Linking thermal variability and change to urban growth in Harare Metropolitan City using remotely sensed data.
Mushore, Terence Darlington.
MetadataShow full item record
Urban growth, which involves Land Use and Land Cover Changes (LULCC), alters land surface thermal properties. Within the framework of rapid urban growth and global warming, land surface temperature (LST) and its elevation have potential significant socio-economic and environmental implications. Hence the main objectives of this study were to (i) map urban growth, (ii) link urban growth with indoor and outdoor thermal conditions and (iii) estimate implications of thermal trends on household energy consumption as well as predict future urban growth and temperature patterns in Harare Metropolitan, Zimbabwe. To achieve these objectives, broadband multi-spectral Landsat 5, 7 and 8, in-situ LULC observations, air temperature (Ta) and humidity data were integrated. LULC maps were obtained from multi-spectral remote sensing data and derived indices using the Support Vector Machine Algorithm, while LST were derived by applying single channel and split window algorithms. To improve remote sensing based urban growth mapping, a method of combining multi-spectral reflective data with thermal data and vegetation indices was tested. Vegetation indices were also combined with socio-demographic data to map the spatial distribution of heat vulnerability in Harare. Changes in outdoor human thermal discomfort in response to seasonal LULCC were evaluated, using the Discomfort Index (DI) derived parsimoniously from LST retrieved from Landsat 8 data. Responses of LST to long term urban growth were analysed for the period from 1984 to 2015. The implications of urban growth induced temperature changes on household air-conditioning energy demand were analysed using Landsat derived land surface temperature based Degree Days. Finally, the Cellular Automata Markov Chain (CAMC) analysis was used to predict future landscape transformation at 10-year time steps from 2015 to 2045. Results showed high overall accuracy of 89.33% and kappa index above 0.86 obtained, using Landsat 8 bands and indices. Similar results were observed when indices were used as stand-alone dataset (above 80%). Landsat 8 derived bio-physical surface properties and socio-demographic factors, showed that heat vulnerability was high in over 40% in densely built-up areas with low-income when compared to “leafy” suburbs. A strong spatial correlation (α = 0.61) between heat vulnerability and surface temperatures in the hot season was obtained, implying that LST is a good indicator of heat vulnerability in the area. LST based discomfort assessment approach retrieved DI with high accuracy as indicated by mean percentage error of less than 20% for each sub-season. Outdoor thermal discomfort was high in hot dry season (mean DI of 31oC), while the post rainy season was the most comfortable (mean DI of 19.9oC). During the hot season, thermal discomfort was very low in low density residential areas, which are characterised by forests and well maintained parks (DI ≤27oC). Long term changes results showed that high density residential areas increased by 92% between 1984 and 2016 at the expense of cooler green-spaces, which decreased by 75.5%, translating to a 1.98oC mean surface temperature increase. Due to surface alterations from urban growth between 1984 and 2015, LST increased by an average of 2.26oC and 4.10oC in the cool and hot season, respectively. This decreased potential indoor heating energy needed in the cool season by 1 degree day and increased indoor cooling energy during the hot season by 3 degree days. Spatial analysis showed that during the hot season, actual energy consumption was low in high temperature zones. This coincided with areas occupied by low income strata indicating that they do not afford as much energy and air conditioning facilities as expected. Besides quantifying and strongly relating with energy requirement, degree days provided a quantitative measure of heat vulnerability in Harare. Testing vegetation indices for predictive power showed that the Urban Index (UI) was comparatively the best predictor of future urban surface temperature (r = 0.98). The mean absolute percentage error of the UI derived temperature was 5.27% when tested against temperature derived from thermal band in October 2015. Using UI as predictor variable in CAMC analysis, we predicted that the low surface temperature class (18-28oC) will decrease in coverage, while the high temperature category (36-45oC) will increase in proportion covered from 42.5 to 58% of city, indicating further warming as the city continues to grow between 2015 and 2040. Overall, the findings of this study showed that LST, human thermal comfort and air-conditioning energy demand are strongly affected by seasonal and urban growth induced land cover changes. It can be observed that urban greenery and wetlands play a significant role of reducing LST and heat transfer between the surface and lower atmosphere and LST may continue unless effective mitigation strategies, such as effective vegetation cover spatial configuration are adopted. Limitations to the study included inadequate spatial and low temporal resolution of Landsat data, few in-situ observations of temperature and LULC classification which was area specific thus difficult for global comparison. Recommendations for future studies included data merging to improve spatial and temporal representation of remote sensing data, resource mobilization to increase urban weather station density and image classification into local climate zones which are of easy global interpretation and comparison.