CONTRIBUTION INDEX OF LAND COVER AND LAND SURFACE TEMPERATURE CHANGES IN UPPER HILL NAIROBI, KENYA

Urban heat island is the difference in thermal temperature between rural and urban areas. The urbanization process alters the material type with impervious surfaces being absorbers of incoming radiation during the day and emitting it at night. The research involved the use of time-series satellite imagery from Sentinel, Landsat, ASTER and MODIS for the period 1986, 1995, 2000, 2005, 2011, 2015 and 2017 over the Upper Hill, Nairobi. Morning, afternoon and night land surface temperatures (LST) were calculated for each of these years and analyzed together with the land cover. The mean albedo was calculated to determine the relationship between each land cover and mean LST. The contribution index was calculated to determine whether a land contributed positively or negatively to the mean LST in Upper Hill. Results indicated that built-up land cover had increased from 1986 to 2017 by 0.86% per annum while forest land cover had decreased by 0.99% per annum. Sparse grassland had higher albedo and LST values of 0.81 and 27.9 C respectively, whereas water had lower albedo and LST values of 0.09 and 25.1 C. Water had the lowest mean LST during the day but highest mean LST in the afternoon and night in each of the years due to its high thermal capacity. Bare ground tends to have a higher contribution index compared to other land covers, while forest land cover has a negative contribution index, indicating the impact land cover types have on LST and the urban heat island effect.


INTRODUCTION
Human activities have led to irreversible changes to the ecosystem and it is likely that further damage will occur (Schipper et al., 2010). Perceived changes in temperature, sea-level rise have been reported in many cities and this has been attributed to climate change (Carmin et al., 2012). Whether or not causes of climate change can be attributed to anthropogenic activities, it is evident that there are new climate and weather patterns. This potentially puts urban residents at risk, affecting their health and wellbeing, while stressing the environment, buildings and other assets (Carmin et al., 2012).
The earth's energy budget is affected by human activities that alter properties of the land surface and pollutant emissions (Wu & Lung, 2016). In cities, urban heat islands (UHI) has been intensively studied and involves computing the differences in surface and air temperature between nearby rural and urban areas (Zhan et al., 2015). Boundary layer urban heat island (BUHI), canopy urban heat island (CUHI) and surface urban heat island (SUHI) are the main UHI identified, with SUHI being great both day and night (James, 2000). Differences in cooling between urban and rural areas results to formation of urban heat (UH) in the air (Voogt, 2007) . The land surface temperature (LST), which considered an important parameter in urban climate, directly controls UHI (Feizizadeh & Blaschke, 2013).
Spatial metrics enable correct interpretation about patterns and structure of changes such as in patch sizes, landscape fragmentation etc. (Koukoulas et al., 2008). Spatial metrics are impactful tools when combined with remote sensing techniques to link urban landuse patterns and dynamic processes (Zhao & Murayama, 2011). Studying patterns of land cover changes enables one to quantify the effects and patterns of these changes. The urbanization process alters the surface albedo, leading to increased air temperatures compared to rural areas, resulting to UHI effect (Bhargava & Bhargava, 2018). Vegetation is thus considered an important part of the urban landscape due to its effects on air quality, temperature and air humidity (Mishra, 2009). Tree cover and natural vegetation, through the evapotranspiration process, effectively cools surrounding areas (Grover & Singh, 2015).
The effects of increased urbanization, on land surface temperature needs to be evaluated. Contribution of these changes on the UHI is important in developing adaptive and mitigative measures on the climate change challenges facing the world in the 21 st century and in future.

Study Area
The study area is Upper Hill, Nairobi Kenya and is 4.2 square kilometers and has an average altitude of 1700 meters above sea level. Nairobi has two rainy seasons: 'long' rainy season in March-May (MAM) and a 'short' rainy season in October-December (OND). Predominant winds are easterlies which are linked to moisture inflow from the Indian ocean into the country (Ongoma et al., 2016). Average temperature and rainfall data between 1984-2013 from Wilson meteorological station indicates that the coldest season occurs in June-August (JJA) (Figure 1) due to the advection of cold air from the southern hemisphere (Ongoma et al., 2016). Higher temperatures occur in February-March ( Figure 1).

Figure 1: Mean monthly temperature and rainfall from Wilson
Station, Nairobi Upper Hill has seen rapid transformation from residential use to commercial, office and government institutions, with high-rise buildings being the main development (Mwangi et al., 2020). Changes in zoning policies on plot ratio (PR) and ground coverage ratio (GCR) have contributed to land use changes and eventual lover changes. Proximity to the central business district (CBD) has also contributed to its transformation with pressures from businesses relocating and eventual construction of high-rise buildings (Mwangi et al., 2020;Kiai et al., 2008). The transformation has resulted in urban greenery being transformed to impervious surfaces (Karanja & Matara, 2013), resulting to increased surface temperatures. Effects of land cover changes on the albedo and eventual surface temperature needs to be modelled for necessary mitigative measures. Lack of indicators on the interaction of land cover with incoming radiation becomes a challenge in formulating policies on dealing with changing climate in urban areas.

Processing of Satellite Imagery
Satellite imagery was downloaded from USGS website and processed in respect to the datasets required for the analysis. A summary of the image processing processes is shown in figure 3. Satellites acquired for the analysis included: Landsat 5, 7 and 8; ASTER; MODIS and Sentinel-2 imagery. Sentinel-2 was only used in land cover analysis as it does not have thermal bands. All downloaded imagery was transformed from UTM WGS84 Zone 37 N/S projection system to UTM Arc1960 South in ArcGIS 10.7, after which the area of interest (AOI) (Figure 2) was clipped.

Land Cover Classification
Cloud-free level-1 tier satellite imagery was acquired from USGS for different sensors on different days in the months of January and February. These are considered warms month in the year in Kenya (Table 1). These datasets were also used in determining the LST and albedo over Upper Hill. Cloud-free satellite imagery collected during the cold seasons was not available within the same years as datasets collected in table 1. Hence only imagery collected in January and February were used. Five land cover were identified following the Food and Agriculture Organization (FAO) land cover classification system (LCCS) in Upper Hill namely: forest, built-up, water, open grassland and sparse grassland. These were used to define training sites. Cropped imagery and training sites were imported in R studio for land cover classification using random forests. Classification accuracy was determined using the confusion matrix. Land cover changes were carried out between the years 1986-1995, 1995-2000, 2000-2005, 2005-2011 and 2011-2017. Land cover changes included: vegetated to built-up, forest to open grassland, forest to sparse grassland and no change. A minimum mapping unit (MMU) was applied to eliminate small land cover (LC) pixels, where 1986 and 1995, 1995 and 2000 an MMU of 0.09 Ha was used as the spatial resolution of Landsat is 30m. In 2000-2005 and 2005-2011 an MMU of 0.045 Ha was applied as ASTER has a spatial resolution of 15m. In 2011-2017 an MMU of 0.0225 Ha was applied as Sentinel's spatial resolution is 10m. LST changes between the same year intervals were spatially linked to the land cover changes. This was to determine whether changes in land cover had subsequent effects on the changes in LST in areas where changes had been mapped.

Land Surface Temperature Computation
Day, afternoon and night LST was analyzed to determine changes in minimum, maximum and mean temperatures. Satellite imagery indicated in table 1 were used to determine the morning LST, except for Sentinel-2.  (Cristóbal et al., 2018). ASTER imagery was processed in ENVI 5.1 where all thermal bands were analyzed together using the three-step method to calculate the LST. Landsat 5 TM was converted to Landsat 7 ETM+ equivalent data to compute the top of atmosphere (TOA) (Equation 1). This has been elaborated by Vogelmann et al., (2001), with Firl & Carter (2011) describing how one can compute to Landsat 5 TM from Landsat 7 ETM+.

Date
Where DN7 and DN5 are the digital numbers of Landsat 7 ETM+ and Landsat 5 TM respectively.
The slope and intercept values are band specific and these have been calculated by Firl & Carter (2011). Band 6, the thermal band, was not recalculated.

Converting to Top of Atmosphere (TOA) from DN:
The digital number (DN) spectral information was transformed into TOA radiometric values for Landsat 7 (Equation 2) and Landsat 8 (Equation 3). Band 6a in Landsat 7 was used for the analysis as it has a low radiance variance compared to band 6b (Patricia Wanjiku Mwangi, Karanja, & Kamau, 2018).

Albedo Computation
The albedo, which is the surface material type of a surface, for each of the five land cover classes for the years 1986, 1995, 2000, 2005, 2011 and

Contribution Index Computation
The contribution index (CI) indicates whether a land cover type contributes positively or negatively to the surface temperature in the study area. A positive value indicates that the land cover positively contributes to the LST in the study area while negative values indicate that it negatively contributes to the surface temperature and therefore lowers it. CI was determined using equation 16.

CI = A * TD
Where A is a ratio between the land cover area and the total area of the study area, while TD is the difference in mean temperature between a land cover and the study area.

Land Cover Change
Land cover changes from 1986 to 2017 showed an increase in builtup land cover of 0.86% per annum while forest land cover decreased by 0.99% per annum. Water bodies did not show any significant changes as there is only one within the study area used for recreational purposes. Open and sparse grassland showed alternating fluctuations between the years due to changes in climatic conditions (Table 3).    This was attributed to dry conditions in each of these years, with 2010 to 2011 drought being the worst drought in 60 years (Mbogo et al., 2015). 1986 had lower mean LST which indicated increased mean LST thus changing climatic conditions (Table 4).   However, the amount of moisture in the soil affects the surface temperature (Tian et al., 2012). Forest land cover and water bodies had a low LST with a mean of 23.5 0 C in 1986. Forests have a cooling effect due to evapotranspiration while water takes longer to warm up due to its large thermal heat capacity thus taking longer to warm up. Water had the lowest temperature ranges indicating minimal differences between the maximum and minimum calculated LST, while forests had the largest ranges in all years ( figure 6).

LST and Land Cover
a. b. c. d. e.
f.  (Table 5). This was attributed to the dry conditions experienced in 2000,2005 and 2011 which indicates a similar trend with the morning LST (Table 4).  Figure  7). Results showed that water had the highest LST in all years, both afternoon and night. Its high thermal capacity allows it to absorb and store heat from incoming solar radiation, slowly emitting it in form of long wave radiation. This creates a warming effect in the surrounding environments. This is contrary to LST of water taken in the morning at 10:00 am where it had the lowest mean LST ( Figure 6).  In the afternoon, forests LST was slightly higher than built-up, which had the lowest LST in 2005 and 2011, while in 2017, forest LC had the lowest LST at 40.37 0 C. At night, forest land cover had higher LST than built-up, open and closed grassland. Forests also have a warming effect as seen in the figure 7, as they release heat energy that has been stored during the day (Li et al., 2015). However, this warming effect tends to be lower than its cooling effect during the day. Trapped humidity and heat within urban canopy layers results in cool day-time and warmer night-time temperatures of forests (Sodoudi et al., 2018). In the afternoon, sparse and open grassland had lower LST than water since heat is conducted downwards during the day in soils and conducted upwards at night. This means that at night, open grounds tend to cool faster than other land covers (Sodoudi et al., 2018). This can be noted in night LST where open and sparse grassland have the lowest LST in all years (Figure 7).

2011-2017
Temp ( (Trlica et al., 2017). The amount of solar insolation is reflected back into the atmosphere is determined by the albedo (Odunuga & Badru, 2015). Figure 8 shows that forest land cover had the highest negative contribution index indicating that it does not contribute to the increased LST but instead reduces the surface temperature. However, its CI decreased from -0.40 in 1986 to -0.15 in 2017. This could be attributed to a reduction in percentage coverage in the area (Table 3).

Figure 8: Contribution index of land cover
Sparse grassland had the highest positive contribution index indicating that it contributed to the increased surface temperature over Upper Hill (Figure 8). Water land cover did not show any significant contribution in 1986 and 1995. However, between the years 2000 and 2005 it had negative values of -0.02 and -0.01, indicating it had a cooling effect over the surface. Open grassland has a positive contribution index between 1986 and 2000 indicating that it had an overall warming effect over the land as it was higher than the mean LST over Upper Hill. In 2005 and 2011, it has a negative contribution index of -0.09 and -0.02 respectively, indicating that it had a lower surface temperature than the mean LST over Upper Hill. Built-up land cover has a positive contribution index from 1986 to 2011. However in 2017, it has a negative contribution index of -0.02 due to increased impervious surfaces which increases the thermal storage of the area (Yang et al., 2017). Built-up land cover, during the day, absorbs and stores thermal energy resulting to negative contribution index.

CONCLUSION
This paper investigated the relationship between changes of land cover on land surface temperature and its effect on the contribution index. Multi-temporal satellite imagery are important in change detection and analyzing patterns of development, which enables planning decisions based on environmental conditions. The study was limited to analyzing imagery available during the warmer periods of the year but not the colder seasons due to cloud cover. The study determined that policy changes on built-up development have encouraged expansion of urban areas. This has mostly affected vegetated land cover with its reduction with increased urban development. Zoning policies in urban areas should consider performance zoning that would ensure that development does not occur at the detriment of vegetated land cover. There is need for a comprehensive environmental management plan in urban areas that would handle issues of greening in urban areas. Open spaces are important in urban areas and there is need of their protection and maintenance due to their importance in mitigating effects of heat islands. Forest land cover has experienced the greatest percentage decrease corresponding to increases in built-up density, resulting to a corresponding increase in LST. Increases in temperature occur where land cover changes from vegetated to impervious surfaces due to increases in mean albedo. Albedo influences the land surface temperature with vegetated areas having a lower albedo than builtup areas. Surface albedo is thus an important parameter in the studies of urban heat island as it affects the surface energy budget of a material. The interaction between land cover and incoming radiation is affected by the time of day due to the inherent characteristics of the land cover. It is therefore important to analyze both day and night surface temperatures to determine the relationship between material type and LST. This is comparable with Zhao et al., (2015) where high albedo roofing materials in urban areas provide a cooling effect. Reducing the urban greenness of a city can have adverse effects and its potential effects can be determined on its residents from the contribution index. Increasing built-up densities increases the thermal storage within urban areas, resulting to warmer nights. Water bodies and tree cover are important as they reduce the surface and ambient temperature of an area. Analyzing the interaction of land cover and incoming radiation at different times of the day ensures a better understanding of the dynamics that changing landscapes can have on the thermal comfort of urban residents, and impacts of climate change.