CHAPTER 2: A REVIEW OF IMPLICATIONS OF URBAN GROWTH ON INDOOR
2.7 Future recommendations
Although significant progress has been achieved in remote sensing based LULC mapping, accuracy in mapping urban landscapes stills needs further improvement. The urban landscape is characterized by complex changes in LULC in space and time. Such detection usually suffers from the mixed pixel challenge. In order to improve LULC mapping in urban areas, studies should focus on algorithms, data enhancement and data merging techniques using medium and high resolution remote sensing data sets. Future efforts must include LULC classification using data from recently launched freely available medium resolution sensors such as Sentinel and Landsat 8. Specifically, Landsat 8 has improvements which include improved noise-signal ratio, high radiometric resolution and improved spectral range. Additionally, the sensor has two thermal infra-red bands whose inclusion has the potential to increase mapping accuracy, given the strong relationship between LULC and LST (Xu et al., 2013; Larsen & Gunnarsson- Östling, 2009; Yuan & Bauer, 2007). The options for improving urban mapping include merging optical bands with indices such as NDBI, NDBaI, NDWI and NDVI (Chen, et al.
2006). Other high resolution data such as the 12m Germany TerraSAR-X with a global coverage also need to be tested even in mapping LULC of African and other cities. The potential of remote sensing based LCZ mapping to improve understanding of link between urban surface characteristics and near-surface temperatures needs to be tested especially in developing countries.
Lack of high spatial resolution thermal imagery has significantly constrained surface temperature characterization. Previous urban surface temperature studies have relied heavily on medium resolution sensors such as ASTER and Landsat series with spatial resolutions of at least 60m. As a result, urban surface covers and land cover indices are mapped at a higher resolution than thermal information which strongly affects attempts for their correlation.
Engineering solutions should involve development of thermal sensors with higher spatial resolution. Options may also include launching other polar orbiting satellites with thermal sensors at lower altitude to reduce Instantaneous Field of View (IFOV), thereby increasing spatial resolution. Furthermore, using the relationship between LST and higher resolution variables such as vegetation indices may improve spatial resolution at which thermal conditions are characterized.
32
Although medium resolution thermal datasets are useful for urban surface temperature analysis, low temporal resolution limits their applications for temporal analysis. For example, the number of Landsat images available per period is limited by the 16-day revisit time. This is further worsened by the requirement for cloud free images in order analyze surface conditions (Sattari & Hashim, 2014). This makes it difficult to relate satellite observations from medium resolution sensors with in-situ air temperature data due to comparatively few coinciding observations. Seasonal analysis, for instance, requires a representative number of observations in order to calculate average conditions and avoid errors due to basing on random single date imagery to represent entire season. However, due to gaps caused by limited data quantity, it is difficult to accurately represent seasonal and long term surface temperature properties using medium resolution thermal data. High temporal resolution thermal data characterize low spatial resolution sensors such as geostationary satellites (Sattari & Hashim, 2014). For instance, SEVIRI sensor has a 15 minutes temporal resolution, but due to a very coarse spatial resolution (3km), the thermal data are not useful for characterizing thermal conditions in urban landscapes. Studies should focus on integrating low and medium resolution sensors to build thermal datasets with improved spatial and temporal resolution. This is capable of monitoring spatial and temporal variations of temperature in heterogeneous urban landscapes. This will improve temperature analysis at seasonal and longer time scales critical also for prediction of future thermal conditions.
Previous studies have attributed long term changes in urban surface temperature to LULC transitions without resolving the contribution of other factors such as global warming.
Similarly, global and regional climate models have mostly left out urban meteorological stations and assumed that their contribution is negligible (Nayak & Mandal, 2012). Where urban areas are considered, emphasis is placed on the contribution of anthropogenic emissions to greenhouse gas concentration levels. However, studies like Pielke et al. (2011) have shown that impacts of LULC changes is superimposed on already changing temperatures due to among others the general global warming patterns. For example, Ogrin and Krevs (2015);
Pielke et al. (2011) observed that long term changes in urban temperature were faster than other areas. Future studies need to develop methods to separate effects of LULC conversion from other causes of temperature increases such as changes in GHG concentrations and ozone
33
depletion in urban environments. This has potential to improve urban planning, growth policies and strategies as well as modeling of the urban micro-climatic changes.
Impact assessments such as extreme heat vulnerability mapping should enhance use of remotely sensed contributing factors and improve the spatial resolution. Available literature such as Johnson, et al. (2014) produced heat vulnerability maps at low spatial resolution for instance, at district and census block scale. However, socio-economic status and physical exposure are not uniform within each sub-division. Hence there is need to improve spatial details of input factors in order to produce maps that show variations found even with census blocks. Remote sensing, especially using medium and high spatial resolution sensors should be capable of mapping heat exposure factors at higher resolution than census blocks. Studies need to identify heat vulnerability factors detectable using medium and high resolution dataset and establish methods to combine them with other data to improve the mapping of heat vulnerability. Due to the link between surface properties and temperature, remotely sensed variables such as NDVI and NDBI have been incorporated in vulnerability studies. However, where this was done, focus was not placed on improving the spatial resolution of the maps generated. Detailed heat vulnerability maps are important for tailoring interventions based on levels of need, which eliminates poor strategies caused by generalization at coarse resolution.
Assessments of thermal discomfort and impact on air conditioning energy demand have most widely been executed using in-situ observations of temperature (Vardoulakis, et al., 2013).
However, research has shown that surface temperature modulates near surface (2m) air temperature and there is strong correlation between the two (Pielke, et al., 2011; Marland et al., 2003; Cai, et al., 2011). Future studies should use this relationship to upscale temperature observations. In that case, the use of medium resolution thermal data will enable mapping of the spatial distribution of temperature as well as derivation of responses such as discomfort and energy consumption. For example, up-scaling temperature observation will enable analysis of the variations in degree days, hence air conditioning energy between urban landscapes. The use of medium resolution remotely sensed thermal imagery in estimating temperature has strong potential in mapping spatial variations with higher reliability than point data. This is also important for analysis of thermal conditions in resource constrained cities where stations are usually undesirably sparse.
34
Remote sensing based thermal studies have mostly been focused on single date, seasonal and long-term but historical analysis of the relationship between urban LULC patterns and LST (Xu, et al., 2013; Larsen & Gunnarsson-Östling, 2009; Yuan & Bauer, 2007). Land cover indices have also been used to show the quantitative relationship between LULC conversion and temperature (Weng, et al., 2004; Chen, et al., 2006; Yuan & Bauer, 2007; Tran, et al., 2006;
Xiao et al., 2007; Zhang, et al., 2012; Hung, et al., 2006). Thus, previous studies have mostly stressed on historical changes in temperature in response to urban growth and proved that surface alterations result in warming (Yuan & Bauer, 2007; Hu & Jia, 2010; Valsson & Bharat, 2009; Odindi, et al., 2015). However, there is paucity of literature on future LULC and temperature patterns using remote sensing. Due to improvements in sensor technology, data availability and better spatial coverage than meteorological stations, there is need to use freely available medium spatial resolution remote sensing data sets such as Landsat and Sentinel-2 Multi-Spectral Instrument (MSI) in models to predict future LULC and temperature distribution. This is important for planning, especially in developing countries often characterized by rapid urban growth and limited spatial data.
Ahmed et al. (2013) used NDVI as a predictor of future LST using the Markov Chain Analysis.
Although the approach is unique, NDVI has limitations which include saturation when vegetation fraction is high. The study did not test the potential of other indices, hence there is need to test other indices like NDBI and NDWI or a combination of various indices in multi- variate regression to predict future temperature. In cases where more than a single index is used, there is need to identify a set of indices that most accurately predict LST with optimal errors and minimal multiple collinearity issues. This will complement efforts made by greenhouse gas concentration based models to predict future temperatures. This will also enhance understanding of the implications of urban growth patterns and associated LULC changes on future climate even if greenhouse gas concentration levels are to remain constant or be reduced in effect.