Introduction

The land use land cover (LULC) changes, and their impact on land surface temperature (LST) have become a serious environmental issue worldwide1,2. The LULC changes are directly related to urbanization, which include the replacement of natural surfaces, such as vegetative cover, and porous surfaces, such as water bodies, with impervious surfaces like urban buildings constructed with warm concrete3,4. These urban buildings have more potential for absorbing heat, influencing the LST of built-up areas5. The change of vegetative cover and waterbodies into built-up areas or barren land also increases the LST6,7. However, proper land use development that reduces urban expansion and encourages vegetation by altering barren land causes the cooling effect8,9.

The 2016 Global Climate Observing System Implementation Plan has suggested the LST as a key climatic factor, similar to near-surface air temperature. The LST and near-surface air temperature are interconnected. The higher the surface temperature, the higher the surrounding air temperature, and vice versa10. The LST data shows helpful conditions when near-surface air temperature data are scarce because of insufficient weather stations. Furthermore, the satellite-based LST data is easier to access for wide areas of land, which is beneficial for planning cities11. In the twentieth century, there was a worldwide growth in urbanization, and as a result, we are currently facing the effects of high temperatures in urban areas12,13. The adverse impacts of rapid urbanization are not limited to LST but also affect the growth in housing, transportation, and industrial sectors, thus resulting in higher demands of energy, mainly coming from fossil fuels14. This increases carbon dioxide (CO2) emissions, deteriorating environmental conditions and promoting global warming15,16. The increasing number of concrete buildings in urban areas and the rising global temperature make urban residents and infrastructures more vulnerable to the adverse impacts of climate change17.

Tan et al.18 stated that the expansion of urbanized and dry-land areas between 1995 and 2013 resulted in a 3.5 °C rise in the average winter temperature of Dongting Lake, China. Similarly, Shang et al.14 also reported a 0.5 °C rise in the average LST of Hangzhou Metropolitan, China, due to LULC changes between 2005 and 2018. The district of Rajshahi in Bangladesh also experienced about 13 °C increase in LST due to LULC changes between 1997 and 201719. Similarly, the LST of Luis Potosí Basin, Mexico, raised approximately 11 °C between 2007 and 2020 due to LULC changes 20. The Karachi City of Pakistan also experienced approximately a 0.04 °C increase in LST between 2009 and 2017 due to fast urban expansion21. Hussain and Karuppannan,22 reported a 0.5 °C increase in the annual mean LST of Khanewal City, Pakistan, due to uncontrolled urban sprawl during the last four decades. According to Traore et al.23 the LST of Bangui, Central Africa, increased by 1 °C due to a 130% increase in urban regions from 1986 to 2017. Likewise, the built-up areas of Multan City increased by 3.5%, which ultimately increased the city mean LST by about 0.9 °C from 1990 to 202024.

The current research examines LULC changes and their impact on LST in the industrial province of Kabul, Afghanistan. Afghanistan shows the maximum rate of urban expansion among Central Asian countries, with a rate of 34% between 2003 and 201325. Approximately 41% of Afghanistan’s urban population resides in Kabul, with 82% living in 55 informal settlements26. Hence, Afghanistan’s urban areas are growing unplanned to meet the basic demands of the increasing urban population and provide their fundamental daily needs. Afghanistan as a whole, and Kabul province in particular, lacks the proper urban planning. Generally, modeling-based suitability studies are not carried out in construction planning or urban growth activities. Increased urbanization and the engulfment of agricultural and green regions by industrial growth and more migrants seeking better prospects are contributing to this trend27.

Therefore, mitigating pressure on Kabul province’s limited land resources can be achieved by analyzing spatial trends in LULC changes and their impact on changing LST and applying machine learning to predict future patterns in LULC changes and LST variations. The majority of LULC studies that have been done until now are quantitative. The novel aspect of this study lies in utilizing machine learning methods, such as SVM and CA-LR models, to more accurately anticipate the spatial elements of LULC changes and LST variations for the future based on past patterns. However, to the best of our knowledge, no work has been carried out to investigate the possible impacts of LULC changes on the LST of Kabul province, which is economically important for producing farming products and its industries, which might offer insightful information for urban planning. The present work is planned to cover the following objectives: (i) to widely investigate the spatiotemporal changes in LULC between 1998 and 2022, (ii) to study the association between the LULC changes and its impact on the LST of Kabul province in summer, spring, fall and winter season and (iii) to apply a CA-LR model to estimate possible variations in LULC and their related impact on LST for the years 2034 and 2046 to offer insights for the developing of urban areas and related climatic consequences.

Methodology

Description of the study area

The current study focuses on Kabul, a significant and rapidly developing urban province in Afghanistan. Kabul is located at 34°31′31′′ North latitude and 69°10′42′′ East longitude (Fig. 1). Kabul is the largest province in Afghanistan in terms of people living in urban areas, with about 6 million population28,29. The total land area of Kabul is 4,523 km2. The average elevation of Kabul province above sea level is 1803 m. The Kabul City region has a total built-up area of 40,143 ha, of which 17,335 ha are used for housing purposes, and 1006 ha are used for commercial purposes. Most of Kabul's land area is mountainous, with only around 38% flat. Kabul has four distinct seasons, with the most precipitation from February to April26,30 with the maximum rainfall reported annually by 400 mm26.

Figure 1
figure 1

Location map of the study area (Kabul province, Afghanistan) (Software: ArcGIS v. 10.8).

Full size image

Kabul province experiences semi-arid to dry climates with severe cold, and the temperature can drop below 11 °C and rise approximately 40 °C during hot summers30. The natural landscapes in Kabul province have suffered greatly in recent decades due to high immigration rates and rapid population growth. The rapid urbanization in the last few years and about 70% of illegal settlements without proper planning and management have made the path for the LULC changes and LST in Kabul province. Thus, Kabul was properly selected as the research study area for the present work.

Data collection and preprocessing

The current study used remote sensing satellite data from the Multi-temporal Landsat satellite 5-TM, 7-ETM+, and 8-OLI sensors to evaluate the previous LULC and LST variations over the study area from 1998 to 2022. Twelve-year intervals of Landsat data were used to observe the changes in LULC and LST. The satellite images were downloaded every data year in May to minimize seasonal variations. Because of the generally clear weather in May (less than 20% cloud cover), it was selected for our study to increase classification accuracy. Using ArcGIS software, the images were mosaicked, and the study area was extracted from the mosaicked image. The details are given in (Table 1).

Table 1 Details of satellite data used in this study.
Full size table

An advanced and standard pre-treatment method was used to derive LULC and LST maps for the Landsat imageries. Due to sensor issues with the Landsat images, the pre-treatment method includes radiometric calibration, atmospheric correction, and removing lines corresponding to Landsat 7-ETM+. The Landsat data was corrected radiometrically to remove the effects of light, sensor calibration, view angle, and topography and atmospherically to remove the impacts of atmospheric scattering31.

After pre-treatment, the SVM algorithm produced the LULC maps for 1998, 2010, and 2022. SVM offers several advantages, including handling high-dimensional data, non-linear relationships, and high accuracy and generalization capabilities. However, SVM also has some limitations, such as being computationally expensive for large datasets and sensitive to parameter tuning. As a supervised learning algorithm, SVM uses kernel functions to transform data into a higher-dimensional space to find the optimal hyperplane that maximizes the margin between classes. Compared to algorithms like Random Forest, Gradient Boosting, and Neural Networks, SVM excels in handling non-linear relationships and high-dimensional data, making it a popular choice for complex classification tasks.

During the same data period, thermal bands from Landsat 5-TM, Landsat 7-ETM+, and 8-OLI satellite data were used to produce the LST maps. The LST from Landsat 8 was measured using bands 10 and 11, respectively. After determining the LST for both thermal bands, the cell statistics tool in ArcMap 10.5 software was applied to get the average LST values. Furthermore, the future LULC changes and LST variations were predicted using CA-LR for 2034 and 2046, respectively, as shown in (Fig. 2).

Figure 2
figure 2

A systematic framework involves the main steps in using the CA-LR model.

Full size image

The cellular automata-logistic regression (CA-LR) model

The Cellular Automata-Logistic Regression (CA-LR) model is a hybrid method that integrates the spatial analysis capabilities of Cellular Automata (CA) with the predictive power of Logistic Regression (LR). In this model, CA is used to figure out urban growth and land use changes by iteratively applying transition rules to neighboring cells. At the same time, LR is employed to estimate the probability of land use changes based on historical data and spatial variables. The CA component captures the spatial dynamics and complex interactions between cells. In contrast, the LR component incorporates the statistical relationships between variables, enabling the forecast of future land use changes and urban growth patterns. Combining these two techniques, the CA-LR model provides a strong framework for simulating and predicting urban expansion and land use changes.

Land use land cover (LULC) classification

The classification process is an essential step in preparing LULC data. Numerous methods for LULC classification have been developed worldwide. This study used the preprocessed Landsat images to produce LULC maps, which were divided into four different LULC classes such as built-up areas, bare soil, vegetation, and water bodies. Typically, classification can be divided into two categories: supervised and unsupervised. While analyzing remote sensing image data quantitatively, the supervised classification technique might be promising and helpful compared to unsupervised classification. However, the unsupervised classification method offers multiple imagery classes for grouping images without requiring any training samples. In the present research, an SVM, a machine learning algorithm, was used to categorize all the Landsat images in Google Earth Engine (GEE). The change detection and transition matrices were calculated after classification. These changes were calculated in QGIS before future prediction. The transition and change matrices were calculated by subtracting two classified images for two time periods such as initial (1998) and final (2022) and so on for the respective time duration.

Assessment of land surface temperature (LST)

LST is a critical parameter for ensuring energy and water stability at both local and global scales. Employing remote sensing (RS) sensors have significantly enhanced the precision of evaluating the impacts of LULC and LST. The Landsat sensors represent thermal data using Digital Numbers (DNs). Using the Landsat satellite images from 1998 to 2022, the thermal bands were used to calculate the LST of Kabul province. The Landsat 5-TM and 7-ETM+ used band 6 as their thermal band, while the Landsat-8 used bands 10 and 11. The following four basic steps were applied to convert these DNs into LST in the current work. Other several researchers have applied the same method to calculate the LST32,33,34.

First step: Converting DNs to radiation

The first step involved using Eq. (1) to convert DNs into radiance.

$$L\lambda = (Lmin+ (Lmax - Lmin) \times DNs/255)$$
(1)

In this Eq. (1), Lλ = spectral radiance; Lmin = 1.238; Lmax = 15.30 (Landsat 5); Lmin = 1.238 and Lmax = 15.600 (Landsat 7); while Lmin = 0.10033 and Lmax = 22.00180 (Landsat 8 with bands 10 and11) and DNs = Digital Numbers.

Second step: Conversion of radiance into brightness temperature (TB)

The radiance was converted into TB using Eq. (2) in the second step.

$$TB=K2/In ((K1 / L\lambda ) +1)$$
(2)

In Eq. (2), TB = brightness temperature; K1 and K2 = calibration constant 1 and 2 with the three sensors values as follows: K1 = 607.76; K2 = 1260.56 (Landsat 5); K1 = 666.09; K2 = 1282.71 (Landsat 7); K1 = 774.89; K2 = 1321.08 (Landsat 8 with band 10); and K1 = 480.88; K2 = 1201.14 (Landsat 8 with band 11).

Third step: TB (Kelvin) conversion into degree Celsius (°C)

Next, applying Eq. (3), the TB (Kelvin) based on the earlier step was converted to degrees Celsius (°C).

$$TB\left(^\circ \text{C}\right)=TB-273(K)$$
(3)

In Eq. (3), where TB (°C) = temperature in degrees Celsius, and TB = temperature in Kelvin.

Fourth step: TB (°C) conversion into LST

Finally, in the last step, the Eq. (4) was used to change the TB (°C) into LST.

$$LST=TB/ [1+ (\lambda *TB/p) *ln (\varepsilon )]$$
(4)

In Eq. (4), where λ = wavelength of emitted radiation; p = h * c / s; h = Planck’s constant = 6. 626 * 10−34 J s; c = speed of light = 2.998 * 108 m/s; s = Boltzmann’s constant = 1.38 * 10−23 J/K and ε = surface emissivity.

For 1998, 2010, and 2022, the NDVI was used to calculate the proportion of vegetation (PV) by applying Eq. (5) given below, and the obtained PV was used to estimate the surface emissivity.

$${\text{PV}} = \left( {\left( {{\text{NDVI}} - {\text{NDVI}}_{{{\text{min}}}} } \right)/\left( {{\text{NDVI}}_{{{\text{max}}}} - {\text{NDVI}}_{{{\text{min}}}} } \right)} \right)^{2}$$
(5)

Finally, the surface emissivity was calculated with the help of Eq. (6).

$$\text{Surface emissivity }(\upvarepsilon ) = 0.004 *\text{ PV }+ 0.98$$
(6)

Standardization of LST

LST standardization was carried out to compare LST images as seasonal and topographic variations occur due to images from different years. Direct evaluation and comparison of LST between seasons and years is inappropriate. The significant objective of standardization is to bring every factor into balance. The LST data for 1998, 2010, and 2022 were standardized using Eq. (7) and the image from 2022 as a reference for every change to be proportional to one another in the current study 35.

$${\text{LSTfs}} = \left( {{\text{LSTm}} {-} {\text{LSTk}}/ {\text{LST}}\beta {\text{j}}} \right) {\text{LST}}\alpha {\text{i}} + {\text{LSTx}}$$
(7)

In Eq. (7), where LSTfs is the standardized LST value for each pixel for the year m (i.e., 1998 or 2022), LSTm is the LST value of each pixel in the initial image before standardization; LSTk represents the image-specific mean LST values for the year m (i.e., 1998 or 2022); LSTβj is the standard deviation for each image especially for the LST values for the year m (i.e., 1998 or 2022); LSTαi is the standard deviation for each image especially for the LST values for the reference year i (i.e., 2022); and LSTx shows the images specific mean LST values for the year 2022.

Retrieving of biophysical parameters

Several biophysical indices were calculated to increase the understanding of changes in a specific land cover. The relation between these indices and LST was also examined to understand their effect on LST. However, NDVI, NDBI and NDWI which indicate vegetation or greenery, built-up areas or impervious surfaces, water bodies and bare soil without greenery or any object of any area, were calculated respectively using the following equations.

$$\text{NDVI}=\left({B}_{\text{NIR}}-{B}_{\text{R}}\right)/\left({B}_{\text{NIR}}+{B}_{\text{R}}\right)$$
(8)
$$\text{NDBI}=\left({B}_{\text{SWIR}1}-{B}_{\text{NIR}}\right)/\left({B}_{\text{SWIR}1}+{B}_{\text{NIR}}\right)$$
(9)
$$\text{NDWI}=\left({B}_{\text{G}}-{B}_{\text{MIR}}\right)/\left({B}_{\text{G}}+{B}_{\text{MIR}}\right)$$
(10)

where NIR is near-infrared, R is red band, SWIR1 is shortwave infrared1, G indicates green band and B is blue band.

Zonal base LST classification

A total of seven (07) classes were used to categorize the LST of Kabul province, such as < 6 °C, 6–11 °C, 11–16 °C, 16–21 °C, 21–27 °C, 27–32 °C, and > 32 °C. This classification evaluated the study area’s geographic differences and patterns (area-based distribution) in different temperature regions.

Future LULC change and LST prediction using the CA-LR model

Various models are employed to predict future LULC and LST variations26,34. The selection of a suitable model or simulation technique depends on several variables specific to each context. These variables comprise but are not limited to, the characteristics of land alteration processes, the availability of information, the primary objectives of the study, and the accuracy of the prediction techniques36. The present research, the CA-LR model is applied to predict LULC changes and LST variations for 2034 and 2046, respectively. The input parameters of the CA-LR model comprise both dependent and independent variables. The dependent variables, such as the LULC change map, LST distribution map, and independent variables, include the distance from roads, residential, educational, and commercial places, elevation and slope for LULC prediction, and NDVI, NDBI, and NDWI considering specifically for LST simulation. Transition matrices were created by computing the difference in LULC and LST between two time periods. Initially, the model was first trained using the LR method and then CA was used to predict the future LULC changes and LST variations for 2034 and 2046. Other kappa statistics were produced to validate accuracy, such as percent correctness and standard kappa. All transitions were accurately modeled using a single set of explanatory variables due to the similarity in driving forces for all transitions. In the first step, the estimated changes in land cover (such as transitions from other LULC types to built-up areas) were treated as dependent variables. The second step involved considering various spatial variables, such as distance to roads, water bodies, vegetation, bare soil, slope, and elevation, as independent variables37. Figure 2 presents a methodological flow chart that outlines the steps involved in the CA-LR process.

Accuracy assessment

The accuracy assessment in this work was conducted using the Kappa coefficient, a reliable statistical metric frequently used to assess the quality of satellite image classification. The Kappa coefficient is a reliable indicator of the agreement between the classified results and the ground truth data. Three primary variables were specifically used in this research to evaluate accuracy: overall, producer, and user accuracy. These measures allowed for a thorough assessment of the classification accuracy of satellite images, providing insight into the accuracy and reliability of the results from classification. To evaluate the classification accuracy of classified images, approximately forty (40) training samples were produced by each LULC class utilizing the baseline data and additional data from many sources (e.g., Google imagery and field study sites). The Kappa index was used to measure the accuracy of the current study’s efficient results. The Kappa index is a method applied to compare LULC classes by calculating how similar the mapped and actual LULC classes are to one another38. The accuracy was determined using the following equations.

$${\text{Producer}}^{\prime}{\text{s accuracy}}\left( \% \right) = \left( {\frac{{x_{kk} }}{x + k}} \right) \times 100$$
(11)
$${\text{User}}^{\prime}{\text{s accuracy}}\left( \% \right) = \left( {\frac{{x_{kk} }}{{x_{k + } }}} \right) \times 100$$
(12)

The Overall accuracy and Kappa coefficient were calculated using the following equations.

$${\text{Overall accuracy}}\left( {{\text{OA}}} \right) = \frac{1}{N}\sum_{K = 1}^{r} n_{i}$$
(13)
$${\text{Kappa coefficient}}\left( {\text{k}} \right) = \frac{{N\sum_{K = 1}^{r} x_{kk} - \sum_{K = 1}^{r} \left( {x_{k + 1} \cdot x_{ + k} } \right)}}{{N^{2} - \sum_{K = 1}^{r} \left( {x_{k + 1} x_{ + k} } \right)}}$$
(14)

Results

Past pattern of LULC class changes

Using the SVM algorithm, the past pattern of LULC classes (from 1998 to 2022) was obtained from Landsat data. Results showed major changes; the built-up area increased while the vegetation and bare soil reduced between 1998 and 2022. The classification accuracy (overall) was 93.74%, 93.78%, and 93.26% for 1998, 2010, and 2022, respectively, as given in (Table 2).

Table 2 Accuracy assessment of the classified images for 1998, 2010, and 2022.
Full size table

Results highlighted the change of bare soil and vegetation into the built-up areas, and both LULC classes decreased by about 7.20% and 2.35%, respectively, in the previous 25 years, as shown in (Table 3) and (Fig. 3).

Table 3 Periodic and net LULC changes in Kabul province from 1998 to 2022.
Full size table
Figure 3
figure 3

shows changes in LULC classes for the years (A) 1998, (B) 2010, and (C) 2022 (Software: ArcGIS v. 10.8).

Full size image

Figure 3 illustrates the spatial distribution findings in Kabul province between 1998 and 2022. According to LULC classification, the built-up areas increased approximately 134 km2 between 1998 and 2010 and 347 km2 between 2010 and 2022. In Kabul province, the overall increase in the built-up areas from 1998 to 2022 was reported to be approximately 481 km2 (Table 3). Throughout the study period, the total bare soil and vegetation area was significantly reduced by 395 and 89 km2, respectively. Figure 4 shows that the bare soil and vegetation have been converted into built-up regions spread throughout the study area from 1998 to 2022.

Figure 4
figure 4

Land use transition over the study period (A) 1998, (B) 2022, and (C) Change detection in the study area (Software: ArcGIS v. 10.8).

Full size image

Figure 4 shows the transition of different LULC classes from 1998 to 2022 in Kabul province. The details are given in (Table 4). The results showed that 5.69 and 3.44% of bare soil and vegetation were converted into built-up areas, thus ultimately impacting the local climate regarding LST, further leading to urban heat islands (UHI). The buildings and roads increase heat absorption and retention by urban materials like concrete and asphalt, elevating anthropogenic heat from human activities. This results in higher temperatures as natural cooling is lost due to the loss of vegetation, and heat is absorbed and slowly released by urban infrastructure.

Table 4 Transition of different LULC (%) in the study area from 1998 to 2022.
Full size table

Figure 5 shows the 2022–2034 and 2022–2046 land use transitions, respectively. The finding revealed that 2.68 and 2.16% of bare soil and vegetation will convert to built-up areas during 2022 and 2034. Similarly, 5.67 and 3.97% of land covered by bare soil and vegetation will be converted to built-up areas for 2022 and 2046. The results show less transition in other land use classes, such as water bodies.

Figure 5
figure 5

Land use transition for the period of (A) 2022–2034 and (B) 2022–2046 (Software: ArcGIS v. 10.8).

Full size image

Past pattern of LST changes

The past pattern of seasonal and annual changes in LST during the study period showed a significant change, as shown in (Fig. 6). From 1998 to 2022, the area covered by high LST classes, such as > 32 °C, constantly increased in Kabul province. However, the area within LST class < 32 °C declined between 1998 and 2022 because this area became a high-temperature area.

Figure 6
figure 6

LST variation for spring, summer, autumn, and winter; (A, B, C, D) 1998, (E, F, G, H) 2005 and (I, J, K, L) 2020 in the study area (Software: ArcGIS v. 10.8).

Full size image

Results in (Fig. 6) showed changes in seasonal and annual LST for 1998. The highest maximum LST was found in the summer at 40.91 °C, followed by spring, autumn, and winter. The yearly average mean and maximum LST were estimated at 24.36 °C and 36.87 °C, respectively, in 1998. In 2010, the highest maximum LST was recorded at 42.49 °C, which was 40.91 °C in 1998. The highest LST trend followed the same order as in 1998. A 1.03 °C increase was found from 1998 to 2010 in annual mean LST in the study area. The yearly average mean and maximum LST were estimated at 25.39 °C and 37.75 °C, respectively, in 2010. LST was continuously increasing from 2010 to 2020. The lowest minimum and highest maximum temperatures increased from 2010 to 2022. The highest summer minimum temperature was recorded at 7.66 °C which was 6.39 °C in 2010. Similarly, the highest maximum summer LST was recorded at 44.89 °C, 42.49 °C in 2010. Approximately 1.10 °C increase was recorded in annual mean LST from 2010 to 2022 in the study area.

The areal LST variation of Kabul province is shown in (Fig. 8). LST was classified into seven (07) classes ranging from < 6 to > 32 °C. Results showed that lower LST classes lost their area from 1998 to 2022. An increasing areal trend in higher LST classes was found in the study area from 1998 to 2022, as given in (Fig. 7). The analysis showed a decreasing trend from 22 to 7% and 39% to 22% in LST classes 16–21°C and 21–27°C from 1998 to 2022 (Fig. 8). The higher LST classes showed an increasing trend over the study area. The LST class 27–32 °C was 23% in 1998, increasing to 51% in 2022. Similarly, the LST class > 32 °C was increased from 2 to 11% during the study duration, as shown in (Fig. 8). An overall 37% increase and 32% decrease occurred in higher (27–32°C and > 32°C) and lower (16–21°C and 21–27°C) LST classes in the Kabul province from 1998 to 2022.

Figure 7
figure 7

Overall changes in the LST of the study area for the years (A) 1998, (B) 2010, and (C) 2022 (Software: ArcGIS v. 10.8).

Full size image
Figure 8
figure 8

Area distribution trend (%) under different LST (°C) classes in the study area from 1998 to 2022.

Full size image

Temperature variations in different LULC classes

The built-up area has the highest temperature, followed by bare soil and vegetation, as shown in (Fig. 9). The current study also showed that the seasonal and annual mean LST increased in all LULC classes since 1998, even in the vegetation.

Figure 9
figure 9

Differences in the mean LST (°C) throughout the study period across various LULCs in the Kabul province.

Full size image

The growth of impervious surfaces and built-up regions substantially contributed to the rise in LST 39,40. For all LULC classes in the study area, the LST increased between 1998 and 2022, as shown in the spatial distribution map (Fig. 9). A large section of the study area experienced higher temperatures in built-up areas, followed by bare soil, vegetation, and water bodies. Though not as much as in the summer, the tendency of rising temperatures is also evident in the spring season. The study's overall findings highlight the impact of urbanization on altering LST patterns and its significant consequences for local climate dynamics.

Projection of the future LULC dynamics

The previous pattern of LULC was significantly changed over the study period (1998–2022) in the Kabul province, as given in (Table 3 and Fig. 3). Therefore, it is important to predict the future pattern of LULC by applying simulation models. The present study, the CA-LR model was applied to estimate the future LULC changes for 2034 and 2046, respectively. This model uses each cell's previous pattern to project the future state according to standardized transition rules41. According to the results of the future projection, approximately 898 and 1215 Km2 of the land area will be transformed into an urban or built-up region in 2034 and 2046, respectively, as shown in (Table 5, Fig. 10). According to the simulation findings, the LULC is likely to change in the future, which could harm the environment, impacting the climate and biodiversity of the area.

Table 5 LULC for the years 2034 and 2046 respectively.
Full size table
Figure 10
figure 10

Simulated LULC maps of the Kabul province for the years (A) 2034 and (B) 2046 (Software: ArcGIS v. 10.8).

Full size image

Simulation of the future LST

Like the LULC, the LST pattern also showed significant changes during the study duration (1998–2022). Therefore, the future LST changes were projected for the near future (2034 and 2046), as shown in (Fig. 11). The LST pattern was predicted using previous data from the CA-LR model to anticipate future patterns in the research area. Due to the strong connection with LST, the graph shows an increasing trend influencing the temperature. The future predicted result showed that 16.32% and 16.47% of the Kabul province will experience the highest temperature zone (i.e., > 32 °C) in 2034 and 2046, respectively, as given in (Table 6), which was 11.04% in 2022. This high temperature is harmful to human beings as well as animal and plant species.

Figure 11
figure 11

Simulated LST maps of the Kabul province for the years (A) 2034 and (B) 2046 (Software: ArcGIS v. 10.8).

Full size image
Table 6 Simulated LST for the years 2034 and 2046, respectively.
Full size table

Discussion

The rapid urbanization and population growth in Kabul province over the past few decades have led to significant changes in LULC patterns, substantially reducing green and fertile land. These changes alter the ecological system and impact natural resource management. The notable variations in LULC patterns have also led to fluctuations in LST, exhibiting seasonal and annual variability. This highlights the need for sustainable urban planning and management strategies to mitigate the adverse effects of urbanization on the environment and natural resources. Understanding these dynamics is crucial for developing effective measures to protect the ecological balance and ensure a sustainable future for Kabul province. The results showed that throughout the previous and expected period, the LST, especially in built-up areas, increased significantly in the study area. The findings highlight that the temperature of the Kabul province has increased due to LULC changes between 1998 and 2022.

Built-up areas increased significantly about 9.37% during the study period. Economic and technological development, environmental changes, population growth, and migration were major factors in changing LULC patterns in Kabul province42. Between 1998 and 2022, the study area's summer and winter temperatures rose approximately 3.98 °C and 2.45 °C due to a rapid increase in built-up areas and a reduction in vegetation cover. This means that throughout the previous time period, the LST of the study area significantly increased due to the construction of concrete-based impermeable structures or built-up areas, which absorbed more heat than vegetation and water bodies. The study predictions for the future revealed that the built-up areas might be increased to 898 and 1215 km2 and vegetation cover will be decreased by about 598 and 437 km2 respectively, which may further influence the LST in 2034 and 2046, respectively. The findings indicated a strong relationship in the predicted and observed estimated LULC variations for the base period.

Increased urbanization and reduced vegetation and water bodies have contributed to Kabul province's seasonal and annual LST variations. Similar patterns in LST and LULC changes have been observed in previous studies. Over the research period, urban growth driven by population expansion, economic development, and migration led to significant land cover changes. Additionally, land cover changes and climate change have contributed to the rise in air temperatures and LST over time.

Similar results have been reported by several researchers, such as Kafy et al.19, who observed an increase in LST of about 7.24 °C in Dhaka City, Bangladesh, due to a 14% increase and 5% decrease in built-up areas and vegetation cover, respectively, between 2000 and 2020. Kafy et al.19 also reported a strong connection between urbanization and LST in Rajshahi, Bangladesh, indicating a 13°C increase in temperature due to vegetation decrease between 1997 and 2017. The fast urbanization growth in the San Luis Potosi Basin, Mexico, contributed to a total 11°C rise in LST between 2007 and 202020.

The outcome of the present work revealed that higher LST classes covered most of the Kabul province in 2022 compared to 1998. Following the present research results, Imran et al.44 also reported that, the majority of Dhaka City in Bangladesh experienced higher temperature classes between 1993 and 2020. The rapid increase in urbanization is highly associated with the Kabul province's local climate. The spatial analysis results for each year between 1998 and 2022 indicated that the LULC changes during the past 25 years had led to an increase in the LST in all four seasons. In general, Afghanistan is facing climate change. Future projections show that Afghanistan will experience a general increase in dryness and almost 2–6 °C increase in average temperature by the year 210045. There would be much less rainfall in the spring season.

Implication of the study

Significant implications of this research exist for environmental management, urban planning, and climate adaptation techniques. This research provides important insights by addressing the impact of urbanization on LST using machine learning techniques (SVM and CL-AR). To build sustainable cities and reduce the negative effects of heat stress on urban residents, it is essential to understand how urbanization alters LST trends. The results provide policymakers and urban planners with practical insights to guide land use decision-makers, prioritize the development of green infrastructure, and implement heat mitigation measures in fast-growing metropolitan areas such as Kabul.

This study also improves our understanding of how urbanization and climate change interact. The heat produced by urban activities worsens the temperature rise, causing difficulties for climate resilience and adaptation efforts. The study improves our capability to project the potential temperature rises while developing practical climate adaptation strategies by measuring the association between urbanization and LST patterns. Finally, the information obtained from this study can help to develop integrated urban planning strategies that prioritize both environmental sustainability and human well-being in the face of increasing urbanization and climate change in the study area.

Conclusion

This study aimed to analyze the impact of LULC changes on seasonal and annual LST in Kabul province, Afghanistan, from 1998 to 2022. The study used the SVM algorithm on Landsat satellite images to categorize the study area into four distinct LULC classes, i.e., built-up, bare soil, vegetation, and water bodies. Seasonal and annual LST data were obtained from the thermal band of Landsat satellite images. LULC changes and LST variations were projected for 2034 and 2046 using the CA-LR model. The results showed significant changes in LULC and LST in Kabul province from 1998 to 2022.

  • Built-up areas showed about 9.37% increase, while bare soil and vegetation decreased by 7.20% and 2.35% during the study duration.

  • Built-up areas showed the maximum annual mean LST (27.53 °C), followed by bare soil (27.11 °C), vegetation (24.51 °C), and water bodies (17.83 °C) in 2022.

  • The maximum seasonal temperature was estimated in summer at 40.91 °C, 42.49 °C, and 44.89 °C in 1998, 2010, and 2022, respectively.

  • The results of future prediction showed that the built-up areas will increase to 17.08% and 23.10% in 2034 and 2046, respectively, which was 11.23% in 2022.

  • Similarly, the area associated with the LST class (≥ 32 °C) is likely to increase by about 27.01% and 43.05% in 2034 and 2046, respectively, which was 11.21% in 2022.

This research highlights the need for stakeholders to implement measures to control unplanned growth in Kabul province and mitigate the adverse impacts on human health and the environment. Moreover, implementing urban greening initiatives in the city centers represents the optimal solution for mitigating the adverse effects of high temperatures. Future research should address the unplanned and planned urban growth of LST in Kabul and other major cities of Afghanistan.