Intra-urban cooling in the city of Ouagadougou, capital of Burkina Faso in the Sahel zone of West Africa, was studied during the dry seasons in 2003, 2004 and 2007. The aim was to see how vegetation, built structure and position within the built-up area influenced the nocturnal cooling. Cooling was divided into two phases. In Phase 1 (16:00 –20:00 hours LST = CET), cooling was very different between the sites while in Phase 2 (20:00 –06:00 hours LST), cooling rates differed insignificantly and the whole area cooled almost at the same rate. Thus the temperature differences between the sites developed during these few hours in Phase 1 were preserved during the rest of the night. In Phase 1, Evening Evapotranspirative Cooling was intensive at vegetated sites that cooled almost twice as fast as sparsely vegetated. This was indicated by a humidity rate (increase of specific humidity per hour) that was high at a vegetated site, but considerably lower at a sparsely vegetated. In Phase 2 the humidity rate was slightly negative with little difference between the sites. After a division in vegetated and sparsely vegetated sites built structure (sky view factor) were shown to influence cooling, but no influence of the position within the built-up area was traced. Thus, the site-specific properties dominated cooling, giving large intra-urban temperature differences. The study also showed the importance of considering a large enough source area to account for micro-scale advection.
In warm climates or during heat waves, night time cooling is essential to human health and well-being since it offers a relief from the heat. Instead of installing air conditioners, which create an additional demand for electricity and release of heat into the city itself, climate-responsive design utilizing urban geometry, vegetation, surface and building material, colour, and so on could be used. In developing countries, where electricity is limited and often very expensive, climate-sensitive design offers a good alternative for mitigating heat both day and night. A better understanding of nocturnal cooling and the processes behind it is thus of great importance when developing design guidelines and strategies.
Nocturnal Urban Heat Island (NUHI) intensities in the canopy layer are generally lower in (sub)tropical cities compared to those cities located in the mid-latitudes with comparable population (Wienert and Kuttler, 2005; Roth, 2007). This is mainly supposed to be a result of the differences in urban form, including building dimensions and spacing, thermal properties and amount of vegetation, but also the input of anthropogenic heat. Many (sub)tropical cities show a clear seasonal difference in NUHI intensities, with 0.5–3 K lower intensity during the wet season compared to the dry season (Okoola, 1990; Jauregui et al., 1992; Jauregui, 1997; Robaa, 2003; Jonsson, 2004; Chow and Roth, 2006; Roth, 2007; Balogun et al., 2009). The lower NUHI intensities during the wet season are explained by the higher amount of cloud cover, but also by thermal admittance due to increased soil-water (Jauregui, 1997; Jonsson, 2004; Jonsson and Lindqvist, 2005; Roth, 2007). The importance of soil wetness to cooling was emphasized by Jonsson and Lindqvist (2005) who compared cloud-free nights during the dry and the wet season, respectively.
In contrast to the comparatively small urban–rural temperature differences, large intra-urban differences have been reported from some dry tropical cities, such as Mexico city (Jauregui, 1997), Gaborone, Botswana (Jonsson, 2004) and Ouagadougou (Lindén, 2010). Also in the dry city of Phoenix, Arizona, (temperate climate but with very hot summers) intra-urban differences larger than urban–rural temperature were found (Buyantuyev and Wu, 2010). In (sub)tropical as well as in temperate cities, the vegetation fraction of urban areas (Grimmond and Oke, 2002) has been shown to effectively reduce heat storage uptake during daytime through evapotranspirative cooling and shading from trees, hence reducing air temperature and the heat island and creating large daytime intra-urban temperature differences (Jauregui, 1997; Spronken-Smith and Oke, 1998; Jonsson, 2004). The cooling effect of vegetation also exists during the night. In fact it has been shown to be stronger at night than during the day (Lee et al., 2009; Yokobori and Ohta, 2009; Bowler et al., 2010). Further, Li and Roth (2009) showed a fairly good relation between vegetation fraction and NUHI intensity. At night, the cooling effect is supposed to be a result of increased outgoing radiation due to higher sky view factor (SVF) compared to built-up areas, although this is counteracted by the higher soil-water content (higher thermal admittance) in vegetated areas compared to built-up areas (Spronken-Smith and Oke, 1998). On the contrary, Pochter et al. (2008) found that irrigated garden trees and grass were colder at night than unirrigated. They also observed that the irrigated area was cooler than the surrounding open desert while the unirrigated was warmer. Similarly, Chow et al. (2011) put forward nocturnal evapotranspiration to explain that grass was colder than bare soil in a park in Phoenix, Arizona. Hardware modelling of evaporation and its influence on the energy balance were performed by Perlmutter et al. (2009).
By calculating cooling rates (air temperature change per hour), the development of NUHI or intra-urban temperature differences can be analysed. Several studies have shown large differences in cooling rates between urban and rural sites around sunset, with more intensive cooling at the rural site than at the urban (Oke and East, 1971; Oke and Maxwell, 1975; Upmanis et al., 1998; Chow and Roth, 2006; Giannopoulou et al., 2010). Later, cooling rates decreased and showed a similar, decreasing pattern at all sites throughout the night. Intra-urban comparisons also showed a similar pattern with two different cooling phases (Upmanis et al., 1998; Chow and Roth, 2006; Erell and Williamson, 2007; Holmer et al., 2007). Holmer et al. (2007) proposed that site-dependent differential cooling around sunset was dominated by radiative divergence and sensible heat flux cooling, which both are influenced by the geometry of the buildings that can be expressed as the SVF. However, as the evening progresses a capping inversion which controls the radiative cooling develops. In this second phase, cooling seems to be independent of built geometry (SVF) and surface material (thermal admittance).
To compare temperatures from different sites it is necessary to classify both the physical properties of the sites as well as the meteorological conditions. The imprecise use of ‘urban’ has long been recognized as a problem in UHI studies (Chandler, 1962). In addition, the shifting properties of ‘rural’ have been discussed (Jauregui et al., 1992; Grimmond et al., 1993). Stewart and Oke (2009) proposed a classification in ‘local climate zones’ (LCZ) to describe urban and rural sites. An altered version which included tests to classify towns in temperate climates was presented a year later (Stewart and Oke, 2010). The weather classification is often based on wind speed and cloudiness (Sundborg 1950; Runnalls and Oke 2000). Another approach is to use the relation between NUHI and nocturnal rural lapse rate that is strongly influenced by wind speed and cloudiness (Ludwig, 1970; Lee, 1975, 1979). The Diurnal Temperature Range (DTR) is also influenced by wind and cloudiness (Dai et al., 1999; Liu et al., 2004), and in the absence of cloud and lapse rate data we will in this study use DTR as a measure to study the influence of weather on nocturnal cooling.
The overall objective of this study is to examine the intra-urban cooling in a city which is located in a tropical steppe climate. Specific objectives are (i) to study the temporal development of cooling during the night to see if there are two cooling phases with different cooling rates, (ii) to analyse the intra-urban differences depending on building density (expressed as SVF), vegetation (especially trees) and location within the built-up area (expressed as distance from the city centre) and (iii) to introduce the use of DTR to characterize nocturnal weather in order to evaluate the role of weather in the cooling process. The study was performed in Ouagadougou, the capital of Burkina Faso in the Sahel region of Africa during the dry seasons of 2003, 2004 and 2007.
2. Study area
Ouagadougou, the capital of Burkina Faso (12°22′N, 1°31′W, 300 m above sea-level) is situated on a plain in the Sahel, south of the Sahara desert (Figure 1). The difference in altitude in the urban area is less than 15 m. The population of Ouagadougou has grown from 59 000 inhabitants in 1962 (Skinner, 1974) to over 1.3 million in 2005. As Africa's proportionally fastest growing large city, this number is expected to reach 4.3 million by 2025 (UN_Habitat/UNEP, 2010).
The land cover/use classification presented in Figure 1 is based on visual interpretation of satellite images that were retrieved from Google Earth (pixel resolution ∼4 m) in 2004. This classification largely agrees with a spectral classification of SPOT-XS images of Ouagadougou presented by de Jong et al. (2000). The denotations are taken from the LCZs by Stewart and Oke (2010).
The urban structure is open and dominated by low buildings, sparse vegetation and many open areas spread out over the city. The urban centre is of LCZ open-set mid-rise type and consists mainly of two to five-storey buildings, with some taller constructions being around ten-storeys high. Mainly, modern building materials such as concrete, asphalt and metals are used. Vegetation in the urban centre consists mainly of scattered large trees. The urban centre is surrounded by high-income residential areas, hotels and commercial areas, mainly of LCZ open-set low-rise type. In these areas vegetation cover is higher, consisting of irrigated lawns, small parks, trees, restaurants and hotel gardens. Residential areas outside of the urban centre are also of the LCZ open-set low-rise type, but generally constructed with a higher percentage of local materials such as clay bricks and vegetation is scarce or non-existent. Rapidly growing in the outskirts of the city are areas with informal spontaneous settlements of LCZ light-weight, low-rise mixed with LCZ scarce or non-existent vegetation. These densely inhabited areas of houses made of earth bricks with tin roofs cover an area of approximately 35 km2 or 20% of the total urban area. Only a few main roads are paved outside of the city centre.
A large shallow reservoir, approximately 2 km2 in size and constructed in the 1960s stretches along the north side of the city centre (Figure 1). Around the reservoir are more vegetated areas, used for growing food crops and other plants as well as for animal grazing grounds. A large protected but unirrigated park/forest (∼2 km2) with natural forest vegetation is situated east of the reservoir. The park/forest consists of dense, large trees and relatively dry ground vegetation. The rural areas surrounding Ouagadougou are characterized by heavy foraging for firewood, food and grazing that has left these areas bare and dry, with very scarce and dry vegetation (de Jong et al., 2000).
According to the Köppen classification, Ouagadougou has BSh climate (hot steppe climate), but close to the border of Aw climate (Savannah climate). There is a dry period from October to April with the Harmattan wind from north and northeast blowing in from the Sahara and a wet period from May to September. The average rainfall is 815 mm. Average monthly temperatures range from 24.8°C to 32.7°C. During November and December when the study was conducted, the average temperatures are 28.1°C and 25.4°C, respectively, and the precipitation 2.8 mm and 0.3 mm (data from Global Historical Climatology Network, GHCN). The weather during the dry season is characterized by clear skies and a large diurnal range of temperature, rather high wind speed daytime and very low at night, and atmospheric stability changing from unstable during the day and very stable at night. However, sometimes there are spells of weather with increased cloudiness and wind.
The air is heavily polluted by suspended particles (Boman et al., 2009; Eliasson et al., 2009; Lindén et al., 2011). However, the particles seem to have little influence on the downwelling long-wave radiation (Jonsson et al., 2006). Offerle et al. (2005) found the latent heat flux at a suburban site with sparse tree vegetation to be greater than expected, but explained this to be a result of the urban water consumption. The dry season urban climate is characterized by large intra-urban temperature differences (Lindén, 2010) and at night wind speed is generally low, especially before midnight (Eliasson et al., 2009). These circumstances give opportunities for an intra-urban wind system to develop during stable nights (Lindén and Holmer, 2011). The wind system creates winds from the southeast in the central parts of Ouagadougou, while a regional wind blows from the northeast. At times the wind at the surface and also the atmospheric stability appear to be influenced by a nocturnal low-level jet (NLLJ) such as the one centred at 400 m above ground level, with wind speeds of around 15 ms−1 and reaching maximum speed at approximately 05:00 hours UTC as described by Lothon et al. (2008). In November and December the NLLJ generally comes from the northeast and is present in over 90% of the nights. The NLLJ was not observed in Burkina Faso, but at locations both north (Niamey, 13°N), and south (Nangatchori, 09°N) of Ouagadougou.
3.1. Sites, equipment and data
Meteorological data were collected during three field studies: October–December 2003, November to December 2004 and November to December 2007. Five measurement sites located within the Ouagadougou metropolitan area were chosen in 2003, four sites in 2004 and five in 2007. The locations of the sites are given in Figure 1. In 2007, when the purpose was to study the influence of cooling by the SVF the sites were placed from the city centre and northwards and some sites had more than one temperature logger to catch different SVFs. The sites were classified according to Stewart and Oke (2010) and descriptions of land cover, built structure, SVF and amount of vegetation are given in Table 1. The area covered with green vegetation was calculated for a circle with a radius of 400 m. The percent vegetation in such a circle was shown by Lindén (2010) to give the highest correlation with temperature. In 2007, SVFs were obtained by digital fisheye images captured at sensor height with a Nikon Fisheye Converter FC-E8. The lens has an opening angle of 182°; so in the subsequent data processing with the GIS programme IDRISI according to the method by Holmer et al. (2001) a cropping to 180° was performed. After cropping the image the diameter was 1644 pixels. SVFs for 2003 and 2004 were estimated from ordinary ground level images and with fisheye images from similar places as calibration.
Table 1. Study areas described in order of increasing distance to the urban centre. For locations, see Figure 1. The LCZ classifications are given in italics
Air temperature was monitored with Tiny-tag loggers (Gemini Data Loggers Ltd.) and Rotronic temperature-humidity sensors (Rotronic AG). The Tiny-tags were inter-compared in room temperature and in a heated room (up to 50°C) before and after field campaigns. Only instruments which differed less than ±0.3 K at 30°C were used in measurements. The Rotronics were calibrated in a climate chamber (temperature between −20°C and 20°C) and showed instrumental differences of T: ±0.2 K and Rh: ±0.6%. Temperature and humidity sensors were mounted at a height of 2.5–3 m on lamp posts or similar poles, as far away from surrounding walls and trees as possible to minimize influence of the close surroundings. The radiation shields were tubes of hard plastic (thickness 2 mm, diameter 9 cm) with a reflective outside surface. In 2003 and 2004 the tubes were mounted in a 30° angle from the horizontal plane, with the highest point to the north, to avoid any direct sunlight to reach the instruments. In 2007 they were placed vertically with the top end shielded from direct sunlight. A small fan powered by solar cells was also placed at the top end of the tube, forcing a constant air flow around the sensor. In order to minimize conduction of heat through the radiation shield a layer of 5 mm plastic drinking straws were also fitted on the outside of the tube and covered by an additional reflective outer surface.
Temperature sensors have been shown to be somewhat overheated by short-wave radiation in naturally ventilated radiation shields, but this effect is reduced by wind speed (Anderson and Baumgartner, 1998; Nakamura and Mahrt, 2005; Mauder et al., 2008). If the findings by Mauder et al. (2008) are applied to our radiation shields there will be an over-heating with about 0.3 K or less 2 h before sunset but it will diminish to 0 at sunset. The changing amount of over-heating prior to sunset corresponds to a virtual cooling of −0.2 K h−1. However, no corrections have been applied since the local wind speeds at the sites are not known. As a result the observed cooling before sunset probably is somewhat exaggerated. After sunset the influence of the naturally ventilated radiation shield has been shown to be < 0.05 K (Nakamura and Mahrt, 2005).
At Site 1 Urban Centre (LCZ Open-set mid-rise) and Site 5 DMN (LCZ Open-set trees) continuous measurements of wind, radiation and humidity were collected in addition to the air temperature measurement. Site 5 DMN is used as a reference station in our study and it was also used as reference by Offerle et al. (2005), Eliasson et al. (2009) and Lindén (2010). Details of the instruments are given in Table 2. All data are recalculated to 20-min averages in order to obtain a sufficiently good time resolution, but not too much scatter. In 2003, 35 days with simultanous data is used; in 2004, 20 days; and in 2007, 11 or 5 days.
Table 2. Instrument information
Tiny-tag Plus Gemini Data Loggers
2003 2004 2007
T: ±0.25 K
Rotronic Hygrometer MP 100A
Air temperature and humidity
1d and 5
10–15 s (lower at higher wind speeds)
T: ±0.3 K Rh: ±2%
Four cup anemometer
Wind speed: ±0.1 m s−1 Threshold 0.2 m s−1
Ultra sonic anemometer Young 8100
1 and 5
Wind speed: ±1% Direction: ±2°
Incoming and outgoing long- and short-wave radiation
1 and 5
Daily totals ±10%
3.2. Weather classification
To distinguish the cloudy and windy weather spells in an objective way, in spite of lacking systematic cloud observations, we as a first step calculated atmospheric stability. The basic idea was that stability is influenced by both windspeed and cloudiness; compare with the Pasquill–Gifford–Turner stability classification scheme (Mohan and Siddiqui, 1998). Available data made it possible to calculate the Richardson radiation number (Rirad) according to the method used by Mahrt and Ek (1984):
where g is the gravitation force (m s−2), z the height of measurements above ground surface (m), θ the potential temperature (K) and u the horizontal wind speed (m s−1). R is defined as
where Rn is the net downward radiative flux (W m−2), S the soil heat flux (W m−2), ρ the air density (kg m−3) and Cp the specific heat of air (J kg−1K−1). Data from the reference station at DMN (Site 5) were used for the calculation (20 min-averages from 2007).
The relationship between the Radiation Richardson number and atmospheric stability depends on the actual evaporation rate. Both S and the actual evaporation rate are unknown here but as suggested by Mahrt and Ek (1984), S would in most practical situations be neglected. The actual evaporation rate contributes relatively little and further it is likely to counteract (condensation is very unlikely since the difference between air and dew point temperature was always at least 15 K) the possible influence of S, hence these two factors were considered negligible. This may result in slight errors in the transition time between stable and unstable conditions, but general patterns will not be affected. These approximations allowed us to use Rirad as a general indication of the atmospheric stability as well as for comparing the days included in the study. A positive value indicates a stable atmosphere while a negative value implies instability.
The next step was to make a regression of the nocturnal average Rirad for each night and the corresponding DTR using the 2007 data at Site 5 DMN. Then this relation was used to get a proxy of the stability of the nights also in 2003 and 2004. In 2003 air temperature data from Site 5 (DMN) were available, but in 2004 when Site 5 had no logger Site 9 (INERA Nature) was used instead. These two sites were monitored simultaneously in 2007 and showed to have very similar temperatures. DTR is defined as the difference between the daily maximum temperature and the minimum temperature the following night.
4.1. Nocturnal weather
On the basis of the data from November to December 2007 at Site 5 DMN weather was classified either as: (i) days with an average nighttime Rirad ≥ 0.35, referred to as extremely stable nights or (ii) days with nighttime Rirad < 0.35, referred to as moderately stable nights. These two stability classes we estimate to roughly correspond to the G and F stabilities in the Pasquill–Gifford–Turner classification (Mohan and Siddiqui, 1998). According to our wind and long-wave radiation monitoring and unsystematic cloud observations no less stable situations occurred. Averages of the two weather types show that in daytime Rirad was slightly negative, i.e. the air was unstable (Figure 2(c)). During extremely stable nights the Rirad increased rapidly at sunset, i.e. the air became very stable, and lay on a high level all night although there were fluctuations. During moderately stable nights the Rirad peak was delayed by about 1 h compared to the extremely stable nights and around midnight the stability was weak (0 < Rirad < 0.01). The nocturnal variations in stability were mainly dependent on the variations in wind speed.
The fluctuating pattern of downwelling long-wave radiation (Figure 2(b)) during moderately stable nights, especially during the second half of the night, could be explained by the presence of more and/or denser clouds during these nights compared to the extremely stable nights due to higher emissivity of clouds compared to air. Wind speed (Figure 2(a)) showed a pronounced diurnal variation with considerably lower wind speed in the night. The lowest wind speed occurred at or shortly after sunset. There was an evident difference in wind speed between moderately and extremely stable nights.
There was a highly significant (p < 0.001) relationship (R2 = 0.70) between Rirad and DTR (Figure 3). This is reasonable since DTR is also sensitive to nocturnal net radiation (influenced by cloudiness) and wind speed. The DTR data points are divided in two groups representing the two weather types observed in the field. Moderately stable nights (Rirad < 0.35) had a DTR below 16 K. Thus DTR could be used to divide the nights according to weather. DTR was then used to classify also the days (nights) in 2003 and 2004. Table 3 shows the number of moderately stable and extremely stable nights used in this study.
Table 3. Number of studied nights during the three field studies
In Figure 4 average temperatures and cooling rates are given for extremely stable and moderately stable nights for all sites used in 2003, 2004 and/or 2007. Each curve represents the ensemble average at one site each year. Thus Site 1b Urban Centre has three curves as it was used during all three years, Site 5 DMN two curves and the others one curve each as they were used during only one year. The three curves at Site 1b are remarkably similar during the extremely stable nights so it is very difficult to differentiate them. The average nighttime temperatures differ only 0.4 K at Site 1b and the same is valid for the two curves at Site 5. In the case of moderately stable nights temperatures differ more between the years depending on differences in wind and cloudiness, but the shape of the temperature curves is similar which indicates that the cooling rates were not affected. It can thus be justified to pool all data and use them together and not consider the year in which the data were collected.
During periods with extremely stable nights (Figure 4(a)) daytime air temperatures were within a narrow range while the nighttime temperatures were divided into two groups by a gap of 5–6 K. These groups differed according to the amount of vegetation surrounding the sites (>40% vegetation and <10% vegetation, respectively). During the spells with moderately stable nights (Figure 4(b)), characterized by the presence of clouds and higher wind speed, daytime temperatures differed more as a result of varying insolation. In the night, cooling was reduced, especially after midnight, probably due to an increase in wind speed and cloudiness. Nonetheless, it is still possible to see a division in cooling rate between vegetated and sparsely vegetated sites around sunset.
As shown in Figure 4, vegetated sites cool more rapidly around sunset (16:00 –20:00 hours LST, Phase 1) than sparsely vegetated sites. After these 3–4 h of intense sitelong-wavedependent cooling, all sites cool more or less at the same but diminishing rate until sunrise (20:00 –06:00 hours LST, Phase 2). One consequence is that the temperature differences between the sites develop in Phase 1 and are almost the same throughout the night. The twolong-wavephase cooling pattern applies both for the moderately and extremely stable nights. For some sites, a wavy pattern can be seen in the cooling rate. The waviness is probably the result of an interaction of the local surface wind and the regional wind above as described by Lindén and Holmer (2011), but will not be discussed further in this study.
Figure 5 shows mean air temperatures and cooling rates for the moderately and extremely stable nights for the sites in Figure 4(a) and (b). Daytime air temperatures were about the same in both cases but at sunset temperatures started to diverge and continued to do so all night. As a result, cooling was stronger during extremely stable nights. This was valid not only in the intense cooling in Phase 1 around sunset, but also during the entire night, although the cooling rates did not differ much. On average, the cooling rates in Phase 1 were −1.7 and −2.4 K h−1 for moderately and extremely stable nights, respectively. In Phase 2 the corresponding figures were −0.5 and −0.7 K h−1. Note that the average cooling was more than three times faster in Phase 1 than in Phase 2, while the most intensive cooling in Phase 1 was about six times faster. This applies to both classes of stability.
4.3. The influence of vegetation and SVF on nocturnal cooling
Figure 6 shows the average diurnal course of air temperatures and cooling rates during extremely stable nights when the sites are divided into four groups according to percentage green vegetation (as defined above) and SVF (<0.7 or ≥0.7). Sites 2 and 6 are excluded since these sites only have data from moderately stable nights. There is a strong influence of vegetation while the effect of SVF is weak. At the time of the most intensive cooling (18:00 hours LST) vegetated sites with high SVF cooled at a rate of −5.8 K h−1 while sparsely vegetated with high SVF only cooled −2.7 K h−1. The corresponding cooling rates with low SVF were −5.1 K h−1 and −1.9 K h−1, respectively. Thus the difference in SVF changed the cooling rates with 0.7–0.8 K h−1 while the difference in vegetation gave a change of 3.1–3.2 K h−1. Contrary to this, later in night (Phase 2) the differences between the groups were more or less the same which means that neither vegetation nor SVF longer have any discernible influence on cooling of the air.
The influence of vegetation is further analysed in Figure 7 where the relations between percentage of green vegetation (mostly trees) and average cooling rate for Phases 1 and 2 are shown divided on weather types. There are strong linear relations in Phase 1 for both stability classes, but the cooling influence of vegetation was more powerful for extremely stable nights. In Phase 2, the influence of amount of vegetation was negligible, but there was slower cooling during moderately stable nights.
4.4. Evening evapotranspirative cooling
As shown above, the amount of vegetation is important to the cooling process. A possible cause of this is that latent heat is consumed by evapotranspiration. Figure 8 shows air temperature, specific humidity (g m−3), cooling rate (K h−1) and specific humidity rate (g m−3 h−1) at Site 1d Urban Centre—roof top (LCZ Open-set mid-rise with sparse vegetation) and Site 5 DMN (LCZ Bush/shrub and closelong-waveset trees) in November to December 2007. Due to the uncertainty of the humidity measurements, 1 h moving averages are used to reduce the scatter. Canopy layer temperatures at Site 1 were not available for the whole period with humidity monitoring (11 d). However, a comparison between the roof top temperature (Site 1d) and the canopy layer temperatures (Site 1a Open yard and Site1b Street) for 5 days showed that the yard was −0.2 K colder and the street 0.5 K warmer than the roof during the night. Thus the roof top temperature developed in between the two canopy layer temperatures can be used to represent the canopy layer.
During the extremely stable nights the specific humidity increases rapidly (Figure 8(a)) as air temperature and wind speed decreased around sunset, i.e. when atmospheric stability shifted from unstable to stable. The increase in specific humidity was mainly a result of the increased supply of water vapour when the dilution by turbulent processes ceased. Specific humidity remained high during the entire night and decreased rapidly 1–2 h after sunrise, when the morning inversion broke up and turbulence increased. During the night, specific humidity was higher and air temperature lower at the vegetated site (Site 5 DMN) compared to the sparsely vegetated urban centre (Site 1).
The vegetated site showed a pronounced peak in specific humidity rate as the cooling rate increased (Figure 8(b)). This is to be compared with the sparsely vegetated urban sites, which showed a much smaller peak in specific humidity rate. Since the soil surface was very dry the increase in specific humidity must be interpreted as a result of transpiration. Evening evapotranspirative cooling (EEC) therefore seems to explain much of the difference between the sites. On average, the humidity rate in Phase 2 was slightly negative at both sites. As for the cooling rate, the specific humidity rate was more or less the same at the different sites in the second phase. During moderately stable nights, specific humidity was about the same at the two sites (Figure 8(c)). This was also the case for air temperature. As a result there were small differences in the nocturnal cooling and specific humidity rates (Figure 8(d)).
4.5. Intra-urban differences of cooling rates
As shown in Figures 5-8, vegetation seems to explain much of the difference in cooling rates between the sites. Other possible explanations are the SVFs at each site, soil heat flux and location within the builtlong-waveup area. In Figure 9 the cooling rates at the sites are plotted according to distance from the city centre. Each year is treated as a separate dataset; Site 1 Urban Centre at 0 km has three replicates, for example. In Phase 1, cooling rates differed considerably between the sites. The sparsely vegetated Site 1 Urban Centre (LCZ Open-set mid-rise) has a cooling that is only slightly less intensive than the sparsely vegetated sites at the city border about 7 km from the city centre—Site 7 CILSS (LCZ Bare soil) and Site 8 Spontaneous settlement (LCZ Light-weight low-rise). Furthermore, the vegetated sites at about 4 km [Site 3 Forest/reservoir (LCZ Close-set trees), Site 4 Modern residential (LCZ Open-set low-rise) and Site 5 DMN (LCZ Bush/shrub and close-set trees)] have cooling rates equal to the ones at Sites 9 and 10 Inera (LCZ Open-set trees) and are situated more than 10 km from the city centre and 6 km outside the perimeter of the built-up area. It is thus clear that distance from city centre has very little to do with the observed temperature differences between the sites.
The slight difference in cooling rate (about 0.3 K h−1) between Site 1 in the city centre and Sites 7 and 8 at 6.4 km and 7.2 km, respectively, on the outskirts of the city could indicate a slight distance effect but a statistical t-test shows that even with α = 0.2 there is no significant slope of the regression line. Nevertheless, from a physical point of view there are differences in the properties of the sites. The SVF at the urban City centre site is 0.64 while the SVFs at the sites at the outskirts are about 0.8. Differences in soil/storage heat flux may also contribute since thermal admittance probably differs due to buildings and paved main streets although sidewalks and smaller streets consist of soil. However, the intra-urban differences are much bigger. Sites 3–5 which are vegetated and situated 3–4 km from the city centre cool much faster than the sparsely vegetated sites 1, 7 and 8 but with a similar rate as the vegetated Sites 9 and 10 at 10.5 km (about 6 km beyond the built-up perimeter. Sites 3 and 4 have almost the same amount of vegetation, but cooling was more intensive at Site 4 in spite of its lower SVF (see Table 2) and higher soil wetness (irrigated). This is interpreted as EEC is more important to cooling in this environment.
The influence of SVF was also studied during 1 week in December 2007. Eight loggers were recording simultaneously. During this period all nights were classified as moderately stable. The relations between SVF and cooling rate for vegetated and sparsely vegetated sites are shown in Figure 10. In Phase 1 there is a linear relation between SVF and cooling rate; this explains 42% and 98% of the variance at the vegetated and sparsely vegetated sites, respectively. The regression lines are almost parallel and the difference between them is about 0.4 K h−1 which can be regarded as the average influence of vegetation around the sites when the influence of SVF is taken into consideration. However, there is no influence of the SVF on cooling rate in Phase 2 for either the vegetated or sparsely vegetated sites. All sites have almost the same cooling rate at this stage.
5. Interpretation and discussion
5.1. Two-phase cooling
The intra-urban two-phase cooling in Ouagadougou is the same pattern that was previously found in the high-latitude temperate city of Gothenburg, Sweden (Holmer et al., 2007) and in the Mediterranean climate cities of Adelaide, Australia (Erell and Williamson, 2007) and Athens, Greece (Giannopoulou et al., 2010). The two-phase cooling is also clearly seen in the data from wet-tropical Singapore (Chow and Roth, 2006). two-phase cooling thus appears to be a common phenomenon of nocturnal cooling independent of the climate zone.
Although the two-phase cooling pattern could be seen in all nights examined in Ouagadougou, the magnitude of cooling is determined by the prevailing weather condition, i.e. clear and calm conditions favour large nocturnal cooling and intra-urban differences. During moderately stable nights, characterized by cloudier and windier weather conditions, radiative cooling is less and vertical mixing and advection of air from the surroundings are stronger, decreasing the cooling and intra-urban differences compared to extremely stable nights. This is mainly important in Phase 1.
According to Holmer et al. (2007), site-dependent sensible heat flux, together with radiative cooling were believed to be the main cooling processes in Phase 1 in Gothenburg. Both are related to the SVF at the site. In Phase 2, cooling in the canopy layer was independent of the SVF and presumed to depend on the radiative balance between the canopy layer and the nocturnal inversion layer above. Consequently, cooling at ground level depends not on the surface characteristics, but on the cooling of the inversion layer by radiative divergence. To what extent is this conceptual cooling model applicable in Ouagadougou? An outline of the energetics is found in Offerle et al. (2005). From their results it can be seen that the sensible flux decreased rapidly before sunset, but remained directed upwards. At the same time net radiation became strongly negative and the storage heat flux changed sign to compensate the energy losses. From about 19:00 hours LST, i.e. 1 h after sunset, the turbulent fluxes were close to 0 and the storage heat flux equalled the net radiation loss. Thus the turbulent fluxes had some importance during Phase 1 while net radiation was dominant during Phase 2.
5.1.1. Phase 1—site-dependent cooling
One obvious difference between Gothenburg and Ouagadougou is the role of vegetation. As shown in Figures 5-9, vegetated areas in Ouagadougou cool much faster than sparsely vegetated areas around sunset, i.e. in Phase 1. The intense cooling of vegetated areas coincides with an increase in specific humidity (Figure 8), i.e. latent heat is consumed in EEC. Sparsely vegetated sites have a small increase in specific humidity in Phase 1 and also less (evaporative) cooling. Due to the low soil-water content, the most probable explanation for the increase in specific humidity is evapotranspiration. It has often been assumed that evapotranspiration ceases at sunset, but several studies have shown that evapotranspiration might continue after sunset, especially in dry season savannah environments (Fisher et al., 2002; Domec et al., 2006; Dawson et al., 2007), at least partly as an effect of the mid-day depression of photosynthesis (Franco and Luttge, 2002; Hu et al., 2009). In Ouagadougou, the level of specific humidity stabilized or began to diminish about 2 h after sunset, indicating that evening evapotranspiration ceases at that time (Figure 8). It is unlikely that the decrease in specific humidity was caused by dew formation, since dew point temperature was far below air temperature. Instead there may have been a slight divergent turbulent upward flux or advection of air from drier surroundings.
The same phenomenon, i.e. an increase in specific humidity around sunset and intensive cooling has been reported from vegetated urban areas in Gaborone, Botswana by Jonsson (2004). He suggested that the increase in specific humidity was a result of evapotranspiration enhanced by irrigation. An increase in specific humidity after sunset has also been observed in temperate urban areas (Hage, 1975; Holmer and Eliasson, 1999), who explained the increase by urban evaporation and/or release of water vapour from automotive exhaustions. In Ouagadougou there are a very large number of vehicles, especially mopeds, which emit water vapour. However, the traffic intensity is much lower at DMN than in the urban centre, yet the specific humidity was higher at DMN. Since the water vapour in the exhausts is created in the engine it will not contribute to cooling, but will instead warm the air by condensation in the colder air outside the engine.
The increased evening-specific humidity concentration also depends on turbulence and stability. Latent heat flux (including transpiration from the vegetation) is highest in daytime as shown by Offerle et al. (2005). However, since turbulence also is effective during daytime vapour concentration will become low in the canopy layer since the evapotranspirated vapour is dispersed in a large volume of air above. At sunset, the latent heat and vapour fluxes are lesser, but in spite of this, vapour concentration increases since turbulence decreases and stability increases towards sunset and thus the effective volume of air that receives water vapour is reduced. In the morning the nighttime stable stratification disappears, turbulence increases and the effective volume to disperse water vapour increases and specific humidity becomes low in spite of increased evapotranspiration when solar energy becomes available.
During extremely stable nights the cooling was −2.9 K h−1 in Phase 1 at DMN, while it was −1.6 K h−1 at the drier Urban Centre. As a crude estimation the vegetation accounted for an extra cooling of −1.3 K h−1 at DMN, i.e. 45% of the total cooling. There may also have been some cooling due to vegetation at the Urban Centre, since an increase in specific humidity also appeared there. However, this increase could partly be the result of advection from more humid surroundings and thus not associated with latent heat change. Since some of the cooling at the Urban Centre depends on net radiation and latent heat flux (evapotranspiration) it may be concluded that sensible heat flux cooling is about the same or less than cooling by latent heat flux at DMN. During moderately stable nights the total cooling in Phase 1 at DMN was −1.4 K h−1 while the Urban Centre cooled −1.1 K h−1. Thus the extra (latent heat) cooling at DMN was −0.3 K h−1 which is 20% of the total cooling. As a result, sensible heat flux cooling becomes more important. This probably depends on the higher wind speeds during moderately stable nights.
Other studies have also shown that parks become colder than the surrounding built-up areas not only by day, but also in the night (Spronken-Smith and Oke, 1998; Li and Roth, 2007; Yokobori and Ohta, 2009; Murphy et al., 2010; Stewart and Oke, 2010). Yokobori and Ohta (2009), Lee et al. (2009) and Bowler et al. (2010) also showed that the influence of vegetation on air temperature was bigger during the night. None of them refer to cooling by evapotranspiration. Instead, it was proposed that radiative cooling enhanced by high SVF and/or soil properties was the cause of the low air temperatures. However, the difference in cooling in Ouagadougou exists also when the SVF is equal between vegetated and sparsely vegetated areas (Figure 10) and thus indicating EEC. Pochter et al. (2008) and Chow et al. (2011) have also pointed out that excessive nocturnal cooling depended on evapotranspiration. Further, Pochter et al. (2008) found that irrigated vegetation cooled faster than unirrigated, i.e. in spite of higher soil wetness and higher thermal admittance and thus better conditions for compensating upward soil heat flux during the evening the irrigated area cooled faster. In our study, we can see a similar relation between the unirrigated forest at Site 3 and the irrigated gardens ant Site 4 (Figure 9). The influence of differences in soil heat flux has not been specifically studied. However, it can be considered to be small in Ouagadougou. This may result from the fact that even in the city centre the sidewalks consist of bare soil also when the carriageways are paved with asphalt. Besides only the main streets are paved.
Several studies have shown that there is a relation between the SVF and the NUHI (see review by Unger 2004). In Gothenburg it was shown that there was a relation between SVF and cooling in Phase 1 (Holmer et al., 2007). Also in Ouagadougou, it was possible to see such a relation (Figure 10) when the sites were separated into vegetated and sparsely vegetated. Over the range of SVFs studied in Ouagadougou (from 0.3 to 0.9), the average cooling rate doubled from −0.8 to −1.6 K h−1. As data were available only for moderately stable nights, it is presumed that a still larger influence of SVF on cooling rate will be found during extremely stable nights. In general, the SVF is used in connection with radiation but it can also be regarded as an index of wind and turbulence. Consequently, the SVF is also important to cooling by the sensible heat flux. This means that as long as there is some wind and temperature decreases with height, there will be a cooling by sensible heat flux that depends on the space between buildings and trees. The SVF can also be considered as a (reversed) index of the ‘complete aspect ratio’ as discussed by Christen and Vogt (2004), i.e. the surface enlargement in built-up areas that enhances the release of heat stored in the buildings to the canyon air.
Cooling due to vegetation seems to be independent of SVF since the regression lines for vegetated and sparsely vegetated sites in Figure 10 are parallel. Contrary, hardware modelling by Perlmutter et al. (2009) shows that the early evening latent heat flux (i.e. evaporation) decreased with increasing SVF. Maybe the different influence of SVF is caused by the free water surfaces in the model and that the ‘complete vegetation fraction’ becomes smaller with decreasing SVF while the biological control of the transpiration is more important in Ouagadougou.
In most cities the NUHI intensity is highest in the city centre and diminishes towards the outskirts—or more precisely, the NUHI increases from the urban boundary towards the city centre (Fortuniak, 2003; Balazs et al., 2009). This may depend on the SVFs that are increasing from the city centre (Gal et al., 2009) or on the advection of cool air from rural areas. In Ouagadougou there is no such clear urban–rural spatial pattern. Instead, the intra-urban differentiation is much stronger, determined mainly by the amount of vegetation and the SVF. In addition, rural areas surrounding Ouagadougou have little tree vegetation (de Jong et al., 2000) due to the dry climate and foraging for firewood, which result in small urban–rural temperature differences. This can be compared with the evident NUHI in the medium-sized city of Akure, Nigeria surrounded by rainforest (Balogun et al., 2009).
In the light of the results of this study it seems probable that the rural site in Singapore (Chow and Roth, 2006) was influenced by EEC (secondary rain forest covering 95%). This may also explain the high intensity of the NUHI in relation to the tropical hot/wet climate and population that was found in Singapore. Another possible effect of evening cooling can be seen in Akure, (Balogun et al., 2009). During the dry season the temperature difference between the sparsely vegetated city centre and the rural airport surrounded by rainforest grew by 2.3–3.0 K around sunset. Still another probable effect of EEC can be found in Nagano, Japan, where Stewart and Oke (2010) showed seasonal differences in nighttime temperature in an Open-set tree area. The temperature departure from the traverse mean was clearly lower in spring and summer compared to winter. Since the leaf canopy decreases the SVF, the probable explanation for the temperature difference is EEC in the spring and summer.
5.1.2. Phase 2—spatially uniform cooling
From 2 h after sunset and throughout the rest of the night cooling is virtually the same at all sites in Ouagadougou. Thus the differential cooling in Phase 1 has lost its importance and any enhanced effect on cooling by vegetation, SVF or thermal admittance has disappeared. A fundamental change between Phases 1 and 2 seems to be the development of a capping inversion over the city. Such an inversion is not verified by direct measurements in our study, but considering the weather situation, with almost calm and clear skies coupled with the strong stability shown by our Rirad data an inversion is very likely to be present. Further, Offerle et al. (2005) remarked that neutral to stable conditions were observed at their mast in Eastern Ouagadougou during nights with weak wind. In temperate areas Godowitch et al. (1985) and Uno et al. (1988) have observed the development of an elevated inversion about 2 h after sunset. The latter group also remarked on the weak sensible heat flux in the urban boundary layer (UBL) under the inversion. Both in Ouagadougou (Offerle et al., 2005) and Mexico City (Velasco et al., 2011) the turbulent fluxes ceased 2 or 3 h after sunset and the net radiation was almost entirely balanced by the storage heat flux that became within ±5% of the net radiation (Velasco et al., 2011). Low turbulent fluxes some hours after sunset also applies to temperate areas (Christen and Vogt, 2004; Newton et al., 2007). It should, however, be noted that it is possible with an upward latent heat flux (evaporation) at the same time as the sensible heat flux is downward due to the inversion, e.g. Mexico city and Tucson, Arizona in Figure 4 in Roth (2007), but both will be small since turbulence is small. Thus long-wave radiation fluxes govern the energy exchange in the UBL and cooling becomes the result of radiative divergence. Since the atmosphere is by no means transparent to long-wave radiation, cooling of the air layer above the city also governs cooling of the city with a similar rate over the whole area.
Using pairs of pyrgeometers Fuggle and Oke (1976) and Nunez and Oke (1976) have shown that the calculated cooling of the air through radiative divergence was much larger than the actual cooling of the air. As compensation there was an upward convergent flow of sensible heat. Yap and Oke (1974) have shown that the convergence reaches about 20 m above the roofs. With a two-channel radiometre Nunez and Wilson (2006) found that on a clear and calm night almost all cooling through radiative divergence was compensated by heat stored in the ground.
5.2. The use of DTR for classification of nocturnal weather
The common way to handle the influence of weather in urban climate studies is to divide data according to wind speed and cloudiness thus there are many investigations dealing with clear and calm weather. Another approach is to use the nocturnal lapse rate in NUHI studies (Ludwig, 1970; Lee, 1975, 1979). Further, if combining two diagrams in Jauregui et al. (1992) such a relation is also evident. However, in Mexico City Jauregui (1997) did not find a significant correlation between lapse rate and NUHI, but instead a relation between inversion depth and NUHI. Possibly that relation was due to that the site used (the airport) was almost entirely surrounded by built areas so there was not a rural boundary layer. A linear relationship between upwind rural temperature inversion and NUHI intensity was also found in a numerical modelling study by Yu and Wagner (1975).
The rural lapse rate is influenced by both wind speed and cloudiness and is an expression of the potential energy of the atmosphere. To also include the kinetic energy the Richardson number (the ratio of potential and kinetic energy) can be used. In our case we had data to use the radiation Richardson number (Rirad) and were able to divide the 2007 data set in two groups that differed in wind speed and (indirectly via downwelling long-wave radiation) cloudiness. As it was shown, there was a highly significant relation between Rirad and DTR we used this relation as a new method to classify weather and applied it for all three field study years.
Rural lapse rate and any Richardson number contain information about the vertical structure of the boundary layer, about the turbulence and vertical exchange of heat. These properties develop in an interaction between air and ground surface. DTR also develop in an interaction between air and ground surface. For example, if the thermal admittance is low cooling of the air is enhanced so if there are low wind speed and clear skies DTR will be large. At the same time there will develop an intense inversion and strong stability. The weather information will not be so detailed if only wind and cloudiness are used to classify the weather. An example of the importance of the surface is shown by Jonsson and Lindqvist (2005) who found lesser NUHI intensity during the wet season than in the dry season when comparing calm and clear nights, i.e., the weather was the same but not the surface properties and as a consequence the processes that influenced the NUHI did not got the same intensity.
5.3. Source areas—LCZ
The purpose of the LCZ presented by Stewart and Oke (2006, 2010) is primarily to define typical environments to make it possible to compare UHIs in different cities as well as intra-urban differences. However, the LCZs may also be considered as source areas. In Ouagadougou, Lindén (2010) obtained the best relationship between intra-urban temperatures and source areas with a radius of 400 m. This is smaller than the circle with 500 m radius used by Chow and Roth (2006) but a larger area than the 300 m grid used by Hart and Sailor (2009) and Yokobori and Ohta (2009) or 180 m sector in the wind direction used by Murphy et al. (2010). Li and Roth (2007) used both 100 and 500 m radius and found best correlation between vegetation cover and UHI intensity for the smaller circle.
Source areas need to be large enough to allow for micro-advection from the surroundings and thus to blend the effects of differences within the LCZ. The influence of parks on air temperature has been observed to reach up to 1 km from the park boundary (Jauregui, 1991; Spronken-Smith and Oke, 1998; Upmanis et al., 1998; Shashua-Bar and Hoffman, 2000). An example from Ouagadougou on the importance of considering the size of the source area is Site 5 DMN. This site was assessed as bare soil by Offerle et al. (2005). The closest surroundings consist of bare soil and scattered trees, but within the 400 m circle around the site there is an abundance of trees (more than 40% of the area), especially in the northeastern sector of the site; this is also the common wind direction during the dry season. Thus air temperature and specific humidity at the site adapt to the influence of vegetation in spite of that the bare soil close to the site. The effective size of a source area depends on wind and stability, but also on built structure and vegetation. According to Murphy et al. (2010) source sectors describe these properties better than circular source areas. However, the presence of local wind systems such as the ones found in Ouagadougou (Lindén and Holmer, 2011) makes this difficult to handle. Another example of the importance of advection was given by Giannopoulou et al. (2010). In narrow canyons the temperature development differed depending on the ambient air temperature in the surroundings of the canyons.
We encountered some difficulties in using the LCZs. As shown above vegetation has a major influence on the cooling process and the intra-urban temperatures. Since the amount of vegetation can vary a lot between neighbourhoods with similar built structure, it is necessary to take the amount of vegetation in the different zones into account, particularly the built-up ones. For example, in Ouagadougou ‘LCZ Open-set low-rise’ includes Site 3 which is a wealthy area with lots of trees and irrigated gardens, but also Site 6 which is a traditional area with sparse vegetation. Comparable data are only available for moderately stable nights but then the average temperature difference was about 3 K. Thus, by consideration of vegetation in built-up areas, one would improve the means of comparing intra-urban temperatures and UHIs of different cities and climate zones.
In tropical climates there are often considerable differences in soil wetness and thus thermal admittance over the year. Consequently, the UHI differs by season. In a previous version of the LCZs (Stewart and Oke, 2009), this was accounted for but not in the latest (Stewart and Oke, 2010). On the basis of our results we find it necessary to reintroduce soil wetness to handle the seasonal difference of UHI.
5.4. Final remarks
Despite the high daytime temperatures in Ouagadougou at this time of year (35–40°C), the rapid night time cooling in vegetated areas causes considerably lower nocturnal temperatures (minimum air temperature 16–18°C in vegetated areas compared to 21–23°C in sparsely vegetated). Though Burkinabe locals often claim that nocturnal temperatures during this period are too low for comfort, the use of vegetation in urban planning may be an effective method of mitigating nocturnal heat stress where this is desired.
One possible cause of the increase in wind speed and cloud cover during moderately stable nights could be the NLLJ over West Africa, described by Lothon et al. (2008) and Schrage and Fink (2010). During NLLJ events, both the near ground wind speed and the amount of low clouds increase. The development of a NLLJ also shows that a rather thick inversion developed prior to the NLLJ (Blackadar, 1957; Kallistratova et al., 2009). One consequence of the probable connection of moderately and extremely stable nights in Ouagadougou to features in the regional meteorology in West Africa is that night time cooling can be better predicted. This might be of particular importance to people without air conditioning, since it would enable them to get nighttime relief from the intensive daytime heat during parts of the dry period. It can also be included in the modelling of future climates and estimations of the increased heat stress induced by a warmer climate.
There are large intra-urban differences in the nocturnal air temperatures in Ouagadougou. Sites with abundant tree vegetation (>40% of the surface) have temperatures that are 5–6 K lower than sites that are sparsely vegetated (<10%).
Nocturnal cooling of air temperatures during anticyclonic weather is divided in two phases: In Phase 1, around sunset cooling is site-dependent and differs greatly between the sites (−6 to −2 K h−1 at sunset on extremely stable nights). In Phase 2, there is slow area-dependent cooling where all sites cool at a very similar but decreasing rate from about −1.2 to −0.4 K h−1 during extremely stable nights. Thus the temperature differences developed in Phase 1 are preserved until the morning. This pattern seems to be independent of city structure and climate zone.
The difference between sparsely vegetated and vegetated areas gave a cooling rate difference of 3.1–3.2 K h−1 at the time of maximum cooling for extremely stable nights. The corresponding figures for low and high SVF were 0.7–0.8 K h−1.
EEC is very important in vegetated areas as indicated by a larger increase in specific humidity in vegetated than sparsely vegetated sites. In Phase 1, especially during extremely stable nights, EEC is equal to or larger than the sensible heat cooling. During moderately stable nights its importance is strongly reduced and sensible heat flux cooling is instead dominant. The SVF of a site does not appear to be important to the EEC. In Phase 2, EEC is fractional.
In Phase 1, the net radiation appears to be of secondary importance to cooling compared to the turbulent fluxes but in Phase 2 it is the main cooling process. In Phase 2, the development of a very stable stratification (most probably inversion) makes turbulent fluxes very small.
Distance from city centre is of little importance in Ouagadougou and consequently also of little importance to the urban–rural temperature difference. Instead, it is the site-specific properties, especially the green vegetation, that are important.
The cooling rates during moderately stable nights are weaker than during extremely stable nights, but vegetated sites cool yet more than sparsely vegetated site. This is valid for Phase 1 as well as Phase 2 as a consequence of higher wind speed and more clouds during moderately stable nights.
The DTR showed to be useful as a weather classification.