Determining the Influence of Long Term Urban Growth on Surface Urban Heat Islands Using Local Climate Zones and Intensity Analysis Techniques

Abstract

Urban growth, typified by conversion from natural to built-up impervious surfaces, is known to cause warming and associated adverse impacts. Local climate zones present a standardized technique for evaluating the implications of urban land use and surface changes on temperatures of the overlying atmosphere. In this study, long term changes in local climate zones of the Bulawayo metropolitan city were used to assess the influence of the city’s growth on its thermal characteristics. The zones were mapped using the World urban Database data, while the Access Portal Tool procedure was used to determine multi-temporal change. Data were divided into 1990 to 2005 and 2005 to 2020 temporal splits and intensity analysis used to characterize transformation patterns at each interval. Results indicated that growth of the built local climate zones (LCZ) in Bulawayo was faster in the 1990 to 2005 interval than the 2005 to 2020. Transition level intensity analysis showed that growth of built local climate zones was more prevalent in areas with water, low plants and dense forest LCZ in both intervals. There was a westward growth of light weight low rise built LCZ category than eastern direction, which could be attributed to high land value in the latter. Low plants land cover type experienced a large expansion of light weight low rise buildings than the compact low rise, water, and open low-rise areas. The reduction of dense forest was mainly linked to active expansion of low plants in the 2005 to 2020 interval, symbolizing increased deforestation and vegetation clearance. In Bulawayo’s growth, areas where built-up LCZs invade vegetation and wetlands have increased anthropogenic warming (i.e., Surface Urban Heat Island intensities) in the city. This study demonstrates the value of LCZs in among others creating a global urban land use land cover database and assessing the influence of urban growth pattern on urban thermal characteristics.

Full text

Citation: Mushore, T.D.;
Mutanga, O.; Odindi, J. Determining
the Influence of Long Term Urban
Growth on Surface Urban Heat
Islands Using Local Climate Zones
and Intensity Analysis Techniques.
Remote Sens. 2022,14, 2060. https://
doi.org/10.3390/rs14092060
Academic Editor: Xuecao Li
Received: 28 January 2022
Accepted: 18 April 2022
Published: 25 April 2022
Publisher’s Note: MDPI stays neutral
with regard to jurisdictional claims in
published maps and institutional affil-
iations.
Copyright: © 2022 by the authors.
Licensee MDPI, Basel, Switzerland.
This article is an open access article
distributed under the terms and
conditions of the Creative Commons
Attribution (CC BY) license (https://
creativecommons.org/licenses/by/
4.0/).
remote sensing
Article
Determining the Influence of Long Term Urban Growth on
Surface Urban Heat Islands Using Local Climate Zones and
Intensity Analysis Techniques
Terence Darlington Mushore 1, 2, *, Onisimo Mutanga 1and John Odindi 1
1Discipline of Geography, School of Agricultural, Earth and Environmental Sciences, University of
KwaZulu-Natal, Scottsville, Pietermaritzburg 3209, South Africa; mutangao@ukzn.ac.za (O.M.);
odindi@ukzn.ac.za (J.O.)
2Department of Space Science and Applied Physics, Faculty of Science, University of Zimbabwe,
630 Churchill Avenue, Mt Pleasant, Harare 00263, Zimbabwe
*Correspondence: tdmushore@science.uz.ac.zw
Abstract:
Urban growth, typified by conversion from natural to built-up impervious surfaces, is
known to cause warming and associated adverse impacts. Local climate zones present a standardized
technique for evaluating the implications of urban land use and surface changes on temperatures of
the overlying atmosphere. In this study, long term changes in local climate zones of the Bulawayo
metropolitan city were used to assess the influence of the city’s growth on its thermal characteris-
tics. The zones were mapped using the World Urban Database and Access Portal Tool (WUDAPT)
procedure while Landsat data were used to determine temporal changes. Data were divided into
1990 to 2005 and 2005 to 2020 temporal splits and intensity analysis used to characterize transforma-
tion patterns at each interval. Results indicated that growth of the built local climate zones (LCZ)
in Bulawayo was faster in the 1990 to 2005 interval than the 2005 to 2020. Transition level intensity
analysis showed that growth of built local climate zones was more prevalent in areas with water, low
plants and dense forest LCZ in both intervals. There was a westward growth of light weight low rise
built LCZ category than eastern direction, which could be attributed to high land value in the latter.
Low plants land cover type experienced a large expansion of light weight low rise buildings than the
compact low rise, water, and open low-rise areas. The reduction of dense forest was mainly linked
to active expansion of low plants in the 2005 to 2020 interval, symbolizing increased deforestation
and vegetation clearance. In Bulawayo’s growth, areas where built-up LCZs invade vegetation and
wetlands have increased anthropogenic warming (i.e., Surface Urban Heat Island intensities) in the
city. This study demonstrates the value of LCZs in among others creating a global urban land use land
cover database and assessing the influence of urban growth pattern on urban thermal characteristics.
Keywords:
WUDAPT; thermal environment; urban climate; local climate zones; intensity analysis;
urban growth
1. Introduction
Urban areas continue to expand in population and built-up extent, with faster rates
in developing countries [
1
–
6
]. Whereas urban growth is often considered as a sign of
economic vitality [
3
], its adverse impacts that include increased air pollution, Surface
Urban Heat Islands, dust and haze, significantly influence urban micro- and macro-climate
and affect urban environmental quality and human health [
7
,
8
]. Globally, urbanization
has caused climate modifications, most evident in higher temperatures in urbanized than
the non-urbanized surroundings [
9
–
13
]. Such growth often exacerbates heat stress in the
already warming (due to global climate change) urban areas, leading to deterioration of
outdoor thermal comfort [
14
]. Urban growth and associated surface changes induce near-
surface warming, which increases energy and water demand due to search for indoor and
Remote Sens. 2022,14, 2060. https://doi.org/10.3390/rs14092060 https://www.mdpi.com/journal/remotesensing

Remote Sens. 2022,14, 2060 2 of 22
outdoor thermal comfort [
15
–
18
]. Hence, urban growth assessment techniques that consider
both surface and near surface characteristics and atmospheric gas/pollutant emissions are
valuable in determining the influence of anthropogenic processes on the urban thermal
emissions to inform sustainable urban growth.
Remote sensing offers a variety of data for analyzing spatial and temporal effects
of urban land surface changes on the urban thermal environment. This has largely been
facilitated by advances in sensor development that has improved the quality and avail-
ability of remotely sensed data. For instance, missions such as Landsat offer large archival
data spanning as far back as 1972 at reasonable and improved radiometric, spectral and
spatial resolution, valuable for assessing both large scale and localized temporal and multi-
temporal landscape and environmental patterns [
19
–
23
]. A large body of literature has
investigated the impact of land use land cover (LULCs) changes on surface and near surface
temperature, e.g., Kumar and Shekar [
24
] and Uddin et al. [
25
]. These studies have indi-
cated that impervious, bare and built-up surfaces result in urban warming while vegetation
and water bodies act as thermal sinks. However, whereas LULC-based techniques and
surface characterization schemes account for the contributions of land surface changes to
temperature dynamics, they often exclude other anthropogenic contributions such as emis-
sion, which are key drivers of temperature changes associated with urban growth. Hence,
schemes that account for both land surface characteristics and gas emissions/pollutants as
drivers of changes in the thermal environment are necessary to adequately explain climatic
changes in such complex environments.
Due to the dependence on LULC schemes, studies on effects of urbanization on
temperature have mostly defined Surface Urban Heat Island (SUHI) as the difference
between “urban” and “rural” temperature [
26
–
31
]. In such studies, rural and urban are
vaguely defined by differences in population and built-up extent in a manner that is not
universal [32–34]
. However, this separation is no longer always clear cut as traditional and
non-traditional urban and rural land uses increasingly continue to coexist [
33
]. The tradi-
tional classification scheme is area specific, making it difficult to make global comparisons
as LULC characteristics vary between cities of the same country and between countries.
However, analysis based on Local Climate Zones (LCZ) provide understanding of urban
structures and land uses in a globally standardized manner, useful for understanding the
influence of urbanization on urban climate [
33
–
35
]. The LCZ scheme is local and climatic in
nature considering surface cover, three dimensional surface structures (such as height and
density of buildings and vegetation) as well as anthropogenic thermal emissions [
8
,
35
–
39
].
LCZs provide useful information for assessing adherence of cities to the 2030 agenda for
sustainable development Goal 11 [
40
] both in the form of LULC transitions and anthro-
pogenic emissions effects on climate. Hence, LCZ provide a standard basis upon which the
impacts of urban growth on the thermal environment can be monitored and assessed.
The LCZ scheme emphasizes the difference in temperature among the categories
within and between cities, thus directly linked to climate of an area, while contributing to
the creation of the global urban database [
11
,
32
]. Close association between LCZ and LST
shows that LCZs are helpful for examination of evolution of SUHI over time [
41
]. LCZs are
more conducive to analysis and less prone to confusion because they highlight common ex-
posure characteristics and invite physically based explanations of SUHI
magnitude [42–49]
.
Studies which used LCZs in SUHI analysis mostly focused on short temporal scales
such as diurnal, seasonal and annual, e.g., United Nations General
Assembly [40]
and
Ardiyansyah et al. [49]
. Focus on long term interactions between LCZs and SUHI have
remained understudied, especially in Africa. Furthermore, although Stewart and Oke [
33
]
showed the effectiveness of LCZs in defining SUHI in cities, application of inter-LCZ
temperature difference to define SUHI has remained minimal, especially in the analysis of
long-term changes. Most of the studies that analyzed the relationship between LCZs and
SUHI either used LST to directly represent SUHI [
33
,
46
,
49
], reclassified LST into different
SUHI categories [
45
], converted LST into other forms such as Distribution Index [
41
] or
normalized temperatures [
47
,
50
], or used urban to rural temperature
gradient [29,48,51,52]
.

Remote Sens. 2022,14, 2060 3 of 22
Of the few studies that used temperature difference between LCZ types to quantify SUHI,
most of them used the Low plants LCZ as a reference against which LSTs of other LCZs
were compared [
50
,
53
–
55
]. A number of studies on LCZ have mainly focused on the current
state and short-term variations as well as contributing to World Urban Database and Access
Portal Tool (WUDAPT), e.g., Cai et al. [
32
], Cai et al. [
36
],
Danylo et al. [37]
,
Qiu et al. [39]
,
and Demuzere et al. [
56
], with little effort towards understanding and explaining their long
term changes that affect a city’s SUHI intensity. In long term analysis, single date imageries
are commonly used to develop static LCZ for each year, ignoring the value of combining
multi-seasonal data for the same analysis. Long term changes provide a better under-
standing of the contribution of human activities to local climate change and assessment of
adherence of city growth patterns to Agenda 2030 for Sustainable Development Goal 11 [
39
].
Hence, there is a need to use multi-season imageries to generate LCZ and LCZ-based SUHI
to determine long term effects of urban growth on a city’s thermal environment.
Cities in developing countries have been experiencing growth characterized by mas-
sive changes in land surface characteristics and intensification of activities, which have the
potential to pollute the atmosphere and exacerbate changes in local thermal environments.
Although the trends have been observed in different parts of the world, actual changes
vary between and within countries, triggering the need for detailed and city-specific assess-
ments. Hence, in-depth understanding of a city’s specific influences on LCZ is important
for ensuring that further development is climate smart and sustainable. However, available
literature on long term LCZ changes [
52
,
57
–
59
] uses the traditional “from to” change de-
tection approach which lacks in depth analysis to provide detailed understanding of long
term LCZ transitions and their potential impacts on local climate. Although not yet applied
to understand long term LCZ transitions, in depth analysis of changes based on transition
matrices of different periods is better done using intensity analysis than the traditional
“from to” change detection approach. Intensity analysis is useful for effecting classifica-
tions of different time intervals to understand sizes and intensities of temporal changes
among categories [
60
–
63
], as it provides details on whether transition from one category to
another deviates from a uniform process [
61
]. It also identifies time intervals when rate
of change was fast or slow, identifies whether category changes were active or dormant
in a time interval and whether a category was targeted or avoided by changes during an
interval [62,64,65]
. Furthermore, it analyzes land changes relative to size of category to
identify systematic transitions over time [
62
]. As such, it reveals information such as un-
derlying processes associated with changes which ordinary change detection
conceals [66]
.
For instance, Alo and Pontius [
67
] revealed that protected areas in Ghana experienced
systematic transitions from closed forest to bare and bush fires, while
Ekumah et al. [66]
revealed that between 1985 and 2017, human induced LULCs grew at the expense of natu-
ral categories in the Densu Delta, Sukumo II and Muni Pomodze Ramssar sites in Ghana.
Intensity analysis is thus valuable for obtaining an in-depth and detailed analysis and un-
derstanding of long term LCZ changes, especially for cities such as Bulawayo where spatial
and temporal temperature patterns are not yet documented. Combining intensity analysis
and SUHI retrieval in the context of LCZ will therefore provide a detailed understanding
of the effect of urban growth patterns on the thermal environment of cities.
Hence, the aim of this study was to integrate intensity analysis and LCZ-based SUHI
retrieval to provide detailed analysis of the impact of urban growth on the thermal en-
vironment in Bulawayo metropolitan city in Zimbabwe. Specifically, this study sought
to determine long term effect of urban growth on the thermal environment using LCZ
between 1990 and 2020 in Bulawayo city, Zimbabwe. The study also utilized intensity
analysis for an in-depth assessment of the changes in LCZ in Bulawayo between 1990 and
2020. Additionally, and contrary to the broad literature that uses the between rural–urban
difference in temperature, this study enhanced the use of inter-LCZ temperature difference
approach to quantify SUHI intensity and their long term changes.

Remote Sens. 2022,14, 2060 4 of 22
2. Methodology
2.1. Description of the Study Area
Bulawayo is the second largest city in Zimbabwe (Figure 1). It is located to the
southeast of the country (Figure 1a) at an elevation of approximately 1358 m above sea
level. The period between October and March is hot and wet with a minimum of 16
◦
C
and maximum of 30
◦
C, with an average temperature of 25
◦
C, while the rest of the year is
cool and dry, with a minimum of 10
◦
C, maximum of 25
◦
C and average temperature of
15 ◦C [68]
. Generally, the area receives erratic rainfall, with annual average precipitation of
600 mm that ranges from 199.3 mm to 1258.8 mm, typical of a semi-arid climate. Bulawayo
(Figure 1b) lies in the subtropical steppe (Bsh) according to Koppen climate classification.
The period from December to February is the wettest. Most rain falls from December
to February and the area is vulnerable to droughts due to proximity to the Kalahari
Desert [69,70].
Remote Sens. 2022, 14, x FOR PEER REVIEW 4 of 22
2020. Additionally, and contrary to the broad literature that uses the between rural–urban
difference in temperature, this study enhanced the use of inter-LCZ temperature differ-
ence approach to quantify SUHI intensity and their long term changes.
2. Methodology
2.1. Description of the Study Area
Bulawayo is the second largest city in Zimbabwe (Figure 1). It is located to the south-
east of the country (Figure 1a) at an elevation of approximately 1358 m above sea level.
The period between October and March is hot and wet with a minimum of 16 °C and
maximum of 30
°C, with an average temperature of 25 °C, while the rest of the year is cool
and dry, with a minimum of 10 °C, maximum of 25 °C and average temperature of 15 °C
[68]. Generally, the area receives erratic rainfall, with annual average precipitation of 600
mm that ranges from 199.3 mm to 1258.8 mm, typical of a semi-arid climate. Bulawayo
(Figure 1b) lies in the subtropical steppe (Bsh) according to Koppen climate classification.
The period from December to February is the wettest. Most rain falls from December to
February and the area is vulnerable to droughts due to proximity to the Kalahari Desert
[69,70].
Figure 1. Map of Africa showing the location of Zimbabwe and Bulawayo (a) and map of Bulawayo
showing distribution of training areas (b)—training sites not visible.
2.2. Field Observations of Local Climate Zones in Bulawayo
Since the WUDAPT places little emphasis on field data collection, a survey to identify
and obtain ground truth samples of LCZ categories in the study area was important to
guide digitizing of training polygons in Google Earth Field observations, which allowed
for identification of inter- and intra-category variabilities that could not be adequately
captured from Google Earth. The sample coordinates of each LCZ category (ground truth
data) were obtained between 18 and 27 October in 2020. This experience also guided se-
lection of training areas for the historical periods using Google Earth in the absence of
field measurements for that period. Field observations increased the validity of the anal-
ysis instead of exclusive reliance on Google Earth retrievals. Generally, 8 LCZ categories
were identified in the study area that fit into the description of LCZs provided by Stewart
and Oke [33]. The categories were three land use-based LCZs, namely Compact low rise
(LCZ3), Open low rise (LCZ6) and Light weight low rise (LCZ7), as well as three land
cover-based LCZs, which were Dense forest (LCZA), Low plants (LCZD) and Water (LCZ
Figure 1.
Map of Africa showing the location of Zimbabwe and Bulawayo (
a
) and map of Bulawayo
showing distribution of training areas (b)—training sites not visible.
2.2. Field Observations of Local Climate Zones in Bulawayo
Since the WUDAPT places little emphasis on field data collection, a survey to identify
and obtain ground truth samples of LCZ categories in the study area was important to
guide digitizing of training polygons in Google Earth Field observations, which allowed
for identification of inter- and intra-category variabilities that could not be adequately
captured from Google Earth. The sample coordinates of each LCZ category (ground truth
data) were obtained between 18 and 27 October in 2020. This experience also guided
selection of training areas for the historical periods using Google Earth in the absence of
field measurements for that period. Field observations increased the validity of the analysis
instead of exclusive reliance on Google Earth retrievals. Generally, 8 LCZ categories were
identified in the study area that fit into the description of LCZs provided by Stewart and
Oke [
33
]. The categories were three land use-based LCZs, namely Compact low rise (LCZ3),
Open low rise (LCZ6) and Light weight low rise (LCZ7), as well as three land cover-based
LCZs, which were Dense forest (LCZA), Low plants (LCZD) and Water (LCZ G). The study
used LCZs definitions and pictorials provided by Stewart and Oke [
33
] as reference to
identify similar classes in the study area for global comparability.

Remote Sens. 2022,14, 2060 5 of 22
2.3. Multi-Temporal Remotely Sensed Datasets
Multi-temporal Landsat 5, Landsat 7 and Landsat 8 Operational Land Imager datasets
were downloaded from the United States Geological Survey’s (USGS) earth explorer website
for analysis. Cloud free imageries for the wet and dry vegetation periods were downloaded
to minimize compromising effects of atmospheric noise on image radiometric values and
LCZ mapping accuracy. Table 1shows the imageries used for 1990, 2005 and 2020. In this
study, Landsat thermal, panchromatic as well as bands for monitoring cirrus clouds and
coastal aerosols were not used for analysis. For each year, dry and wet periods were selected
in order to capture seasonal variations in LCZ, especially in areas with vegetation. The post
rain period was chosen to represent the wet biomass period because during that period,
trees and grasses are vibrant after a rainy season. The rainy season was avoided to attain
both temporal and multi-temporal cloud free imagery. Amorim [
71
] indicated that Surface
Urban Heat Island intensity is influenced by responses of vegetation to rainfall patterns
before the imagery date. Amorim [
71
] observed that Heat Island Intensities increased
during periods of high biomass, which reduce temperatures in vegetation areas. Therefore,
precipitation patterns of up to 10 days prior to overpass (amounts and rainy days) are
shown in Table 1. Generally, the number of rainy days prior to overpass was higher in
the post-rain than other seasons, with the lowest number in the cool season. Cumulative
rainfall amounts in 10 days before overpass were also low in all seasons (less than 20 mm).
Table 1. Multi-temporal and multi-spectral remote sensing imagery used in the study.
Imagery Date Season
Days to Recent
Precipitation before
Overpass (Days)
Rainy Days in 10 Days
before Overpass (Days)
Precipitation in 10
before Overpass (mm)
Landsat 5 27 April 1990 Post rain 1.0 1.0 4.8
Landsat 7 12 April 2005 Post rain 4.0 6.06 13.1
Landsat 7 21 April 2020 Post rain 1.0 10.0 16.9
Landsat 5 14 June 1990 Cool 27.0 0.0 10.0
Landsat 5 7 June 2005 Cool 22.0 0.0 0.0
Landsat 7 24 June 2020 Cool 2.0 4.0 1.4
Landsat 5 20 October 1990 Hot 1.0 1.0 7.0
Landsat 7 21 October 2005 Hot 99.0 0.0 0.0
Landsat 8 OLI 15 October 2020 Hot 3.0 5.0 6.5
2.4. Mapping of LCZ Using Dry and Wet Season Imagery
The advantage of the LCZ scheme is that their mapping follows an easy and standard-
ized approach for mapping LCZ, which involves downloading of imagery, digitizing of
training sites (for classification and accuracy assessment) on Google Earth and supervised
classification using the random forest (RF) classifier [
37
,
72
,
73
]. Local Climate Zones for
Bulawayo were thus mapped following the WUDAPT L0 procedure [
8
,
36
,
56
,
73
,
74
]. The
procedure involves downloading suitable imagery of the study area, on-screen selection
and digitizing of training areas on Google Earth, supervised image classification using the
Random Forest (RF) Classifier and post classification accuracy assessment in SAGA GIS.
The procedure was adopted due to its use of readily available and freely downloadable im-
ageries as well as easy access to the SAGA GIS software for implementation. Additionally,
the steps have been followed in different parts of the world, making it very easy to follow.
Use of the RF classifier makes the procedure attractive as it can perform bootstrapping
analysis which is used to assess quality of the LCZ database [
56
,
75
]. The RF model is a
collection of decision trees and each tree is made up of a subset of training dataset for a
subset of predictors [
73
,
76
,
77
]. It is termed RF because its subsets are randomly formed. In
RF classification, the predicted value is the mode of the predictions from all trees. Main
advantages of RF are the use of both categorical and numerical values, the evaluation of
the precision of prediction, the robustness in the presence of outliers, noise, and overfitting.
The RF model can quantify the contribution of each predictor to the total spatial variability

Remote Sens. 2022,14, 2060 6 of 22
of the target and assigns a variable importance score to each predictor. Random forest
requires a small amount of training data, yet provides competitive results and can handle a
large volume of input data without deletion, while still capable of identifying important
variables for classification [
78
–
83
]. In addition, RF is not sensitive to over-training or noise
and is desirable for multi-source remote sensing and geographical information systems
data [
75
,
81
]. LCZ maps were produced for the years 1990, 2005 and 2020 using the same
training areas. These were collected from locations whose LULC category did not change
over time in order to eliminate the effect of differences in ground truth data on mapping
accuracy. The use of the same training areas was made possible by availability of historical
Google Earth imagery where the areas could be clearly identified at different periods. In
order to adhere to the definition of LCZ, which requires that they cover at least a hundred
meters to several kilometers to influence temperature [
8
,
11
,
41
], the maps were resampled
using a 5 by 5 pixels window. A LCZ must be large enough to influence temperature of
an area. In order to quantify the effect of seasonality on mapping accuracy, LCZ maps
were also produced using data for the hot and dry season for comparison with analysis
based on a combination of data from the post rain, cool dry and hot dry periods. Figure 2
provides a summary of the procedure followed. The broken arrow in the figure shows
the approach taken by previous studies which largely skip the iterative step of qualitative
accuracy assessment, further improving accuracy before change detection.
Remote Sens. 2022, 14, x FOR PEER REVIEW 6 of 22
[56,75]. The RF model is a collection of decision trees and each tree is made up of a subset
of training dataset for a subset of predictors [73,76,77]. It is termed RF because its subsets
are randomly formed. In RF classification, the predicted value is the mode of the predic-
tions from all trees. Main advantages of RF are the use of both categorical and numerical
values, the evaluation of the precision of prediction, the robustness in the presence of out-
liers, noise, and overfitting. The RF model can quantify the contribution of each predictor
to the total spatial variability of the target and assigns a variable importance score to each
predictor.
Random forest requires a small amount of training data, yet provides competi-
tive results and can handle a large volume of input data without deletion, while still ca-
pable of identifying important variables for classification [78–83]. In addition, RF is not
sensitive to over-training or noise and is desirable for multi-source remote sensing and
geographical information systems data [75,81]. LCZ maps were produced for the years
1990, 2005 and 2020 using the same training areas. These were collected from locations
whose LULC category did not change over time in order to eliminate the effect of differ-
ences in ground truth data on mapping accuracy. The use of the same training areas was
made possible by availability of historical Google Earth imagery where the areas could be
clearly identified at different periods. In order to adhere to the definition of LCZ, which
requires that they cover at least a hundred meters to several kilometers to influence tem-
perature [8,11,41], the maps were resampled using a 5 by 5 pixels window. A LCZ must
be large enough to influence temperature of an area. In order to quantify the effect of
seasonality on mapping accuracy, LCZ maps were also produced using data for the hot
and dry season for comparison with analysis based on a combination of data from the
post rain, cool dry and hot dry periods. Figure 2 provides a summary of the procedure
followed. The broken arrow in the figure shows the approach taken by previous studies
which largely skip the iterative step of qualitative accuracy assessment, further improving
accuracy before change detection.
Figure 2. Flowchart showing summary of procedures used in this study.
2.5. Accuracy Assessment
The WUDAPT procedure automatically splits training areas into 50% for classification
and 50% for accuracy assessment. A confusion matrix is formulated in a tabular form for the

Remote Sens. 2022,14, 2060 7 of 22
purposes of comparing reference class labels (ground truth) with labels for corresponding
pixels on a derived classification of remote sensing imagery [
83
]. The diagonal values on
the matrix indicate where categories assigned on the classified map correctly corresponded
with ground truth. For example, the matrix shows the number of pixels which were
assigned same LCZ value as observed on the ground as well as those that are misallocated
following classification of remote sensing data. The confusion matrix was used to generate
indicators of accuracy at class level (Producer Accuracy and User Accuracy) as well as
at entire study area level (Overall Accuracy—OA—and Kappa—K). The use of OA and
K with the inclusion of ground data is the most common and reliable way of assessing
accuracy [
84
]. Accuracy assessment is important for the separation of real changes from
changes due to errors for LCZ maps of different periods. Error analysis was done for all
the study years. Accuracy was also assessed qualitatively by overlaying the produced LCZ
maps with a corresponding Google Earth imagery in combination with expert judgement.
The number of training areas was objectively and iteratively increased in areas where
marked mismatches were observed to capture for intra- and inter-class variabilities using
Google Earth.
2.6. Detection of Long-Term Changes in LCZ in Bulawayo
A 30-year period (1990 to 2020) was selected, as the study aimed at using LCZ dynam-
ics as a proxy for temperature changes in Bulawayo. The World Meteorology Organization
(WMO) recommends a minimum of 30 years for a representative climate change analysis.
The period was further split into two 15 year periods (i.e., 1990 to 2005 and 2005 to 2020) for
understanding of rapid changes that occur at local scale and for intensity analysis purposes.
Additionally, Coppin and Bauer [
85
] recommended a period of at least 3 years for change
detection involving forests and other vegetation types. In an urban setting, there is a mix
of rapid and slow LCZ making a period of at least 15 years enough to detect effect of all
change trajectories on the climate of an area. A post classification change detection ap-
proach was used. The approach produces a change matrix/table which shows the number
of pixels which were converted to other LCZ types or remained in the same categories in
the considered interval. For instance, the table indicated the number of pixels which were
in LCZ1 at the beginning and remained in the same as well as those that were changed to
other LCZ categories during the same time interval. Although the change matrix is useful
in depicting the quantities and directions of change, it does not adequately explain the
changes [60,85,86], hence the need for intensity analysis.
2.7. Intensity Analysis for In-Depth Characterization of LCZ Changes
Intensity analysis was used to obtain a better understanding of LCZ transitions be-
tween 1990 and 2020 in Bulawayo. It was used to assess locations and intensities of
temporal changes among categories. The analysis was done at the interval, category and
transition levels [
87
–
90
] using freely available software on Pontius Clarke University web
page (https://www2.clarku.edu/faculty/rpontius/ (accessed on 15 April 2021)). The site
provides an easy to use Excel sheet where the change matrix for a given time interval is
entered. Varga et al. [
91
] provide descriptions and defining equations used in the analysis
that were adopted in this study.
2.7.1. Interval Level Intensity Analysis
Interval level was used to determine overall changes per time period for the 1990 to
2005 and 2005 to 2020 time intervals. Overall changes in the interval 1990 to 2005 were
compared with those for the interval 2005 and 2020. This was important to identify which
interval had changes characterized as fast or slow. The change percentage for an interval t
is defined as in Equation (1) [91]:
St=(size o f changes during interval t)×100%
size o f studyarea in which changes are occurring (1)

Remote Sens. 2022,14, 2060 8 of 22
S
t
is the uniform intensity in interval t. An interval change is fast if it exceeds uniform
intensity and slow if otherwise.
2.7.2. Category Level Analysis
Changes in gross loss or gain in intensity among different categories was described
by category level intensity analysis during time interval t(where tseparately represents
the two interval periods—1990 to 2005 and 2005 to 2020). Category level loss (L
ti
) and gain
(Gtj) during the interval tare obtained using Equations (2) and (3) as
Lti =(size o f lo ss o f categ ory i durin g in terval t)×100
size o f i at the start o f interval t (2)
Gtj =(size o f gain o f category j during interval t)×100
size o f category j at the end o f interval t (3)
According to [
64
], calculation of category persistence (P
ti
) is done using Equation (4):
Pti =(size that has maintained category i during interval)×100%
size o f spatia l e xtent (4)
St=Lti =Gtj
for all categories iand jif changes are uniformly distributed across
spatial extent. Uniform transition assumes category iuniformly changes to other categories
during the time interval. If
Lit >St
then the loss of category iis active in the interval t
while loss is dormant if
Lit <St
. Dormant implies that the loss of category islowed down
or stopped within the interval t. Similarly, gain of category jin the interval tis active if
Gtj >St
and dormant if
Gtj >St
. When two intervals are considered and the status of a
change as dormant or active is same in both intervals, then the category’s loss or gain is
said to be stationary.
2.7.3. Transition Level Analysis
The analysis describes the variation in intensity with which the gain of a particular
category transitions from other categories within a time interval [
60
]. Transition level
intensity analysis was used to determine whether change of a category avoids or targets
other categories. If the intensity of the change from category ito category jexceeds uniform
intensity, then category itargets jotherwise it avoids.
2.7.4. Retrieval of Changes in SUHI in Response to Long Term LCZ Dynamics
Thermal data of Landsat 5, 7 and 8 were used to compute land surface temperature
(LST) for 1990, 2005 and 2020. Initially, the data were corrected of differences in solar zenith
angles. While Landsat 8 has two thermal infrared bands, a single channel technique was
applied for all the periods to minimize effects of differences in computation algorithms
on LST variations between time periods. Digital numbers of thermal data were converted
to radiances, which were then used to determine brightness temperature (T
b
) and surface
temperature (Ts) using Equations (4) and (6), respectively [91–94].
Tb=K2
lnK1
Lλ+1(5)
where K
1
takes a values of 607.76, 666.09 and 774.89 W/(m
2
sr
µ
m), while K
2
has values of
1260.56, 1282.71 and 1321.08 W/(m
2
sr
µ
m), using Landsat 5, Landsat 7 and Landsat 8 data,
respectively. A method based on spectral and blackbody radiance of the thermal infrared
band was used to obtain pixel-based land surface emissivity map (
ε
) [
95
]. Emissivity

Remote Sens. 2022,14, 2060 9 of 22
correction was applied on brightness temperature to obtain actual land surface temperature
using Equation (6) [96].
Ts=TB
1+λTB
ρln ε
(6)
where λis the central wavelength of emitted thermal radiance (11.5 µm for Landsat 5 and
Landsat 7 and 10.9
µ
m for band 10 of Landsat 8) and
ρ
is equal to 1.438
×
10
−2
mK. The
procedure above was used to retrieve LST on two different dates so that independent sets
were used for training and accuracy assessment of the developed estimation algorithm.
The spatial structure of LST intensities were used for visual and quantitative analysis of
changes which occurred between 1990 and 2020.
Stewart and Oke [
33
] defined SUHI as the difference in LST between LCZs. In this
study, we adopted an approach by Dimitrov [
97
], which defined SUHI as the maximum
temperature difference between LCZs. For each year, the Surface Urban Heat Island was
computed as the difference between the average T
s
of LCZ category and the mean surface
temperature of the water LCZ (LCZ G), which was identified as giving the highest LST
difference with other LCZs consistently in all years using Equation (7).
SUHILCZ =LSTLCZ X −LSTLCZ Y (7)
SUHI
LCZ
is the SUHI derived from LST difference between other LCZs (X) and the
Water LCZ (Y). Although studies such as Stewart and Oke [
33
] used LCZ D as reference,
it was not applicable in this study due to the varying and opposing effects of the LCZ
in different places and seasons for daytime analysis. For instance, agricultural areas had
thermal values that vary in space and time and between seasons, rendering them as heat
sinks in some instances and heat sources in others. As such, during the growing season,
they acted as heat sinks while in the dry season, their heat mitigation value was reduced. In
other areas, they were completely removed, as they turned to dry biomass or bare soil areas.
Similarly, grasslands (also in LCZ D) vary in heat mitigation value depending on season
and maintenance, making them another example of inconsistency of LCZ D. The LCZ G
was chosen as a reference since it was the coolest and more stable than the vegetation based
categories, which have broad seasonal and long term variations in characteristics. Mean
LST per LCZ category was obtained using the Zonal Statistics overlay function in ArcGIS
version 10.2 for each year. Changes in SUHI per LCZ strata were monitored and linked
with observed changes in LCZ from 1990 to 2020 in 15-year intervals.
3. Results
3.1. LCZ Maps Based on Multi-Seasonal Image Analysis
The use of imagery for the wet and dry vegetation periods reduced the confusion
between light weight low rise and low plants in the western areas (Figure 3). The overall
classification accuracies were 98%, 98.2% and 95%, for 1990, 2005 and 2020, respectively.
Visual inspection shows that between 1990 and 2020, light weight low rise LCZ was
spread westwards into areas formerly occupied by low plants. All built local climate zones
increased in spatial coverage while low plants and dense forest LCZ decreased in coverage.
The water LCZ also decreased in coverage during the study period.
Compact low rise increased by 4.33 km
2
between 1990 and 2005 and by 2.50 km
2
between 2005 and 2020 (Table 2). Similarly, light weight low rise expanded by 21.27 km
2
between 1990 and 2005 and by 14.25 km
2
between 2005 and 2020. The open low rise LCZ
also showed the same pattern of larger increase in the 1990 to 2005 interval than in the
2005 to 2020 interval. On the other hand, the expansion rate of dense forest between 1990
and 2005 (2.91 km
2
in 15 years) was smaller than the depletion rate of the LCZ between
2005 and 2020 (15.25 km
2
in 15 years). Low coverage diminished faster in the 1990 to 2005
interval than in the 2005 to 2020 interval.

Remote Sens. 2022,14, 2060 10 of 22
Remote Sens. 2022, 14, x FOR PEER REVIEW 10 of 22
Figure 3. LCZ maps produced using multi-seasonal remotely sensed data.
Compact low rise increased by 4.33 km
2
between 1990 and 2005 and by 2.50 km
2
be-
tween 2005 and 2020 (Table 2). Similarly, light weight low rise expanded by 21.27 km
2
between 1990 and 2005 and by 14.25 km
2
between 2005 and 2020. The open low rise LCZ
also showed the same pattern of larger increase in the 1990 to 2005 interval than in the
2005 to 2020 interval. On the other hand, the expansion rate of dense forest between 1990
and 2005 (2.91 km
2
in 15 years) was smaller than the depletion rate of the LCZ between
2005 and 2020 (15.25 km
2
in 15 years). Low coverage diminished faster in the 1990 to 2005
interval than in the 2005 to 2020 interval.
Table 2. Coverage of LCZ categories in 1990, 2005 and 2020.
LCZ Category Coverage of LCZ Categories in km
2
(% in Bracket)
1990 2005 2020
Compact low rise 12.84 (3.0) 17.17 (4.0) 19.67 (4.5)
Dense Forest 38.41 (8.9) 41.32 (9.5) 26.07 (6.0)
Light weight low rise 39.60 (9.1) 60.87 (14.0) 75.12 (17.3)
Open low rise 45.13 (10.4) 74.00 (17.1) 88.56 (20.4)
Water 2.06 (0.5) 1.45 (0.3) 1.26 (0.3)
Low plants 295.42 (68.2) 238.66 (55.1) 222.77 (51.4)
3.2. Changes in LCZs for Bulawayo Using Multi-Temporal (Dry and Wet) Datasets
Table 3 shows that Compact low rise increased by 1.5% between 1990 and 2005 and
a further 2.2% between 2005 and 2020, giving a 30-year expansion of 3.7%. A significant
decrease in coverage was observed in the dense forest LCZ, which experienced a net re-
duction of 42% between 1990 and 2020, with greater change in the 1990 to 2005 than 2005
to 2020 periods. Generally, all built-up LCZ increased in coverage between 1990 and 2020,
with larger expansion in the light weight low rise than other built-up LCZs. Sustained
contraction was recorded in dense forest and water LCZs. The low plants LCZ, which in
Figure 3. LCZ maps produced using multi-seasonal remotely sensed data.
Table 2. Coverage of LCZ categories in 1990, 2005 and 2020.
LCZ Category Coverage of LCZ Categories in km2(% in Bracket)
1990 2005 2020
Compact low rise 12.84 (3.0) 17.17 (4.0) 19.67 (4.5)
Dense Forest 38.41 (8.9) 41.32 (9.5) 26.07 (6.0)
Light weight low rise 39.60 (9.1) 60.87 (14.0) 75.12 (17.3)
Open low rise 45.13 (10.4) 74.00 (17.1) 88.56 (20.4)
Water 2.06 (0.5) 1.45 (0.3) 1.26 (0.3)
Low plants 295.42 (68.2) 238.66 (55.1) 222.77 (51.4)
3.2. Changes in LCZs for Bulawayo Using Multi-Temporal (Dry and Wet) Datasets
Table 3shows that Compact low rise increased by 1.5% between 1990 and 2005 and
a further 2.2% between 2005 and 2020, giving a 30-year expansion of 3.7%. A significant
decrease in coverage was observed in the dense forest LCZ, which experienced a net
reduction of 42% between 1990 and 2020, with greater change in the 1990 to 2005 than 2005
to 2020 periods. Generally, all built-up LCZ increased in coverage between 1990 and 2020,
with larger expansion in the light weight low rise than other built-up LCZs. Sustained
contraction was recorded in dense forest and water LCZs. The low plants LCZ, which
in this study included croplands, grasslands and parks increased by 4.3% over the entire
period, except a 0.9% decrease recorded between 2005 and 2020.
3.3. Intensity Analysis
3.3.1. Category Level Intensity Analysis for 1990 to 2005 and 2005 to 2020 Intervals
All other LCZs except compact low rise were gainers or losers in the 1990 to 2005 and
2005 to 2020 intervals (Figure 4a,b). The gain in Compact low rise LCZ was dormant in
the 1990 to 2005 interval, implying the gain stopped or slowed along the interval. In the
interval 2005 to 2020, the gain of compact low rise was active in the 2005 to 2020 interval.

Remote Sens. 2022,14, 2060 11 of 22
Low plants were active losers in the 1990 to 2005 interval and became dormant losers (the
loss of the class was slow or stopped along the interval) in the 2005 to 2020. The water LCZ
was an active loser in both intervals. The light weight low rise LCZ was a dormant gainer
in the initial interval, but turned into an active gainer in the interval 2005 to 2020.
Table 3. LCZ changes from 1990 to 2020 in Bulawayo.
LCZ Category LCZ Category Changes in km2(% in Bracket)
1990 to 2005 2005 to 2020 1990 to 2020
Compact low rise 4.33 (33.7) 2.50 (14.6) 6.83 (53.2)
Dense Forest 2.91 (7.6) −15.26 (−37.0) −12.35 (−32.1)
Light weight low rise 21.27 (53.7) 14.27 (23.5) 35.54 (89.7)
Open low rise 28.88 (64.0) 14.50 (19.7) 43.43 (96.2)
Water −0.61 (−29.8) −0.19 (−13.2) −0.81 (−39.1)
Low plants −56.77 (−19.2) −15.89 (−6.7) −72.66 (−24.6)
Remote Sens. 2022, 14, x FOR PEER REVIEW 11 of 22
this study included croplands, grasslands and parks increased by 4.3% over the entire
period, except a 0.9% decrease recorded between 2005 and 2020.
Table 3. LCZ changes from 1990 to 2020 in Bulawayo.
LCZ Category LCZ Category Changes in km2 (% in Bracket)
1990 to 2005 2005 to 2020 1990 to 2020
Compact low rise 4.33 (33.7) 2.50 (14.6) 6.83 (53.2)
Dense Forest 2.91 (7.6) −15.26 (−37.0) −12.35 (−32.1)
Light weight low rise 21.27 (53.7) 14.27 (23.5) 35.54 (89.7)
Open low rise 28.88 (64.0) 14.50 (19.7) 43.43 (96.2)
Water −0.61 (−29.8) −0.19 (−13.2) −0.81 (−39.1)
Low plants −56.77 (−19.2) −15.89 (−6.7) −72.66 (−24.6)
3.3. Intensity Analysis
3.3.1. Category Level Intensity Analysis for 1990 to 2005 and 2005 to 2020 Intervals
All other LCZs except compact low rise were gainers or losers in the 1990 to 2005 and
2005 to 2020 intervals (Figure 4a,b). The gain in Compact low rise LCZ was dormant in
the 1990 to 2005 interval, implying the gain stopped or slowed along the interval. In the
interval 2005 to 2020, the gain of compact low rise was active in the 2005 to 2020 interval.
Low plants were active losers in the 1990 to 2005 interval and became dormant losers (the
loss of the class was slow or stopped along the interval) in the 2005 to 2020. The water
LCZ was an active loser in both intervals. The light weight low rise LCZ was a dormant
gainer in the initial interval, but turned into an active gainer in the interval 2005 to 2020.
Figure 4. Category level intensity analysis for 1990 to 2005 (a) and 2005 to 2020 (b).
3.3.2. Transition Intensity of Gaining Categories Encroaching into Losing Categories
The expansion of compact low rise LCZ targeted water, open low rise and light
weight low rise between 1990 and 2005 with open low rise LCZ as the most intensely
targeted (Figure 5a). The growth of compact low rise avoided low plants and vegetation
LCZs during the same period. In the 2005 to 2020 interval, the Compact low rise LCZ
continued to target open low rise and lightweight low rise areas (Figure 5b). The intensity
of transition of water to compact low rise was greater in the interval 1990 to 2005 than
2005 to 2020. In both intervals, the gain of compact low rise avoided low plants and dense
forest LCZ areas.
Figure 4. Category level intensity analysis for 1990 to 2005 (a) and 2005 to 2020 (b).
3.3.2. Transition Intensity of Gaining Categories Encroaching into Losing Categories
The expansion of compact low rise LCZ targeted water, open low rise and light weight
low rise between 1990 and 2005 with open low rise LCZ as the most intensely targeted
(Figure 5a). The growth of compact low rise avoided low plants and vegetation LCZs
during the same period. In the 2005 to 2020 interval, the Compact low rise LCZ continued
to target open low rise and lightweight low rise areas (Figure 5b). The intensity of transition
of water to compact low rise was greater in the interval 1990 to 2005 than 2005 to 2020. In
both intervals, the gain of compact low rise avoided low plants and dense forest LCZ areas.
The gain of light weight low rise between 1990 and 2005 targeted low plants and
avoided dense forests and open low rise water and compact low rise LCZ areas (Figure 6a).
Between 2005 and 2020, the gain of Light weight low rise LCZ continued to target low
plants while avoiding other LCZs (Figure 6b).
The gain of the dense forest LCZ in the 1990 to 2005 interval targeted low plants
(Figure 7a). This could imply growth of trees in grasslands such as parks in addition to
other tree planting efforts. The expansion of low plants LCZ targeted dense forest, open
low rise and light weight low rise, while it avoided water and compact low rise in the 2005
to 2020 interval (Figure 7b). The study also noticed slight spectral confusion between light
weight low rise and low plants, especially in the western parts of the study area.

Remote Sens. 2022,14, 2060 12 of 22
Remote Sens. 2022, 14, x FOR PEER REVIEW 12 of 22
Figure 5. Transition intensity given gain of Compact low rise from (a) 1990 to 2005 and (b) 2005 to
2020.
The gain of light weight low rise between 1990 and 2005 targeted low plants and
avoided dense forests and open low rise water and compact low rise LCZ areas (Figure
6a). Between 2005 and 2020, the gain of Light weight low rise LCZ continued to target low
plants while avoiding other LCZs (Figure 6b).
Figure 6. Transition intensity given the gain of Light weight low rise form (a) 1990 to 2005 and (b)
2005 to 2020.
The gain of the dense forest LCZ in the 1990 to 2005 interval targeted low plants (Fig-
ure 7a). This could imply growth of trees in grasslands such as parks in addition to other
tree planting efforts. The expansion of low plants LCZ targeted dense forest, open low rise
and light weight low rise, while it avoided water and compact low rise in the 2005 to 2020
interval (Figure 7b). The study also noticed slight spectral confusion between light weight
low rise and low plants, especially in the western parts of the study area.
Figure 5.
Transition intensity given gain of Compact low rise from (
a
) 1990 to 2005 and
(b) 2005
to 2020.
Remote Sens. 2022, 14, x FOR PEER REVIEW 12 of 22
Figure 5. Transition intensity given gain of Compact low rise from (a) 1990 to 2005 and (b) 2005 to
2020.
The gain of light weight low rise between 1990 and 2005 targeted low plants and
avoided dense forests and open low rise water and compact low rise LCZ areas (Figure
6a). Between 2005 and 2020, the gain of Light weight low rise LCZ continued to target low
plants while avoiding other LCZs (Figure 6b).
Figure 6. Transition intensity given the gain of Light weight low rise form (a) 1990 to 2005 and (b)
2005 to 2020.
The gain of the dense forest LCZ in the 1990 to 2005 interval targeted low plants (Fig-
ure 7a). This could imply growth of trees in grasslands such as parks in addition to other
tree planting efforts. The expansion of low plants LCZ targeted dense forest, open low rise
and light weight low rise, while it avoided water and compact low rise in the 2005 to 2020
interval (Figure 7b). The study also noticed slight spectral confusion between light weight
low rise and low plants, especially in the western parts of the study area.
Figure 6.
Transition intensity given the gain of Light weight low rise form (
a
) 1990 to 2005 and
(b) 2005 to 2020.
Remote Sens. 2022, 14, x FOR PEER REVIEW 13 of 22
Figure 7. Transition intensity given (a) gain of dense forest from 1990 to 2005 and (b) low plants from 2005 to
2020.
The gain of open low rise LCZ in the 1990 to 2005 interval targeted low plants (Figure
8a), while in the interval 2005 to 2020 it targeted low plants, water, dense forest and com-
pact low rise (Figure 8b). The gain targeted dense forest more than low plants LCZ, which
could be due to the nature of the LCZ that consists of a few and well-spaced buildings
surrounded by trees and grass. Expansion into water reveals an adverse environmental
impact, with growth intrusion into wetland areas.
Figure 8. Transition intensity given gain of Open low rise from (a) 1990 to 2005 and (b) 2005 to 2020.
3.4. Long Term Changes in the Two Dimensional LST in Response to LCZ Changes
Figure 9 shows expansion of high temperature surfaces between 1990 and 2015. In
1990, LSTs in the 37.8 to 43.8 °C range (Figure 9a) dominated the city, while LSTs below
43.8 °C became uncommon in 2005 (Figure 9b). Visual inspection shows that in 2020, the
LSTs became even higher, with most areas recording values above 46.8 °C (Figure 9c). The
water areas were the most stable, with LSTs in the 16.8 to 37.8 °C range in 1990, 2005 and
2020.
Figure 7.
Transition intensity given (
a
) gain of dense forest from 1990 to 2005 and (
b
) low plants from
2005 to 2020.

Remote Sens. 2022,14, 2060 13 of 22
The gain of open low rise LCZ in the 1990 to 2005 interval targeted low plants
(
Figure 8a
), while in the interval 2005 to 2020 it targeted low plants, water, dense forest and
compact low rise (Figure 8b). The gain targeted dense forest more than low plants LCZ,
which could be due to the nature of the LCZ that consists of a few and well-spaced buildings
surrounded by trees and grass. Expansion into water reveals an adverse environmental
impact, with growth intrusion into wetland areas.
Remote Sens. 2022, 14, x FOR PEER REVIEW 13 of 22
Figure 7. Transition intensity given (a) gain of dense forest from 1990 to 2005 and (b) low plants from 2005 to
2020.
The gain of open low rise LCZ in the 1990 to 2005 interval targeted low plants (Figure
8a), while in the interval 2005 to 2020 it targeted low plants, water, dense forest and com-
pact low rise (Figure 8b). The gain targeted dense forest more than low plants LCZ, which
could be due to the nature of the LCZ that consists of a few and well-spaced buildings
surrounded by trees and grass. Expansion into water reveals an adverse environmental
impact, with growth intrusion into wetland areas.
Figure 8. Transition intensity given gain of Open low rise from (a) 1990 to 2005 and (b) 2005 to 2020.
3.4. Long Term Changes in the Two Dimensional LST in Response to LCZ Changes
Figure 9 shows expansion of high temperature surfaces between 1990 and 2015. In
1990, LSTs in the 37.8 to 43.8 °C range (Figure 9a) dominated the city, while LSTs below
43.8 °C became uncommon in 2005 (Figure 9b). Visual inspection shows that in 2020, the
LSTs became even higher, with most areas recording values above 46.8 °C (Figure 9c). The
water areas were the most stable, with LSTs in the 16.8 to 37.8 °C range in 1990, 2005 and
2020.
Figure 8.
Transition intensity given gain of Open low rise from (
a
) 1990 to 2005 and (
b
) 2005 to 2020.
3.4. Long Term Changes in the Two Dimensional LST in Response to LCZ Changes
Figure 9shows expansion of high temperature surfaces between 1990 and 2015. In 1990,
LSTs in the 37.8 to 43.8
◦
C range (Figure 9a) dominated the city, while LSTs below 43.8
◦
C
became uncommon in 2005 (Figure 9b). Visual inspection shows that in 2020, the LSTs
became even higher, with most areas recording values above 46.8
◦
C (Figure 9c). The water
areas were the most stable, with LSTs in the 16.8 to 37.8 ◦C range in 1990, 2005 and 2020.
Remote Sens. 2022, 14, x FOR PEER REVIEW 14 of 22
Figure 9. Spatial structure of LSTs in Bulawayo in (a) 1990, (b) 2005 and (c) 2020.
In each year, SUHI intensities between the Compact low rise and Low plants LCZs
were comparable, although in 2020 the Low plants were slightly warmer (Figure 10). SUHI
intensities increased with built-up density as well as density of tall buildings evidenced
by largest intensity in Compact low rise (e.g., 11.7 °C in 2020) and lowest in Open low rise
(10 °C). Dense vegetation LCZ areas were cooler than built LCZ in all the periods.
Figure 10. Changes in SUHI intensities across LCZs from 1990 to 2020 in Bulawayo.
0
2
4
6
8
10
12
14
Compact low
rise
Dense forest Light weight low
rise
Open low rise Low plants
Surface Urban Heat Island Intensity (oC)
1990 2005 2020
Figure 9. Spatial structure of LSTs in Bulawayo in (a) 1990, (b) 2005 and (c) 2020.

Remote Sens. 2022,14, 2060 14 of 22
In each year, SUHI intensities between the Compact low rise and Low plants LCZs
were comparable, although in 2020 the Low plants were slightly warmer (
Figure 10
). SUHI
intensities increased with built-up density as well as density of tall buildings evidenced by
largest intensity in Compact low rise (e.g., 11.7
◦
C in 2020) and lowest in Open low rise
(10 ◦C). Dense vegetation LCZ areas were cooler than built LCZ in all the periods.
Remote Sens. 2022, 14, x FOR PEER REVIEW 14 of 22
Figure 9. Spatial structure of LSTs in Bulawayo in (a) 1990, (b) 2005 and (c) 2020.
In each year, SUHI intensities between the Compact low rise and Low plants LCZs
were comparable, although in 2020 the Low plants were slightly warmer (Figure 10). SUHI
intensities increased with built-up density as well as density of tall buildings evidenced
by largest intensity in Compact low rise (e.g., 11.7 °C in 2020) and lowest in Open low rise
(10 °C). Dense vegetation LCZ areas were cooler than built LCZ in all the periods.
Figure 10. Changes in SUHI intensities across LCZs from 1990 to 2020 in Bulawayo.
0
2
4
6
8
10
12
14
Compact low
rise
Dense forest Light weight low
rise
Open low rise Low plants
Surface Urban Heat Island Intensity (oC)
1990 2005 2020
Figure 10. Changes in SUHI intensities across LCZs from 1990 to 2020 in Bulawayo.
4. Discussion
Mapping accuracy was very high, exceeding 95% for 1990, 2005 and 2020 LCZ maps.
Accuracy of at least 95% in LCZ mapping were also recorded by Danylo et al. [
37
] in
Kyiv and Lviv in Ukraine for city specific analysis, which was higher than the below 75%
accuracy they achieved using training data for multiple city mapping. The differences
in accuracy between city level and large scale LCZ mapping approaches demonstrate
that better LCZ maps are generated with city specific efforts than when training areas
of another city are used for LCZ mapping [
55
]. The high accuracy achieved stresses the
high performance of random forest classifier in comparison to other classifiers such as
support vector machine [
75
,
80
,
81
,
83
]. RF uses a set of classifiers that make it superior to
individual classifier based approaches [
81
]. The use of multi-seasonal data also enhanced
discriminability of LCZs in this study, which resulted in high yearly accuracies. LCZ maps
did not show most of the linear features such as roads and rivers due to the large filter used.
As noted by Kotharkar and Bagade [
8
] and Gal et al. [
72
], an LCZ must be large enough to
affect an area’s temperature, while use of a large filter removes linear features and expands
urban area. According to Kotharkar and Bagade [
8
] and Gal et al. [
72
], shifting to a coarser
scale results in loss of a number of LCZs.
Consistent with global trends, there was a general increase in built LCZ in Bul-
awayo between 1990 and 2020. This is a characteristic of city growth globally typified
by conversion from natural land covers to urban fabric. For instance, in Kigali, Rwanda,
Akinyemi et al. [60]
observed an increase in built area from 1% in 1981 to 19% in 2002,
followed by a slight decrease to 18% in 2014. Similar to Bulawayo, and indeed other global
cities trends,
Akinyemi et al. [60]
showed a general increase in built-up area over a 33 year
period. Whereas the expansion of Built LCZ continued throughout the study period, it
was faster between 1990 and 2005 than between 2005 and 2020. For instance, expansion of

Remote Sens. 2022,14, 2060 15 of 22
compact low rise, which coincides with the Central Business District, was slower than other
built LCZs in both intervals. Between 1990 and 2020, lightweight low rise LCZ expanded
faster than other built LCZ. Generally, in Bulawayo, other built LCZ grow around the
Central Business District with light weight low rise spreading to the west and spacious
built-up LCZ spreading to the east. This limits space for further expansion of the Central
Business District (CBD), which explains slowed growth of the corresponding LCZ between
2005 and 2020.
All other LCZs were gainers or losers in the 1990 to 2005 and 2005 to 2020 intervals,
except compact low rise LCZ that was dormant in the 1990 to 2005 interval, implying the
gain stopped or slowed along the interval. In the 2005 to 2020 interval, the gain of compact
low rise was active. Low plants were active losers in the 1990 to 2005 interval and became
dormant losers (the loss of the class was slow or stopped along the interval) in the 2005
to 2020. The water LCZ was an active loser in both intervals. The light weight low rise
LCZ was a dormant gainer in the initial interval, but was an active gainer in the 2005 to
2020 interval. Sustained contraction was recorded in dense forest and water LCZs. The
low plants LCZ, which in this study included croplands, grasslands and parks increased
over the entire period (by 4.3%), although they experienced a decrease (by 0.9%) between
2005 and 2020. This shows that although the LCZ had expanded due to activities such as
deforestation, it was affected by the expansion of built-up between 2005 and 2020.
The expansion of built LCZs between 1990 and 2005 was most prevalent in the open
low rise LCZ. The growth of compact low rise avoided low plants and vegetation LCZs
during the 1990 to 2005 period. In the interval 2005 to 2020, the Compact low rise LCZ
continued to target open low rise and lightweight low rise areas. This may signify growth
of the industrial and CBD to serve the surrounding residential areas, which were also
expanding. The intensity of transition of water to compact low rise was greater in the
interval 1990 to 2005 than 2005 to 2020. In both intervals, the gain of compact low rise
also avoided low plants and dense forest LCZ areas. This may signify adherence to
the Environmental Management Act, which was signed into law in 2013 [
98
]. The Act
has increased protection of wetlands and the general environment with non-compliance,
especially of business enterprises attracting heavy fines. Avoidance of water areas by
expansion of compact low rise may also be a result of costs associated with construction
in wetlands.
The gain of light weight low rise between 1990 and 2005 targeted low plants and
avoided dense forests, open low rise water and compact low rise LCZ areas. Between
2005 and 2020, the gain of Light weight low rise LCZ continued to target low plants while
avoiding other LCZ. The shift from low plant to light weight low rise indicates the change
from primary production to industry based economy as the city grows. This may have
resulted in expansion of low-income residential areas, which comprise the Light weight
low rise occupied by most of the people who work in the industries and Central Business
District (Compact low rise). The expansion of light weight low rise affecting low plant
LCZ in the western direction may also indicate spreading of low income residential areas
from the Central Business District (compact low rise area), where cost of land is high.
Beside the agricultural land, grasslands are also part of the low plants that were targeted
by the expansion of the densely built-up light weight low rise. Although transition level
shows avoidance, the expansion of light weight low rise into open low rise areas during
both intervals (1990 to 2005 and 2005 to 2020) could indicate increase in built-up density
in an area which previously had few buildings surrounded by enough vegetation to be
classified as open low rise. Such areas have few buildings during early stages of land
allocation with densities increasing with time, thus changing from an open to a densely
packed built-up setting.
The gain of the dense forest LCZ in the 1990 to 2005 interval which targeted low
plants could imply growth of trees in grasslands such as parks in addition to other tree
planting efforts. Over a period of 15 years, the planted and naturally growing trees can
increase in canopy size, density and leaf area, enough to be separable from low plants. The

Remote Sens. 2022,14, 2060 16 of 22
expansion of low plants LCZ targeted dense forest, open low rise and light weight low
rise, while it avoided water and compact low rise in the 2005 to 2020 interval. The spread
into dense forests indicates deforestation, which turns formerly dense forest LCZ into
low plants areas such as grasslands and croplands. As the light weight low rise occupied
formerly low plant areas, demand for areas such as peri-urban agriculture increased. This
may cause communities to spread activities into unused areas by clearing some of the
dense forests. Low plants targeting light weight low rise areas could be associated with
growth of vegetation within the built-up LCZ. The vegetation includes edible vegetation in
small gardens as well as lawns around households. The study also noticed slight spectral
confusion between light weight low rise and low plants, especially in the western parts
of the study area. However, most of the confusion was eliminated through the use of
multi-season data, which increased inter-class discriminability in supervised classification.
The gain of open low rise LCZ in the 1990 to 2005 interval targeted low plants while
in the 2005 to 2020 interval it targeted low plants, water, dense forest and compact low
rise. The gain targeted dense forest more than low plants LCZ, which could be due to the
nature of the LCZ that consists of a few well-spaced buildings surrounded by trees and
grass. Expansion into water demonstrates adverse environmental impact with growth
intrusion into wetlands. The expansion of open low rise into low plant areas could indicate
development of spacious settlements into formerly grassland, bare and agricultural areas.
Furthermore, the expansion into other LCZs could indicate the advantage of wealth, as
this LCZ is mostly occupied by medium to high income strata which can afford land
and develop in any area. The expansion of this spacious LCZ could also be a sign of
economic emancipation which enables residence of the city to purchase tracts of large land.
In Zimbabwe, this includes existing land owners in densely built-up low income areas
but who prefer accommodation in low density built-up areas. Due to increased demand
for such spacious settings, developments encroach into formerly protected LCZs such as
wetlands and dense forests.
Intensity analysis provided details of LCZ transitions in Bulawayo between 1990 and
2020 beyond the usual “from to” change detection analysis. It revealed the tendency of
built-up growth, which was at the expense of vegetation areas, especially low plants. The
analysis also showed that the growth of light weight low rise targeted low plant areas and
completely avoided compact low rise and open low rise areas. This could be associated
with the cost of land in the compact low rise and open low rise, mainly found in the central
business district and low-medium density residential (largely occupied by medium to
high income strata), respectively. The findings of this study emphasize the argument by
Niya et al. [
63
] that intensity analysis clarifies substantial causes and processes of land
use changes. Additionally, in agreement with Huang et al. [
66
], intensity analysis can
assess evidence of a particular change and help develop hypothesis concerning processes
of change.
High temperature surfaces expanded while LST temperatures increased between 1990
and 2015. According to Nayak and Mandal [
99
], urbanization causes temperature change
due to both alteration of land use land cover and greenhouse gas concentrations. Similarly,
in this study, we attributed surface warming to both LCZ transitions and background cause
by anthropogenic activities such as industrial emissions. Blake et al. [
100
] also reported that
Harare was warming, despite cooling in the decade from 1900 to 2002. The expansion of
high LST areas and increase in LST intensities was largely due to replacement of natural land
covers with built-up LCZs. Buildings and impervious surfaces have high heat absorption
capacities which cause elevation of LSTs, especially where vegetation fraction and surface
wetness are low. High LSTs were recorded in compact low rise areas especially as their
coverage increased over time. This is in agreement with the sentiment that large areas of
densely packed buildings create homogeneous areas with high LST [
49
]. Therefore, the
growth patterns of Bulawayo have caused warming due to massive replacement of natural
surfaces with buildings and impervious surfaces.

Remote Sens. 2022,14, 2060 17 of 22
Each year, SUHI intensities between the Compact low rise and Low plants LCZs
were comparable, although in 2020, the Low plants were slightly warmer. This is because
field observations showed that during the hot and dry seasons, low plant LCZ areas are
characterized by dry vegetation and even bare or close to bare ground in areas that are
cleared at the end of farming seasons. On the other hand, the link between heat stress and
built-up extent is also complex, as buildings also provide shading while on the other hand
reducing ventilation leading to opposite effects on thermal comfort [
100
,
101
]. Additionally,
vegetation within built-up LCZs has a heat mitigation effect, which can reduce differences
in SUHI intensities between built-up and natural (land cover based) LCZs.
Surface Urban Heat Island intensities increased with built-up density as well as
density of tall buildings evidenced by largest intensity in Compact low rise. According
to Stewart and Oke [
33
], built-up LCZs vary in air temperature depending on factors
that include density and height of buildings as well as type and density of vegetation
within built-up areas. Consistent with this study, Lelovics et al. [
11
] stressed that LCZ2
is warmer than LCZ3, which is warmer than LCZ6. They also concluded that LCZ maps
can distinguish areas based on degree of LULC modification. Contrasts in temperature
between classes with differences in geometry/cover can exceed 10
◦
C, while classes with
few physical differences can be less than 2
◦
C [
33
]. Similarly, Lau et al. [
14
] recorded the
highest temperature (38.9
◦
C) and lowest (29.9
◦
C) in land cover LCZs in Hong Kong.
According to Qiu et al. [
39
], LCZ scheme considers three-dimensional surface structure and
anthropogenic parameters such as heat from human activities that influence temperature.
Based on this understanding, comparatively very high land surface temperatures and SUHI
were observed in the compact low rise area of the city, followed by the densely packed light
weight low rise while the open low rise LCZ was the coolest of the built LCZ.
Dense vegetation LCZ areas were cooler and had lower SUHI intensities than built
LCZ in all the periods. Even as vegetation cover declined, their heat mitigation value
remained remarkably high within artificial LCZs that cause SUHI intensification as the city
grows. Low plants LCZ areas were warmer than Open low rise areas in all the periods.
This could be due to the fact that the Open low rise areas constitute residents of the
middle to high income strata who have resources to ensure that vegetation around their
homes is well-maintained and healthy throughout the year. This is because vegetation in
urban areas serves as temperature refugees in streets and parks providing cooling effect
through evapotranspiration and shading [
14
]. According to Lu et al. [
59
], vegetation within
buildings reduces patch sizes of built-up LCZs thus lowering their thermal effect on the
surrounding LCZs. High SUHI in low plants LCZ contradicted with other studies such
as Shi et al. [
102
] and Lu et al. [
59
] which had low plants as heat sinks. The disparity was
because open low rise includes natural grassland areas, which in Bulawayo experienced
drying of vegetation during the hot dry season. Low plants also include bare areas and
croplands whose cover during the dry season could be bare or dry vegetation, reducing
the surface cooling effect of latent heat transfer. Therefore, spatial and temporal variations
in the thermal characteristics of low plants significantly reduced their heat mitigation
value in Bulawayo. Higher UHI in 2005 and 2020 may partly be explained by higher
average precipitation around satellite overpass dates in those years than in 1990. This is
also in tandem with Amorim [
71
], that the heat mitigation value of urban greenery and
SUHI varies with seasons and are increased during wet periods around overpass when the
vegetation biomass is increased.
5. Conclusions
In order to understand the effect of urban growth on the thermal environment, the
study used multi-temporal Landsat data to map LCZ with very high accuracy and retrieve
SUHI for Bulawayo metropolitan city in Zimbabwe for 1990, 2005 and 2020. LST of the
water LCZ were used as reference for quantifying SUHI intensities instead of the subjective
traditional “rural-urban” LST difference. The high LCZ mapping accuracies were attributed
to precise generation of training data and the robustness of the RF classifier, an ensemble

Remote Sens. 2022,14, 2060 18 of 22
based technique, compared to single classifier-based approaches. Furthermore, the use
of dry and wet biomass periods significantly improved LCZ mapping accuracy in all
years. In both intervals, i.e., 1990 to 2005 and 2005 to 2020, built LCZs monotonically
expanded at the cost of vegetation- and water-based LCZs. Intensity analysis showed that
the growth of lightweight low rise mainly targeted low plant areas. Deforestation in the city
was expressed by the gain of low plants, which targeted dense forests. Intensity analysis
also showed that the growth of compact low rise occupied mostly by low income strata
avoided eastern direction where there are compact low rise and open low rise generally
characterized by high cost per land unit. Due to expansion of built and polluting LCZs,
the SUHI intensities rose monotonically during the study period. SUHI intensities varied
between LCZs as they intensified with built-up proportion and density of tall building
while decreasing with abundance of healthy vegetation. Based on the findings, the study
concluded that human activities and growth induced LCZ changes have continued to
trigger warming in Bulawayo. SUHI retrieval based on LCZ scheme proved effective in
determining effects of urban growth on the thermal environment.
Author Contributions:
Conceptualization, T.D.M., O.M. and J.O.; methodology, T.D.M.; software,
T.D.M.; validation, T.D.M., O.M. and J.O.; formal analysis, T.D.M.; investigation, T.D.M., O.M. and
J.O.; resources, T.D.M., O.M. and J.O.; data curation, T.D.M., O.M. and J.O.; writing—original draft
preparation, T.D.M.; writing—review and editing, T.D.M., O.M. and J.O.; visualization, T.D.M.;
supervision, O.M. and J.O.; project administration, O.M. and J.O.; funding acquisition, T.D.M., O.M.
and J.O. All authors have read and agreed to the published version of the manuscript.
Funding:
We acknowledge the German DAAD climapAfrica and the National Research Foundation
of South Africa for funding the research. The research of this article was supported by DAAD within
the framework of the climapAfrica program of the Federal Ministry of Education and Research. The
publisher is fully responsible for the content. The work and article processing charge was also funded
by the National Research Foundation of South Africa (NRF) Research Chair in Land Use Planning
and Management (Grant Number: 84157).
Data Availability Statement:
Remotely sensed data used in this study can be freely downloaded
from United States Geological Survey (USGS) Earth Explorer website (www.earthexplorer.usgs.gov
(accessed on 27 January 2022)). The training data used to map LCZ have been uploaded on the
WUDAPT website (https://lcz-generator.rub.de/factsheets/3eb90c0ab4210886bdd0b23f1fc7dccae8
93c005/3eb90c0ab4210886bdd0b23f1fc7dccae893c005_factsheet.html) (accessed on 27 January 2022).
Acknowledgments:
We acknowledge the Climate Modeling Group of the climapAfrica fellowship
for inputs during virtual presentations which contributed to the quality of this manuscript. We
thank the Discipline of Geography, School of Agricultural, Earth and Environmental Sciences in
Pietermaritzburg, South Africa for availing a fruitful research environment. The Department of Space
Science and Applied Physics at University of Zimbabwe also provided a working environment for
this research.
Conflicts of Interest: The authors declare no conflict of interest.
References
1.
Ahmed, B.; Ahmed, R. Modeling urban land cover growth dynamics using multioral satellite images: A case study of Dhaka,
Bangladesh. ISPRS Int. J. Geo-Inf. 2012,1, 3–31. [CrossRef]
2.
Grimmond, S. Urbanization and global environmental change: Local effects of urban warming. Geogr. J.
2007
,173, 83–88.
[CrossRef]
3.
Hegazy, I.R.; Kaloop, M.R. Monitoring urban growth and land use change detection with GIS and remote sensing techniques in
Daqahlia governorate Egypt. Int. J. Sustain. Built Environ. 2015,4, 117–124. [CrossRef]
4.
McDonald, R.I.; Green, P.; Balk, D.; Fekete, B.M.; Revenga, C.; Todd, M.; Montgomery, M. Urban growth, climate change, and
freshwater availability. Proc. Natl. Acad. Sci. USA 2011,108, 6312–6317. [CrossRef]
5.
Nyamekye, C.; Kwofie, S.; Ghansah, B.; Agyapong, E.; Appiah, L. Land Use Policy Assessing urban growth in Ghana using
machine learning and intensity analysis: A case study of the New Juaben Municipality. Land Use Policy
2020
,99, 105057. [CrossRef]
6.
Seto, K.C.; Güneralp, B.; Hutyra, L.R. Global forecasts of urban expansion to 2030 and direct impacts on biodiversity and carbon
pools. Proc. Natl. Acad. Sci. USA 2012,109, 16083–16088. [CrossRef]

Remote Sens. 2022,14, 2060 19 of 22
7.
Heaviside, C.; Vardoulakis, S.; Cai, X.M. Attribution of mortality to the urban heat island during heatwaves in the West Midlands,
UK. Environ. Health A Glob. Access Sci. Source 2016,15, 49–59. [CrossRef]
8.
Kotharkar, R.; Bagade, A. Urban Climate Local Climate Zone classification for Indian cities: A case study of Nagpur. Urban Clim.
2018,24, 369–392. [CrossRef]
9.
Bekta¸s Balçik, F. Determining the impact of urban components on land surface temperature of Istanbul by using remote sensing
indices. Environ. Monit. Assess. 2014,186, 859–872. [CrossRef]
10.
Chow WT, L.; Brennan, D.; Brazel, A.J. Urban heat island research in Phoenix, Arizona. Bull. Am. Meteorol. Soc.
2012
,93, 517–530.
[CrossRef]
11.
Lelovics, E.; Unger, J.; Gál, T.; Gál, C.V. Design of an urban monitoring network based on Local Climate Zone mapping and
temperature pattern modelling. Clim. Res. 2014,60, 51–62. [CrossRef]
12.
Odindi, J.O.; Bangamwabo, V.; Mutanga, O. Assessing the value of urban green spaces in mitigating multi-seasonal urban heat
using MODIS land surface temperature (LST) and Landsat 8 data. Int. J. Environ. Res. 2015,9, 9–18.
13.
Saaroni, H.; Ben-Dor, E.; Bitan, A.; Potchter, O. Spatial distribution and microscale characteristics of the urban heat island in
Tel-Aviv, Israel. Landsc. Urban Plan. 2000,48, 1–18. [CrossRef]
14.
Lau, K.K.; Ching, S.; Ren, C. Outdoor thermal comfort in different urban settings of sub-tropical high- density cities: An approach
of adopting local climate zone (LCZ) classification. Build. Environ. 2019,154, 227–238. [CrossRef]
15.
Ormandy, D.; Ezratty, V. Thermal discomfort and health: Protecting the susceptible from excess cold and excess heat in housing.
Adv. Build. Energy Res. 2016,10, 84–98. [CrossRef]
16.
Radhi, H.; Sharples, S. Quantifying the domestic electricity consumption for air-conditioning due to urban heat islands in hot
arid regions. Appl. Energy 2013,112, 371–380. [CrossRef]
17.
Radhi, H.; Fikry, F.; Sharples, S. Impacts of urbanisation on the thermal behaviour of new built up environments: A scoping study
of the urban heat island in Bahrain. Landsc. Urban Plan. 2013,113, 47–61. [CrossRef]
18.
Zhou, Y.; Weng, Q.; Gurney, K.R.; Shuai, Y.; Hu, X. Estimation of the relationship between remotely sensed anthropogenic heat
discharge and building energy use. ISPRS J. Photogramm. Remote Sens. 2012,67, 65–72. [CrossRef]
19.
Jianya, G.; Haigang, S.; Guorui, M.; Qiming, Z. A review of multi-temporal remote sensing data change detection algorithms.
Int. Arch. Photogramm. Remote Sens. Spat. Inf. Sci. 2008,37, 757–762.
20.
Lee, L.; Chen, L.; Wang, X.; Zhao, J. Use of Landsat TM/ETM+ data to analyze urban heat island and its relationship with land
use/cover change. In Proceedings of the 2011 International Conference on Remote Sensing, Environment and Transportation
Engineering, Nanjing, China, 24–26 June 2011; pp. 922–927. [CrossRef]
21.
Gobakis, K.; Kolokotsa, D.; Synnefa, A.; Saliari, M.; Giannopoulou, K.; Santamouris, M. Development of a model for urban heat
island prediction using neural network techniques. Sustain. Cities Soc. 2011,1, 104–115. [CrossRef]
22.
Keramitsoglou, I.; Kiranoudis, C.T.; Ceriola, G.; Weng, Q.; Rajasekar, U. Identification and analysis of urban surface temperature
patterns in Greater Athens, Greece, using MODIS imagery. Remote Sens. Environ. 2011,115, 3080–3090. [CrossRef]
23.
Kolokotsa, D.; Psomas, A.; Karapidakis, E. Urban heat island in southern Europe: The case study of Hania, Crete. Sol. Energy
2009,83, 1871–1883. [CrossRef]
24.
Kumar, D.; Shekhar, S. Statistical analysis of land surface temperature-vegetation indexes relationship through thermal remote
sensing. Ecotoxicol. Environ. Saf. 2015,121, 39–44. [CrossRef]
25.
Uddin, S.; Al Ghadban, A.N.; Al Dousari, A.; Al Murad, M.; Al Shamroukh, D. A remote sensing classification for land-cover
changes and micro-climate in Kuwait. Int. J. Sustain. Dev. Plan. 2010,5, 367–377. [CrossRef]
26.
Li, H.; Liu, Q. Comparison of NDBI and NDVI as indicators of surface urban heat island effect in MODIS imagery. Int. Conf. Earth
Obs. Data Process. Anal. 2008,7285, 728503.
27.
Jamei, Y.; Rajagopalan, P.; Sun, Q. Spatial structure of surface urban heat island and its relationship with vegetation and built-up
areas in Melbourne, Australia. Sci. Total Environ. 2019,659, 1335–1351. [CrossRef] [PubMed]
28.
Sung, C.Y. Mitigating surface urban heat island by a tree protection policy: A case study of The Woodland, Texas, USA. Urban For.
Urban Green. 2013,12, 474–480. [CrossRef]
29.
Rasul, A.; Balzter, H.; Smith, C. Spatial variation of the daytime Surface Urban Cool Island during the dry season in Erbil, Iraqi
Kurdistan, from Landsat 8. Urban Clim. 2015,14, 176–186. [CrossRef]
30.
Sheng, L.; Lu, D.; Huang, J. Impacts of land-cover types on an urban heat island in Hangzhou, China. Int. J. Remote Sens.
2015
,
36, 1584–1603. [CrossRef]
31.
Wu, C.D.; Lung, S.C.C.; Jan, J.F. Development of a 3-D urbanization index using digital terrain models for surface urban heat
island effects. ISPRS J. Photogramm. Remote Sens. 2013,81, 1–11. [CrossRef]
32.
Cai, M.; Ren, C.; Xu, Y.; Dai, W.; Wang, X.M. Local Climate Zone Study for Sustainable Megacities Development by Using
Improved WUDAPT Methodology–A Case Study in Guangzhou. Procedia Environ. Sci. 2016,36, 82–89. [CrossRef]
33.
Stewart, I.D.; Oke, T.R. Local climate zones for urban temperature studies. Bull. Am. Meteorol. Soc.
2012
,93, 1879–1900. [CrossRef]
34.
Fenner, D.; Meier, F.; Bechtel, B.; Otto, M.; Scherer, D. Intra and inter ‘local climate zone’ variability of air temperature as observed
by crowdsourced citizen weather stations in Berlin, Germany. Meteorol. Zeitscrift 2017,26, 525–547. [CrossRef]
35.
Bechtel, B.; Alexander, P.J.; Böhner, J.; Ching, J.; Conrad, O.; Feddema, J.; Mills, G.; See, L.; Stewart, I. Mapping local climate zones
for a worldwide database of the form and function of cities. ISPRS Int. J. Geo-Inf. 2015,4, 199–219. [CrossRef]

Remote Sens. 2022,14, 2060 20 of 22
36.
Cai, M.; Ren, C.; Xu, Y.; Lau, K.K.; Wang, R. Urban Climate Investigating the relationship between local climate zone and land
surface temperature using an improved WUDAPT methodology–A case study of Yangtze River Delta, China. Urban Clim.
2018
,
24, 485–502. [CrossRef]
37.
Danylo, O.; See, L.; Bechtel, B.; Schepaschenko, D.; Fritz, S. Contributing to WUDAPT: A Local Climate Zone Classification of
Two Cities in Ukraine. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 2016,9, 1841–1853. [CrossRef]
38.
Perera NG, R.; Emmanuel, R. Urban Climate A “Local Climate Zone” based approach to urban planning in Colombo, Sri Lanka.
Urban Clim. 2018,23, 188–203. [CrossRef]
39. Qiu, C.; Mou, L.; Schmitt, M.; Xiang, X. Local climate zone-based urban land cover classification from multi-seasonal Sentinel-2
images with a recurrent residual network. ISPRS J. Photogramm. Remote Sens. 2019,154, 151–162. [CrossRef]
40.
United Nations General Assembly. Sustainable Development Goals. SDGs Transform Our World 2030; United Nations General
Assembly: New York, NY, USA, 2015.
41.
Zhao, C.; Jensen JL, R.; Weng, Q.; Currit, N. Use of Local Climate Zones to investigate surface urban heat islands in Texas Use of
Local Climate Zones to investigate surface urban heat islands in Texas. GISci. Remote Sens. 2020,57, 1083–1101. [CrossRef]
42.
Nassar, A.K.; Blackburn, G.A.; Whyatt, J.D. International Journal of Applied Earth Observation and Geoinformation Dynamics
and controls of urban heat sink and island phenomena in a desert city: Development of a local climate zone scheme using
remotely-sensed inputs. Int. J. Appl. Earth Obs. Geoinf. 2016,51, 76–90. [CrossRef]
43.
Zhou, X.; Xu, L.; Zhang, J.; Niu, B.; Luo, M.; Zhou, G.; Zhang, X. Data-driven thermal comfort model via support vector machine
algorithms: Insights from ASHRAE RP-884 database. Energy Build. 2020,211, 109795. [CrossRef]
44.
Eldesoky AH, M.; Gil, J.; Berghauser, M. Urban Climate The suitability of the urban local climate zone classification scheme for
surface temperature studies in distinct macroclimate regions. Urban Clim. 2021,37, 100823. [CrossRef]
45.
Dutta, K.; Basu, D.; Agrawal, S. Evaluation of seasonal variability in magnitude of urban heat islands using local climate zone
classification and surface albedo. Int. J. Environ. Sci. Technol. 2021, 1–22. [CrossRef]
46.
Badaro-saliba, N.; Adjizian-gerard, J.; Zaarour, R.; Najjar, G. Urban Climate LCZ scheme for assessing Urban Heat Island intensity
in a complex urban area (Beirut, Lebanon). Urban Clim. 2021,37, 100846. [CrossRef]
47.
Mandelmilch, M.; Ferenz, M.; Mandelmilch, N.; Potchter, O. Urban Spatial Patterns and Heat Exposure in the Mediterranean City
of Tel Aviv. Atmosphere 2020,11, 963. [CrossRef]
48.
Lee, Y.; Lee, S.; Im, J.; Yoo, C. Analysis of Surface Urban Heat Island and Land Surface Temperature Using Deep Learning Based
Local Climate Zone Classification: A Case Study of Suwon and Daegu, Korea. Korean J. Remote Sens. 2021,37, 1447–1460.
49.
Ardiyansyah, A.; Munir, A.; Gabric, A. The Utilization of Land Surface Temperature Information as an Input for Coastal City
The Utilization of Land Surface Temperature Information as an Input for Coastal City. IOP Conf. Ser. Earth Environ. Sci.
2021
,
921, 012004. [CrossRef]
50.
Zhang, Y.; Li, D.; Liu, L.; Liang, Z.; Shen, J.; Wei, F.; Li, S. Spatiotemporal Characteristics of the Surface Urban Heat Island and Its
Driving Factors Based on Local Climate Zones and Population in Beijing, China. Atmosphere 2021,12, 1271. [CrossRef]
51.
Fricke, C.; Pongrácz, R.; Gál, T.; Savi´c, S.; Unger, J. Using local climate zones to compare remotely sensed surface temperatures in
temperate cities and hot desert cities. Morav. Geogr. Rep. 2020,28, 48–60. [CrossRef]
52.
Dian, C.; Pongrácz, R.; Dezs˝o, Z.; Bartholy, J. Urban Climate Annual and monthly analysis of surface urban heat island intensity
with respect to the local climate zones in Budapest. Urban Clim. 2020,31, 100573. [CrossRef]
53.
Budhiraja, B.; Gawuc, L.; Agrawal, G. Seasonality of Surface Urban Heat Island in Delhi City Region Measured by Local Climate
Zones and Conventional Indicators. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 2019,12, 5223–5232. [CrossRef]
54.
Shi, L.; Ling, F.; Foody, G.M.; Yang, Z.; Liu, X.; Du, Y. Seasonal SUHI Analysis Using Local Climate Zone Classification: A Case
Study of Wuhan, China. Int. J. Environ. Res. Public Health 2021,18, 7242. [CrossRef]
55.
Bechtel, B.; Demuzere, M.; Mills, G.; Zhan, W.; Sismanidis, P.; Small, C.; Voogt, J. Urban Climate SUHI analysis using Local
Climate Zones—A comparison of 50 cities. Urban Clim. 2021,28, 100451. [CrossRef]
56.
Demuzere, M.; Bechtel, B.; Mills, G. Urban Climate Global transferability of local climate zone models. Urban Clim.
2019
,
27, 46–63
.
[CrossRef]
57.
Gusso, A.; Cafruni, C.; Bordin, F.; Veronez, M.R.; Lenz, L.; Crija, S. Multitemporal Analysis of Thermal Distribution Char-
acteristics for Urban Heat Islands Management. In Proceedings of the 4th World Sustainability Forum, Basel, Switzerland,
1–30 November 2014. [CrossRef]
58.
Wang, R.; Cai, M.; Ren, C.; Bechtel, B.; Xu, Y.; Ng, E. Urban Climate Detecting multi-temporal land cover change and land surface
temperature in Pearl River Delta by adopting local climate zone. Urban Clim. 2019,28, 100455. [CrossRef]
59.
Lu, Y.; Yang, J.; Ma, S. Dynamic Changes of Local Climate Zones in the Guangdong–Hong Kong–Macao Greater Bay Area and
Their Spatio-Temporal Impacts on the Surface Urban Heat Island Effect between 2005 and 2015. Sustainability
2021
,13, 6374.
[CrossRef]
60.
Akinyemi, F.O.; Pontius, R.G., Jr.; Braimoh, A.K. Land change dynamics: Insights from Intensity Analysis applied to an African
emerging city. J. Spat. Sci. 2017,62, 69–83.
61.
Aldwaik, S.Z.; Pontius, R.G., Jr. Map errors that could account for deviations from a uniform intensity of land change. Int. J.
Geogr. Inf. Sci. 2013,27, 1717–1739. [CrossRef]
62.
Aldwaik, S.Z.; Gilmore, R.P., Jr. Landscape and Urban Planning Intensity analysis to unify measurements of size and stationarity
of land changes by interval, category, and transition. Landsc. Urban Plan. 2012,106, 103–114. [CrossRef]

Remote Sens. 2022,14, 2060 21 of 22
63. Niya, A.K.; Huang, J.; Karimi, H.; Keshtkar, H.; Naimi, B. Use of Intensity Analysis to Characterize Land Use/Cover Change in
the Biggest Island of Persian. Sustainability 2019,11, 4396. [CrossRef]
64.
Feng, Y.; Lei, Z.; Tong, X.; Gao, C.; Chen, S.; Wang, J.; Wang, S. Spatially-explicit modeling and intensity analysis of China’s land
use change 2000–2050. J. Environ. Manag. 2020,263, 110407. [CrossRef]
65.
Huang, B.; Huang, J.; Gilmore, R.P., Jr.; Tu, Z. Comparison of Intensity Analysis and the land use dynamic degrees to measure
land changes outside versus inside the coastal zone of Longhai, China. Ecol. Indic. 2018,89, 336–347. [CrossRef]
66.
Ekumah, B.; Armah, F.A.; Afrifa, E.K.; Aheto, D.W.; Odoi, J.O.; Afitiri, A.R. Assessing land use and land cover change in coastal
urban wetlands of international importance in Ghana using Intensity Analysis. Wetl. Ecol. Manag.
2020
,28, 271–284. [CrossRef]
67.
Alo, C.A.; Pontius, R.G. Identifying systematic land-cover transitions using remote sensing and GIS: The fate of forests inside and
outside protected areas of Southwestern Ghana. Environ. Plan. B Plan. Des. 2008,35, 280–295. [CrossRef]
68.
Mutengu, S.; Hoko, Z.; Makoni, F.S. An assessment of the public health hazard potential of wastewater reuse for crop production.
A case of Bulawayo city, Zimbabwe. Phys. Chem. Earth 2007,32, 1195–1203. [CrossRef]
69.
Gumbo, B.; Mlilo, S.; Broome, J.; Lumbroso, D. Industrial water demand management and cleaner production potential: A case of
three industries in Bulawayo, Zimbabwe. Phys. Chem. Earth 2003,28, 797–804. [CrossRef]
70.
Muchingami, I.; Hlatywayo, D.J.; Nel, J.M.; Chuma, C. Electrical resistivity survey for groundwater investigations and shallow
subsurface evaluation of the basaltic-greenstone formation of the urban Bulawayo aquifer. Phys. Chem. Earth
2012
,50–52, 44–51.
[CrossRef]
71.
Amorim, M.C.D.C.T. Spatial variability and intensity frequency of surface heat island in a Brazilian city with continental tropical
climate through remote sensing. Remote Sens. Appl. Soc. Environ. 2018,9, 10–16. [CrossRef]
72.
Bechtel, B.; Langkamp, T.; Böhner, J.; Daneke, C.; Oßenbrügge, J.; Schempp, S. Classification and Modelling of Urban Micro-
Climates Using Multisensoral and Multitemporal Remote Sensing Data. ISPRS-Int. Arch. Photogramm. Remote Sens. Spat. Inf. Sci.
2012,39, 463–468. [CrossRef]
73.
Gal, T.; Bechtel, B.; Unger, J. Comparison of two different Local Climate Zone mapping methods. In Proceedings of the ICUC9—
9th International Conference on Urban Climate Jointly with 12th Symposium on the Urban Environment, Toulouse, France,
20–24 July 2015; pp. 1–6.
74.
Shi, Y.; Lau, K.K.; Ren, C.; Ng, E. Urban Climate Evaluating the local climate zone classi fi cation in high-density heterogeneous
urban environment using mobile measurement. Urban Clim. 2018,25, 167–186. [CrossRef]
75.
Verdonck, M.L.; Okujeni, A.; van der Linden, S.; Demuzere, M.; De Wulf, R.; Van Coillie, F. Influence of neighbourhood
information on ‘Local Climate Zone’ mapping in heterogeneous cities. Int. J. Appl. Earth Obs. Geoinf.
2017
,62, 102–113. [CrossRef]
76.
Gislason, P.O.; Benediktsson, J.A.; Sveinsson, J.R. Random Forest Classification of Multisource Remote Sensing and Geographic
Data. In Proceedings of the IGARSS 2004. 2004 IEEE International Geoscience and Remote Sensing Symposium, Anchorage, AK,
USA, 20–24 September 2004; pp. 1049–1052.
77.
Novack, T.; Stilla, U. Classification of Urban Settlements Types based on space-borne SAR datasets. ISPRS Ann. Photogramm.
Remote Sens. Spat. Inf. Sci. 2014,2, 55–60. [CrossRef]
78.
Choe, Y.J.; Yom, J.H. Improving accuracy of land surface temperature prediction model based on deep-learning. Spat. Inf. Res.
2020,28, 377–382. [CrossRef]
79.
Meinel, G.; Winkler, M. Long-term investigation of urban sprawl on the basis of remote sensing data-Results of an international
city comparison. In Proceedings of the 24th EARSeL-Symposium: New Strategies for European Remote Sensing, Dubrovnik,
Croatia, 25–27 May 2004.
80. Pal, M. Random forest classifier for remote sensing classification. Int. J. Remote Sens. 2005,26, 217–222. [CrossRef]
81.
Rodriguez-galiano, V.F.; Ghimire, B.; Rogan, J.; Chica-olmo, M.; Rigol-sanchez, J.P. An assessment of the effectiveness of a random
forest classifier for land-cover classification. ISPRS J. Photogramm. Remote Sens. 2012,67, 93–104. [CrossRef]
82.
Gislason, P.O.; Benediktsson, J.A.; Sveinsson, J.R. Random Forests for land cover classification. Pattern Recognit. Lett.
2006
,
27, 294–300. [CrossRef]
83.
Sumaryono, S. Assessing Building Vulnerability to Tsunami Hazard Using Integrative Remote Sensing and GIS Approaches.
Ph.D. Thesis, LMU München, Faculty of Geosciences, München, Germany, 2010.
84.
Sheykhmousa, M.; Mahdianpari, M.; Ghanbari, H. Support Vector Machine vs. Random Forest for Remote Sensing Image
Classification: A Meta-analysis and systematic review. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens.
2020
,13, 6308–6325.
[CrossRef]
85.
Coppin, P.R.; Bauer, M.E. Digital change detection in forest ecosystems with remote sensing imagery Digital Change Detection in
Forest Ecosystems with Remote Sensing Imagery. Remote Sens. Rev. 1996,13, 207–234. [CrossRef]
86.
Feng, H.; Zhao, X.; Chen, F.; Wu, L. Using land use change trajectories to quantify the effects of urbanization on urban heat island.
Adv. Sp. Res. 2014,53, 463–473. [CrossRef]
87.
Pontius, R.G.; Gao, Y.; Giner, N.M.; Kohyama, T.; Osaki, M.; Hirose, K. Design and Interpretation of Intensity Analysis Illustrated
by Land Change in Central Kalimantan, Indonesia. Land 2013,2, 351–369. [CrossRef]
88.
Adu, I.K.; Tetteh, J.D.; Puthenkalam, J.J.; Antwi, K.E. Intensity Analysis to Link Changes in Land Use Pattern in the Abuakwa
North and South Municipalities Ghana from 1986 to 2017. World Acad. Sci. Eng. Technol. Int. J. Geol. Environ. Eng.
2020
,
14, 225–242.

Remote Sens. 2022,14, 2060 22 of 22
89.
Enaruvbe, G.O.; Gilmore, R.P., Jr. Influence of classification errors on Intensity Analysis of land changes in southern Nigeria.
Int. J. Remote Sens. 2015,36, 244–261. [CrossRef]
90.
Yang, Y.; Liu, Y.; Xu, D.; Zhang, S. Use of Intensity Analysis to Measure Land Use Changes from 1932 to 2005 in Zhenlai County,
Northeast China. Chin. Geogra Sci. 2017,27, 441–455. [CrossRef]
91.
Varga, O.G.; Pontius, R.G., Jr.; Kumar, S.; Szabó, S. Intensity Analysis and the Figure of Merit’s components for assessment of a
Cellular Automata–Markov simulation model. Ecol. Indic. 2019,101, 933–942. [CrossRef]
92.
Chen, X.L.; Zhao, H.M.; Li, P.X.; Yin, Z.Y. Remote sensing image-based analysis of the relationship between urban heat island and
land use/cover changes. Remote Sens. Environ. 2006,104, 133–146. [CrossRef]
93.
Srivanit, M.; Hokao, K.; Phonekeo, V. Assessing the Impact of Urbanization on Urban Thermal Environment: A Case Study of
Bangkok Metropolitan. Int. J. Appl. Sci. Technol. 2012,2, 243–256.
94.
Stathopoulos, T.; Wu, H.; Zacharias, J. Outdoor human comfort in an urban climate. Build. Environ.
2004
,39, 297–305. [CrossRef]
95.
Yang, J.S.; Wang, Y.Q.; August, P.V. Estimation of Land Surface Temperature Using Spatial Interpolation and Satellite-Derived
Surface Emissivity. J. Environ. Inform. 2015,4, 40–47. [CrossRef]
96.
Weng, Q.; Liu, H.; Lu, D. Assessing the effects of land use and land cover patterns on thermal conditions using landscape metrics
in city of Indianapolis, United States. Urban Ecosyst. 2007,10, 203–219. [CrossRef]
97.
Dimitrov, S.; Popov, A.; Iliev, M. An Application of the LCZ Approach in Surface Urban Heat Island Mapping in Sofia, Bulgaria.
Atmosphere 2021,12, 1370. [CrossRef]
98.
Chatira-Muchopa, B.; Tarisayi, K.S.; Chidarikire, M. Solid waste management practices in Zimbabwe: A case study of one
secondary school. TD J. Transdiscipl. Res. South. Afr. 2019,15, 1–5. [CrossRef]
99.
Nayak, S.; Mandal, M. Impact of land use and land cover changes on temperature trends over India. Land Use Policy
2019
,
89, 104238. [CrossRef]
100.
Blake, R.; Grimm, A.; Ichinose, T.; Horton, R.; Gaffin, S.; Jiong, S.; Bader, D.; Cecil, L.D. Urban climate: Processes, trends
and projections. In Climate Change and Cities: First Assessment Report of the Urban Climate Change Research Network; Cambridge
University Press: Cambridge, UK, 2011; pp. 43–81.
101.
Lin, T.P.; De Dear, R.; Hwang, R.L. Effect of thermal adaptation on seasonal outdoor thermal comfort. Int. J. Climatol.
2011
,31,
302–312. [CrossRef]
102.
Shi, Y.; Xiang, Y.; Zhang, Y. Urban Design Factors Influencing Surface Urban Heat Island in the High-Density City of Guangzhou
Based on the Local Climate Zone. Sensors 2019,19, 3459. [CrossRef]