There has been an increase in the number of urban dwellers, together with an accompanying expansion of built-up area globally [1
]. Urban areas are strategic areas economically, as well as from an administrative perspective. They are important for issues such as the improvement of education and health delivery of a nation. Despite their socio-economic importance, urban areas and characteristic complex land use and land cover (LULC) spatial structure also pose a variety of environmental changes [2
]. According to Acharya et al. [5
], the benefits of urban growth in developing countries include opportunities for employment, specialization, and the better production of goods and services. The challenges, however, include air pollution and water pollution in industrialized areas, while flash flooding is prevalent in highly impervious areas. Another notable challenge of urban development is temperature elevation, especially in densely built-up areas [6
]. Studies have shown that urban areas are comparatively warmer than undisturbed surroundings such as rural areas; a phenomenon called Urban Heat Island (UHI) [8
]. According to Gusso et al. [8
], cities use construction materials such as concrete and asphalt, which do not allow water to penetrate and absorb a large amount of heat, thereby increasing urban temperatures. Elevated temperature results in increased outdoor and indoor human thermal discomfort, as well as increased heat-related health risk [13
]. Urban heat islands have maximized the number of heat wave days and tropical-like night conditions in several main cities, including Paris, Baltimore, Washington D.C., and Shanghai, during the summer [17
]. Furthermore, the Intergovernmental Panel on Climate Change (IPCC) [20
] stressed that land cover changes have the potential to raise air temperatures of urbanized areas by 4 °C by 2100. The changes and associated adverse impacts seriously threaten the sustainable development of urban areas [21
]. Urban land use and land cover heterogeneity, as well as changes, result in the complex and varied spatial structure of heat intensities which also vary from city to city. It is thus important to establish city specific land surface temperature patterns in order to derive relevant mitigation and response strategies.
Remote sensing offers a variety of options for monitoring both LULC and LST spatial structure. Unfortunately, space-borne sensors detect thermal infra-red at either a low (e.g., above 500 m such as METEOSAT) or medium (e.g., 30–500 m such Landsat, ASTER and MODIS), but not high, spatial resolution (e.g., below 30 m such as SPOT). This results in mismatch in the resolution between retrieved LULC and LST maps. High resolution thermal data is often obtained from air-borne missions. Generally, high spatial resolution datasets are expensive to gather, have a low temporal resolution, usually lack a thermal infra-red component, and have very limited historical archives not sufficient for long term analysis [22
]. Medium resolution multi-spectral datasets are often reliable for urban LULC and LST analysis. For example, Landsat has large stores of visible, infra-red, and thermal data archives spanning from as early as 1972 to present [6
]. Recently, studies showed that Landsat data are effective and very accurate in mapping urban LULC distribution, as well as changes thereof [22
]. For example, using Landsat data, Mushore et al. [9
] retrieved LULC spatial and temporal patterns in Harare between 1984 and 2015 at overall accuracies greater than 80%. Studies have also proved the effectiveness of Landsat thermal data in mapping land surface temperature variations, including those in complex urban settings [26
]. Recently, multi-temporal Landsat data was used to develop a model to predict future urban surface temperatures in Harare [29
]. Mushore et al. [29
], showed that if historical growth patterns will persist, land surface temperatures will increase by as much as 5 °C by 2045. Therefore, the utility of medium resolution datasets in quantifying the impact of urban growth on LST patterns needs to be continually exploited. This is necessary in cities of low Gross Domestic Product countries such as in Africa, especially where similar studies have not yet been done; for example, in Sierra Leone.
In Africa, the studies have been confined to a few cities mainly in South Africa, Zimbabwe, and Nigeria. For example, Odindi et al. [7
] investigated the impact of seasonality of urban greenery on heat island patterns in the Ethkwini municipality in South Africa. However, although they used 30 m multispectral Landsat 7 data for LULC mapping, surface temperatures were retrieved from course resolution (1 km) MODIS thermal data. Other studies in Africa were also confined to a single city; for example, Mushore et al. [9
] only focused on Harare in Zimbabwe, while in West Africa, Abegunde and Adedeji [30
] focused on Ibadan in Nigeria. Given the projected urban growth which must be faster in developing countries, there is thus a need to understand the implications in other parts of Africa [19
]. While Odindi et al. [31
] compared LST patterns in coastal cities of South Africa, there is a general paucity of literature on comparing LST patterns between two cities of an African country. Precisely, there is a lack of literature comparing LST patterns of two cities, especially with one being inland and the other being coastal, such as Freetown and Bo town in Sierra Leone. As such, there is the need for a novel study to understand urban growth patterns, as well as responses of LST, in Sierra Leone, in West Africa. Such analysis is important for understanding both the differential effect of urban growth and of global warming between a coastal and an inland city in West Africa. Adaptation and mitigation strategies derived from such an analysis will take into account the position of a city relative to the ocean. Furthermore, the Contribution Index (CI) has not yet been used to compare growth patterns of two cities, as well as to explain the impacts of growth on surface temperatures in West Africa. To the best of our knowledge, the index has only been successfully tested on the African continent in South Africa [7
] and in Zimbabwe [9
]. Odindi et al. [31
] used CI to compare LULC and LST patterns between coastal cities of South Africa, but did not compare a coastal city with an inland city. Although Odindi et al. [31
] compared LST variations in two cities; they used course resolution MODIS data, leaving a gap on comparison analysis using Landsat data in Africa. Liu and Weng [32
] also found the 30 m visible and infrared, as well as the 90 to 120 m resolution thermal infra-red, Landsat data to be optimal in the analysis of the relationship between LULC and LST patterns.
The objectives of this study are thus to (1) use remote sensing to determine urban growth patterns in Sierra Leone; (2) quantify the effect of urban growth on spatial and temporal LST patterns in two major cities of Sierra Leone using the CI; and (3) understand the differences in responses of LST to urban growth and global warming between a coastal city (Freetown) and an inland city (Bo town) in Sierra Leone. The study hypothesizes that urban growth patterns should differ between Freetown and Bo town and thus influence LST spatial and temporal changes to differ between the two cities.
The study obtained a high classification accuracy both in a coastal city (Freetown) and an inland city (Bo town). The overall classification accuracy reached the 85% recommendation by Anderson [49
], because even at a 30 m resolution of Landsat optical data, the mixed pixel problem did not significantly affect the quality of the LULC maps produced. Despite the complexity of classification in urban areas due to surface heterogeneity, the mapping accuracies are also higher than the 80% overall accuracy recommended by Omran [50
]. The high level of accuracy can be justified by Voogt and Oke [51
], who noticed that improvements that have occurred in satellite sensors over the years provide detailed and accurate land surface representation at a low cost. The high classification accuracy could also be attributed to the renowned performance of the Support Vector Machine algorithm [22
]. According to Jia et al. [41
], the Support Vector Machine (SVM) algorithm was found to outperform other common classifiers such as ANN, maximum likelihood, and Mahalanobis distance. The algorithm was also used for multi-temporal Landsat-based classification in an urban setting in Harare, where overall accuracies above 80% were also obtained. These findings show the value of freely available medium resolution space-borne remotely sensed datasets for monitoring urban extent and growth, especially in resource-constrained nations.
The population increased by almost ten-fold in both Freetown and Bo town between 1963 and 2015, while the population densities also increased. In all the periods considered, the population size of Freetown has always far exceeded that of Bo town. Most of the economic and administrative activities of Sierra Leone are concentrated in Freetown, hence the larger population size and faster growth than Bo town. Furthermore, the beauty of the sea seems to make residents prefer to concentrate along the coastal margins of Freetown than to spread further inland towards the dense vegetation area. Besides increasing population sizes, built-up areas are also expanding in both cities. Growth patterns observed in both cities agree with earlier observations and predictions that urban population is growing, globally [19
]. Expansion of the built-up area in Freetown has been mainly concentrated along the coast and is most notable in the northern, eastern, and western parts of the city. This growth along the northern margins of Freetown explains why the dense vegetation area in the central part of the city is not diminishing even as the built-up area is expanding. A different pattern is observed in Bo town, where the built-up area is expanding from central locations outwards. Unlike in Freetown, the growth of Bo town has led to a reduction in the area of the densely vegetated LULC category between 1985 and 2015. As observed in Bo town, in most studies, the proportion of total area occupied by dense vegetation decreases with continuous urban expansion [26
]. Kamusoko et al. [54
] observed that the expansion of built-up areas in Harare Zimbabwe pushed most dense vegetation locations outwards to the peripheries of the city.
As expected, temperature responded strongly to spatiotemporal dynamics of LULC in both Freetown and Bo town. High temperatures in both cities were observed in built-up areas and their extent increased with time as the cities were expanding. The influence of buildings explains why high surface temperatures (above 30 °C) were recorded in northern and eastern parts of Freetown. Over the years, surface temperatures in this regime have also been spreading southward along the western margin of the city following the expansion of the built-up area. The shape of high surface temperature areas in both Freetown and Bo town closely mimics that of the built-up area, indicating their strong warming influence. This concurs with Sha and Ghauri [28
], who observed that surface urban heat island expands with expansion in a built-up area. Buildings reduce heat removal by advection and reduce the sky view factor, thus limiting heat escape to space, while walls and pavements absorb and emit heat [28
]. This results in large amounts of stagnant heat and high temperatures, especially in closely packed and high rise buildings. The warming in both cities could also be explained by increased anthropogenic activities supported by an increasing population size in both cities over time, which increases long wave radiation in the lower atmosphere. Nayak and Mandal [3
] and Grimmond [57
] also attributed urban warming to both LULC changes and other anthropogenic effects such as greenhouse gas emissions. The rising temperature in response to the growth of both cities can be captured by the explanation that, as population grows, urbanization increases and the magnitude of the urban heat island also expands [58
]. Similar findings were obtained in Australia between 1951 and 2003, where land cover changes produced statistically significant warming [59
Vegetation cover has been indicated to be a strong mitigation measure against the elevation of surface temperatures in both cities. For example, in Freetown, low surface temperatures (below 22 °C) remained characteristic of the central and southwestern parts of the city where buildings have not yet replaced vegetation cover. Similarly, low temperature areas (below 20 °C) surround an expanding hot spot in the central parts of the city of Bo town. The heat mitigation value of vegetation was also captured by a strong negative Contribution Index (between −0.5 and −1) in areas with dense and sparse vegetation. This concurs with Odindi et al. [7
] who in the EThekwini municipality, South Africa, showed that the temperature reduction effect of vegetation increases with the percentage of total area covered. Although water bodies also have a cooling effect (negative Contribution Index [CI]), their contribution has remained minimal over the years due to the low proportion of the cities they occupy in both cities. Based on CI, the cooling effect of dense vegetation was more than of sparse vegetation, which echoes the suggestion by Zhang et al. [60
] that not only vegetation types but also spatial structure affects LST distribution. Vegetation cover promotes surface cooling due to latent heat transfer.
In both Freetown and Bo town, agriculture areas were causing warming of the city, as indicated by a positive Contribution Index (CI) in all periods. This could be because, during the dry seasons, agriculture areas will either be covered by drying crop residue or will be semi-bare/bare, thus absorbing a considerable amount of heat. This is in agreement with the findings of Mushore et al. [61
] in Harare, which showed that, during the hot dry season, croplands act as a heat source as they absorb and release large amounts of heat due to negligible evaporation. Although areas under agriculture have reduced between 1998 and 2015, the CI has remained positive and increased, implying an increased warming contribution to the city. This could be because the temperature of these areas has increased over the years with the changes attributed to global warming. Early planting of crops means that by the dry season the residues will be completely dry, resulting in high heat absorption, which could also be another explanation. However, the decrease in area under agriculture may indicate a shift of agriculture to the secondary industry and services in both cities. In other cities such as Harare [9
], growth is also characterized by the major replacement of dense vegetation and agriculture areas with building and impervious surfaces, resulting in warming. Therefore, the surface warming mostly of Freetown between 1998 and 2015 can be attributed to global warming, the warming effect of dry agricultural land, and increase in the built-up area which absorbs a significant amount of heat. This agrees with Jiang and Tian [62
], who demonstrated that the construction of buildings leads to the transition of an area from a dense vegetation low temperature to sparse vegetation high temperature zone.
Even in coastal cities where the water table is presumed to be high and sea breezes cool the atmosphere, a high density of buildings still causes warming. Although Freetown is larger in population size as well as built-up extent and also growing faster, it was cooler than Bo town in all periods (by about 2 °C). The difference could be a result of surface moisture and cold air advection due to proximity to the sea. Surface wetness reduces the temperature of a surface due to increased evaporation and latent heat transfer [56
]. According to Rasul et al. [56
], green areas and water bodies act as urban cool islands, hence the low temperature of Freetown despite being larger in size than Bo town. Besides being close to the sea, the proportion of dense vegetation cover is greater in Freetown than Bo town, which reduces the average temperature of the city. According to Sithole and Odindi, green spaces act as heat sinks, tend to be porous, and assimilate heat. Due to the influence of the sea, dense buildings and high surface temperature are found along the coast in Freetown. This has also led to the sustenance and expansion of a tongue of dense green area and low temperature in the central part. Water and vegetation which surround the built-up area of Freetown act as a sink to these gases, which may also explain the lower temperature there than in Bo town. According to Odindi et al. [7
], the heat contribution of dense vegetation is similar to that of water, hence Freetown is surrounded by cool areas resulting a in lower mean surface temperature than Bo town.
We have compared urban growth and land surface temperature patterns between a coastal city (Freetown) and an inland city (Bo town) in Sierra Leone in this paper. Multi spectral Landsat data are used to quantify land use and land cover, as well as surface temperature, changes between 1998 and 2015. Based on the findings of the study, we conclude that multi-spectral Landsat data and the Support Vector Machine algorithm retrieve LULC spatial patterns and urban growth with a high accuracy. The growth patterns of Freetown are concentrated along city margins at the coast, while Bo town expanded from the center outwards. The abundance of dense vegetation and proximity to ocean makes Freetown cooler, although it is larger in population and is expanding in terms of the built-up area faster than Bo town. However, even in cool areas such as at the coast, built-up areas have warmer surface temperatures than non-built-up areas such as dense vegetation areas. Expansion of the built-up area from the city core pushes out vegetation towards the margin, resulting in a high temperature towards the center, as in Bo town. Overall, the built-up area expansion increases urban temperature, in addition to the effect of global warming, while vegetation has a strong heat mitigation effect. The Freetown-Bo town scenario has indicated that it is possible for a small city to be warmer than larger and faster growing cities within the same country. Temperature patterns depend heavily on position relative to ocean, as well as the size and spatial structure of dense vegetation area. Therefore, even vegetation and water patches around a built-up area (not only those within) have an influence on its temperature. Although the study managed to convincingly link urban growth induced LULC changes with LST dynamics, future efforts must address some limitations which could hamper the reliability of the findings. The study depended on medium spatial resolution Landsat datasets, whose temporal resolution of 16 days is low. This, together with the cloud free image requirement for surface analysis, resulted in a limited amount of data available for the study. In the presence of sufficient data, averages could have been computed to eliminate the effects of randomness associated with the use of single date images to represent an entire month. Due to the low temporal resolution of Landsat data, it is difficult to obtain in-situ meteorological data at the exact time of satellite overpass for a comparison of temperatures obtained from remote sensing with in-situ observations of air temperature in Sierra Leone. Meteorological operations in Sierra Leone are still manned; taking observations at World Meteorological Organization (WMO) prescribed synoptic hours which do not coincide with the overpass times of Landsat missions. Limited access to in-situ meteorological data inhibited the analysis to test the validity of the findings of this study, although they agreed with global trends. Reflective bands of Landsat are at a higher spatial resolution than the thermal dataset (for example 30 m versus 100 m for Landsat 8). This mismatch has the potential to increase the mixed pixel problem on LST retrievals, thus compromising the link between LULC (30 m resolution) and LST (100 m), even though thermal data is downloaded at a resolution of 30 m after resampling. Other factors which affect thermal properties such as differences in building material and roof types between Freetown and Bo town were not investigated in this study.