Spatio temporal dynamics of the urban heat island in singapore 4

The LULC variables have significant relationships with nocturnal mean UHIraw,daytime mean UHIraw and maximum UHIraw.. However, none of these vari-ables have strong relationships with mini

Between 21:00 to 23:00 hrs, the highest intensities are found in the CBD, with

a warm belt stretching across the south-western coast of Singapore and a secondarypeak found over the western industrial estates These findings are consistent withthe study conducted three decades ago by the Singapore Meteorological Services(1986), which took measurements at 22:00 hrs (Figure 4.20) A notable feature notpresent in the 1986 study is the existence of warm spots in the north and north-east,which are expected due to rapid development of high-rise residential estates in thepast few decades

The cool islands in the central catchment area and rural north-west also pear to have diminished in influence, possibly caused by the development of theBukit Timah Expressway (beginning in 1983) and the Kranji Expressway (begin-ning in the early 90s), along with new residential and industrial estates along theexpressway, found between the two cool zones Another factor that may have con-tributed to the northward migration of the cool center in the rural north-west is thedevelopment residential and industrial areas in the Jurong West Extension in theearly 1990s The 1997 study by Goh and Chang (1999) found that the residentialestates in Jurong West have the highest heat island intensities among the 17 townssampled in Singapore

ap-The growth of another secondary heat island in the east is discussed in Gohand Chang (1998), a period during which new developments in the east were com-pleted In the present study, large parts of the east have high UHI intensities withthe exception of a small cool spot over an airfield, noticeably different from threedecades back

Figure 4.20: Isothermal maps of Singapore during the NE (top) and SW tom) monsoons produced with data collected over nine days between 1979 and

(bot-1981 Source: Singapore Meteorological Services (1986)

4.5.2 Spatial variation of ensemble mean monthly UHI across

a seasonal cycle

The spatial pattern of UHIraw also experiences seasonal variations (see Section4.3.2) In January, during the wet and cool NE monsoon season, mean UHIraw gra-dients are small and both heat island and cool islands are not very developed (Fig-ure 4.21) The highest mean intensities during this month are ∼2◦C In February,UHI intensities are generally lower, with a more pronounced cool island (<0◦C) inthe central catchment and weaker heat islands (∼1.5◦C) in most parts of the island

At the end of the NE monsoon in March and April, heat islands begin to growstronger (∼2.5◦C) and thermal gradients are increasing Values of mean UHIraw

continue to increase, peaking around May and June (∼3◦C) During the south-westmonsoon, heat islands are less developed than the months before but remain strong.Towards the end of the south-west monsoon (September and October), UHI inten-sities increase a little, particularly in the the heat island in the south (∼3.5◦C) Asthe north-east monsoon looms, mean intensities across the island return to lowervalues, most significantly in November (∼2.5◦C) and December (∼2◦C)

The relative spatial differences between NE and SW monsoon are not unlikethose found in the study by Singapore Meteorological Services (1986) (Figure 4.20)

In the NE monsoon, strong heat islands are absent, with warm belts over the urbanareas The cool islands are distinctly larger during the NE than the SW monsoonperiods Strong heat islands are present during the SW monsoon in both studiesalthough some differences arise due to new urban developments discussed in theprevious section

4.6 Urban effects on UHI

Land use and land cover

The land use and land cover (LULC), in terms of built-up ratio (BUP)and vegetation ratio (VP), are calculated for the surroundings of each station(100m and 500m radii) (Appendix B) To determine the relationship between theabove variables and the UHI-related dependent variables, linear least-square re-gression is used Four types of statistical linear functions are used for curve fit-ting, namely, straight line (y=x), quadratic (y=x2+x), logarithmic (y=log(x)) andsquared (y=√

x) For each pair of dependent and independent variables, the tion that yields the lowest p-value is chosen as the optimal relationship 35 stationsfrom S01 to S40, with the exception of stations with limited data (S16, S26, S27,S33 and S35) are used in the regression analysis Data used are from the entireobservation period (Feb 2008 to Jul 2011) Daytime UHI values are defined as 07:00

func-to 18:50 hrs and nocturnal UHI values are defined as 19:00 func-to 06:50 hrs

Table 4.10, and Figures 4.22 and 4.23 show the relationships of built-up tio at a 100 m radius (BUP100) and a 500 m radius (BUP500), vegetation ratio

ra-at a 100 m radius (VP100) and a 500 m radius (VP500) against the dependentvariables of UHIraw and UHImax As UHImax has daytime values filtered, daytimemean and minimum UHImax values are irrelevant and thus left out Comparing theresults, relationships of LULC variables with UHIraw are consistently better thanthose with UHImax This is likely due to the small size of data available for thelatter As such, discussion will revolve around UHIraw from here on

The LULC variables have significant relationships with nocturnal mean UHIraw,daytime mean UHIraw and maximum UHIraw In particular, VP100 and VP500 ex-plain most of the variances in nocturnal mean UHIraw (R2 > 0.6) On the other

hand, their predictive strengths for daytime mean UHIraw are notably weaker thanBUP100 and BUP500 Maximum UHIraw is best explained by VP500 but the otherLULC variables have relatively high R2 values too However, none of these vari-ables have strong relationships with minimum UHIraw, although BUP100 is weaklycorrelated with it (p < 0.05).

Table 4.10: Urban variables and their relationships with dependent variables.Observations from 35 stations for the entire observation period (Feb 2008 to Jul2011) are used

y=x 0.627***

y=x 0.677***

y=log(x) 0.351**

y=log(x) 0.111

y=x 2 +x 0.290 DM

UHI raw

y= √x

0.499***

y=x 0.385***

y=x 0.290***

y= √x0.244**

y= √x0.243*

y=x 0.213**

y=x 2 +x 0.478** Max

UHI raw

y=x

0.513***

y=x 0.547***

y=x 0.492***

y=x 0.608***

y=log(x) 0.321**

y=log(x) 0.085

y=x 2 +x 0.170 Min

UHI raw

y=log(x)

0.149*

y=x 0.022

y=x 2 +x 0.074

y=x 2 +x 0.077

y=x 0.123

y=x 0.177*

y=log(x) 0.180*

y=x 0.611***

y=x 0.632***

y=log(x) 0.310**

y=log(x) 0.088

y=x 2 +x 0.189 Max

UHI max

y=x

0.475***

y= √ x 0.469***

y=x 0.480***

y=x 0.551***

y=log(x) 0.360**

y=log(x) 0.080

y=x 2 +x 0.112

*** = p < 0.001; ** = p < 0.01; * = p < 0.05; + = p < 0.1;

The equations of the best performing LULC variables with strongly cant relationships (p < 0.001) are as follows:

signifi-Nocturnal mean UHIraw =−0.033 VP500+ 3.826 (4.2)

Daytime mean UHIraw = 0.130�

BUP100− 0.238 (4.3)

Maximum UHIraw =−0.039 VP500+ 6.562 (4.4)

The similar slopes of the linear equations for Equations 4.2 and 4.4 (-0.033 and-0.039 respectively) suggest that nocturnal mean UHIraw and maximum UHIraw

are influenced at similar rates by the ratio of vegetated surfaces in a 500 metre dius from each station For every 10% increase in vegetated surface ratio, nocturnalmean and maximum UHI decreases by 0.3 to 0.4◦C The base value difference (when

ra-VP500 = 0) of just under 3◦C is the main differentiating factor between Equations4.2 and 4.4 As for daytime mean UHIraw, when BUP100 = 0, the base value is-0.405◦C For every 25% increase in BUP100, the daytime mean UHIraw increases

Canyon geometry

The canyon geometry factors, H/W, zH/W and SVF (see Section 3.5.2),have been calculated for the surroundings of each station (Appendices B and D).Among the three canyon geometry variables, H/W ratio has the highest predic-tive strength (p < 0.01) for nocturnal mean UHIraw and maximum UHIraw (Table4.10) In general, zH/W ratio has the lowest predictive strength for these two de-pendent variables SVF has the strongest relationship with daytime mean UHIraw

(p < 0.01) Interestingly, zH/W and SVF have weakly significant relationships withminimum UHIraw (p < 0.05)

The equations of the best performing canyon geometry variables with strongrelationships (p < 0.01) are as follows:

Nocturnal mean UHIraw = 0.293 log10H/W + 3.426 (4.5)

Daytime mean UHIraw = 1.299 SVF− 1.885 (SVF)2+ 0.6 (4.6)

Maximum UHIraw = 0.309 log10H/W + 6.076 (4.7)

As with LULC variables, the variable H/W best predicts both nocturnalmean UHIraw and maximum UHIraw The coefficient for base-10 logarithm of H/W

is similar between the two variables (0.293 and 0.309) with the main difference ing the base values when log10H/W = 0 The base values here are also similar tothose from LULC suggesting the effects of land use, land cover and urban geometrymay be linked and difficult to separate All in all, the canyon geometry variables

be-do not account for UHI variance as well as the LULC variables

A logarithmic relationship between height-width ratio and UHI intensity isalso identified by Oke (1981) when comparing H/W in city centres against theirrespective UHIM AX However, little or no studies have verified this relationship and

a past study on Singapore by Goh and Chang (1999) also questions the logarithmicrelationship However, the present study reports a relatively better performance ofH/W in predicting night time UHI values in this study (R2 = 0.351) as compared

to the study by Goh and Chang (R2 = 0.285) which uses a median H/W across theestate to predict heat island intensity in a straight-line function As the H/W ratioused in the present study focuses on the immediate proximity of the sensor, it may

be indicative of the relative importance of microscale variables

Urban metabolism

In the previous discussion on QF, two possible means of QF influencing UHIare identified There is the spatial variation of QF dependent on the land use andfunction, as well as the temporal variation due temporal differences in patterns ofanthropogenic activity (e.g Sailor, 2011; Quah and Roth, 2012) As comprehensivedata for anthropogenic flux for each station is not available, surrogate variables areused to identify any significant effects Spatial variation of QF is difficult to detachfrom existing urban variables such as built-up ratio and is omitted With regards

to temporal variability, significant variations in QF within a week (weekdays vsweekends) are identified by (Quah and Roth, 2012) If QF plays an important role

in influencing UHI in the study area, comparing weekday (Mon to Fri) and weekend(Sat and Sun) observations would yield identifiable differences as most Singapore-ans have a 5-day working week

Figure 4.25 shows the relationship between weekend and weekday mean

Table 4.11: Distribution of stations and a comparison of their maximumUHIraw values for weekdays and weekends Data are taken from all stationsacross the entire observation period (Feb 2008 to Jul 2011) WE = weekends,

WD = weekdays

S31, S33, S42, S44, S45, S46 Low-rise

QF from anthropogenic activities (Chow and Roth, 2006; Quah and Roth, 2012).For this purpose, Table 4.11 was drawn up All of the commercial sites had higherweekday than weekend maximum UHIraw intensities, consistent with the hypoth-esis Industrial stations are also expected to have lower QF during weekends andthree of these stations (S02, S12 and S36) had higher maximum weekday UHIraw

intensities than weekend maximum UHIraw intensities Only one industrial station(S25) had the opposite relationship For residential stations (both low-rise andhigh-rise), three stations (S05, S06 and S32) had higher weekend maximum UHIrawintensities than weekday maximum UHIraw intensities However, eight other resi-dential stations had the opposite relationship There is almost an equal number ofstations sited in rural areas, parks and coastal areas in both categories

4.7 Weather effects on monthly UHI

Earlier sections studied the variation of air temperature and consequently UHI,given ”mean” conditions across the entire study period and filtering for weatherconditions This section investigates the influence of synoptic weather conditions

on mean monthly UHI intensities across all stations As we are interested in thelonger-term effects of weather, the non-filtered definition of UHI, ΔTu −r, will beused Due to a lack of synoptic data with high spatio-temporal resolution, monthlydata from Changi Meteorological Station will be used as a surrogate

On a day-to-day basis, the effects of extreme weather conditions have a morepronounced effect but it is difficult to isolate the effects of synoptic conditions onUHI For example, in an ideal situation, the same amount of rain has to fall atthe same intensity over both the urban and rural sites, ceteris paribus Any asyn-chronous occurrence of discrete weather conditions, such as rainfall events, willbring about artificial increases or decreases in UHI When averaging across entiremonths, relationships between weather conditions and UHI become more clear

Air temperature

Regression plots of monthly air temperature at Changi Meteorological tion against monthly mean and monthly mean maximum UHI (Figure 4.26) showsignificant positive relationships (p < 0.001) with R2values above 0.5 Based on theregression equations, mean and maximum monthly UHI increase by 0.274◦C and0.411◦C, respectively, with every degree increase in air temperature Air tempera-ture, however, may not be the direct factor that influences the UHI but correlatedalong with other influential factors such as solar radiation and drier conditions,

Sta-similar to the differences between winter and summer UHI in temperate countries(Oke, 1982).

Rainfall

Interestingly, there are no significant relationships (p > 0.1; R2 < 0.1) tween total monthly rainfall, and either mean or maximum UHI intensities Asecond monthly variable, number of rain days, was also tested against mean andmaximum UHI intensities and similarly yielded no significant relationships A pos-sible explanation is that soil moisture (and hence thermal admittance) is a moreimportant variable and has a lagged relationship with rainfall events

be-Wind speed

Monthly mean wind speeds have significant negative relationships with monthlymean and maximum UHI intensities (p < 0.001 and p < 0.01 respectively), withcoefficients of determination at 0.298 and 0.414 respectively With every ms−1 in-crease in wind speed, monthly mean and maximum UHI intensities are expected todecrease by 0.284 and 0.485 ◦C respectively This finding is consistent with previ-ous research showing the influence of wind on UHI intensities (e.g Oke, 1998) andwith the study on Singapore by Chow and Roth (2006)

Air temperature at Changi Met Station(°C)

Monthly rain at Changi Met Station(mm)

Wind speed at Changi Met Station(m/s)

Figure 4.26: Regression of monthly mean UHI intensity against (top) air perature, (middle) total monthly rainfall, and (bottom) mean wind speed Notethat UHI intensity here is ΔTu−r Observations from 35 stations for the entireobservation period (Feb 2008 to Jul 2011) are used Shaded region representsthe 95% confidence bands