2% of the total datasets for the 2-yr study period.
To check for potential drift problems, the data were also screened for unreasonably large deviations of individual site temperatures from the spatially averaged temperatures. At first, the temperature perturbation for each 5-min average temperature record r at each station i were calculated according to
with the spatial average temperature 〈Tr〉 defined as
where N is the total number of available records from the 26 sites. Then, the QA criterion 2 flags were applied to daily records that violate the criterion |dTi,r|daily max ≥ 8°C or |dTi,r|daily mean ≥ 4°C; however, this second QA screening did not trigger any further flags.
As a next step, the QA-checked records were visually compared with the mesonet temperatures T1.5m in daily plots depicting the temperature time series measured at all 26 HOBO sites, the spatially averaged micronet temperature, and the mesonet temperature (Fig. 3). Although on many days, for example, 16 March 2005, the temperatures measured at the different HOBO sites agree very well with each other and with the mesonet data (Fig. 3c), a high spatial variability of temperatures across the micronet, particularly during nighttime hours, is observed on many other days, such as on 12, 13, and 17 March 2005. (Figs. 3a–3d). Differences can also be noted in the daytime temperatures; however, these variations will not be further investigated in the present study, as we currently cannot quantify potential measurement errors of the daytime readings caused by radiative heating.
To check for drift errors during the 2-yr-long observation period, measurements at the 26 micronet sites and the Norman mesonet site were compared for all nights for which the nighttime average Richardson number was less than 0.05. This criterion was chosen to select nights with strong mechanical mixing during which horizontal temperature gradients should be small (LeMone et al. 2003). For the 2-yr-long record, the criterion < 0.05 was met for 65 cases. For these cases, the nighttime 5-min temperature records (2:00–7:55 UTC) were selected for each of the 26 sites and compared against the corresponding spatial average 〈Tr〉 (Fig. 4a). The data from each site agree very well with the spatial average 〈Tr〉, following almost perfectly the one-to-one line with very little scatter. The larger scatter when comparing 〈Tr〉 with the Norman mesonet temperatures TNRMN (Fig. 4b) can be explained by the distance between the two measurement sites (17 km) and the different time constants of the sensors used at the micronet (34 min in still air) and mesonet (10 s in still air). The 65 cases with < 0.05 included some nights with frontal passages (e.g., on 11 January 2005; Fig. 4c) for which both the geographical site separation and the different sensor time constants can cause a temporal shift in the temperature trends. However, a linear regression between 〈Tr〉 and TNRMN resulted in the coefficients slope a = 0.997 and intercept b = 0.121 and the correlation coefficient R = 0.997, which shows that the agreement between the micronet and mesonet temperatures is overall very good. Linear regressions for each of the sites against 〈Tr〉 resulted in a mean slope a of 1.000 (minimum value 0.991 for site 24, maximum value 1.006 for site 25), mean intercept b of −0.004 (minimum value −0.189 for site 11, maximum value 0.195 for site 23) and a correlation coefficient R = 1.000 for all sites. The values for the intercept b indicate that some sites may have a small warm or cold bias, but the possible bias values are still within the sensors’ specified accuracies of ±0.2°C.
Given the good agreement of the micronet temperatures for nights with strong mechanical mixing, it was concluded that the observed spatial variations of nighttime temperatures are not caused by measurement errors but are related to CP development. Figure 3 illustrates that under certain conditions, significant temperature variations persist over the whole nighttime period (0200–0755 UTC) or even longer but that shorter and interrupted CP events are also common.
To classify the nocturnal cold events, the average NT temperature perturbation for each site, dTi,NT, is defined as the mean value of all 72 individual 5-min records collected from 0200 to 0755 UTC, that is,
Monthly averages of dTi,NT (Fig. 5a) clearly show that the lowest temperatures are observed at stations 20–22, which are located in the southern part of the micronet, near the lowest elevations (Fig. 1). Stations 20 and 21 are in a clearing to the north of a thick forest bordering the lake, while station 22 is within the forest. Stations 23–27 have a tendency for positive (warm) temperature perturbations, particularly during the summer months. Of these stations, 25–27 are close to the lake and in the forest, whereas station 24 is in a clearing sheltered by forest on two sites and at a slightly higher elevation than stations 20–22.
On the basis of these results, the average temperature perturbations for stations 20–22 are used to identify the peak cold-pool temperature perturbation
and a CP index (CPI) is defined as the average nighttime temperature perturbation for stations 20–22:
Category 1: −0.5°C ≤ CPI: No CP case.
Category 2: −1.5°C ≤ CPI < −0.5°C: Weak CP case.
Category 3: CPI < −1.5°C: Strong CP case.
Although the threshold CPI values for the three categories are somewhat arbitrary, they appear to capture the observed phenomena quite well. During summer, CPI values less than −1.5 are rare; however, for the rest of the year, the distribution is almost bimodal, with two peaks for CPI ≈ ±0.5 and CPI ≈ −2. Also, days with persistent, strong CPs, such as 17 March 2005 (Fig. 3d), are characterized as strong CP cases, while days with shorter or interrupted CP development, such as on 12 and 13 March 2005 (Figs. 3a and 3b), typically fall into the weak CP category, even though these cases were very strong for portions of the night. The monthly frequencies of weak and strong CP events during 2005 and 2006 are shown in Fig. 6. Although the number of weak CP events does not show a clear seasonal trend, strong CP events, as expected from Fig. 5b, are less frequent during summer months. The remaining analysis focuses on identifying the environmental conditions that favor CP formation and clarifying which mechanisms play a major role in CP formation.
3. Results for the intensive observation period in spring 2005
a. General trends
Our analysis first focused on the period 12 March–31 May 2005, for which wind measurements with sonic anemometers were obtained at the micronet tower. During this period, 15 days were classified as strong cold-pool cases (Table 1). The astronomical sunrise and sunset times as well as onset and peak times of the CP development for these days are shown in Table 2. Cloud cover data from the Oklahoma City (KOKC) Automated Surface Observing System (ASOS) site at Will Rogers World Airport in Oklahoma City and infrared satellite data showed that skies were nearly clear (cloud cover fraction <1/8) at sunset for all but one case. For 20 March 2005, some scattered midlevel clouds were present from 2200 to 0600 UTC. While the clouds reduced radiative cooling, an area of high pressure centered over Kansas maintained stable conditions at the surface. During the night, cloud cover became more variable for five cases; however, for most of these nights, increases in cloud cover did not significantly affect the strength of the CP. But on 14 March 2005, the CP was interrupted at 0700 UTC as cloud cover increased, light precipitation began to fall, and the associated decrease in stability and an increase in wind speeds caused the CP to weaken. Overall, it can be concluded that CP formation occurred primarily on clear nights that supported rapid radiative cooling.
The peak temperature perturbations dTp [see Eq. (5)] are also listed for each case in Table 2. The cold-pool onset is defined as the time that the average temperature perturbation at stations 20–22 drops below −1.5°C, whereas the CP peak is defined as the time that the average temperature perturbation reaches its lowest value dTp in the CP region. For 10 out of the 15 cases, the CP develops in less than one hour after sunset; in 3 cases, CP onset occurred even before astronomical sunset; in the 2 remaining cases, it developed 110 and 148 min, respectively, after astronomical sunset. The timing of the CP formation is thus very similar to the findings of previous studies (Acevedo and Fitzjarrald 2001; LeMone et al. 2003), and the EET of the boundary layer apparently plays an important role in the CP formation at the micronet. However, the time between CP onset and peak is more variable, with the peak being reached just 25 min after CP onset on 17 March 2005 but 5 h and 20 min after CP onset on 16 April 2005.
The average nighttime wind speed and wind direction data for the Norman mesonet site at 10 m and the micronet tower at 1.5 and 15 m (Table 1) illustrate that for all strong CP cases, fairly low wind speeds prevailed. With the exception of one case, the 10-m winds at the Norman mesonet site were <3 m s−1; for the majority of strong CP days, the winds at this site were <2 m s−1.
At the micronet tower, the wind direction at 1.5 m was within ±30° of due north in 12 out of the 15 strong CP cases. We would expect the downslope (northerly) wind component to be stronger near the surface than higher up for a pronounced katabatic flow signature (i.e., locally forced katabatic flow with weak synoptic-scale forcing). This could occur with (i) northerly wind directions at both the 1.5- and 15-m level but higher wind speeds at 1.5 m than at 15 m or (ii) lower wind speeds at 1.5 rather than at 15 m but significant wind veering with northerly near-surface winds and the 15-m wind directions being from a different sector. On the basis of the average nighttime wind speed data given in Table 1, condition (i) was never met, but the wind directions recorded at 15 m were frequently outside the northerly sector. Taking into account that for the micronet, a stronger downslope component is equivalent to a larger negative value of the υ velocity component, condition (ii) is satisfied for the seven cases highlighted in bold in Table 1. We will refer to these cases as potentially katabatic flow cases. However, when using this terminology, we do not imply that the CP formed as a result of the downslope pooling of cold air. As will be shown next, the same mechanism (radiative cooling) that leads to CP formation in sheltered regions also induces downslope flows; however, these flows are associated with pronounced warm-air advection on the micronet slope.
For the remaining eight strong CP events, either the wind direction at the lowest tower level was not within the northerly sector or the directional shear between the lowest and the top tower level was so small that the magnitude of the downslope wind component near the surface was less than at the 15-m measurement level. For these cases, the wind data do not provide a compelling signature for katabatic flow, and we will refer to such cases as nonkatabatic cases. It should be noted that for two of these cases, 3 and 17 April 2005, the wind speeds at 1.5 and 15 m were both southerly, which clearly excludes drainage flow as a factor in CP formation for these two days.
Curiously, peak CP temperature perturbations observed in katabatic and nonkatabatic cases are very similar (Table 3). One might expect that the strongest katabatic flow events occur during nights with the strongest radiative cooling and that simultaneously the temperature perturbations increase as radiative cooling becomes stronger. Intuitively, a stronger temperature perturbation signature should thus be observed for the potentially katabatic flow cases. One possible explanation why this is not the case could be that the katabatic flow causes warm-air advection into the CP region, similar to the down-valley transport of warm air that leads to reduced cooling rates in well-drained valleys (Whiteman 1990). A second, related explanation is that the katabatic current is warmer than the CP and that the katabatic current flows over the CP instead of penetrating it. In this latter case, mixing of the air at the horizontal interface between the CP and the overlying warm air would act to moderate the CP intensity. The typical northward-directed quasi-horizontal temperature gradient north of the CP in these cases suggests that either explanation is plausible.
It is thus instructive to quantitatively consider the role of advection in the thermodynamic energy budget in the surface layer at the tower site for the CP cases that had a pronounced katabatic signature. The relevant equation can be written in terrain-following coordinates with unit vectors it pointing eastward across the slope, jt pointing northward along the slope, and kt pointing upward in the slope-normal direction as
where T is air temperature, ut = utit + υtjt is the along-slope wind vector, ∇t is the gradient operator along the slope surface, wt is the slope-normal velocity component, zt is the slope-normal coordinate, and Q̇ is the residual forcing term, which includes contributions from radiative and turbulent flux divergences. It should be noted that the anemometers were carefully leveled so that the measured wind components actually indicate the flow components due east, north, and in the antigravity directions rather than the components in a terrain-following system. Near the ground, in the cases we are considering, the along-slope wind vector is basically northerly (ut ≈ 0, υt < 0); since the slope angle α is small (∼5.5° slope angle at the tower, calculated from the elevations of the sites immediately north and south of the tower), we can approximate υt ≈ υ (error less than 3%), which results in ut ≈ υj, with υ being the south–north-oriented wind component in the anemometer coordinate system. Accordingly, (7) becomes
Focusing on the typical behavior at 0300 UTC (when both the near-surface temperature near the tower site is dropping and the CP has been established and is further intensifying), we estimate a typical cooling rate of 1.6°C h−1. It is significant that in our dataset, in every CP case with a katabatic signature, the along-slope temperature increased toward the north (∂T/∂yt > 0) while υ < 0. The average value (across seven cases) for the along-slope advection term, estimated as −υ(∂T/∂yt) ≈ 26.6°C h−1, would thus tend to contribute to very strong warming at the tower site. The fact that the temperature at the tower site was decreasing indicates that the slope-normal advection and/or the residual forcing term Q̇ would need to be large and negative. Unfortunately, accurate measurements of the slope-normal temperature gradient at the tower site were not available for spring 2005. However, assuming a typical value of 1°C m−1 for the slope-normal temperature gradient in the stable near-surface layer (Mahrt et al. 2001; Sun et al. 2003), the value of the slope-normal velocity component wt required for −wt(∂T/∂zt) to exceed −υ(∂T/∂yt) would only need to be larger than ∼0.007 m s−1. Given a mean value (based on the wind data for 0200–0400 UTC for the seven katabatic cases) of υ = −0.37 m s−1 and noting that wt = −υ sinα + w cosα, the requisite vertical velocity w at the tower could actually be as low as −0.03 m s−1. This value is significant, because it indicates that very weak subsidence (with respect to the true vertical) can, nevertheless, be associated with a sufficiently strong positive slope-normal velocity component to produce the requisite, compensating cooling. The subsidence observed at the tower site at z = 1.5 m was actually slightly larger, with the average vertical velocities measured during 0200–0400 UTC for the seven katabatic cases being w = −0.04 m s−1, which would indicate that the slope-normal advection term also contributes to warming. However, it should be noted that the accuracy of the sonic wind velocity measurements is ±1% rms and ±0.05 m s−1, which exceeds both the measured average values and the requisite compensating/exceeding value for w. Moreover, regardless of the assumed vertical temperature gradient, given our υ and w data, misalignment of the sonic anemometer as small as 1° could yield a positive slope-normal velocity component wt. Difficulty in calculating the vertical advection terms as a result of sensor limitations has also been noted in Sun et al. (2003). Whiteman (1990) also addresses the often-significant errors in estimates of the individual budget terms. Given these uncertainties with the slope-normal temperature advection, we did not feel comfortable evaluating the residual forcing term. However, important results of our budget analysis are that (i) along-slope advection would consistently and strongly oppose the local cooling at the tower site in every katabatic case; and (ii) it is plausible that slope-normal advection could compensate the along-slope advection, but it is extremely difficult to quantify.
For the CP region, where the typical cooling rate at 3 UTC is also −1.6°C h−1 (evaluated for site 21), the lack of wind measurements further complicates a budget analysis. However, if one assumes that the katabatic flow observed at the tower site penetrates down into the CP region, the main horizontal advection term −υ(∂T/∂y) would result in an average warming rate of ≈52.7°C h−1 in that region. This increased warming rate is associated with an average 80% larger surface temperature gradient ∂T/∂y at the edge of the CP region (estimated using data from sites 13 and 21, this warming rate, based on inspection of the temperature patterns, is likely a conservative estimate) than near the tower (gradient between sites 5 and 11). Given that the penetration of the katabatic jet would result in dramatic warming in the CP region rather than the observed cooling, we speculate that the katabatic flow does not penetrate into the CP but flows over it.
b. Case studies
In this subsection, the temperature and wind patterns observed during three of the cold-pool days are discussed to provide a better picture of the nature of the CP and its development. Results for 3 April 2005, the day with the lowest CPI and peak temperature perturbations (Tables 1 and 2), are also shown in Shapiro et al. (2009).
The first case chosen is 17 March 2005, for which the temperature time series from all surface stations are shown in Fig. 3d. Selected temperature time series and the wind data from the Norman Mesonet site and the Micronet tower are presented in Fig. 7. The CP started to form shortly after 0000 UTC; at 0040 UTC, just 2 min after astronomical sunset (Table 2), the temperature perturbation threshold of −1.5°C at stations 20–22 was reached, with the lowest temperatures recorded at station 22. As previously mentioned, the CP reached its peak strength with dTp = −3.11°C just 25 min later, at 0105 UTC. During the next eight hours (Figs. 8c–8f), the strength of the CP and its location remained relatively unchanged. During this period, the wind speeds recorded at the Norman mesonet and the micronet tower are rather low (Fig. 7), and a clear directional shift at the different measurement levels can be noted. Wind roses (Fig. 9), based on 5-min wind speed and direction at the 1.5- and 15-m level of the micronet tower during the period 0000–0955 UTC, further illustrate that a situation suggestive of katabatic flow persisted: winds at the 15-m level were predominantly westerly and near the ground, they were mostly north-northwesterly (downhill). The CP lasted until wind speeds started to increase shortly after sunrise. The increase in wind speed, presumably a result of convective mixing, coincides with the vanishing of katabatic flow.
The second case of a potentially katabatic CP day is 1 May 2005. The diurnal temperature and wind speed variations (Fig. 10), and the temperature perturbation maps (Fig. 11) for this day illustrate that the conditions were very similar to the ones on 17 March 2005. Interestingly, the onset of the CP at 0055 UTC was 19 min before the actual astronomical sunset for that day (Table 2). The CP peak strength (dTp = −2.59°C) occurred at 0155 UTC, one hour after the onset time. Relative to 17 March 2005, the CP starts to weaken about two hours earlier, shortly after 1000 UTC, but vanishes at around the same time, at 1400 UTC, when a marked increase in the wind speeds can be noted. The region in which the CP starts to form and the general area where it becomes strongest also agree with the findings for 17 March 2005, but the lowest temperatures are observed at station 21 in the clearing just north of the wooded area adjacent to the lake, whereas they previously occurred at station 22 inside the wooded area. Throughout the night, the wind speeds are very low, with substantial directional shear between the near-surface tower level (MTL1) and the 15-m tower level (MTL5). Although the near-surface winds are, again, predominantly northerly, the 15-m level winds are easterly. Since westerly winds at MTL5 are observed on 17 March 2005 (Fig. 12), it appears that katabatic flow conditions may develop for either westerly or easterly 15-m level winds.
A clearly nonkatabatic CP case was observed on 17 April 2005 (Figs. 13 –15). CP onset occurred 22 min after sunset at 0125 UTC (Table 2), and the CP peak (dTp = −2.48°C) is reached at 0525 UTC, that is, four hours after CP onset. One interesting feature of this case is that the CP starts to weaken at around 0600 UTC, but persists, though weaker, for several hours and even regains strength shortly after 1000 UTC before it vanishes at 1400 UTC. During the period of diminished temperature perturbations from 0600 to 1000 UTC, the wind speeds increased by more than a factor of two. In general, the highest nighttime tower wind speeds of all strong CP days were recorded on that day (Table 1 and Fig. 15). It is also interesting that the wind direction on all tower levels was consistently from the south (Figs. 13 and 15), that is, downhill drainage flow was not occurring. It is likely that wind sheltering by vegetation played a role in the CP evolution on that day. A strip of fairly dense vegetation with trees and shrubs runs along the lake just south of the clearing in which sites 20 and 21 are located (Fig. 1). For southerly wind directions, sites 20 and 21 are thus in an area in which wind sheltering by the trees is likely. It can also be noted that the temperature time series at station 22, which is within the wooded area but closer to the lake and thus for southerly flow is not as sheltered as sites 20 and 21, shows a wave-like pattern of fluctuating temperatures. Following the arguments of Gustavsson et al. (1998), the sudden warming events that interrupt the CP evolution near site 22 during that night could be caused by periods of enhanced vertical mixing in the more exposed area closer to the lake. Another possibility could be that “pockets” of warmer air get advected from the lake toward station 22 by southerly wind gusts. The strong warm bias of station 27, located closest to the lake shoreline, supports the notion that the lake has already experienced sufficient warming throughout the spring season and that nocturnal warm air from just above the lake surface is plausible.
4. Average cold-pool characteristics in 2005 and 2006
Although the analysis of the spring 2005 intensive observation period (IOP) data allowed us to identify environmental conditions that favor CP formation, the statistical representativeness of the findings is questionable because of the limited amount of data. To obtain better statistics, all available micronet datasets from 2005 and 2006 have been further analyzed. Table 3 provides an overview of the climatology for the three CP categories in 2005 and 2006. The average temperature perturbations for all no, weak, and strong CP days during 2006 are shown in Fig. 16 (the plots for 2005 are very similar to those for 2006; thus, they were not included in the paper). Even for the no CP cases, a small cool bias can be noted in the lower portions of the micronet, which is likely caused by the selection criterion of the no CP cases. All cases with a CPI > −0.5°C are classified as no CP events; this criterion allows for small negative temperature perturbations to exist in the general CP region. A clear CP signature exists for the weak CP category, for which the individual results are quite variable with respect to CP duration and strength. Figure 16 shows how the CP intensity increases for the three categories, while the region of CP formation does not change much. A similar trend can be noted in the temporal evolution of the CP, as illustrated by the average temperature perturbation time series at selected sites for the three categories (Fig. 17). Although the CP duration slightly increases when comparing the no, weak, and strong CP cases, it is mostly the magnitude of the temperature perturbations that becomes more prominent for the strong CP events.
The analysis of the IOP data has clearly shown that very light winds are common during strong CP events. The average wind conditions for the three CP categories in 2005 and 2006 further confirm the strong negative correlation between wind speed and CP formation (Table 3 and Fig. 18). During strong CP events, the average wind speed at the 10-m level of the NRMN mesonet is roughly a factor of 2 less than the average wind speeds on no CP days. The wind roses (Figs. 18 and 19), which are based on the daily average nighttime wind conditions, further illustrate that during weak and strong CP events, southerly wind directions at NRMN are prevalent (Fig. 18). The fact that CP events primarily occur for southerly wind directions is particularly interesting, as northerly winds, even though less frequent than southerly winds, are quite common for the period with the highest frequency of CP events (September–April). Northerly wind directions, however, are typically associated with higher wind speeds and thus less favorable for CP development. Additionally, these results suggest that flow sheltering by vegetation, which is expected to be most effective for southerly wind directions when the CP region is located in the wake of a wooded area (Fig. 1b), plays a major role in CP formation at the micronet site. For northerly wind directions, the upwind terrain of the CP region has some individual trees but the vegetation is overall sparser (Fig. 1b) and vegetation sheltering is thus less likely. At the 15-m level of the micronet tower, such a clear trend of southerly wind directions being favorable for CP formation is not evident; however, during strong CP events, northerly winds clearly become dominant (Fig. 19) at the micronet slope near the surface (1.5-m level). This, again, indicates that drainage flows do occur; however, given the observed temperature patterns, they do not start or strengthen the micronet CP. As discussed above, they may in fact weaken it because of warm-air advection into the CP region and possibly also by enhancing vertical mixing.
Finally, the influence of atmospheric stability was assessed by analyzing nighttime averages of Ri, computed using the NRMN data for wind speed and temperature according to Eq. (1). The average Ri increased from 0.14 to 0.18 for no CP events, to 0.37–0.45 for weak CP events, and to 0.75–0.89 for strong CP events (Table 3). The tendency of CP formation being favored under more stable conditions is further demonstrated by Ri histograms computed for the three CP categories (Fig. 20). Taking into account that Ric = 0.25 is often specified as the critical value up to which flows remain turbulent under stable conditions (Stull 2000), it becomes obvious that the majority of strong CP events occurred for conditions with very limited vertical mixing. Locally, Ri values were likely even larger because compared to NRMN, mechanical mixing was further reduced in the CP region because of the already-discussed wind sheltering by the micronet vegetation. However, for improving short-term minimum temperature predictions for similar sites that are important for agricultural and recreational interests, regional stability estimates can prove useful.
5. Conclusions and future studies
The analysis of the 2-yr-long temperature records from a Lake Thunderbird Micronet site on gently sloped terrain with patchy vegetation has shown that strong cold pools are frequently observed at this site. During such CP events, surface temperature differences of the order of 3°–4°C were typical with an elevation change as small as 6 m over 70–100-m distances. The CP location consistently coincides with the area of local minimum in elevation in the southeast corner of the micronet. Previous findings (Acevedo and Fitzjarrald 2001; Gustavsson et al. 1998) that show that CP onset is linked to the EET of the boundary layer and that conditions with weak winds and strong stability are favorable for CP formation are confirmed. In situ cooling in the presence of wind sheltering due to topography and vegetation appears to be important for CP formation. This mechanism was also pointed out in other studies (e.g., Gustavsson et al. 1998; Clements et al. 2003), but drainage flow is still often used as additional explanation for the development of nocturnal cold pools (e.g., LeMone et al. 2003; Sun et al. 2003). Although signatures of katabatic flow were also identified in about 50% of strong CP events during a 2005 IOP, thermodynamic energy budget calculations for these katabatic flow cases have clearly demonstrated that the observed near-surface wind and temperature patterns promote significant warm-air advection into the micronet’s CP region. Therefore, pooling of cold air as a result of drainage flow can clearly be excluded as a factor causing the CP development at the micronet. The lack of accurate information about all relevant terms in the budget equation prevented us from closing the budget. However, since the estimated warming rates in the CP region due to horizontal temperature advection were on average ≈52.7°C h−1, we speculate that the katabatic jet does not actually penetrate into the CP but flows over it. We further concluded that the likelihood of katabatic flow during CP events reflects that the same mechanism that causes CP formation in areas with reduced mixing and strong radiative cooling also causes katabatic flow to develop along the micronet slope. The role of katabatic flow in CP development may be stronger in regions with more significant elevation changes and particularly in mountain valleys with thermally driven slope and valley winds. However, even for basins in mountainous terrain, Thompson (1986) and Clements et al. (2003) also concluded that drainage flows were not an important factor in CP development.
The obtained findings of frequent and significant CP events in gentle terrain are of great relevance for agriculture, as temperature variations of a few degrees can cause severe loss in crop and fruit harvests. Our analysis also reveals that information from a nearby operational monitoring site of the Oklahoma Mesonet (McPherson et al. 2007) can be used to identify the regional-scale atmospheric conditions favorable for CP formation at the micronet. The gradient Richardson number estimated with routinely collected mesonet temperature and wind data provides skill in characterizing the conditions for strong CP formation. The results of this analysis may provide useful information for future improvements in minimum temperature prediction for similar terrain and can potentially be used to issue warnings to farmers about the likelihood of CP events. For recreational activities, knowledge of such small-scale microclimate effects may be useful in selecting a desirable camp site, as precise location may have a significant effect on nighttime comfort.
Additionally, the results are also of relevance for numerical weather prediction (NWP) models. As the spatial resolution of the models further increases and the assimilation of surface data becomes more widespread, the issue of spatial representativeness of data must be addressed. As shown in the present study, moving a measurement within a radius of 30 m can cause very large temperature variations. Strategies to allow assimilation routines in NWP models to handle such locally influenced measurements should be developed. While one may argue that the Lake Thunderbird Micronet is far from being an ideal meso- or synoptic-scale monitoring site and thus not a representative example for data assimilation, CP events have also been observed at well-exposed monitoring sites, such as at the Oklahoma Mesonet site in El Reno (Hunt et al. 2007). Fiebrich and Crawford (2001) also discuss the challenges of CP events for automated quality control of operational surface networks.
A number of open questions concerning the processes involved in the CP formation remain. Previous experiments have shown that negative net radiation values are conducive for the formation of large temperature gradients (Mahrt 2006) and strongly influence CP onset and average CP duration. For future experiments at the micronet, net radiation measurements should thus be made to further clarify the role of radiative processes in CP evolution. Other factors that influence surface radiative processes (cloud cover, vegetation) could also be further analyzed to better understand what governs the CP formation, and turbulent fluxes measured at the micronet tower should be integrated in the analysis. Within and around the CP region, additional temperature and wind measurements should be conducted to study the influence of drainage flows on the west–east slope and the interaction between wind sheltering and katabatic flow effects. Furthermore, more attention should be paid to the role of lake effects. Warm-air advection from the lake potentially limits the CP development, particularly inside the wooded area, and one can expect to observe very interesting flow phenomena developing at the boundary between the CP and the warm air to its south over the lake. Flow visualization experiments with smoke would be an ideal start to get a better picture of the nocturnal flow and mixing processes in the CP area.
We thank Ed Jessup for allowing us to use his land for establishing the Lake Thunderbird Micronet and for helping us on many occasions. Without his generosity, this research would not have been possible. Seed funding for the micronet was provided by an award from the University of Oklahoma Research/Creative Activity Program (principal investigators Alan Shapiro, Bruce Hoagland, and Frank Gallagher), with matching funds from the School of Meteorology. The meteorological tower was obtained through the Federal Excess Personal Property (FEPP) program. The research was also supported through the NSF career award ILREUM of the second author (NSF ATM 0547882). This grant provided an undergraduate research fellowship for the first author. Data from the Norman mesonet site were provided by the Oklahoma Climatological Survey, which also continues to support the micronet through instrument donations. We are also grateful to Evgeni Fedorovich (University of Oklahoma, School of Meteorology), Sherman Frederikson (National Severe Storms Laboratory), Joe Grego (University of Oklahoma FEPP manager), Susan Hendon (Lake Thunderbird State Park), and numerous undergraduate and graduate students who have assisted with the Lake Thunderbird Micronet project since its start in 2002.
Corresponding author address: Petra M. Klein, School of Meteorology, University of Oklahoma, 120 David L. Boren Blvd., Norman, OK 73072-7307. Email: email@example.com