1 Introduction

Microclimate is the most important element in modern global climate studies because worldwide climate change is the cumulative result of the impact of local microclimates [1]. Earth’s surface temperature is a result of the balance between incoming solar energy and outgoing radiation energy [2]. Urban heat islands (UHIs) are events resulting from rapid urbanization and are described as urban areas with significantly warmer temperatures than nearby rural areas [3]. There are a number of contributing factors that play significant roles in the creation of UHIs, such as low-albedo materials, wind blocking, air pollutants, human gatherings, and increased use of air conditioners [4].

Rapid urbanization driven by population growth and economic development has led to drastic and widespread changes in the Earth’s surface, causing the replacement of natural surfaces such as vegetation with impervious surface materials such as concrete, asphalt and buildings [5]. Increased replacements of natural greenery areas with urbanized areas have led to significant changes in local climate conditions [6]. A more practical method of mitigating UHIs is the strategic planting of vegetation in urban areas and the design of green technology approaches [7]. Urban greenery also acts as a natural agent against air pollution in the urban environment [8]. The urban planning, preparation, and implementation strategies allocated 30%, 30%, and 40%, respectively, to roads, infrastructure, green areas and shared public use, and building construction in their urban land management plans [9].

In Ethiopia, urbanization causes extreme changes in vegetation cover, hydrological services, and nearby climate scales. The most obvious climatic impact of urbanization is an increase in LST in urban areas relative to that in surrounding rural areas [10]. Most studies neglect the cooling effect of all urban green space patches for mitigating microclimates. In addition to other urban parameters, built-up and green space play a significant role in LST dynamics [11]. By applying the geospatial technique of the land use transfer matrix method (LUTM), the cooling impact of green space patches on UHI mitigation can be properly analyzed [12]. For a long time, Addis Ababa has been the capital city and a hive of economic activities in the country. The city is experiencing a construction boom [13], and observing the scale of development and construction activities occurring on Addis Ababa, Young [14] described the entire city as one construction site. Especially in recent years, the city has witnessed a number of changes with new development and redevelopment programs—the construction of mega structures such as malls and business centers, international hotels, and high ways to create new settlement areas. According to the RUAF Foundation, Addis Ababa is the thirty fastest growing city in the world, with an annual growth rate of 3.4 from 2006–2020.

The urban LST is highly dependent on the physical or thermal properties of an object, and it is vital to determine the size, thickness, and dissemination of green spaces in urban regions to determine their effect on UHIs and determine their connection with the LST. Therefore, the aim of this study was to assess the effects of urban land use and land cover change on microclimate dynamics in Addis Ababa city by considering the parameters of LULC, LST, UHI, NDVI, and NDBI to provide an understanding of the land surface temperature (LST), urban heat island (UHI) and LULC status of the area as inputs for planning and decision-making.

2 Methods and materials

2.1 Description of the study area

Addis Ababa is the capital and the largest city in the central highland of the Ethiopian federal government and on the western edge of the Rift valley in the Eastern Africa region [15]. Addis Ababa covers about 540 Km2 and Geographically, located 38° 40′ 00″–38° 52′ 30″ E and 8° 52′ 30″–9° 50′ 00″ N, with topographic variations ranging from 1700 to 2000 m above sea level (Fig. 1). According to the National Metrological Agency, the mean annual rainfall amounts at the Kotebe, Bole, and Akaki stations are 960 mm, 998 mm and 955 mm, respectively. During the summer from July to August acquired maximum rainfall and minimum during the winter December to February [16] Addis Ababa receives average annual rainfall of about 971 mm [17]. The wet season is from June to August which receives maximum rainfall and the driest season is from December to April.

Fig. 1
figure 1

Location, elevation and estimated population of study area

Full size image

The minimum and maximum mean annual temperatures in Addis Ababa are approximately 12 °C and 24 °C, respectively [16, 18]. The nighttime and daytime temperatures of the city are 10–15 °C and 20–24 °C, respectively, during the dry period [18]. By 2037, Ethiopia’s urban population is expected to triple to 42.3 million [19]. Addis Ababa hosts an estimated 3.2 million people, which is 17% of Ethiopia’s total urban population [19].

2.2 Data sources and description

Landsat image TMs from 1990, Landsat ETM+ images from 2005, and Landsat OLI/TIRS images from 2021 were used for this study (Table 1; Fig. 2). The Landsat images were downloaded from the USGS Earth Explorer website (https://earth-explorer.usgs.gov/), which is freely available to users. All the Landsat images were downloaded with little cloud cover (< 10%) and during the dry season (January, February). The Landsat images were used for LULC change analysis and calculation of LST, NDVI, and UHI.

Table 1 Data sources and descriptions
Full size table
Fig. 2
figure 2

Schematic framework of the methodology

Full size image

2.3 Method

2.3.1 Land use/land cover classification

Landsat images of multispectral bands from 1990, 2005, and 2021 were used for LULC classifications. Landsat images were classified by using supervised classification with the maximum likelihood algorithm [20,21,22]. The Fast Line-of-sight Atmospheric Analysis of Hypercube (FLAASH) module in ENVI 5.3 software was used for atmospheric and radiometric correction by setting some important information from the metadata of respective Landsat images such as the senor type, sensor altitude, flight date and time, pixel size and atmospheric model. The LULC types of the study area were classified as bare land, farmland, vegetation, and built-up area (settlement). The detailed information on the LULC types is provided in Table 2.

Table 2 Description of LULC and UMTs in the study area
Full size table
2.3.1.1 Accuracy assessment

The accuracy aim is to evaluate the quality of the classification output error matrix based on an assessment of the overall accuracy; the producer accuracy, user accuracy, and kappa coefficient (Eq. 1) were utilized to evaluate the pixel-based classification output for LULC classification [23]. A total of 284 sample data points from the field were obtained using GPS for the recent period of 2021, and a digitized region of interest (ROI) polygon from Google Earth pro of each LULC types were prepared for the period 1990 and 2005. The result of an accuracy assessment provides an overall accuracy map based on an average of the accuracies for each class in the map [24].

$${\varvec{OAC}} = \frac{\Sigma Xij}{N}*100,\;\user2{UAC } = \frac{Xii}{{X_{ + i} }}*100,\;{\varvec{PAC}} = \frac{Xii}{{X_{i + } }}*100\;{\text{and}}\;{\varvec{Khat}} = \frac{{\left( {Obs {-} exp} \right) }}{{\left( {1 {-} Exp} \right) }}$$
(1)

where OAC is overall accuracy, UAC is user accuracy, PAC is producer accuracy, Khat is the kappa statistic and represents the total number of samples, Xij is the diagonal value, xi + is the total number of columns, X + i is the total number of rows, r is the number of categories, Obs = represents the accuracy reported in the error matrix (overall accuracy), and Exp = represents correct classification.

2.3.1.2 Trends and rate of LULC change

Trends in LULC classes were the main indicator of how much LULC types had changed in the study area over time [25]. Trends in the LULC types in the study area were calculated by using the final and initial years of the LULC classes. Greater values of final LULC classes were related to positive trends and vice versa (Eq. 2).

$${\mathbf{Trends}} \, {\mathbf{of}} \, {\mathbf{LULC}}\left( {{\mathbf{ha}}} \right) = {\mathbf{X}}2 - {\mathbf{X}}1$$
(2)

where X1 is the initial year and X2 is the final year.

2.3.1.3 Rate of LULC change

The extent and degree of LULC change during the study period in the study area were assessed using the rate of LULC change. In this study, LULC types from 1990, 2005 and 2021 were applied to assess the rate of LULC change in Addis Ababa city. According to [25, 26],the rate of LULC change was calculated to evaluate the degree of LULC change in Addis Ababa city from 1990 to 2020 (Eq. 3).

$${\mathbf{Rate}} \, {\mathbf{of}} \, {\mathbf{change}} = {\varvec{Y}}2 - {\varvec{Y}}1/{\varvec{Z}}$$
(3)

where Y1 and Y2 are the area coverage of LULC in the initial year (Y1) and final year (Y2), respectively, and Z is the time interval between two years.

2.3.1.4 Land use and land cover change matrix

In the present study, LULC change detection for Addis Ababa city was applied by using the land use transfer matrix from 1990 to 2021. The rates of change, loss and gain of LULC within the study period were calculated by using the change matrix in the study area [27, 28].

2.3.2 LST extraction process

2.3.2.1 Conversion of DN in to radiance

Digital numbers are manually converted to at-sensor radiances, then to brightness temperature. The TM and ETM + DN values range between 0 and 255 (for Landsat 5 and 7)

$${\mathbf{L\lambda }} = \frac{{{\mathbf{LMAX\lambda }} - {\mathbf{LMIN\lambda }} }}{{{\mathbf{QCALMAX}} - {\mathbf{QCALMIN}}}} \times \left( {{\mathbf{DN}} - {\mathbf{QCALMIN}}} \right) + {\mathbf{LMIN\lambda }}{ }$$
(4)

where QCAL is the quantized calibrated pixel value in DN, LMINλ is the spectral radiance that is scaled to QCALMIN in watts/(meter squared * ster * μm), LMAXλ is the spectral radiance that is scaled to QCALMAX in watts/ (meter squared * ster * μm), QCALMIN is the the minimum quantized calibrated pixel value (corresponding to LMINλ) in DN, 1 for LPGS products, 1 for NLAPS products processed after 4/4/2004, 0 for NLAPS products processed before 4/5/2004, QCALMAX is the maximum quantized calibrated pixel value (LMAXλ) in DN is the 255. The Digital Numbers (DNs) of bands 10 and 11 from the Landsat 8 OLI was first converted to spectral radiance. Landsat 8 with header file data on Radiance Multiplier (ML) and Radiance Add (AL), the thermal infrared (TIR) band was converted into Spectral radiance (Lλ). Using the approach provided by Chander and Markham [29] and the Landsat 8 OLI science Data Users Handbook. Top of Atmosphere Spectral Radiance (TOA) can be calculated using the formula;

$${\mathbf{L\lambda }} = \left( {{\mathbf{ML}} \times {\mathbf{QCal}}} \right) + {\mathbf{AL}}$$
(5)
2.3.2.2 LST retrieval for Landsat 5 and 7 (Mono Window Algorithm)

Where Lλ is the Top of Atmosphere (TOA) spectral radiance (Wm − 2 sr − 1 μm − 1), ML is the Band-specific multiplicative rescaling factor from the metadata (RADIANCE_MULT_BAND_x, where x is the band number), AL is the Band-specific additive rescaling factor from the metadata (RADIANCE_ADD_BAND_x, where x is the band number), and QCal is the Quantized and calibrated standard product pixel values (DN). According NASA CEOS LST Protocol [30] the land surface temperature of this study was extracted from the Landsat images of 1990, 2005, and 2021 G.C. (Eq. 6).

$${\mathbf{s}} = \left\{ {{\mathbf{a}}6 \left( {1 - {\mathbf{C}}6 - {\mathbf{D}}6} \right) + \left( {{\mathbf{b}}6 \left( {1 - {\mathbf{C}}6 - {\mathbf{D}}6} \right) + {\mathbf{C}}6 + {\mathbf{D}}6} \right) {\mathbf{T}}6 {-} {\mathbf{D}}6{\mathbf{Ta}}} \right\}/{\mathbf{C}}6 = 6{{\varvec{\uptau}}}6 {\mathbf{D}}6 = \left( {1 - {{\varvec{\upvarepsilon}}}6} \right) \left[ {1 + \left( {1 - {{\varvec{\upvarepsilon}}}6} \right) {{\varvec{\uptau}}}6} \right]$$
(6)

where T6 is the brightness temperature, ε6 is the land surface emissivity, τ is the atmospheric emissivity in the thermal infrared band and Ta is the average atmospheric temperature, which can be calculated via the parameter estimation method of the mono-window algorithm [31]. Additionally, a6 = − 67.355351, b6 = 0.458606, C6 and D6 are intermediate variables, and Ts is the land surface temperature (LST) to be calculated.

This is the effective at-satellite temperatures of the viewed Earth-atmosphere system under an assumption of unity emissivity and using pre-launch calibration constants (Eq. 7).

$${\mathbf{T}} = \frac{{{\mathbf{K}}2}}{{\user2{ ln}\left( {\frac{{{\mathbf{K}}1}}{{{\mathbf{L\lambda }}}} + 1} \right)}}$$
(7)
$${\mathbf{BT}} = {\mathbf{K}}2 {\mathbf{ln}} \left( {{\mathbf{K}}1 {\mathbf{L\lambda }} + 1} \right) { }$$
(8)

where BT is the effective at-sensor brightness temperature (K); K2 is the calibration constant 2 (K); K1 is the calibration constant 1 (W/(m2 *sr * μm)); Lλ is the spectral radiance at the sensor aperture (W/(m2 * sr * μm)); and Ln is the natural logarithm. To compute the land surface emissivity (LSE), it is essential to know the characteristics of the Earth’s surface and change the thermal radiance energy during the calculation of the LST [32, 33]. According to [34], the emissivity is calculated using (Eq. 9).

$${{\varvec{\upvarepsilon}}} = {\mathbf{0}}{\mathbf{.004}} *{\mathbf{PV}} + {\mathbf{0}}{\mathbf{.986}}$$
(9)

where PV is the vegetation proportion; PV = [NDVI − NDVImin NDVImax − NDVImin]2 NDVI is the normalized difference vegetation index; and NDVI max is the maximum normalized difference vegetation index and NDVI min is the minimum normalized difference vegetation index. The calculated radiant surface temperature is corrected for emissivity using the following equation (Eq. 10).

$${\mathbf{TB}} 1 \left( {{{\varvec{\uplambda}}} {\mathbf{TB}}/{\mathbf{P}}} \right) {\mathbf{In\varepsilon }}$$
(10)

where LST is the land surface temperature, TB is the radiant surface temperature in K, λ is the wavelength of the emitted radiance (11.5 μm), ρ is a constant (6.26 *10–34 J s), and ε is the land surface emissivity.

2.3.2.3 LST retrieval for Landsat 8/OLI

The thermal infrared band (TR-10) of Landsat–8 was used to retrieve LST. Cloud–free Landsat-8 images were acquired to avoid any bias in the extraction of the temperature data using Eq. (11). A radiative transfer model based on the single channel method was used to estimate the LST.

$${\mathbf{LST}}_{{{\mathbf{RTEBi}}}} < - {\mathbf{EiTi}} + \left( {\left( {{\mathbf{1}} - {\mathbf{Ei}}} \right){\mathbf{Downwelling}}} \right) + {\mathbf{Upwelling}}$$
(11)

Ei is the surface emissivity of band i. Ti: spectral radiance. Downwelling is the downwelling path radiance. Upwelling is the upwelling path radiance. According to Plank’s law, ground radiance (Ti) can be expressed using Eq. (12).

$$Ti < - \frac{{C_{1} }}{{Wavelength_{{Bi}} {^{5}} \left( {\exp \left( {\frac{{C_{2} }}{{Wavelength_{{Bi}} }}Ts} \right) - 1} \right)}}$$
(12)

where C1 and C2 are Plank’s radiation constant (C1 is 1.19104 * 108 Wμm4 m-2 sr-1 and C2 is 14,387.7 μm k), the wavelength of the TIRS band 10 = 10.602 and Ts is the surface temperature derived by using Eq. (13).

$$Ts_{Bi} < - \frac{{K2_{Bi} }}{{\log \left( {\left( {\frac{{K1_{Bi} }}{Li}Ts} \right) + 1} \right)}} - 273.15$$
(13)

2.3.3 Normalized difference vegetation index (NDVI) estimation

The NDVI is used for monitoring the growth and health of vegetation and detecting any stress or damage. In addition to mapping and categorizing different vegetation types, NDVI values can be used to evaluate changes in vegetation cover over time [35, 36]. Consequently, it has been used to calculate the abundance of vegetation cover on the Earth’s surface [37]. The NDVI value was calculated by using multispectral bands from Landsat images taken in 1990, 2005, and 2021. Bands 4 and 3 of Landsat 5 and 7 were utilized to measure the NIR and red, respectively. Furthermore, in Landsat 8, bands 5 and 4 were employed to assess the NIR and red, respectively (Eq. 14). the NDVI of the study area using the following formula.

$${\mathbf{NDVI}} = \frac{{{\mathbf{NIR}} - {\mathbf{R}}}}{{{\mathbf{NIR}} + {\mathbf{R}}}}$$
(14)

where NIR is the pixel digital number of TM Band 4 and Band 5 for Landsat 7 and 8, respectively, and R is the DN of TM Band 3 and Band 4 for Landsat 7 and 8, respectively [38].

2.3.4 Extraction of the normalized difference built-up index (NDBI)

The normalized difference built-up index (NDBI) is often mixed with plant noise. NDBI values range from − 1 to 1. The greater the NDBI is, the greater the proportion of built-up land and the greater the area of construction land [39]. A study conducted by Zha et al. [40] indicated that the application of the NDBI in identifying general built-up areas is effective. Even if the reflectances of built-up land and bare land are similar and the NDBI considers both to be the same, it is advantageous to determine the high reflectance area for land surface temperature analysis. The NDBI is derived from Landsat ETM and Landsat 8 images from reflectance measurements in the red and mid-infrared (MIR) regions of the spectrum. The NDBI value is obtained using the following equation and applied to identify the urban built area. The NDBI is derived from Landsat ETM and Landsat 8 images from reflectance measurements in the red and mid-infrared (MIR) regions of the spectrum. The NDBI value is obtained using the following equation [39]:

$${\mathbf{NDBI}} = \frac{{{\mathbf{MIR}} - {\mathbf{NIR}}}}{{{\mathbf{MIR}} + {\mathbf{NIR}}}}$$
(15)

where MIR is the mid-infrared band 5 or band 6 for Landsat 7 and NIR is the near-infrared band 5 for Landsat 8 [56].

2.3.5 LST validation (accuracy assessment)

The LSTs estimated from the Landsat images were cross-validated against the measured air temperature data received from the Ethiopian National Meteorological Principal Station in the Addis Ababa city administration. Correlation analysis was used to cross-validate the estimated and measured temperatures within the study area. A correlation function is employed to evaluate the relationship between the estimated average temperature and the measured average temperature from the measured weather station (Eq. 16).

$${\mathbf{CC}} = {\text{ corr }}\left( {{\text{T1}},{\text{ T}}0} \right)$$
(16)

where CC is the correlation coefficient, T1 is the average temperature from meteorology station, and T0 is the average temperature estimated from satellite image.

2.3.6 Transformation of land surface temperature into urban heat islands (UHIs)

It is not appropriate to compare multiple data images from different years because of seasonal variations and different atmospheric conditions within the same period. Therefore, to compare the seasonal variations in urban heat islands (UHIs) on different dates, surface temperature normalization methods were applied via the following equation [41], (Eq. 17):

$${\mathbf{UHI}} = \frac{{{\mathbf{T}}{\varvec{s}} - {\mathbf{T}}{\varvec{m}}}}{{{\varvec{SD}}}}$$
(17)

where Ts is the land surface temperature, Tm is the mean of the land surface temperature of the study area, and SD is the standard deviation.

2.3.7 Variation in the UHI effect in response to land use/land cover change

The equation was developed by [42,43,44] to determine where urban heat is characterized in critical areas based on the LST distribution and available vegetation. LST > μ + 0.5* is considered to indicate a UHI area and LST ≤ μ + 0.5*δ is characterized as a non-UHI area, where the standard deviation is the LST and the mean is the μ value.

2.3.8 Extraction of the thermal value of the urban morphology type

Both satellite-based and field measurements were utilized to extract information about the urban morphology of the study area. Apartment (Condominium), villas and mud houses are the basic and most dominant selected urban morphological types in the study area. Satellite-based information was extracted from Google Earth by digitizing morphology types and converting them into shapefiles using ArcGIS (10.8). Approximately 240 sample plots were taken. For each type of morphology (Apartment, Villa and Mud house), 4 samples with 2 replications (with similar structure, design and colors of residential type) were taken within each subcity inhibited with the selected morphology. The sampling technique was supported by ground truth field measurements (GPS data). Comparative assessments of land surface temperature were made for the three morphological types of residential areas.

2.3.8.1 Percent surface cover analysis of the UMTs

The percent surface cover was determined in each UMT polygon by using random sampling points in ArcGIS 10.8 to quantify the percent land cover of each morphology type and to compare the land cover percent between each morphology type and the determined land cover type in each polygon: asphalt, graveled road, roof, vegetation and bare land.

2.3.8.2 Land surface temperature and urban morphology type relationships

The exact location of the three different urban morphological types (Apartment, Villa and Mud-house) was extracted from Google Earth by visually interpreting and then converting kml to shp in ArcMap. The correlated with the LST had been tested. One-way ANOVA was also used to evaluate the differences among the morphology types of the MSTs.

2.3.9 Zonal statistics, correlation and regression analysis

Correlation and regression analyses between LST and NDVI associated with urban LULC types were computed using different techniques, such as linear regression methods, by integrating zonal statistical data to identify correlations. The impact of green spaces and built-up areas on LST was quantitatively described (SPSS V23 and RStudio).

3 Results and discussion

3.1 Land use/land cover map for the periods 1990, 2005 and 2021

The land use and land cover of the study area were classified into four classes: built-up, vegetation, farmland and bare land.

Among these LULC types, the largest share of land cover type was farmland in the first period (1990), and it changed to a different class in the next time period (2005–2021). In 1990, 2005 and 2021, the farmland cover was 262.3 km2 (47.3%), 228.5 km2 (41.2%) and 91.5 km2 (16.5%), respectively; vegetation cover was 120.4 (21.7%), 96.67 km2 (17.4%) and 76.3 km2 (13.7%); bare land also covered 54.9 km2 (9.9%), 36.5 km2 (6.5%) and 32.9 km2 (5.9%); and built-up land covered 117.2 km2 (21.1%), 213.5 km2 (38.5%) and 333.7 km2 (60.2%) in 1990, 2005 and 2021, respectively (Fig. 3, Table 3).

Fig. 3
figure 3

LULC maps for 1990, 2005 and 2021

Full size image
Table 3 Area of actual LULC and changes in the year from 1990 to 2021
Full size table

Among the four major LULC classes, built-up areas rank 1st in terms of size and percentage increase from 1990 to 2021. This number is rapidly increasing in accordance with the context of rapid urbanization in Addis Ababa city. Moreover, the proportions of farmland and urban green spaces (vegetation) decreased by − 170.8 km2 (30.8%) and − 44.14 km2 (8%), respectively, from 1990 to 2021. However, the built-up area increased by + 216.5 km2 (39.03%). which was the most significant change in the land cover type proportion. Dagnachew [45] noted that built-up areas consume a considerable amount of land from vegetation cover and farmland during urban development.

3.2 Accuracy results

The overall classification accuracies of the LULC precision appraisals for 1990, 2005, and 2021 were 93.2%, 94.6%, and 98.6%, respectively. As a result, for the think about periods 1990, 2005, and 2021, the kappa coefficients were 0.89, 0.92, and 0.97, respectively (Tables 4, 5, 6).

Table 4 Accuracy assessment using confusion matrix for the year 1990
Full size table
Table 5 Accuracy assessment using confusion matrix for the year 2005
Full size table
Table 6 Accuracy assessment using confusion matrix for the year 2021
Full size table

3.3 LULC change matrix

The pattern of land use/land cover in the study area showed a significant change during the investigation period (1990 to 2021). The LUTM method was employed to drive the quantitative description of the state transition system analysis (Fig. 4). The LULC matrix was produced by overlaying two LULC maps of the same area to show the probability that one particular LULC category changed into another category. In this study, transitional land cover matrices were produced for every three periods from the initial to final year. The results show that each class change from the first column represents the initial state of the LULC categories to the second represents the final state of the LULC categories (Table 7).

Fig. 4
figure 4

Map of LULC conversion from the specified study years)

Full size image
Table 7 LULCC matrix from 1990 to 2005, 2005 to 2021 and 1990 to 2021
Full size table

A study conducted by referenced [46] revealed that there is a high rate of urban expansion in the peri-urban area of Addis Ababa, and this urban expansion is the major factor influencing land use/land cover change in the area and the case of declining farmland and vegetation. [47] also noted that the urban expansion of Addis Ababa was due to the high demand for houses in the city, and massive housing program construction activities are listed as the cause of the decline in agricultural land in the city.

Built-up land was one of the lands uses classes that exhibited a high increase throughout the study period, and there was a small change in the other classes during the study years from 1990 to 2021. In this study, the results revealed that the two largest percentages of land conversion from one type to another were vegetation [62.2 km2 (49.84%), 25.7 km2 (33.2%), 64.05 km2 (51.31%) and farmland (52.92 km2 (23.20%), 82.91 km2 (70.5%), 160.11 km2 (70.2%)] mainly in built-up areas and, to some extent, in other land use classes from 1990 to 2005, 2005 to 2021 and 1990 to 2021, respectively.

3.4 Results of the land surface temperature

The minimum temperatures in the study years 1990, 2005 and 2021 were 12.42 °C, 14.96 °C, and 16.15 °C, respectively. This temperature was mainly observed in the northern part of the study area, where the vegetation cover is high; in the mountains with natural forests (e.g., Entoto, Yerer, and others); and in the parks and around the riverbanks, where riverine vegetation exists (Fig. 5). The results showed that maximum temperatures of 32.58 °C, 38.56 °C, and 39.49 °C were recorded in 1990, 2005, and 2021, respectively.

Fig. 5
figure 5

a Map of LST (°C), b spatiotemporal variation in UHI concentration for the years 1990, 2005 and 2021

Full size image

The highest rates of urban expansion were observed in the southern and western directions. Factors accelerating the increase in this direction were the suitability of land, the accessibility of transport and other demographic factors. These areas are mostly dominated by both government- and private-led housing programs, such as condominiums; real estate; and other constriction programs, such as roads, industries, and factory zones, leading to high LST increases throughout the study period. The spatiotemporal trend of the land surface temperature variation was clearly identified from the map in Fig. 5. In 1990, the greatest extent of Addis Ababa was dominated by temperatures between 23 and 26 °C in 2005. Several studies have been conducted to analyze UHIs by generating LSTs, which represent the relative warmth of cities through the use of land-based observation stations to measure air temperature [48, 49] conducted research in Indianapolis and showed that the trend of hot spot areas was directly related to continual urban expansion.

Similarly, the results of this research showed that maximum LST expansion was observed in most parts of Addis Ababa due to rapid urban expansion in all directions except for the northern escarpment bounded by mountains.

3.5 Land surface temperature changes in response to LULC dynamics

The average surface temperature of Addis Ababa city increased from 24.68 °C in 1990 to 28.28 °C in 2005 and 30.25 °C in 2021 at a rate of 1.8 °C per decade. The study clearly revealed that an increase in surface temperature was mainly caused by the decrease in the amount of green space replaced by impervious surfaces. It was found that 213.5 km2 (38.49%) of Addis Ababa city was covered by built-up areas in the year 2005 and 333.74 km2 (60.15%) in the year 2021 (Table 8). More than half percent of the city was packed with man-made features, which made the temperature higher than that of the surrounding regions. This result is in agreement with [39], who reported that distinctive land surface temperature patterns are associated with the thermal characteristics of land cover classes. [50, 51] also showed evidence that climate change, including the occurrence of drought, rising temperature, floods, reduced annual rainfall, and rising sea levels, resulted from LULC dynamics.

Table 8 LULC and LST distributions in 1990, 2005 and 2021.
Full size table

3.6 Urban heat island variation in response to LULC dynamics

The minimum UHI variations in the last three decades were observed in the northern part of the study area, which has natural and plantation forests, in the parks and around the riverbanks, and where riverine vegetation existed. This finding is in line with the result referenced [52] whereas the result did not consider assessing the heat Contribution Index (CI) of each LULC of the years. In this research the UHI profile was extracted from the northern to the central part of the city (Supplementary Material 1: Figure S1). The results indicate that the UHI effect increased with increasing frequency from the northern part toward the center of the city. In particular, the spatiotemporal variations in the UHIs in the southern and southwestern/eastern parts of the city were very high due to the expansion of industries, infrastructure, commercial centers, and built-up areas. The results also indicated that the maximum UHI effect increased spatially at a rate of 231.6 km2 from 1990 to 2021 (Table 9).

Table 9 Spatiotemporal variation in the UHI effect for the periods 1990, 2005 and 2021.
Full size table

This spatiotemporal variation was related to urban development and its LULC change. Therefore, these results are consistent with past research findings [53,54,55], which revealed that high UHI effects are inhibited by dunce built-up areas, industrial zones, and nearby roads.

3.7 Land Surface Temperature and NDVI

The status of vegetation cover determines greenness and non-greenness. The calculated NDVI values within the study area ranged from − 0.13–0.68, − 0.41–0.64, and − 0.16–0.56 in 1990, 2005 and 2021, respectively.

3.7.1 Correlation between land surface temperature and NDVI

The correlation between the NDVI and LST was found to be negative in all the specified years., i.e., R2 = 0.917 (91%), 0.920 (92%), and 0.9291 (92%). Thus, if the area is densely vegetated, the LST is found to be lower. This correlation is said to be indirect. A negative correlation does not indicate a weakness of the correlation; rather, it reflects a decrease in the number of NDVI values and an increase in the LST value (Tables 1, 10).

Table 10 Relationships between land surface temperature and NDVI
Full size table

These findings are in agreement with those of a number of studies. For instance [10, 56] showed that green spaces can lower surface and air temperatures by providing shade that prevents land surfaces from being directly heated from sunlight.

3.7.2 Land Surface Temperature and NDBI

The NDBI maps of Addis Ababa city are shown in Fig. 7 (b) for each time point (1990, 2005, and 2021) and were taken into consideration to depict the spatial pattern of the NDBI distribution. The maximum NDBI value was 0.36 in 1990 and 0.50 in 2005, and it increased to 0.65 in 2021 (Fig. 7b).

The built area greatly expanded at rates of + 17.4%, + 21.66%, and + 39.03% from 1990–2005, 2005–2021, and 1990–2021, respectively. Figure 6 shows the distribution of built-up areas in Addis Ababa city for the specified study years. Several other studies have revealed that a sharp decrease in green spaces and rapid increase in built-up areas decrease the cooling effect of green spaces [57].

Fig. 6
figure 6

Shows the distribution of built-up areas in Addis Ababa city. Spatial expansion of built-up areas and relation between LST

Full size image

A positive value in Fig. 7a illustrates the potential of an area to be under built-up conditions (light green and red). However, a negative value or an area in dark blue indicates the absence of built-up spaces. This is due to the variation between the class values, which is widening in the 2021 NDBI map. According to Fig. 7, the vegetation cover in the study area around mountains and rivers has the lowest NDBI. However, due to the presence of a densely built-up area with a concrete roadway network, the remaining part of the study area exhibited a high NDBI value.

Fig. 7
figure 7

a Map of NDVI, b NDBI for the years 1990, 2005 and 2021

Full size image

3.7.3 Mean LST and NDVI UMT values (2021)

According to Cavan [58], apartments have higher land surface temperatures than other types of residential morphology, and the projected land surface temperature is greater than 40 °C. The mean LST and NDVI values of each UMT were derived from the estimated Land surface temperature and NDVI map of the study area with zonal statistical analysis. The results showed that there was a significant variation between the UMT values of the land surface temperature and the NDVI. The mean land surface temperatures of the UMTs were 37.12 °C, 31.94 °C and 35.34 °C for the apartment, villa and mud house, respectively.

The results of the present study showed that Apartment (Condominium) had the highest mean land surface temperature among the sites. This occurred because most of the sites found in the city had high height increases with colored metal sheet roofs, and most of the roads between the blocks were coble stones and graveled roads. At most sites, there was space for green areas, but most of them were bare/graveled land. Therefore, this morphology evidently contributed to the warming of the surrounding area. In contrast, villas had the lowest mean land surface temperature distribution, which can be partly explained by the existence of green structures in these areas compared with those in the other UMTs. The NDVI also varied due to the vegetation distribution in each morphology type of the study area; as a result, the maximum NDVI values were 0.38, 0.54, and 0.36 for apartments, villas, and mud houses, respectively. This indicates that the mean NDVI value of Villia was greater than that of the other UMTs. Similarly, the villa residential area had a relatively high NDVI value. According to field observations and percent land cover analysis, most villa houses in the study area gardened and had more trees than did the other UMTs. According to Rani et al. [59], a higher NDVI value indicates healthy vegetation, which is important for the provision of surface temperature regulation. The findings of this study indicate that green infrastructure plays an important role in providing temperature regulation services. Urban planning should consider the importance of vegetation and green areas to mitigate the development and increase in land surface temperature. The results of this study agree with those of many other studies [60,61,62], which revealed that physical variables such as housing density, type, structure, roof color, and floor area ratio (FAR) of UMTs affect the temperature of the surface.

3.7.4 Correlations between the mean LSTs of UMTs

The relationships among the mean land surface temperatures of UMTs were evaluated by using one-way ANOVA, and the results showed that an overall P value < 0.000162 was less than 0.05 at the 95% confidence interval. This indicated that there was a significant difference in the mean land surface temperature among the urban morphology types.

3.7.5 Percent of land cover type for each urban morphology type

The percentage surface cover of each urban morphology type is categorized, and the study reveals that in each UMT, the highest proportion is roof and graveled road, while the lowest is asphalt and vegetation. In comparison to the other UMTs, the UMT had a greater vegetation proportion, which was 26% of the total vegetation cover type in the category.

3.7.6 The effects of built-up and green spaces on the LST of Addis Ababa city

The scatter plot generated with SPSS V23 showed a negative correlation between the LST and NDVI and a positive correlation with the NDBI in each of the three years. In this research, multiple linear regression analysis was applied. The model produced through this analysis showed that these two urban parameters (NDBI and NDVI) contributed 92.1% of the LST variations, with R2 = 0.92 and VIF ≤ 10 for the study area. Thus, the LST in Addis Ababa city can be estimated using these two urban parameters with reasonable accuracy. This means that areas with lower vegetation cover experience higher land surface temperatures and vice versa. The equation used to estimate the LST in the study area was as follows:

$${\mathbf{LST}} = {\mathbf{31}}.{\mathbf{22X1}} - {\mathbf{27}}.{\mathbf{402X2}} + {\mathbf{48}}.{\mathbf{17}},$$

where LST is the land surface temperature, X1 is the built-up area (NDBI value), and X2 is the green area (NDVI value) (Table 11).

Table 11 Correlation coefficients of the LST, NDVI and NDBI.
Full size table

Based on the model produced, for areas with the same NDBI, an increase in the NDVI of 5% implies an expected decrease in the LST of 1.4 °C. However, in areas with the same NDVI, an increase of 5% in the NDBI value implies an estimated increase of 1.6 °C in the LST. An increase in the LST can affect the human thermal comfort in the study area. A mature green cover needs to be planted and maintained within an area to balance the negative effect of the built-up area. This result is in line with that of [63], who noted that the two urban parameters NDVI and NDBI have significant effects on the LST and confirmed that built-up areas have a greater influence on the LST than green areas.

3.7.7 Estimation of LST for sub-cities in Addis Ababa

The model produced in this study (LST = 31.22X1—27.402X2 + 48.17) was used to estimate the LST value of each selected hot spot subcategory. The expansion of built-up land (NDBI) and green vegetation (NDVI) accounted for 5%-25% of the total area, leading to extreme LSTs in Addis Ababa by increasing the temperature by more than 1.6 °C–6 °C, while the change in green vegetation (NDVI) was the most suitable measure for reducing LSTs by more than 1.4 °C to 5.3 °C. Mainly, Subcity was inhibited by industrial and factory zones (Akaki, Lafto and Bole). Similarly, in the central part of Addis Ababa (Arada, Liaeta, Chikos and Addis Sun city), the highest LST was observed due to the area being dominated by business centers, extreme dences built up and vehicles along roads and throughout the entire region (Supplementary Material 1: Table S1).

3.7.8 Correlation between the land surface temperature and NDBI

In the context of the effects of the NDBI on the LST, the correlation was found to be positive during all study periods, where the R2 value increased annually by 0.81 in 1990, 0.89 in 2005, and 0.92 in 2021. Therefore, the increase in both the LST and NDBI had strong effects on the increase in LST. These results are in line with those of [64], who reported that there was a strong positive linear relationship between the LST and NDBI. The LST increases with temperature, which is due to the high reflectance of the surface structures, which is the high temperature. The expansion of urban built-up areas has a significant influence on surface temperature (Table 12).

Table 12 Relationships between land surface temperature and NDBI.
Full size table

The mean NDBI increased with increasing LST in the study area. As a result, the LST presented the order from the minimum LST < 18 °C to the maximum LST > 30 °C; likewise, the mean NDBI increased in all the specified LST classes from − 0.11 to 0.09, − 0.13 to 0.19 and − 0.01 to 0.2 in the year 1990, 2005 and 2021, respectively (Table 12).

3.8 The effects of built-up and green spaces on the LST of Addis Ababa city

The scatter plot generated with SPSS V23 showed a negative correlation between the LST and NDVI and a positive correlation with the NDBI in each of the three years. In this research, multiple linear regression analysis was applied. The model produced through this analysis showed that these two urban parameters (NDBI and NDVI) contributed 92.1% of the LST variations, with R2 = 0.92 for the study area. Thus, the LST in Addis Ababa city can be estimated using these two urban parameters with reasonable accuracy. The equation used to estimate the LST in the study area was as follows:

$${\mathbf{LST}} = {\mathbf{31}}.{\mathbf{22X1}} - {\mathbf{27}}.{\mathbf{402X2}} + {\mathbf{48}}.{\mathbf{17}},$$

where LST is the land surface temperature, X1 is the built-up area (NDBI value), and X2 is the green area (NDVI value) (Table 13).

Table 13 Correlation coefficients of the LST, NDVI and NDBI.
Full size table

Based on the model produced, for areas with the same NDBI, an increase in the NDVI of 5% implies an expected decrease in the LST of 1.4 °C. However, in areas with the same NDVI, an increase of 5% in the NDBI value implies an estimated increase of 1.6 °C in the LST. An increase in the LST can affect the human thermal comfort in the study area. A mature green cover needs to be planted and maintained within an area to balance the negative effect of the built-up area. This result is in line with that of [63], who noted that the two urban parameters NDVI and NDBI have significant effects on the LST and confirmed that built-up areas have a greater influence on the LST than green areas.

3.8.1 Estimation of LST for sub-cities in Addis Ababa

The model produced in this study (LST = 31.22X1—27.402X2 + 48.17) was used to estimate the LST value of each selected hot spot subcategory. The LST of Addis Ababa will increase by more than 1.6 °C–6 °C, while the change in green vegetation (NDVI) will be the most suitable measure for reducing the LST from more than 1.4 °C to 5.3 °C. Mainly, Subcity was inhibited by industrial and factory zones (Akaki, Lafto and Bole). Similarly, in the central part of Addis Ababa (Arada, Liaeta, Chikos and Addis Sun city), the highest LST was observed due to the area being dominated by business centers, extreme dences built up and vehicles along roads and throughout the entire region.

3.9 Contribution of LULC type to urban surface heating

The heat contribution is evaluated by computing the average temperature values for the city’s green space in relation to other land use/cover types. Among the four LULC types, only vegetation cover had a negative heat contribution within the study area (− 2.62 °C) in 2021. However, built-up areas and bare land (vacant spaces) had positive contributions. Consequently, built-up areas can be considered the main heat source (4.28 °C), followed by bare land (vacant spaces), which accounts for (+ 1.7 °C). Marginally, dry farmland also had a positive heat contribution (0.23 °C) in 2021 in the study area (Fig. 8).

Fig. 8
figure 8

Heat contribution indices of LULC in 1990, 2005 and 2021

Full size image

This result is in line with the findings of many other studies [64,65,66,67], which showed that urban spaces were more prone to the UHII due to the high share of built-up areas and industrial areas, whereas the UHII was comparatively low due to the presence of greenery. In addition to built-up land, fallow land and barren land contributed significantly to the heat generation (Table 14).

Table 14 Heat Contribution Indices (CIs) of LULC types in 1990, 2005 and 2021.
Full size table

Based on these results, it can be assumed that the heat source/sink role in built-up/vegetation cover plays a vital role in UHI variation. Vegetation cover can therefore be regarded as the most valuable heat sink, whereas built-up areas can be considered the main heat source within the study area. This result agrees with numerous studies that have been conducted on the relationship between UHIs and urban LULC change [68,69,70]. Identifying LULC types with urban microclimate patterns is crucial for understanding and mitigating the impacts of urban LULCC on local microclimate change.

4 Conclusions

The advancements in remote sensing (RS) and geographic information systems (GISs) have led to convincing results in analyzing the environment in which we are living today. In this study, Landsat data for the year 1990 (TM), for 2005 (Lands ETM+) and for 2021 (Landsat 8 (OLI)) were used to generate important information on LULC and to interpret the relationships of LST with NDVI and NDBI for the determination and assessment of the effects of urban land use and land cover change on urban microclimate dynamics in Addis Ababa city, Ethiopia. The study showed that there was rapid expansion of built-up areas, and the deterioration of vegetation cover and agricultural land were the most important deviations that could be major possible causes of the UHI effect in Addis Ababa city. The NDVI in 1990, which had a value > 0.3, decreased from 65.06 km2 (12.35%) to 29.39km2 (5.58%) in 2021. This clearly shows that the expansion of built-up land has caused significant land cover change as well as changes in the LST. In this regard, the LST distribution in Addis Ababa was very closely related to the distributions of vegetation cover (NDVI) and built-up areas (NDBI), with R2 = 0.93 and R2 = 0.92, respectively.

The study also revealed that there is a significant difference in land surface temperature among the urban morphology types. Based on the field observations and percent land cover analysis, villa had the lowest surface temperature, which can partly be explained by the presence of more green structures in the area than in the other UMTs, while apartment (Condominium) had a greater LST distribution than did the other UMTs. The role of green vegetation (NDVI) in the urban climate was also analyzed. Addis Ababa is a Metropolitan city in Ethiopia; LST and UHI issues are very sensitive due to the expansion of the built environment, deterioration of vegetation cover, and small area coverage of urban green spaces, which is worsening these days.