r (OLI) and Thermal Infrared Sensor (TIRS) band data
Figure 2. Flow chart of general methodology.
were converted to Top-Of-Atmosphere (TOA) special radiance rescaling factors provided in the metadata file. The first step was conversion of the DN value to the radiance units using Equation (2).
where, Lλ is the TOA special Radiance(watts/(m2 srad µm)), ML is the Band specific multiplicative rescaling factor from metadata (Radiance_Mult_Band_x where x is band number), AL is the Band specific additive rescaling factor from metadata (Radiance_Add_Band_x where x is band number) and Qcal is the Quantized calibrated pixel value in DNs.
2) Conversion of Radiance to at Satellite Temperature
The next step during quantitative analysis was conversion of radiance to temperature. The satellite temperature was calculated under an assumption of unity emissivity using pre-launch calibration constants K1 and K2 by the following Equation (3):
where, T is the satellite brightness tempreture (K), K2 is the calibration constant (K) and K1 is the calibration constant (W/m2 sr µm).
3) Land Surface Temperature
Land surface temperature was computed from calculated emissivity values and calculated temperature at satellite using the following Equation (4):
where, TLST → Surface temperature (K), TB = at satellite temprature, W is the wavelength of emitted radiance (11.45 µm), ρ = h*c/k (1.438 × 10−2 m・k) with σ = stefan Boltzmann constant (1.38 × 10−23 J/K), h is the plank’s constant (6.26 × 10−34 J・s) and c is the velocity of light (2.998 × 10−8 m/s).
2.2.3. Regression and Factor Analysis
All the five parameters extracted from the physical and secondary datasets were ﬁrst entered to run ordinary list square correlation analysis. To shed some light on their qualiﬁcation on the study, Coefficients have the expected sign, statistically significant, no redundancy among explanatory variables, residuals are normally distributed, residuals are not spatially auto-correlated and strong adjusted r-squared value summarizes their correlation with all the variables that have been evaluated. For land-use/land-cover, the following methods of data analyses were adopted for the study using Maximum Likelihood classification, accuracy assessment with 250 reference points calculation of the area in hectares of the land-use/land-cover types obtained for each study year and subsequently comparing the results and finally annual rate of change (ha/year) = (% of recent land-use/land-cover change −% of previous land-use/land-cover change)/100 * # (15 years).
2.2.4. Overlay Analysis
Overlay Analysis tools included in the spatial analyst extension commonly used to solve multi-criteria problems such as optimal site selection or suitability modeling. It is a technique for applying a common scale of values to diverse and dissimilar inputs to create an integrated analysis  to address which locations are most exposed to environmental degradation. The first step was to model operators using ordinary least squares linear regression to generate prediction to model a dependent variable in terms of its relationships to a set of explanatory variable. Then to generate risk-map of environmental quality, parameter layers were computed by reclassifying as population density, building density, urban heat, greenhouse gas density and land-use/land-cover classification. The overall accuracy for land-use/land-cover of 1986 image classification was 88.80%, of 2000 was 85% and of 2015 was 84.00%. All data groups summarizes as per priority classes, points and influence (weight) values used for weighted overlay to come up with the final environmental risk map of the study area, to show the level of environmental quality of the study area. According to this index, the regions with high greenhouse gas emission, overpopulated, high building density, high heat intensity and low vegetation cover area would get the lowest values, and they will be considered as critical areas. Sparsely populated, low heat, less built-up, low greenhouse gas emission and high vegetation cover areas are considered as environmentally suitable areas.
3.1. Findings from Regression Analysis
Table 1 summarizes the correlation with all the variables that have been evaluated. The correlation coefﬁcients inversely correlated with the dependent variable for most of the tested factors. Probability and robust probability asterisk (*) indicate a coefficient is statistically significant (p < 0.01). Large variance inflation factor (VIF) values (>7.5) indicate redundancy among the explanatory variables. While there was inflation between population (14) and building (17), variable with a VIF > 7.5 have removed by one. The R-squared measure of model fit/performance values showed 92%.
3.2. Findings from Factor Analysis
The results of 1986, 2000 and 2015 classified images are presented in Figure 3. The results show thick vegetation, cropland and bareland have declined whereas the area of settlement has shown increase in extent (Table 2). From the land- use/land-cover distribution of the year 1986, 2000 and 2015, a total of 51,957.92 ha of thick vegetation/forest land had declined from the initial 11,040.5 ha in 1985 to 5238.90 ha in 2000 and 2730.42 in 2015. Urban cropland/grassland extent
Table 1. Summary of the results of urban environmental quality regression analysis.
Note: *indicate probability (b) and Robust-pr (b) variable correlation is significant (P < 0.01).
Table 2. Land-use/land-cover distribution of the year 1986, 2000 and 2015.
Figure 3. Land-use/land-cover change maps for the years 1986, 2000 and 2015.
was 20,705.40 ha in 1986, which declined to 18, 835.10 ha and 10,756.7 ha in 2000 and 2015, respectively. On the other hand, water body has increased from 4416.84 ha in 1986 to 5870.1 ha in 2000 and to 6836.6 ha in 2015. Settlement area had the highest increase from the initial estimate of 7005.60 ha in 1986. The extent has increased to 14, 037.6 ha in 2000 and to 25,715 ha in 2015. This was the most dynamic land-cover in Addis Ababa, which got extended year after year. The extent of bareland showed slight decline from 8789.58 ha to 7976.22 ha and to 5919.2 ha under classification of the years 1986, 2000 and 2015, respectively. The overall accuracy for land-use/land-cover was 88.80% in 1986, 85% for 2000, and 84.00% for 2015. Overall land-use/land-cover changes are given in Table 3.
Figure 4 shows the information derived from landsat images. The shades of
Table 3. Assessment of changes of land-use patterns during 1986-2015.
Note: Annual Rate of Change (ha/year) = (% of recent land-use/land-cover change-% of Previous land- use/land-cover change)/100 * # of Study years (15 no of years).
Figure 4. Normalized difference vegetation index maps for the years 1986, 2000 and 2015.
white represent man-made surfaces (concrete, roofs and roads) and water bodies. Light green represents grassy fields and dark green represents tree covered areas. The pixel values of 1986 NDVI image ranged from around −0.34 for not very vegetated surfaces to values of 0.63 for vegetated surfaces. For 2000, the pixel values for the NDVI images ranged from −0.14 for not very vegetated surfaces to 0.51 for vegetated surfaces. For 2015, the pixel values for the NDVI image ranged from −0.52 for none and not very vegetated surfaces to 0.57 for vegetated surfaces.
Green area distribution patterns in Addis Ababa city are shown in Figure 5 and Figure 6. Yeka and Gulele sub-cities have only about 0.005% of green area, Akaki-Kality sub-city, Nifassilk-Lafto and Bole have only less than 0.002% green area. Arada Kirkose and Lideta sub-cites hold too low green areas and Addis Ketema sub-city has almost no vegetation cover while Kolfe Keranio has 0.03% green area.
Population density in Addis Ababa city in relation to that of the ten sub-cities indicates that the central region is the densest. Addis Ketema with >350 people/ ha is the most populated region in Addis Ababa. Arada and Lideta sub-cites have 300 people/ha, while the Akaki-Kality, Yeka and Bole sub-cities have <17 people/
Figure 5. Map of urban green space in Addis Addis Ababa (2015).
Figure 6. Comparison of urban green space in 10 sub-cities of Addis Ababa.
ha. There is a general increase of surface temperature in the impervious surfaces (building, concreat and asphalt) in comparison with vegetated areas. Maximum average surface temperature recorded was 36.79˚C and minimum was 17.40˚C. Based on zonal mean statistics, the surface temperature in Akaki-Kality and Bole sub-city was 29˚C, followed by Addis-Ketema, Arada, Kirkose, Liedeta and Nifase-Silke (28˚C), Yeka (27˚C) and Kolfe (26˚C), respectively, while Gulele sub-city experienced the lowest surface temperature (24˚C) in Addis Ababa.
Greenhouse gas emission data are shown in Figure 7. Concentration of greenhouse gas emission was high associated in industrial and residential areas and in major roads. Based on this, the maximum greenhouse gas emission recorded was 1.41 mega tone and the minimum was 0.25 mega tone. Based on the zonal mean statistics, Akaki Kality sub-city was the most polluted (0.24 mega tone), followed by Addis-Ketema and Lideta (0.21 mega tone), Kirkose (0.14 mega tone) Gulele (0.04 mega tone), Kolfe Keranio and Yeka (0.03 mega tone), and Bole, Nifasi-Silke and Akaki-Kality (0.02 mega tone).
3.3. Overlay Model
Based on weighted overlay analysis, there was a need to have a universal integration result of the five parameters (land-use/land-cover, population density, surface temperature, greenhouse gas emission and built-up density) to sum up based on the weighting (Table 4 and Table 5) and to isolate the hotspot areas of environmental deterioration the risk-map generated (Figure 8) reclassify as most risk, more risk, risk, marginal and least risk. Based on this classification Addis Ketema belongs its between most risk, Arada more risk, Lideta more risk, Kirkose within risk and Gulele among marginal and least risk and Kolfe, Yeka, Nifas Silke, Bole and Akaki Kality sub-cities belong to least risk. The consistency ratio (CR) indicates the probability that the matrix ratings were randomly generated so that matrices with CR ratings greater than 0.10 should be reevaluated.
Figure 7. Comparison of greenhouse gas emission in the ten sub-cities of Addis Ababa.
Table 4. Derived factor maps and ranking.
Table 5. Derived factor maps weight.
Consistency Ratio (CR) is 0.02.
However, in this study the CR calculated as 0.02 is acceptable. Based on the urban environmental quality indicator rating (Table 5), sub-cities with fewer green spaces, over polluted, high heat accumulation, high greenhouse gas emission and densely built-up areas were identified as critical areas and those with high vegetation, low greenhouse gas emission, low heat island, low built-up area and sparsely populated locations as environmentally safe areas (Figure 8). Based on these results, sub-cities Arada and Addis Ketema were in the category in between most and more critical, Kirkose and Lideta sub-cities were within critical and marginal and Kolfe, Gulele, Yeka, Nifas Silke, Bole and Akaki Kality were least critical.
During the present study, a comparison was made between the increase in impervious surface against the reduction in greenness in Addis Ababa city areas that might signify deterioration in the environmental quality of this African city. Vegetation, crop land/grassland, bare land, settlement and water bodies were either increased or decreased in extent in different rates during 1986-2015 in
Figure 8. Environmental quality map of Addis Ababa city.
Addis Ababa city area. Greenness of the area serves as a means of urban environmental quality    . Information on green area in cities is important for maintaining urban environmental quality. Vegetation is very useful in filtering air, water and sunlight, cooling urban heat, recycling pollutants, moderating local urban climate and providing recreational areas for the people. The ecological functions of vegetation cover in urban parks may filter up to 85% of the air pollutants   .
Settlement areas increased from the initial estimate of 7005.60 ha in 1986 to 25,715.0 ha in 2015 with the change rate of 468.8 ha/year during 1986-2000 and 778.49 ha/year during 2001-2015. This was the most dynamic land-cover change observed in Addis Ababa city area, the cause of which is none other than urbanization and rapid population growth in the recent past. Unbalanced and uncontrolled urbanization lead to environmental degradation in the quality of urban life  . The extent of bareland declined slightly from 8789.58 ha 7976.22 and 5919.2 of the total area under the classification of the year 1986, 2000 and 2015, respectively. This is mainly due to afforestation and rehabilitation conservation programmes. Normalized Difference Vegetation Index is a spectral index that takes into account the amount of infrared rays reflected by plants   . Green plants absorb solar radiation, as a source of energy in the process of photosynthesis. The reason for the relationship of NDVI to vegetation is due to the property of green leaves to reflect 20% or less in 0.5 to 0.7 microrange (green to red) and about 60% in 0.7 to 1.3 microrange (near-infrared). Normalized Difference Vegetation Index itself varies between −1.0 and +1.0. Negative values of NDVI (values approaching −1) correspond to deep water. Values close to zero (−0.1 to 0.1) correspond to bare land areas of rock and sand. The nature and extent of these changes show variations across time and space. Changes attributed to socio-economic and environmental parameters reflect a host of other factors. In Addis Ababa city, Yeka and Gulele sub-cities contains about 0.005% of green area, while Kolfe-Keranio cover 0.03% green area and Akaki-Kality, Nifassilk-Lafto and Bole contain <0.002% green area. Other inner areas in the city such as Arada Kirkose and Lideta hold too low green areas and Addis-Ketema sub-city contains almost zero or no vegetation cover. Green space in Kirkose, Addis-Ketema, Lideta, Arad and kolfe having least green space (4.8 × 10−6 square meter) are the poorest. Gulele and Yeka have <1.5 × 10−5 square meter vegetation area. The suggested standard of green area for cities is 7 to 12 square meters  . This indicates that none of Addis Ababa sub-cities contains the required extent of green area to maintain environmental quality. Over all, the results point to a decline in vegetated areas in Addis Ababa city area, and an increase in settlements, cropland, bare land, population explosion, as well as greenhouse gas emission, which act as threat to the environment of the city. There is a need to have a universal integration of the five parameters analyzed. Overall housing density is a more important negative predictor of the extent of green space and tree cover  . Weighted overlay methodologies can also be customized for reclassification, and map algebra techniques expand the ability of spatial analyses. The accuracy of the study can be enhanced by using high resolution satellite imagery. Not only the urban environmental quality, but also other subjects important for ecological and environmental impact analysis such as air, water, and noise pollution limitations can also be determined using RS and GIS integration. Environmental risk and quality map are also source of data for city and regional planners.
In the light of the present findings, the practical use of geospatial approach involving natural and census data, GIS and remote sensing in tracking urban environment quality change stands appropriate for sustainable environmental planning, considering the little efforts in the past to assess urban environmental quality. Integrated data analysis using remotely sensed satellite imagery and GIS modeling facilitated the analysis of the spatial distribution of environmental changes involving land-use/land-cover classification changes from time to time. Results of this study can influence policy assessments and assist local governments and environmental agencies in monitoring Urban Environmental Quality (UEQ). The UEQ models established in this paper can be applied to assist urban planners not only to evaluate the city’s current UEQ condition, but also to devise efficient development polices to construct a more desirable future UEQ environment at the national level. Land-use planners and policy makers can thus make decisions towards a more sustainable city featuring green areas, which correspond well as a key factor in determining a city’s UEQ condition.
We are grateful to the School of Earth Sciences, Addis Ababa University for access to the laboratory facilities and support for this research programme. We are also thankful to the Central Statics Agency and Addis Ababa Environmental Protection Agency, for their collaboration in delivering valuable data used in this study. We also thank the anonymous reviewers for their valuable comments to improve this manuscript.