Spatio temporal dynamics of the urban heat island in singapore 2

Built-up ratio BUP and vegetation ratio VP Satellite imagery is a common choice for delineating urban land cover types.Two main methods are used.. Figure 3.24: Mosaicked satellite images

3.4 Data quality control

The task of handling the data quality is seldom an easy one In most climate search fields, the main reason for data quality control is to ensure that the dataset

re-is homogeneous (Aguilar et al., 2003) The appeal behind ensuring homogeneity re-isthat it removes any “noise” from sources that may potentially create non-climaticbiases in the data

In the case of urban climate studies, there is a fine distinction between ban influences and other influences The homogeneity in this case refers to climatedata that represent variation due to urban development and possibly some indirectcausation, and the elimination of other inhomogeneities, such as artefacts created

ur-by lagged events The idea is to study the impact of urban development on anotherwise undeveloped location Errors may also occur due to other reasons, such

as instrument error, human error or spikes during data transfer or from externalnon-climatic forces (e.g fires as in the case of S11 or a warm vehicle parking next

to the sensor)

Instrument calibration

Calibrations across all sensors were done prior to mounting in the field inFebruary 2008 The purpose was to ensure that deviations between sensors did notexceed acceptable margins In July 2009, the sensors were taken down for anothersession of calibration Calibrations are done by placing all sensors in a homogeneousenvironment in close proximity (e.g Figure 3.19) In both calibrations, agreementacross sensors was acceptable as differences were < ±0.1◦C, which is less than theaccuracy level of the sensor (±0.2◦C)

Data post-processing

While the determination of erroneous data often requires subjective ment, the large volume of data in this study means that an objective method isfirst needed to systematically scan for parts of data where errors may occur First

judge-a quick sweep of unlikely djudge-atjudge-a points (T > 40◦C and T < 16◦C), to remove realistic extreme values (with reference to Singapore) Next, a despiking approachwas used As air temperature is not normally distributed, the distance of three SDsaway from the mean was used as the lower bound and four SDs away from the meanwas deemed the upper bound All values exceeding the bounds were scrutinisedvisually for likelihood of being erroneous A second net was set by comparing maxvalues with 99th percentile values to determine isolated outliers

un-Scatter plots of two closely-related stations were used to identify any ble errors discussed above Pearson correlation is used to determine reference sitesthat are highly correlated and to form a basis for comparison (Boissonnade et al.,2002; Tayan¸c et al., 1998) A correlation matrix was calculated and “best pairs”(see Figure 3.20) were selected based on the correlation coefficient (R value) Thesepairs were then plotted as scatter plots to identify any obvious non-conformities

possi-Figure 3.21 shows an example of realistic values that escape the first net butbecome obvious when scatter plots of best pairs are plotted In this case, somediscretion has to be used as each pair has different acceptable levels of scatter

In Figure 3.21a it is clear that the stray values at the bottom of the large spreadare artefacts rather than actual occurrences These are most likely measurementsmade when sensors were already unmounted (e.g in a car) but mistakenly stilllogging due to human error As such, they are removed and Figure 3.21b shows thepost-correction scatter plot

0.00 0.02 0.04 0.06 0.08 0.10 0.12

St_01 St_03 St_05 St_06 St_08 St_10 St_11 St_13 St_15 St_16 St_18 St_20 St_21 St_23 St_25 St_27 St_29 St_31 St_32 St_34 St_37 St_38 St_40 St_42 St_43 St_45

Station code

Figure 3.23: RMSE between pre- and post-corrected values for each station

Built-up ratio (BUP) and vegetation ratio (VP)

Satellite imagery is a common choice for delineating urban land cover types.Two main methods are used Spectral analysis of satellite imagery (automated clas-sification) or classification by eye (supervised classification) A popular algorithmfor classification is the NDVI (e.g Botty´an et al., 2005):

N DV I = N IR − R

where NIR = spectral signature of near infrared band and R = spectral signature

of the red band

Figure 3.24: Mosaicked satellite images used for land use classification Source:Microsoft Virtual Earth.

For this study, satellite imagery is used with supervised classification but notthe NDVI algorithm Part of the reason is the unavailability of high-resolution NIR-band imagery A panchromatic SPOT 5 image with 2.5 metre resolution (Figure1.2) is used together with DigitalEye satellite images available on Microsoft VirtualEarthT M

, digitally mosaicked for this purpose (Figure 3.24) Ground-truthing wasconducted to ensure that no major land-use changes had occurred around the sta-tions After the entire study area has been classified, the percentages are calculatedfor the radii of 100 and 500 metres around each station (Appendix B) An example

of the above can be seen in Figure 3.25a and an example of how the percentagesare obtained by pixel counts is available in Figure 3.25b Built-up areas includebuildings, road surfaces, parking spaces and other man-made surfaces Vegetationincludes forests, parks, field, grass patches and other vegetated natural surfaces,excluding bare soil and water bodies

(b)

Figure 3.25: (a) 100 metres (inner) and 500 metres (outer) radii from S02, and(b) calculation of land use percentages at 500 metre for S36

3.5.2 Urban structure

Sky view-factor (SVF)

Similar to H/W and zH/W ratios (discussed later) in attempting to conveysome information on the geometry of an urban canyon, the sky-view factor (SVF)quantifies the fraction of radiation emitted by one surface and captured by another(Oke, 1987; Grimmond et al., 2001) This has strong bearing on the L↑ values

Two main methods are used to determine SVF The first method is to usecomplex geometrical calculations to provide view-factors given the known dimen-sions of the canyon (e.g Oke, 1981; Johnson and Watson, 1984) GIS software can

be used to perform these calculations, although they may not model vegetationwell or provide an accurate results when dealing with complex geometry A secondmethod is to use fish-eye optical equipment Grimmond et al (2001) discuss theuse of a digital camera with fish-eye optical sensor and the LI-COR LAI-2000 PlantCanopy Analyzer This is an empirical method which 180◦ (studies have employedsensors from 140◦ to 189◦) hemispheric images obtained from full circular fisheyelenses The added advantage of fisheye imagery is the ability to account for thesky-view for 360◦ around the point where the photograph is taken, and 180◦ to theaxis of the lens, without the need for many mathematical assumptions

In this study, a Fujifilm IS Pro full-frame DSLR camera body is used with aSigma 4.5mm F2.8 EX DC Circular Fisheye HSM lens (Figure 3.26) The lens has

a documented view-angle of 180◦, in line with the recommendations by Grimmond

et al (2001) The lens also has a quantifiable area/angle projection which makes

it suitable for scientific purposes, in this case, areal calculations For consistency,images are taken with the camera body mounted on a tripod, at a height of 1.2metres A fluid leveller is also used to ensure that the camera body is level when

images were taken.

Figure 3.26: Top left: A Sigma 4.5mm F2.8 EX DC Circular Fisheye HSMlens mounted on the Fujifilm IS Pro full-frame DSLR Top right: A flash hotshoebubble leveller used to level the camera axis Bottom: a tripod

Images were processed using the Gap Light Analyzer (GLA) software written

by the Institute of Ecological Studies and Simon Fraser University (Figure 3.27)

A first round of processing was done to convert the image into a dual-tone imagerepresenting “sky” and “non-sky” pixels The sky view-factor is then obtained as

a proportion of pixel area that is classified as “sky”, noting that pixel area hasalready been weighted based on the projection The fish-eye images taken for thestations in this study can be found in Appendix D

Figure 3.27: User interface of the Gap Light Analyzer (GLA) version 2.0 bythe Institute of Ecological Studies and Simon Fraser University.

Height-to-width ratio (H/W) and roughness height-to-width ratio (zH/W ratio)

The height-to-width ratio (H/W) of an observation site is often used to acterize canyon geometry The ratio of the height of sides of an urban canyon toits width provides this value As with the sky-view factor (SVF), the H/W is oftencited as a factor that promotes heat retention in urban areas High H/W ratiosindicate tall and tightly-packed structures, restricting the degree to which the sky

char-is open to the surroundings of a site (Oke, 1982, 2006) As such, the H/W char-is a rameter which provides an indication of “street canyon” dimensions that influencethe ability of urban areas to radiate heat

pa-In urban climate zone (UCZ) site description scheme by Oke (2006), thegeneric “aspect ratio” is referred to as zH/W While it is conceptually similar, the

zH/W differs from the H/W in that vegetation is considered part of the canyongeometry and is included in the geometric calculations This differs from many

common uses of height-to-width ratio measurements which take into considerationonly buildings and structures in the calculations (e.g Goh and Chang, 1999; Chowand Roth, 2006), thereby not giving “rural” areas a roughness value.

According to Oke, vegetation is included in the calculation of aspect ratio cause it has some form of influence on the flow regime and thermal properties such

be-as roughness length, shading and dissipation of long-wave radiation (Oke, 2006).Roughness height-to-width ratio will be the term used to refer to zH/W ratio inthis report One challenge in determining both ratios is the wide-ranging urbanconfigurations of stations in this study As we are also interested in intra-urbandifferences and UHI in open spaces, not all of which have distinct urban canyons, aspecial method was devised to obtain the ratios Ratios are measured along tran-sects in 4-axes (N-S, E-W, NE-SW and NW-SE) and then averaged to provide anoverall 8-directional mean height-to-width ratio (Figures 3.28 and 3.29)

For each of the transects, to cater to irregular canyons and non-canyons, amean height-to-width ratio is used and vertical surfaces up to 100 metres horizontaldistance from the sensors are considered (Figure 3.28) Note that the height-to-width parameters used in this study are the 8-directional mean values and theindividual transects are merely used to determine them

The same approach is used to obtain zH/W with the exception that tion cover is also considered in the height and width calculations The zH/W tends

vegeta-to be considerably higher than H/W ratios for densely-vegetated areas (e.g forestsand parks); slightly higher for less vegetated areas (e.g residential land use); andidentical in areas without tall vegetation (e.g open fields and open car parks)

Figure 3.28: Determination of height-to-width ratio for each transect.

Chapter 4 Results and Discussion

The main goal of Chapter 4 is to identify and describe distinct patterns of tion in the empirical data collected The definitions used in this Chapter will followclosely to the discussions in Section 2.1 Times listed in this section will refer tolocal standard time (i.e GMT +8) unless otherwise stated Time interval for theair temperature and UHI calculations is 10 minutes unless otherwise stated

varia-Definitions of UHI-related dependent variables

Several calculations of UHI are employed The term UHIraw will refer tothe difference between a value measured at a particular site and the chosen ref-erence site (S16) at a specific point in time, i.e Tu − Tr, excluding hours whichare windy, cloudy and/or wet (i.e when Φm 6= 1 and Φw 6= 1) UHImax (“max”

in lower-case) will refer to the absolute maximum UHI intensity under dry ditions for a given time interval (e.g maximum UHI intensity possible at 21:00hrs) Thus, UHImax calculations is similar to UHIraw except for the added criterion

con-of Φ = 1, meaning no heavy cloud or rainfall events should have taken place at

any point in the day UHIM AX (upper-case) will refer to the absolute maximumUHI intensity (UHIraw or maxmax) measured for any station across all time periods.

UHIraw is mainly used to provide results reflective of actual conditions and toaccount for seasonal and inter-annual weather variations Where an ideal condition

is required, e.g the determination of UHIM AX, UHImax values will be used Calm,clear nights with no antecedent conditions (defined later) will provide a better in-dication of the maximum possible influence of urban development alone

Minimum and maximum UHIraw(or UHImax) are defined as the smallest andlargest value (respectively) of UHIraw (UHImax) for each station across the entirestudy period, unless a specific period is stated For example, monthly maximumUHIraw is the maximum UHI intensity in each month of the year Their inclusionallows evaluation of the influence of various factors on extreme values A subscript,(t), will be used to refer to the number of hours after sunrise during which a cer-tain UHI event occurs, e.g maximum UHIraw (t) hourly ensemble would mean thetime of peak for ensemble hourly UHIraw (t) As was already previously established,daytime and nocturnal UHI are influenced differently, therefore the nocturnal meanUHIraw (NM UHIraw) and the daytime mean UHIraw (DM UHIraw) are selected asdependent variables too

Artefacts in UHI calculations due to asynchronous rainfall events

UHI intensities are calculated from values of two different stations Synopticweather conditions (Φw) affect UHI but they do not always occur simultaneouslyand at equal intensities across all stations This increases the complexity of normal-izing the values as non-relevant factors may lead to misleading results (as discussed

in Section 2.1) For example, a rainfall event that occurs asymmetrically over one

Table 4.1: Rainfall distribution across meteorological stations on 7 July 2010

at 13:00 hrs Note that Tengah Meteorological Station is located approximately

2 kilometres east of the reference station (S16) in north-western Singapore (seeAppendix A)

Meteorological Station Rainfall (mm) on 7 July

2010 at 13:00 hrsTengah Meteorological Station 35

Changi Meteorological Station 0Seletar Meteorological Station 0Paya Lebar Meteorological Station 0

Sembawang Meteorological Station 0

Filtering process

In order to filter the dataset for the effects of Φw and Φm, hourly cloud andrainfall maps for the region were obtained for the entire study period from theWundermap radar map repository (http://www.wunderground.com/wundermap/ )

A shell script using the ImageMagick image processing library was written to tomate the cropping of these maps to the extent of the study area The scriptwas then used to identify days with heavy cloud cover and rainfall over Singapore.These were corroborated using hourly rainfall and wind data from five meteoro-logical stations in Singapore, namely, Tengah, Changi, Seletar, Paya Lebar andSembawang As the meteorological stations do not have a good spatial coverage,the radar map plays an important role in identifying any periods of heavy cloudcover or rain at any location in the study area

au-In the case of UHIraw, to filter for effects of wetted surface (Φm 6= 1), andheavy cloud cover (Φw 6= 1), data points that fall within two hours from the oc-currence of rainfall (hourly rainfall of >0 mm) and heavy cloud cover events thatappear on the radar, are filtered out Isolated data points spanning less than fourconsecutive hours are also removed as they are deemed to be unrepresentative Thecalculation of UHImaxalso uses the same procedure but has added constraints: only