a. Site openness

The ventilation at an observation site is related to the site’s geometrical openness. We defined site openness λ as the aspect ratio of the volume of air above the ground (Fig. 1):

$\lambda =\langle {\displaystyle \frac{x}{h}}\rangle$(1)

$=\langle {\displaystyle \frac{1}{\tan a_i}}\rangle ,$(2)

where h and x denote the height of and distance to obstacles around the sensor, respectively. The temperature sensor is located at the origin of x. The brackets denote the average of the azimuth. The inverse of λ is often used as the aspect ratio of urban street canyons. The aspect ratio is related to the flow regime in urban street canyons (Oke 1978).

View Full Size

Schematic illustration of site openness. The openness is defined by elevation angle (a_i), or height (h) and distance (x) to obstacle.

Citation: Journal of Atmospheric and Oceanic Technology 36, 7; 10.1175/JTECH-D-18-0176.1

• Download Figure
• Download figure as PowerPoint slide

b. Calculation procedure

The screen level is defined as the nearest level to the ground (generally 1–2 m high) in the space between surface obstacles. The screen-level temperature is influenced by many factors at multiple scales, although we focused on changes in site-local environments. The wider-scale environments were assumed to be unchanged. The theoretical consideration included changes in the site’s ground surface temperature, because the screen-level air temperature follows the surface temperature when any factor changes the latter. Its features should not be linear to the surface temperature change, and should be a function of atmospheric stability.

Changes in site openness influence the screen-level air temperature through changes in wind speed (ventilation). Simultaneously, site openness is related to the incoming radiation incident on the ground surface. In this section, the processes caused by site narrowing are described using a daytime case. Narrowing causes a reduction in wind speed at the screen level. The surface heat balance consequently changes to another equilibrium state with a higher surface temperature after the completion of narrowing. This raises the air temperature at the screen level. Moreover, the effects of the shade of surface obstacles on the ground are greater in closed sites, which lowers the air temperature.

Narrowing is further accompanied by increased longwave radiation incident on the ground, because the surface obstacles emit higher amounts of radiation in the infrared wavelength than the sky does. This increases air temperature. The following calculation deals with turbulent heat transfer, solar radiation, and longwave radiation. Note that reduced wind speed could inhibit evaporation on the ground (Sugita 2018) and could therefore change the water content of subsurface soil. However, for simplicity, we did not consider changes in soil wetness, since studying soil wetness requires long-term calculations (up to a few weeks) with a scheme for subsurface water transfer (e.g., Kondo and Saigusa 1994).

Changes in surface temperature due to turbulent heat transfer, solar radiation, and longwave radiation were calculated using a surface heat balance model. This model calculated the surface temperature using input meteorological conditions, and the screen-level air temperature was calculated from the model-calculated surface temperature using a profile function. Figure 2 illustrates the modeled situation. The narrow and open areas were adjacent to each other. The canopy-level conditions, such as air temperature, were common to both areas. However, different temperature profiles were formed for each area. The canopy- and screen-level were respectively 10 and 1.5 m above ground level in this calculation.

View Full Size

Schematic illustration of the modeled situation. Narrow field shows case of λ = 2. The turbulent heat flux (positive upward) is described with a resistance circuit.

Citation: Journal of Atmospheric and Oceanic Technology 36, 7; 10.1175/JTECH-D-18-0176.1

• Download Figure
• Download figure as PowerPoint slide

c. Surface heat budget model

d. Screen-level air temperature

e. Radiation

Site openness influences incoming solar radiation through shade cast by surface obstacles. In the shade, direct solar radiation is obstructed, and diffused solar radiation illuminates the ground. Sites were considered to be fully sunlit or fully shaded. When more than half of a site was covered in shade, the whole area was regarded as fully shaded. Otherwise, the entire site was regarded as fully sunlit. This simplification caused a discontinuity in ground surface temperatures at relatively low sun elevation. Therefore, the model-calculated diurnal course of temperature was analyzed at noon and at night. Note that the shaded area did not cover more than half of the site at noon even at the narrowest point. This simplification could influence the surface heat budget and, thus, the surface temperature at noon through subsurface heat storage, which depends on the diurnal history of sunlight and shade. To have avoided this simplification, it would have been necessary to consider the internal boundary layer formed above the partial shade in a site, and this was outside the scope of this study. The shadows were geometrically calculated assuming a homogeneous distribution of obstacles in the azimuth around the temperature sensor, that is, obstacles surrounding the sensor at all azimuths.

For simplicity, reflection from obstacle surfaces was not considered. Rough evaluation indicates that reflection may increase the incoming radiation incident on the ground by roughly 7%, depending on the reflectance of the obstacle surface. Examples of incoming solar radiation incident on the ground surface at the sites are shown in Fig. 3. Calculations in this study assumed cloud-free conditions. Solar radiation at the canopy level was provided based on the measured values on typical clear weather days.

View Full Size

Incoming shortwave radiation incident on the ground surface in May in Tokyo, Japan. Example of three different values of site openness λ. The asymmetric diurnal course occurs due to the difference between local time and Japan standard time.

Citation: Journal of Atmospheric and Oceanic Technology 36, 7; 10.1175/JTECH-D-18-0176.1

• Download Figure
• Download figure as PowerPoint slide

The sky view factor (ψ) is the solid-angle fraction of the sky in the overlying hemisphere (Oke 1981). The radiation from the cloud-free sky at each incident zenith angle was parameterized with air temperature and humidity (Kondo 1994). The radiation from obstacles was calculated as $\sigma T_o^4$, where the daytime and nighttime obstacle surface temperature T_o was assumed to be T_m + 4°C and T_m − 5°C, respectively, where T_m is the daily mean canopy-level air temperature, given as model input (section 2g).

f. Screen-level wind speed

The screen-level wind speed used in turbulent heat transfer resistance [Eq. (6)] should be related to site openness. The ratio of wind speed at screen and canopy level was evaluated from observations at several sites (Table 1). All sites were routine observation stations administrated by the Japan Meteorological Agency (JMA) and were equipped with aerovanes at canopy level, whose height varied depending on the site. The screen-level wind speed was measured using a 2D ultrasonic anemometer (Windsonic, Gill) at 1.5–2 m (depending on the site). Digital records at 1 s (screen level) and 10 s (canopy level) intervals were averaged for 10-min intervals. We used data when mean wind speed was more than 3 m s⁻¹ at canopy level.

Table 1.

List of wind observation sites. One run lasted 10 min. Site openness λ is the average of sector measurement using weighted wind direction frequency. Terrain slope within 500 m is shown. Displacement height d was estimated as 0.7 times the obstacle height.

View Table

Figure 4 shows the normalized ratio of wind speed R_u/R_u0, where R_u indicates the wind speed ratio at the observation sites:

$R_u={\displaystyle \frac{U_{\text{screen}}}{U_{\text{canopy}}}},$(15)

where U_screen and U_canopy are the measured screen- and canopy-level wind speeds, respectively, and R_u0 is the ratio at an ideal (virtual) flat site with a surface cover of short lawn. It is estimated from the log-law in neutral stratification:

$R_{u0}={\displaystyle \frac{{\displaystyle {\displaystyle \frac{u_{*}}{\kappa }}}\hspace{0.17em}\ln \left({\displaystyle \frac{z_{\text{screen}}}{z_0}}\right)}{{\displaystyle \frac{u_{*}}{\kappa }}\hspace{0.17em}\ln \left({\displaystyle \frac{z_{\text{canopy}}-d}{z_0}}\right)}}.$(16)

View Full Size

Site openness λ and normalized ratio of wind speed. I17 indicates the experimental equation by Ikegaya et al. (2017) with the plane area ratio of roofs (PAR).

Citation: Journal of Atmospheric and Oceanic Technology 36, 7; 10.1175/JTECH-D-18-0176.1

• Download Figure
• Download figure as PowerPoint slide

The surface roughness z₀ was simply set to 0.01 m for all sites, because precise measurement of z₀ was difficult (Grimmond et al. 1998). The displacement height d was determined for each site (Table 1). The reason for using R_u/R_u0 was the variable height of the upper anemometer (z_canopy), depending on the site. Figure 4 shows the linear relationship between the log of site openness (log λ) and the normalized ratio R_u/R_u0. This should be because the size of the eddy on the lee side of obstacles is proportional to the size of the free space at the site, which is similar to the log-shape vertical profile in the surface boundary layer (Kaimal and Finmigan 1994). The figure also shows the experimental results obtained by Ikegaya et al. (2017), who experimentally derived an equation for wind speed in the building canopy based on the data of Razak et al. (2013), Kubota et al. (2008), and Yoshie et al. (2008). The parameter in their equation was the plane area ratio of building roofs and was determined as 0.2 or 0.4 at our observation site based on aerial photographs.

g. Calculation results

The diurnal courses of screen-level air temperature for different site openness values were calculated using the model. The input meteorological conditions were determined considering the climatic normal for May in Tokyo, Japan, where daily total solar radiation and air temperature at the canopy level are 24.7 MJ m⁻² and 20°C, respectively. Relative humidity and wind speed at the canopy level were assumed to be constant throughout the day, at 60% and 3 m s⁻¹, respectively. The parameters for the model are listed in Table 2. The heat transfer resistance was set as slightly larger at night than during the day by changing coefficients a and b in Eq. (6), considering the stable stratification at night. The influence of site openness was analyzed for a mean value of three hours around noon, denoted as daytime (Fig. 5), and three hours after midnight, denoted as nighttime (Fig. 6).

Table 2.

Parameter settings for the model.

View Table

View Full Size

Calculation results for daytime site openness variation in May. (a) Air temperature at the screen level. (b) Surface temperature of the ground. (c) Screen-level wind speed. (d) Incoming longwave radiation flux at the ground surface.

Citation: Journal of Atmospheric and Oceanic Technology 36, 7; 10.1175/JTECH-D-18-0176.1

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Calculation results for nighttime site openness variation in May. (a) Air temperature at screen level. (b) Surface temperature of the ground. (c) Incoming longwave radiation flux at the ground surface.

Citation: Journal of Atmospheric and Oceanic Technology 36, 7; 10.1175/JTECH-D-18-0176.1

• Download Figure
• Download figure as PowerPoint slide

As the sites became narrower, air temperature increased during the daytime (Fig. 5a). The difference in air temperature between λ = 10 and 2 was 1.4°C, which is smaller than the variation in surface temperature. However, these variations in air temperature cannot be neglected in climate change studies because of their large values. The modeled physical processes of temperature change were analyzed by a sensitivity test. The change in screen-level air temperature was strongly correlated to that in surface temperature. Therefore, surface temperature was separately calculated under the influence of each influencing factor (wind speed, shortwave radiation, and longwave radiation). Figure 5b shows that the temperature change was primarily caused by changes in wind speed, which is explained in the following paragraph.

As site openness decreases, screen-level wind speed decreases (Fig. 5c) as shown by Eq. (17), and therefore the thermal resistance increases. Increased resistance leads to higher surface temperature (Fig. 5b), which satisfies the surface heat balance. On the other hand, incoming shortwave radiation decreases due to the shade of obstacles, whose influence on temperature is smallest among the three influencing processes (Fig. 5b) partly because the analysis was conducted at noon. Note that the shade of the obstacles covered less than half of the site area at noon in this calculation condition. As the site openness decreased, the incoming longwave radiation increased (Fig. 5d), because the obstacles shared a greater portion in the sky viewed from the ground and obstacle temperature was higher than that of sky irradiance. These two radiation processes comprised vertical shielding by obstacles, and their influence was smaller than that of ventilation in the range of λ analyzed. Consequently, the ventilation was the most significant factor.

The nighttime results (Fig. 6) showed a smaller range of variation in air temperature (0.1°C for λ = 10–4) than the daytime results did (0.5°C for λ = 10–4). Moreover, air temperature showed a complex trend and was at a minimum at λ = 4. The downward convex shape in Fig. 6a can be explained by the sensitivity test shown in Fig. 6b. The test for solar radiation was included because it has some influence on nighttime temperature through subsurface heat storage. Compared to the sufficiently open site (λ = 10), incoming longwave radiation from obstacles increased at λ < 4 (Fig. 6c), and this increased surface temperature (Fig. 6b). Decreased wind speed (ventilation) at λ > 4, compared to λ = 0, slightly reduced surface temperature (Fig. 6b). The response of nighttime surface temperature to wind speed was the opposite of that during the daytime, because the surface temperature was lower than the air temperature during the night and turbulent sensible heat was transferred from the air to the surface. The turbulence-induced heating from the upper air mitigated surface radiative cooling.

Figures 5 and 6 illustrate the condition of fixed parameters, and we consequently considered the dependence of the parameters. If β, ε, or c_gρ_gλ_g were smaller during the daytime, higher surface temperatures would be required to achieve surface heat balance (Sugawara and Kondo 1995). Screen-level air temperature followed this trend. In this study, β was chosen, and the dependence of air temperature difference between λ = 10 and λ = 5 during the daytime is shown in Fig. 7.

View Full Size

Calculation results for daytime variations in wetness β. (bottom) Air temperature difference between λ = 10 and λ = 5.

Citation: Journal of Atmospheric and Oceanic Technology 36, 7; 10.1175/JTECH-D-18-0176.1

• Download Figure
• Download figure as PowerPoint slide

The temperature difference was positive for all wetness conditions and increases as the surface dries (β decreases), which means that narrow sites were warmer than the more open sites, especially under dry conditions. The reason for increased differences at β > 0.8 is that under very wet conditions sensible heat flux becomes negative to achieve high evaporation (Kondo and Watanabe 1992). The negative sensible heat flux caused air temperature to be higher than the surface temperature through the profile function [Eq. (12)] (Fig. 7a), and increased air temperature is larger at lower wind speeds (narrower sites).

a. Measurement of site openness

Site openness λ was measured with a theodolite, which was placed at the location of the temperature sensor. Elevation angle of the nearest obstacles a_i in Eq. (1) was measured at an azimuth interval of 5°, and $1/\tan a_i$ was averaged for the whole azimuth plane. Narrow obstacles, such as flag poles, were ignored, but a wire fence was included in the measurement, because it could reduce wind speed. At the sloped site, a_i was measured from the mean sloped surface. To achieve continuity of λ in the very open site, a_i was set to 1.9 for all sectors with a_i < 1.9. We further attempted to use hemispherical photographs, but they exhibited large errors due to distortions at low elevation angles. Note that Wada et al. (2016) averaged a_i for the windward sector, which was 40° in width.

b. Measurement of air temperature differences between narrow and open sites

To examine the influence of openness on temperature, air temperature was measured at 1.2 m above ground level and compared at two adjacent points (<100 m separation) with different openness values. We made several measurement runs with different site-openness pairs, as listed in Table 3. The two points were covered with short grass on the flat terrain. Figure 8 shows photographs of the measurement sites.

Table 3.

List of temperature observation sites. One measurement run lasted 2–4 h.

View Table

View Full Size

Photographs of temperature measurement sites. See Table 3 for site name abbreviations.

Citation: Journal of Atmospheric and Oceanic Technology 36, 7; 10.1175/JTECH-D-18-0176.1

• Download Figure
• Download figure as PowerPoint slide

Site HRT was a public park bounded by a 1.8-m-high hedge. We set the open reference point at the center of the park (Fig. 8a). Other narrow points were set near the hedge (Fig. 8b), yielding several types of openness with different distances from the sensor to the hedge. Measurements at site TMU were made by Wada et al. (2016). They took measurements with an intentionally placed curtain in an athletic field (80 m × 160 m). The curtain and some trees (6 m high) around the field were the obstacles. Several measurement runs with different site openness values were taken by changing the distance between the sensor and obstacles. Further details are given by Wada et al. (2016).

Site NIAES was a crop field used for agrometeorological experiments. It had some fences, a few greenhouses, and trees of around 10 m high. An open area in the center of the field was set as the reference point, and areas near the obstacles were selected as the narrow points. Site KMG was a routine observation site operated by the JMA and was surrounded by two- or three-story houses (7–10 m high). Site KTR was also a routine JMA observation site. It was located in a large park and surrounded by trees. An open area in the park was selected as the reference open point, and the narrow point was the field used for routine observations (Fig. 8e). Site NDA was a courtyard (20 m × 60 m) surrounded by buildings (12 m high). The narrow point was set in the courtyard, and the open reference point was set in the short grass field outside the courtyard.

Measurements were taken in mainly March to October from 2012 to 2015. Light wind or calm conditions were selected on days with fair weather or partially cloudy days. These weather conditions were chosen for showing air temperature difference as shown later. The wind speed at the canopy level was 2–3 m s⁻¹ during the daytime and 1–2 m s⁻¹ during the nighttime on the measurement days.

c. Instrumentation

Temperature differences between the narrow and open points were in the order of 0.1°C. The instruments were designed and manufactured specially to capture these temperature differences. The air temperature sensor was a Class A type platinum resistance thermometer of 1000 ohm (Tateyama Kagaku Kogyo Co., Ltd., STP-02), whose diameter was 2.3 mm. A 1000 ohms type thermometer was used rather than a 100 ohm type to reduce the influence of junction resistance at the electric terminal. A datalogger with analog-to-digital converter (TR-55i-Pt, T&D Co., Ltd.) was used.

The laboratory calibration showed that two sets of the same type of sensor agreed within ±0.014°C over the temperature range 3°–33°C. The error range of ±0.014°C was the relative error between the two sensor sets, and it was therefore less than the absolute error of the datalogger (0.3°C). Moreover, it was lower than the resolution of the datalogger (0.1°C) because it was evaluated for an average of a few hours with an interval between raw readings of 20 s (N = 360–540).

The platinum resistance thermometer sensor was installed in a forced ventilation radiation shield, which was newly designed in this study. The shield had triple tubes around the sensor to avoid erroneous heating and cooling by ambient radiation. The innermost tube was painted black, which prevented the reflection of stray sunlight. The ventilation speed was 5 m s⁻¹. Two sensor systems, namely the platinum resistance thermometer sensor, datalogger, and radiation shield, were calibrated under outdoor conditions. The two systems agreed within ±0.03°C for an average of a few hours with an interval of 20 s between raw readings, indicating the total error in temperature difference at two points in this study.

Intervals of the raw reading and averaging period were determined as follows. Screen-level air temperature exhibited significant variation over time due to turbulent mixing. Therefore, we compared the temporal averages at two points in the narrow and open sites in this study, rather than comparing the instantaneous values measured. The temporal standard deviation of screen-level air temperature (σ_T) was experimentally given as 0.25°C for the typical surface boundary layer. The time constant of our sensor was several seconds, and therefore the raw reading interval was set to 20 s. The standard error ${\sigma }_T/\surd N$ was 0.018 and 0.013°C for 1 and 2 h averages, respectively. Therefore, time periods of longer than 1 h are needed to derive the best performance of the sensors (0.014°C), otherwise temporal variations in temperature are larger than the instrumental error. On the other hand, the average over longer periods could include diurnal variations in air temperature, and ${\sigma }_T/\surd N$ could exceed the previous value for larger averaging periods. Therefore, we used averages of 2–5 h around noon and 3–5 h after midnight. Measurement runs with sudden changes in solar radiation due to clouds or wind direction were discarded.

a. Daytime

Figure 9 shows the observed results for the daytime. The x axis represents the difference in openness of two sites in a common logarithm:

$d\hspace{0.17em}\log \lambda =\log \left({\lambda }_{\text{narrow}}\right)-\log \left({\lambda }_{\text{open}}\right),$(18)

where λ_open and λ_narrow represent the openness in the reference open site and narrow site, respectively. An example of the corresponding street-canyon aspect ratio and sky view factor used in urban climatology is shown in Table 4. The aspect ratio is defined as the building height divided by the distance between the buildings.

View Full Size

Air temperature difference dT_a at sites of differing openness. The y axis is the temperature difference (narrow site minus open site). The vertical error bar indicates the standard deviation in multiple runs. See Table 3 for site name abbreviations, except OTM, which was measured by Shido et al. (2015). Note that Shido et al. (2015) showed several months’ average of temperature difference. Lines show calculation result, which is temperature difference between narrow and open site (λ = 10). The gray region indicates the error range in the calculation for March owing to uncertainty of the shear function (see section 5a). For details of model uncertain area shown by an arrow, see section 5a.

Citation: Journal of Atmospheric and Oceanic Technology 36, 7; 10.1175/JTECH-D-18-0176.1

• Download Figure
• Download figure as PowerPoint slide

Table 4.

Example of geometric factors corresponding to $d\hspace{0.17em}\log \lambda$ at λ_open = 10. The sky view factor was calculated for the cross point of the streets (Nakagawa 1996), which is approximately 10% larger than that of the 2D street canyon (Oke 1981).

View Table

The daytime results showed positive differences in air temperature (dT_a), indicating that the narrower sites were warmer. The temperature difference shown in Fig. 9 was significantly larger than the measurement error (0.03°C). The temperature difference increased as the difference in site openness increased (left side of figure). The field experiments showed that it was warmer in the daytime as the site narrowed. Figure 9 further shows the calculation results, which are the calculated temperature difference between narrow and open sites (λ = 10).

Two cases with different weather conditions (Table 5) were calculated with the procedure described in section 2b. The observations were taken mainly between March and October, and the conditions for March and May were used in the calculation. The monthly mean solar radiation was highest in May at these sites, and the other conditions used in the calculation were the same as those in section 2. The calculation results exhibited trends similar to the observations, namely, larger positive dT_a values for larger openness differences. The observed positive dT_a values were due to poor ventilation as described in section 2g. Note that the absolute value of dT_a did not agree between the observation and the calculation because the model inputs were not specific values on observation days.

Table 5.

Weather conditions in two calculated cases (Mar and May). Daily total or averages are shown.

View Table

Sites HRT and KTR had larger dT_a values than the other sites. This could be because the plants surrounding these sites emit larger amounts of sensible heat to the air compared to the other sites, which were surrounded by buildings. Buildings have a large heat capacity and therefore large heat storage and less sensible heat. Some types of evergreen plants at these site (e.g., Cinnamomum camphora and Camellia japonica) do not exhibit much evapotranspiration (Monteith 1975), and such plants could heat the air more than buildings can. The calculation revealed a tendency to overestimate dT_a at $d\hspace{0.17em}\log \lambda <-0.7$ (Fig. 9). This point will be discussed later (section 5a).

b. Nighttime

Depending on the site, nighttime observations consisted of either negative (sites HRT and TMU) or positive (site NDA) differences in dT_a (Fig. 10). Although site NDA had only one run, a positive dT_a was consistent with the calculation results (Fig. 6), where the most open site (λ = 10) had the lowest air temperature. On the other hand, the negative dT_a trend was opposite to that during the daytime. These contradictory results were evaluated using a model sensitivity test. The effect of the addition of longwave radiation (section 2e) was tested in the calculation.

View Full Size

As in Fig. 9, but for nighttime. Four calculation cases, representing different seasons and longwave radiation schemes, are shown by the lines. LR indicates the longwave radiation from obstacles. Observation results are divided according to the amount of cloud cover. The blue area indicates the error range in the calculation for March without longwave radiation (blue line) owing to the uncertainty of the shear function (see section 5a).

Citation: Journal of Atmospheric and Oceanic Technology 36, 7; 10.1175/JTECH-D-18-0176.1

• Download Figure
• Download figure as PowerPoint slide

Four calculation cases are shown as lines in Fig. 10. The broken and solid lines show calculations with and without the addition of radiation, respectively. The calculation with the addition of radiation considered longwave radiation from obstacles and the sky to the ground surface. The calculation without the addition of radiation considered only that from the sky, that is, the sky view factor was set to unity. Both calculations considered the influence of obstacles through wind speed and solar radiation processes. Each solid and broken line represents different weather conditions, denoted as March and May. The addition of longwave radiation made dT_a positive, and the processes of wind speed and solar radiation made dT_a negative. The turbulent exchange caused a negative dT_a, as discussed in section 2g. The calculation results explained why positive and negative dT_a values were found in the observations. Note that the screen level was defined at 1.5 m in the model calculation, although air temperature was measured at 1.2 m. The model level was followed the observation standards of the Japan Meteorological Agency, on the other hand in our observation, a different level was used due to the field administrator’s regulation. A Higher level in the calculation would decrease dT_a by 0.1° and 0.3°C in day and night, respectively.

Variations in dT_a according to site were due to the type of surrounding obstacles at the different sites. Site HRT was surrounded by relatively short plants (1.5 m), and site TMU contained a fabric curtain. Surface temperature approached air temperature at these two sites, potentially resulting in lower longwave radiation transmission to the ground surface from obstacles. At these two sites, dT_a was influenced by turbulent heat exchange rather than longwave radiation.

The narrower point at site NDA was surrounded by buildings. Building walls are heated during the daytime and emit large amounts of longwave radiation during the nighttime (Oke 1981). The addition of longwave radiation causes a positive dT_a (i.e., narrower point is warmer). Temperature variations at site NDA were mainly the result of the addition of longwave radiation. The openness of the narrower point at site NDA was 1.87, which satisfies the condition of λ < 3 shown as the radiation dominant situation in section 2g and Fig. 6.

A comparison of sites HRT and TMU indicates that the former had a relatively larger (more negative) dT_a at night. This tendency was similar to that in the daytime and may have been caused by the sensible heat flux from plant leaves, which cools the air during the night. Note that any differences in weather conditions in different seasons could influence the results. Measurements at site TMU were conducted in August, whereas those at site HRT were conducted during March to November.

Comparison of the observed absolute value of dT_a during the daytime and nighttime indicates that the value during the day is larger. This indicates that dT_a is correlated with incoming radiation incident on the ground surface. The all-wave radiation input (downward shortwave plus downward longwave, minus upward shortwave radiation) in July in Japan is roughly 1000 W m⁻² during the midday and 400 W m⁻² during the nighttime.

Oke (1981) measured screen-level air temperature in cities at night and showed that narrower streets were warmer than more open ones. Calculations with the addition of radiation, shown by the broken lines in Fig. 10, indicate a similar tendency. Therefore, the temperature distribution in cities studied by Oke (1981) is influenced by longwave radiation from building walls. Oke’s measurements were taken in streets with sky view factors of 0.3–0.9. This range of site openness corresponds to $d\hspace{0.17em}\log \lambda$ = −1.5 to −0.7 at λ_open = 10, which exceeds the range of Fig. 10 on the narrower side of the site.

c. Dependence on weather conditions

Plots of site TMU in Fig. 9 and sites HRT and TMU in Fig. 10 were respectively divided into two types depending on the amount of cloud cover during observation. The dT_a value did not seem to be dependent on the amount of cloud cover. However, Wada et al. (2016) noted that dT_a was correlated with solar radiation at site TMU. Our calculation results, which indicate some difference between May and March, suggest a correlation between dT_a and solar radiation. This could not be confirmed in Figs. 9 and 10 because only cloudless or partly cloudy days were used in the observation analysis.

The dT_a value further depended on canopy-level wind speed. The observation runs shown in Figs. 9 and 10 were limited to calm or light-wind conditions, as noted in section 3b, and therefore wind speed dependence could not be confirmed from these figures. Wind speed dependence is depicted in Fig. 11, which is based on measurements at site HRT. Runs of $d\hspace{0.17em}\log \lambda$ = −0.50 or −0.55 were used. Figure 11 shows the calculation results in four lines and observation results in four types of plots, depending on the time of day (daytime/nighttime) and precipitation on the previous day.

View Full Size

Wind speed dependence on temperature difference dT_a between open and narrow points. Observations at $d\hspace{0.17em}\log \lambda$ = −0.50 or −0.55 for site HRT are shown.

Citation: Journal of Atmospheric and Oceanic Technology 36, 7; 10.1175/JTECH-D-18-0176.1

• Download Figure
• Download figure as PowerPoint slide

The dT_a value approached zero as wind speed increased both in the observations and calculations. This is because the more active turbulent heat transfer under high wind speed conditions reduced solar heating of the ground surface during both daytime and nighttime radiative cooling. It was not due to increased horizontal mixing, because two measurement points were located in the canopy at all sites. The dT_a value also seemed to depend on the surface wetness and was lower on days after rain. This could be because the latent heat from wet surfaces prevents surface heating during the daytime. At night, the larger heat capacity of wet surfaces prevented cooling.

a. Model limitations

The model showed a tendency to overestimate dT_a at large openness differences compared to observations (Fig. 9). This could be due to model incompleteness at the narrow sites. There could be four possible reasons. First, the extrapolation of the fitted equation in Fig. 4 could be inadequate. At the narrow sites (λ < 2), the fitted equation [Eq. (17)] underestimated U_screen, which caused an overestimation of dT_a. Relatively lower z₀ in Eq. (16) should also cause underestimation of U_screen. Second, the (constant) coefficient for the turbulent heat transfer parameterization [Eq. (6)] could have been inadequate. The coefficients (a and b) could have depended on λ because obstacles could produce turbulence and compensate for wind speed reductions, which was not considered in this study. Third, the partial shade condition was ignored in the calculation (section 2e). Under actual conditions, shade on the ground increases at narrower sites, which lowers the temperature.

Fourth, the profile function described in section 2d may not be valid at narrow sites. The function was acquired on the assumption of constant flux in the surface boundary layer, in which the vertical eddy flux of momentum and heat is constant at any height. This assumption required horizontal homogeneity and was not always valid in narrow spaces. Surface obstacles that cause drag (surface drag and form drag) and additional heat sources disturb the constant flux layer. Obstacles that emit sensible heat disturb the constant flux assumption. In addition to anthropogenic heat sources, such as air-conditioning facilities, plants, such as hedges, also emit sensible heat to the air. Application limits in the profile function could be restricted to λ < 3. The wake of an obstacle expands on the leeside to 2–3h, where h is the height of the obstacle (Oke 1978), and the flow has complex 3D features in this area. These four uncertainties could reveal that the model would be erroneous at λ < 3, whose range is indicated by an arrow in Figs. 9 and 10.

In boundary layer climatology, the constant flux assumption has been validated within a 10% error (Stull 1988). Figures 9 and 10 show the range of this error in the shear function with the hatched area. The vertical scatter of observation plots was roughly the same as the vertical expansion of the hatched area, indicating that the observations satisfied the constant-flux assumption. Furthermore, the temperature change due to site openness (line curve) was larger than the hatched area, which is why the profile function was used.

b. Long-term trends in temperature and wind speed

We analyzed long-term trends in temperature and wind speed to understand how poor ventilation is expressed in climatic data. A total of 27 sites of routine observations administered by the JMA were selected based on the following conditions: 1) the site’s location was not moved and 2) the site was not located in a big city. Some corrections were made to the data, and details are shown in appendix, along with the full list of sites.

Figure 12 shows the changes in annual mean temperature and wind speed over 9–48 yr for different sites. The change in wind speed (x axis) is shown in the form of a ratio, $\left(U_{\text{old}}-U_{\text{new}}\right)/U_{\text{new}}$, where U_old and U_new are the wind speeds averaged over several years at the beginning and end of the analysis period, respectively. Because even slight changes in land cover could influence temperature data, plots were divided into categories of suspected and little influence of urbanization, according to the degree of urbanization. The degree of urbanization was determined using field surveys and a land-use database (Himiyama 1995).

View Full Size

Change in annual mean wind speed and temperature in historical data from Japan. The x axis is the wind speed ratio $\left(U_{\text{new}}-U_{\text{old}}\right)/U_{\text{old}}$. Plots are divided into two groups of suspected and little urbanization. The lines are fitted to the plots showing little urbanization. The solid line (slope = −0.47° ± 0.13°C) includes one plot in the positive wind ratio, but the broken line (slope = −0.344° ± 0.27°C) does not. See the appendix for a full list of the sites.

Citation: Journal of Atmospheric and Oceanic Technology 36, 7; 10.1175/JTECH-D-18-0176.1

• Download Figure
• Download figure as PowerPoint slide

A larger temperature increase was observed for sites with greater decreases in wind speed (Fig. 12). Fujibe (2009a) conducted a similar analysis to that in Fig. 12, but with many more sites (408) than ours and nonetheless obtained results consistent with ours. We employed stricter selection for sites and included only a small number of sites in our analysis.

These observation sites have been under careful maintenance by the JMA, and the changes in temperature and wind are therefore due to the site ambient environment or even regional-scale climate change. In sites of suspected urbanization, wind speed reduction could have been due to increased building height around the site. On the other hand, wind speed reductions at sites that show little urbanization were caused by the growth of surrounding trees. These sites were located in a rural area and surrounded mostly by parks and forests. For example, our field survey found that in 2007 the Fukaura site was partly surrounded by naturally growing pine trees of 18 m high. Their height in 1970, the beginning of analysis period, was estimated as being roughly 6 m, assuming the typical growth rate of pine trees. There is a site (Okunikkou) where the wind speed increased, as shown in Fig. 12, and this could have been caused by the demolition of an observatory building in 1999.

The temperature increase shown in Fig. 12 may be attributed to poor ventilation at the sites, and the relationship between temperature and wind speed in Fig. 12 was compared to the results in Figs. 9 and 10. We considered changes in site openness λ from 15 to 5 ($d\hspace{0.17em}\log \lambda =-0.477$) because of the growth of trees surrounding the site (3.3–10 m) in an area with a 50 m radius. A change in openness of $d\hspace{0.17em}\log \lambda =-0.477$ caused a temperature change of 0.38°C in the daytime (Fig. 9) and −0.19°C at night (Fig. 10). The simple average temperature during the day and night (0.19°C) would be close to the daily mean value of temperature change. The change in canopy-level wind speed (x axis in Fig. 12) due to tree growth ($d\hspace{0.17em}\left(\log \lambda \right)=-0.477$) was estimated as −0.38 in the log profile of wind speed above the tree canopy.

We assumed that the height of the anemometer was 18 m, which is the average anemometer height in Table A1. We further assumed the displacement height and roughness length of the tree canopy as 0.7h and 0.1h, respectively, based on Kondo (1971) and Brutsaert (1982), where h is canopy height (10 m). Consequently, Figs. 9 and 10 indicate that a change in openness ($d\hspace{0.17em}\log \lambda$ = −0.477) caused a wind speed change of −0.38 (in its ratio) and a temperature change of +0.19°C in terms of the daily mean. These values of change (+0.19°C/−0.38 = −0.5°C) agree roughly with the slope of the line in Fig. 12, which is fitted for plots showing little urbanization.

Therefore, long-term changes in air temperature at historical Japanese climatological observation sites (Fig. 12) could be caused by poor ventilation. Note that Fig. 12 is based on simple annual mean data, which includes cloudy days. However, Figs. 9 and 10 mainly used data for cloudless days. Temperature changes due to poor ventilation are influenced by weather conditions, as noted in section 4c. Thus, the relationship shown here needs to be further investigated in greater detail.

Fujibe (2009b) analyzed historical screen-level air temperature data in Japan and found a positive trend even in sparsely populated areas (100–300 persons km⁻²). As noted by Runnalls and Oke (2006) most common changes in landscape (e.g., irrigation) could cause nocturnal warming even if the changes do not relate to urbanization. The influence of poor ventilation on air temperature may also support this trend revealed by Fujibe (2009b). Warming could develop in a rural area where trees grew and if the measurement site became relatively narrower.

Furthermore, Fujibe (2009b) found a smaller increase of temperature in the daytime than in the nighttime and a positive trend at night. Runnalls and Oke (2006) suggested that tree growth could reduce nocturnal cooling through the alternation of the surface heat balance. These results do not indicate poor ventilation, which is accompanied by a negative (cooling) effect at night. Other processes, such as longwave radiation from obstacles, could cause warming at night, as discussed in section 4b. Moreover, broader-scale urbanization, which causes nocturnal warming, could overcome the influence of poor ventilation.

c. Suggestions for site maintenance and metadata recording

Temperature changes due to poor ventilation could affect analyses of climate change using historical temperature data. These results provide suggestions for maintenance guidelines at historical climatological observation sites. At sites that were previously sufficiently open (λ = 15), additional obstacles or tree growth reaching 2 m and at a horizontal distance of 10 m from the thermometer render the site narrow (λ = 5). This narrowing would cause a ~0.2°C warming in the daily mean recorded temperature, as noted previously. The addition of obstacles that reduce wind speed at the screen level must be avoided, and trees must be trimmed regularly in and around climatological observation sites.

The WMO (2008) has guidelines for site conditions, for example, surface obstacles must be far away from the sensor to prevent it from being in the shade when the sun is higher than 5°. The threshold (5°) corresponds to λ = 11 and is sufficiently open in our estimation. Our suggestion concerns any obstacles at the opposite side of the sun. Even a mesh fence, which does not cast shadows, is problematic in terms of ventilation.

The metadata of observation sites should include records of obstacles regarding their size, distance from the thermometer, and porosity, which are related to ventilation. Additional sensible heat from obstacles and trees could also change screen-level air temperature, but this was not included in our calculations. Furthermore, both plants and anthropogenic heat sources may cause serious temperature changes.