1
Designation of an urban monitoring network based on Local Climate Zone mapping and
1
temperature pattern modelling
2
3
Running page head: Designation of an urban monitoring network
4
5
Lelovics E, Unger J, Gál T
6
7
Department of Climatology and Landscape Ecology, University of Szeged, Szeged, Hungary
8
lelovics.eniko@gmail.com, unger@geo.u-szeged.hu
9
10
Abstract
11
The recently developed Local Climate Zones (LCZ) classification system was not originally designed for mapping, but
12
to classify and standardize urban heat island observation sites. Nevertheless, if the aim is to characterize the areas with
13
different thermal reactions within a wider study area, the mapping seems to be a useful application of the system.
14
Our objectives are: (i) to develop GIS methods to calculate different parameters describing the LCZs for any part of the
15
study area, (ii) to identify and delineate the LCZ types occuring in the study area using the calculated parameters, (iii) to
16
select representative sites of an urban monitoring network using the mapped LCZs and modelled mean annual
17
temperature surplus pattern.
18
The input data were: 3D building, road and Corine Land Cover databases, aerial photographs, topographic map and
19
RapidEye satellite image. The basic area of calculation was the building block with the area belonging to (polygon).
20
These polygons classified with the same or similar parameter values were aggregated to evolve the appropriate size
21
zones. As a result, six built LCZ types were distinguished in the studied urban area.
22
An estimation of the temperature pattern was obtained by an empirical model. In order to designate the 24 stations of
23
the network the sites were selected inside the delineated LCZ areas taking also the modelled pattern into account. The
24
exact places on the lamp posts were determined by field surveys. The bias between the temperature pattern interpolated
25
from the modelled values of the 24 stations and the originally estimated pattern was found to be satisfactory.
26
27
Key words: Urban climate, Monitoring network, LCZs, GIS methods, Modelled temperature
28
pattern, Szeged, Hungary
29
30
1. Introduction
31
32
Owing to the anthropogenic activity, a local climate develops in the built-up areas. Nowadays
33
about half of the human population is affected by the artificial urban environments. This makes
34
studies dealing with the urban impact on climate particularly important. By definition the urban
35
climate is a local climate that is modified by interactions between built-up area and regional climate
36
(WMO 1983). This urban climate is different from the pre-urban (natural) one and is a result of
37
accelerated urbanization: construction of buildings, roads, etc., as well as of the emission of heat,
38
moisture and pollution related to human activities. Among the parameters of the urban atmosphere
39
the near-surface (screen-height: 1.5–2 metres above ground level) air temperature shows the most
40
obvious modification compared to the rural area (Oke 1987).
41
This urban warming is commonly referred to as the urban heat island (UHI) and its
42
magnitude is the UHI intensity. Traditionally, this intensity is interpreted as the difference between
43
values measured in the city centre (urban) and a nearby undeveloped (rural) site (ΔT
u-r
).
44
Nevertheless, in the heat island literature the term “urban” has no single, objective meaning as the
45
areas around the measuring sites could be very different depending on the investigated cities (e.g.
46
park, college ground, street canyon, housing estate, etc.). Similarly, the surroundings of the “rural”
47
sites varied in different studies, it may be e.g. airport, farmland, field, or even a less built suburb.
48
That is, on the one hand, for landscape classification or description of the site surroundings
49
the simple “urban”/“rural” (u/r) is not really appropriate because of the abundant variety of the
50
landscapes according to their surface properties which are reflected in the development of near-
51
surface micro and local climates (Stewart 2007, 2011). This makes (almost) impossible to compare
52
the results obtained in different parts of the world.
53
On the other hand, if the investigation is directed on the detailed monitoring of the
54
representative temperature distribution within the city, this is a difficult task because of the
55
2
complexity and variety of the urban terrain (Oke, 2004). The site locations of an intra-urban station
1
network and thus the question about its appropriate configuration raises an essential problem. This
2
problem is related to the relationship between the intra-urban built and land cover types and the
3
locations of the network sites. Two situations arise:
4
Situation (1): In the case of an already existing network (e.g. Schroeder et al. 2010, Siu &
5
Hart 2013) it may be required to characterize the relatively wider environment around the
6
measuring sites, namely what type of urban area surrounds a given station and whether it can be
7
clearly determined. In other words, how representative is the location of a station regarding a
8
specific, clearly defined urban environment type, that is, whether the data measured at this location
9
are typical for the thermal reactions of a given urban area?
10
Situation (2): In the case of a planned station network (e.g. Unger et al. 2011) the most
11
important questions are what surface types can be distinguished in the given urban area, how
12
precisely they can be delimited, how many they are, and whether their extension is large enough to
13
install a station somewhere in the middle of the area (representing the thermal conditions of this
14
type) while of course taking care to minimize the microclimatic effects of the immediate
15
surroundings (e.g. sunlit walls, AC heat emission). For the accurate positioning some temperature
16
information (earlier measurements, modelling) can be an additional help.
17
To address the questions raised above, Stewart & Oke (2012) developed the “local climate zone”
18
(LCZ) classification system, which describes the exposure and land cover characteristics of a
19
screen-height temperature sensor, that is the LCZs intend to reflect the thermal reactions of a wider
20
environment. It can be applied relative easily in any urban or rural environment. It is based on the
21
earlier classifications of Auer (1978), Ellefsen (1991), Oke (2004) and Stewart & Oke (2009), as
22
well as a world-wide survey of heat island measurement sites and their local settings (Stewart
23
2011). The elements of this system are presented shortly in Section 2.
24
The LCZ classification system was not designed specifically for mapping – it was designed to
25
standardize the classification of urban heat island observation sites, whether urban or rural, fixed or
26
mobile. Nevertheless, if the aim is to establish a new urban observation network, spatial mapping of
27
the urban terrain is a justifiable use of the system to determine areas that are relatively
28
homogeneous in surface properties and human activities, and to identify sites that are representative
29
of those areas. The studies in Hamburg, Germany (Bechtel & Daneke 2012) and Xuzhou, China
30
(Gamba et al. 2012) were among the first steps for automated extraction of LCZ areas in urban
31
environment using applied GIS and remote sensing methods.
32
The present study is connected to an EU-founded project (URBAN-PATH 2013). As a part of
33
this project urban monitoring systems are under development. When completed they will provide
34
online information in a form of maps on temperature, humidity and human comfort conditions
35
within Szeged and Novi Sad (Serbia) (Lazic et al. 2006). The temperature and relative humidity
36
stations (24 items in Szeged and 28 in Novi Sad) of the monitoring system will be located on lamp
37
posts according the above mentioned Situation (2).
38
The objective of this paper is three-fold: (i) to develop GIS methods in order to calculate
39
geometric, surface cover and radiative parameters describing the LCZs for any part of the study area
40
using different databases which are available or created for this purpose, (ii) to identify and
41
delineate the LCZ types which occur in the study area using the calculated surface parameters by
42
the developed methods and (iii) to select representative sites of the urban monitoring network using
43
the mapped LCZs and the modelled mean temperature surplus pattern.
44
45
2. Brief description of the LCZ classification system
46
47
The necessity and ideas of the development of the “local climate zone” classification system
48
and its structure are presented and discussed in detail in Stewart & Oke (2012). Therefore here we
49
highlight only the key features of the system.
50
The primary purpose of the system is to facilitate the characterization of the local
51
environment around a temperature measuring site, in terms of its ability to influence the local
52
3
thermal climate. To this end, the number of types (zones) is not too large and separation is based on
1
objective, measurable parameters. LCZs are defined as “regions of uniform surface cover, structure,
2
material, and human activity that span hundreds of meters to several kilometres in horizontal scale.
3
Each LCZ has a characteristic screen-height temperature regime that is most apparent over dry
4
surfaces, on calm, clear nights, and in areas of simple relief” (Stewart & Oke 2012). Each zone is
5
necessarily “local” in spatial scale because an upwind fetch of typically 200-500 m is required for
6
air at screen-height to become fully adjusted to the underlying, relatively homogeneous surface.
7
The main characters of the types are reflected in their names (Table 1).
8
The LCZ types can be distinguished by the typical value ranges of measurable physical
9
properties (parameters) (Table 2). These parameters largely characterize the surface geometry and
10
cover, but there are also those that reflect the thermal, radiative and anthropogenic energy features
11
of the surface. Stewart & Oke (2012) give typical values for the properties of each zone.
12
The interpretation of the above mentioned parameters are as follows: (1) sky view factor
13
(SVF) is the proportion of the sky dome that is „seen” by a surface, either from a particular point of
14
that surface or integrated over its entire area (Errel et al. 2011), (2) aspect ratio (H/W) is the ratio
15
between the average height of adjacent vertical elements and the average width of the space (Errel
16
et al. 2011), (3) building surface fraction (BSF) is the ratio between the horizontal area of buildings
17
on a given area and the total area, (4) impervious surface fraction (ISF) is the ratio between the
18
horizontal area of impervious surfaces on a given area and the total area, (5) pervious surface
19
fraction (PSF) is the ratio between the horizontal area of pervious surfaces on a given area and the
20
total area, (6) height of roughness elements (HRE) is the average height of the roughness elements
21
on a given area, (7) terrain roughness class (TRC) is the classification of the different urban and
22
natural landscapes into 8 class by the surface roughness increment (Davenport et al. 2000), (8)
23
surface admittance (SAD) is a measure of the ability of a surface to accept or release heat (Oke
24
1987), (9) surface albedo (SA) is the average ratio between the reflected and incident short wave
25
radiation on a given area, (10) anthropogenic heat output (AH) is the heat generated by human
26
activities on a given area.
27
In the context of the new LCZ classification system, the intra-urban UHI intensity is not an
28
“urban-rural” temperature difference (ΔT
u–r
), but an LCZ temperature difference (ΔT
LCZ X–Y
)
29
(Stewart et al. 2013). This difference can take various forms depending on the pairing of two LCZ
30
types. In this way, the application of the LCZ system gives opportunity to objectively compare the
31
thermal reactions of different areas within a city (intra-urban) and between cities (inter-urban).
32
33
3. Study area and earlier temperature measurements
34
35
Szeged is located in the south-eastern part of Hungary (46°N, 20°E) at 79 m above sea level
36
on a flat terrain with a population of 160,000 within an urbanized area of about 40 km
2
(Figs. 1a
37
and 1b). The area is in Köppen's climatic region Cfb (temperate warm climate with a rather uniform
38
annual distribution of precipitation). The annual mean temperature is 10.4ºC and the mean annual
39
amount of precipitation is 497 mm (Unger et al. 2001). The study area covers a rectangle of 10 km
40
× 8 km (80 km
2
) in and around Szeged (Fig. 1c).
41
To validate our results, temperature values originated from our earlier mobile measure
42
campaign were used. These measurements were taken by cars at the same time after sunset on fixed
43
return routes during a one-year period (April 2002 – March 2003) by several times in a grid
44
network (e.g. Unger, 2004). For validation four cases were selected, when the weather was clear
45
and calm in the time of the measurement and in the preceding days too, thus during these nights the
46
weather conditions promoted the surface influence on the thermal conditions in the near-surface air
47
layer.
48
49
4. GIS methods developed for LCZ mapping
50
51
4.1. Parameter calculations for lot area polygons
52
4
1
Using our method we can determine seven properties from the ten geometric, surface cover
2
and radiative ones listed by Stewart & Oke (2012) for any given area inside the study area based on
3
the available databases. From the initial parameters for classification we omitted the H/W since this
4
ratio tends to be too theoretical, it can be clearly calculated just in the case of the regular street
5
network. The surface admittance and the anthropogenic heat output are also lacking, since these
6
data were not available in the study area.
7
During the determination processes of the other seven parameters the basic area of the
8
calculation was the building block and the area belonging to it (lot area polygon). The determination
9
of the building blocks and lot area polygons is based on the 3D building database of Szeged which
10
contains more than 22,000 individual buildings with building height information in ESRI shapefile
11
format (Gál & Unger 2009). Therefore, the buildings touching each other were merged into blocks
12
and then we divided the study area into polygon-shape areas based on these blocks where each
13
polygon consists of the set of points closer to a central building block than to the other blocks. In
14
the case of larger open areas (areas without buildings, e.g. parks, fields, water) the border of the
15
polygon at the edge of the built-up area is at a distance of 100 m from the nearest block (Fig. 2).
16
The calculation processes and the applied databases by parameters were as follows:
17
- SVF: The input was a SVF database with 5 m horizontal resolution originated from our
18
earlier studies (Gál et al. 2009, Unger 2009). It was calculated using the 3D building database of
19
Szeged with a vector based method. The building database contains building footprint areas as
20
polygon-type data, and the building heights which were measured with photogrammetric methods
21
as attributes of them. During the SVF calculation all of the buildings were regarded with flat roofs
22
and the effect of the vegetation was neglected. The SVF values are related to the street level so they
23
calculated for the points not covered by buildings and these values are averaged inside the lot
24
polygon areas.
25
- BSF: The input was also the 3D building database of Szeged which contains the buildings
26
footprints in the study area. BSF is the ratio between the sum of building footprint areas and the lot
27
polygon area.
28
- PSF: The input was a built-up dataset calculated from RapidEye satellite image (RapidEye
29
2012) using NDVI index, a 1:25000 topographic map, a road database and the Corine Land Cover
30
(CLC) (Bossard et al. 2000) database. The RapidEye image was atmospherically corrected
31
(resolution of 5.16 m) and the Normalized Difference Vegetation Index (NDVI) was calculated
32
using bands 3 and 5 (Tucker 1979) and those points were regarded as covered area where the NDVI
33
was below 0.3. The CLC dataset was used to locate the agricultural areas as these areas have small
34
NDVI (like the covered areas) because the amount of plants on them is negligible after harvest. As a
35
second correction the shape of water bodies were digitized from the topographic map because in
36
several cases the water has NDVI values very similar to the values of some building materials. As a
37
last correction the road database was used to locate the asphalt roads in the area because in the
38
urban canyons these roads are usually under tree cover and because the roads which slice
39
agricultural areas do not appear in CLC dataset.
40
- ISF: Its value was calculated using this formula: ISF = 1 – (BSF + PSF).
41
- HRE: The input was also the 3D building database of Szeged. For each lot area the building
42
heights weighted with their footprint areas were averaged.
43
- TRC: For describing the roughness the Davenport roughness classification method was used
44
(Davenport et al. 2000). The principle of the classification process is that the roughness parameter
45
(z
0
) and displacement height (z
d
) values of the studied area are approximately the same as values
46
previously measured on an area with similar surface cover. This widespread method comprises
47
eight classes of roughness. Each polygon was classified into a roughness class with visual
48
interpretation of aerial photographs, the topographical map and the building database.
49
- SA: As input the atmospherically corrected reflectance values of the 5 band (440–510 nm,
50
520–590 nm, 630–685 nm, 690–730 nm, 760–850 nm). RapidEye satellite image were used.
51
5
Broadband albedo was calculated as an average of reflectance values weighted with the integral of
1
the radiation within the spectral range of a given band (Starks et al. 1991, Tasumi et al. 2008).
2
The calculation processes, the necessary databases and the outputs are shown in the upper and
3
left hand parts of Fig. 3.
4
5
4.2. LCZ mapping – aggregation and generalization of lot area polygons
6
7
According to literature, in the urban environment the temperature value measured at a height
8
of 1.5–2 m is influenced by its surroundings with a radius of a few hundred meters as a source area
9
(Oke 2004, Unger 2010). Of course this is only a general approach as the source area depends on
10
the type of “urban” environment. If it is compact urban, the source area may only be tens of metres;
11
if it is open urban, it may be many hundreds of metres. It also depends on weather and stability
12
conditions (Oke 2004).
13
In line with this and the definition of LCZs, the lot area polygons classified into the same or
14
similar LCZ classes were merged into zones of hundreds of meters to several kilometers. In this
15
case, we meet the minimum condition that the measuring site representing an LCZ is at least 250
16
meters from the boundaries of the zone, such that the relatively homogeneous surroundings of the
17
sites constitute a source area with a radius of 250 m or greater.
18
In order to get LCZ areas with appropriate size, the lot area polygons were aggregated into
19
groups by the following procedure:
20
First, the polygons were classified separately.
21
(1) From the obtained surface parameters areal mean or percentage values were calculated to
22
represent the polygons. Seven scores were assigned (Fig. 4) to each LCZ categories by polygons
23
according to its fit into the typical ranges given by Stewart & Oke (2012) and then they were added.
24
Two of the best fitting LCZ categories were assigned to every polygon (for each polygon the best is
25
LCZ
x
and the second best is LCZ
y
), if its scores were high enough (>3.0). In the case where the
26
scores were too low to fit to any LCZ categories then the polygon was considered as unclassified.
27
Second, the lot area polygons were merged according to their LCZ category and their location
28
related to each other.
29
(2) If a lot area polygon was located inside another polygon then the first LCZ class of the
30
small polygon was set to the same as the other polygon.
31
(3) If all of the neighbors of a polygon (or maximum except one of them) were in the same
32
LCZ class then the class of the polygon was modified to the same as these neighbors.
33
(4) If a polygon did not have any neighbor in the same class there were two cases. In one case,
34
if there was a neighbor with same LCZ
x
like the polygon’s LCZ
y
or same LCZ
y
like the polygon’s
35
LCZ
x
, then LCZ
x
of the polygon was set to the same like its neighbor. In the other case, if there was
36
a neighbor with LCZ
x
category similar to the polygon’s LCZ
x
category then the LCZ
x
of the
37
polygon was modified to the LCZ
x
of the neighbor. Table 3 presents the similarity of the LCZ
38
categories: cross (+) indicates the similarity of two LCZ categories in the upper row and the left
39
column, respectively (e.g. for the LCZ 2 “compact mid-rise” similar LCZs are LCZ 1 and LCZ 3
40
because of their density category (“compact”) is equal and they are different with only one height
41
categories, and LCZ 5 also similar as it has the same height category (“mid-rise”)).
42
(5) The LCZ categories of the remaining non classified and non aggregated polygons were
43
defined as the most frequent of the classes of their neighbors.
44
Third, the groups of adjacent polygons with a given LCZ category were investigated
45
according to their spatial extension.
46
(6) If the area of a group covers at least one circle with a radius of 250 m then it was regarded
47
as an independent LCZ area.
48
(7) Polygons of groups which did not satisfy the criterion for the size were merged without
49
considering their properties if they were adjacent. If the obtained group was large enough, the
50
category of the group was set to the most frequent category of its parts; else it was joined to one of
51
the adjoining LCZ areas which have the largest number of contacting lot area polygons with it.
52
