Modeling spatio-temporal assessment of land use land cover of Lahore and its impact on land surface temperature using multi-spectral remote sensing data

Urban sprawl, also widely known as urbanization, is one of the significant problems in the world. This research aims to assess and predict the urban growth and impact on land surface temperature (LST) of Lahore as well as land use and land cover (LULC) with a cellular automata Markov chain (CA-Markov chain). LULC and LST distributions were mapped using Landsat (5, 7, and 8) data from 1990, 2004, and 2018. Long-term changes to the landscape were simulated using a CA–Markov model at 14-year intervals from 2018 to 2046. Results indicate that the built-up area was increased from 342.54 (18.41%) to 720.31 (38.71%) km2. Meanwhile, barren land, water, and vegetation area was decreased from 728.63 (39.16%) to 544.83 (29.28%) km2, from 64.85 (3.49%) to 34.78 (1.87%) km2, and from 724.53 (38.94%) to 560.63 (30.13%) km2, respectively. In addition, urban index, a non-vegetation index, accurately predicted LST, showing the maximum correlation R2 = 0.87 with respect to retrieved LST. According to CA-Markov chain analysis, we can predict the growth of built-up area from 830.22 to 955.53 km2 between 2032 and 2046, based on the development from 1990 to 2018. As urban index as the predictor anticipated that the LST 20–23 °C and 24–27 °C, regions would all decline in coverage from 5.30 to 4.79% and 15.79 to 13.77% in 2032 and 2046, while the temperature 36–39 °C regions would all grow in coverage from 15.60 to 17.21% of the city. Our results indicate severe conditions, and the authorities should consider some strategies to mitigate this problem. These findings are significant for the planning and development division to ensure the long-term usage of land resources for urbanization expansion projects in the future.


Introduction
According to the United Nations statement of "World Organization Prospects 2014," almost half of the worldwide population, approximately 3.7 billion, lives in cities (Prasad et al. 2022). It is projected that the current population decline and unplanned population development will hit 66% (2.5 billion) by 2050 (Caselles et al. 1991;Lopez et al. 2001;Xue and Su 2017). The urban land use information is critical to city population activity monitoring, government policy-making, and urban management. However, the density of urban systems makes it challenging to classify urban functional zones accurately. Most urban land use classification studies have used features extracted from social media data and high, medium, and low spatial resolution satellite images (Peña 2008). Still, only a few have used both features simultaneously because Responsible Editor: Philippe Garrigues there are not any models available for them to use (Baqa et al. 2021).
Land system science (LSS) is at the forefront of research to generate much-needed knowledge that can help find landrelated pathways to sustainable development (Li et al. 2016). According to land system science, human use of land is a social-ecological system with several interconnected components that influence each other (Shi et al. 2019;Baqa et al. 2021;Tariq et al. 2021). Land system scientists examine the implications for sustainable development through the lens of inter and intragenerational justice, emphasizing the importance of incorporating multiple actor perspectives (especially those of local communities) (Hamza et al. 2021). People need to have a good quality of life now and in the future, but they also need to keep the environment safe.
Despite their achievement in predicting trends of population growth, only one analysis used indices of land cover to estimate predictable spreading of land surface temperature (LST) (Saitoh et al. 1996). While the normalized difference vegetation index (NDVI) has been used to estimate residual city typical ecosystems and prospective LST values, the NDVI is thought to soak into large vegetation fractions, resulting in minimal temperature variation. In previous research, the normalized differentiate built-up index (NDBI) and impervious surface areas (ISA) are better predictors of LST than the NDVI (Sobrino et al. 2013). The NDVI was calculated from a single satellite image (Ahmed et al. 2013), which is a method that is vulnerable to error because the season might vary greatly depending on the type of land cover. It is also necessary to alter the methodology by including seasonal estimates of land cover indices. Other analyses used linear regression to calculate LST on several indices, resulting in the NDVI, soil adjusted vegetation index (SAVI), normalized difference built-up area index (NDBI), urban index (UI), built-up index (BI), and normalized differentiate water index (NDWI) (Shao et al. 2019). If several variables are included in a linear regression model, and the collinearity of the explanatory factors is high, the precision of the resulting dependent variable may be compromised (Ahmed et al. 2013). Environment predictions are as significant as they are dependable, but they suggest that a method for appropriately estimating LST without collinearity errors must first be identified.
Markov chain models (Araya and Cabral 2010) have been used to predict land use land cover (LULC) and urban expansion changes. Markov chain analysis for Doha, Qatar, predicted a 21% increase in built-up area growth by 2020 (Hashem and Balakrishnan 2015). Temperature predictions are made using both a global and a local model, which excludes metropolitan trends and considers their impact (Saitoh et al. 1996). These models require further downscaling because they are at a coarse resolution (Hoffmann et al. 2012). Furthermore, global and regional models highlight temperature changes caused by greenhouse gas emissions, including temperature changes due to the impact of LULC changes. A Markov chain-dependent model provides insights into possible thermal surface features due to changes in vegetation (Ahmed et al. 2013). To achieve many of these objectives, competing claims on land are at the heart of many related disputes (Elliott and Frickel 2015). A variety of interests, ranging from residents to multinational corporations, vie for control of and access to land, including the security of their livelihoods, sense of place, economic assets, preservation of natural habitats, and their claim to territorial sovereignty. These competing interests range from the local to the international (Ahmed et al. 2013;Sayemuzzaman and Jha 2014;Sultana et al. 2014). It is possible to see the recent changes in Pakistan's land use and governance patterns as a manifestation of these actors' power dynamics. People cut down trees, made big plantations for commercial monocultures, and set up special economic zones. More non-governmental organizations (NGOs) concerned with forest conservation are some of the effects of these land-use changes in Pakistan, one of the world's biodiversity hotspots (Ma and Tong 2022;Prasad et al. 2022;Zahoor et al. 2022). This article's overall goal is to investigate the links between sustainable development outcomes and recent land-use changes to identify leverage points and priority areas of concern for more sustainable land governance in Lahore City, Pakistan. If Pakistan does not properly manage the many conflicting claims on land, it will not be able to meet its 2030 Agenda goals.
The research is suitable for predicting LST changes at the same spatio-temporal resolution with changes in LULC patterns, hence is able to model local and regional processes such as urban surface dynamics. Because of its prior successes in quantifying LULC alteration-related flexibility, effects, parsimony, and usefulness, the Markov chain model has enormous predicting possible for future LST. By looking at past urban development patterns, this research can help us better understand how to predict future thermal city conditions. Many studies worldwide imply that development contributes to LST changes, but Pakistan currently lacks a body of knowledge on the subject. For the most part, the country's meteorological research has relied on large-scale climate models and in situ meteorological data, focusing primarily on precipitation and its agricultural implications (Mazvimavi 2010;Charles et al. 2014;Manatsa et al. 2017). This area has had very few attempts to use remote sensing technology for climate research, especially at the microclimate levels found in metropolitan settings. Furthermore, urban development estimations based on remote sensing have focused on assessing the latest alterations in the LULC over the long term (Hussain et al. 2022).
Lahore (capital of Punjab; the most populated province of Pakistan) is the second largest city in Pakistan after Karachi.
In the last few decades, the city has experienced extensive developments. Economic growth is increasing steadily due to certain factors such as industrialization and enhanced residential facilities that are a pull factor in the city's urban expansion (Mumtaz et al. 2020). Therefore, the rapid growth pace, its effect on vegetation, agricultural land, and quickchange identification in Lahore City require testing. Local government and community planning can benefit this town's sustainable growth work. This work is essential to figure out the change in urban areas, the transformation of different land uses, and change detection of the urban sprawl in 28 years. The purpose of this research was to identify and predict the urban growth of Lahore City and further provide key insight into the hitherto and upcoming situation of urbanization to the policy makers and the administrative authorities. These studies have a direct impact on urban planning and the prevention of additional sprawl. A well-calibrated CA-Markov technique was used to simulate and forecast future urbanization on three decades of satellite imagery's categorized LULC maps. To ensure that the model was accurate, it was subjected to model validation using kappa statistics and an error matrix. Spatio-temporal modeling and visual urban sprawl assessment of Lahore City LULC maps have not yet been used to identify the type of sprawl and its movement over time. The city's total growth behavior can be derived and the effects on urban systems can be considered by urban planners as a result. The research findings are expected to constitute the basis for the study's mitigation recommendations. This study identifies the land cover indices (NDVI, NDWI, SAVI, NDBI, and BI) using Landsat (TM, ETM + , and OLI) data representing correlations between LST and LULC changes in Lahore City from 1990 to 2018. Furthermore, the research found out which specific indices work well with the CA-Markov chain to predict the LULC and LST.

Study area
Pakistan's second-largest city is Lahore which is located between 31°15′ to 31°45′N and 74°01′ to74°39′ E. According to shapefile, the estimated area is 1860.55 km 2 , with 217 m elevations above sea level ( Fig. 1) (Mumtaz et al. 2020). The District of Sheikhupura is in the north and west of the city of Lahore, bounded by Wagha at east and the district of Kasur in the south. Lahore became the capital of the province of West Punjab as the Indian subcontinent achieved independence in 1947. In 1955, it became the capital of the newly created province of West Pakistan which in 1970 was known as Punjab (Qadir et al. 2008). Lahore has a tropical semi-arid climate with humid, long, and low summers, dry winter, monsoon, and dust storms. Lahore's environment becomes intense during the May, June, and July as temperatures increase to 36-42 °C. The monsoon seasons commence from late June until August, with heavy rainfall throughout the northern and western provinces. The highest average temperature in town was reported on May 30, 2013, at 48.3 °C (118.9 °F) and on June 10, 2016, 48 °C (118 °F) (Tariq et al. , 2022. The heat index in direct sunlight was recorded at 55 °C (131 °F) when the weather service in the shade officially recorded this temperature. The minimum temperature reported at Lahore City on January 13, 1967, is −1.1 °C (30 °F). The maximum reported 24-h rainfall in the city is 221 mm (8.7 in), which occurred on August 13, 2008 ).

Data acquisition and image processing
Remote sensing data combined with satellite imagery provides spectral, spatial, and temporal analysis for urban sprawl investigations and identifies LULC, LST, and spectral indices. Medium resolution multispectral satellite images were needed to measure the extent of the urban sprawl, LST, and LULC transition. Therefore, Landsat 5, 7, and 8 imagery were obtained from the online site of USGS-EROS (United States Geological Survey-Earth Explorer) website (https:// www. usgs. gov/) with a cloud cover of less than 10% (Tariq and Shu 2020). Table 1 displays the path, row, and acquisition dates of downloaded USGS datasets, accessible and readily available. We used the Fast Line-of-Sight Atmospheric Analysis of Spectral Hypercubes (FLAASH) module to make the atmosphere correction in ENVI v5.4 (Lehoczky et al. 2017). Image analysis was performed after retrieving the images from the satellite, and image processing was conducted to use such images. Layer staking had been the first step in image processing. In ERDAS Imagine 2016, layer staking was performed to unite all the bands to shape a multispectral image. Landsat 7 data has missing data in straps form that should be removed before use. The 2004 satellite images SLC reflectors collected information from the surrounding pixels focal analysis from the spatial toolbox to fill the missing data (Weng et al. 2007). Land surface variations were predicted using satellite images from 1990, 2004, and 2018. These data were used to make the actual prediction. We used the same satellite images for the model efficiency according to Table 1. LULC and LST distributions were predicted using only one satellite image each year, which allowed the calculation of LST and non-urban indices.

Land use land cover mapping and accuracy assessment
Statistical grouping or grouping data image values (spectral pattern identification) into thematic groups or feature groupings are defined as classification. This technique aims to assemble and position all pixels with the same value in a single category/class. LULC maps for 1990, 2004, and 2018 were obtained from Landsat data using maximum likelihood classification (MLC) algorithms. Four LULC categories: barren land, vegetation, built-up area, and water were acquired from each image as shown in Table 2  .
Assuming customarily distributed statistics for each class in each band, MLC estimates the likelihood that a given pixel belongs to one of many classes. ERDAS Imagine 2016 was used to classify the land use categories. In the training stage, analysis for each class was used to develop a statistical characterization signature editor (Hoffmann et al. 2012). These signatures are then used for supervised classification through signature allocation tools in the second stage. The final findings were then analyzed to examine 28 years of improvements in LULC (Xu et al. 2013). We identified the changes of LULC and crossed match with Google Earth images. City shapefile were obtained from the Urban Unit for 1990, 2004, and 2018. They were checked by Google Earth, marked as built-up, and then exported to Arc Map  10.8 (Yuan and Bauer 2007). Classified images for each year were assessed and displayed urban sprawl transition with its corresponding city boundary. Minimum thirty representative GPS locations per class were collected during an April-June 2018 field survey. We used training (70%) and testing (30%) samples for the accuracy assessment of LULC. Using shapefiles of each sample instead of points increases the accuracy of validation samples. LULC data from earlier research, aerial photographs, and topographical maps were used to generate ground-truth regions for assessing classification accuracy. Kappa coefficients (K) and overall accuracy (OA) were employed to evaluate the accuracy of LULC classifications. Every land cover class in Lahore was analyzed using post-classification (Jensen 1983) variations between 1990 and 2018. Overall accuracy was determined by separating the cumulative number of pixels correctly identified from the cumulative number of pixels (Mumtaz et al. 2020) that could be written as N and Xii . Therefore, Xii = number of correctly labeled pixels, or the diagonal value, and N = cumulative number of pixels in the matrix. Kappa statistics incorporate the off-diagonal elements of the error matrices . It was calculated by using the following Eq. 1: where r is the number of rows in confusion matrix; Xii the number of observations in row i and column i (along the major diagonal); x i+ is the marginal total of row i; x +i are the marginal totals of column i ; and N is the total number of observations.

Estimation of LST using Landsat data
Thermal bands from Landsat 5 and 7 (B# 6) and Landsat 8 (B# 10) were used to determine LST for each year. Images taken in May and June were utilized to minimize the impact of seasonality. Landsat 8 has two thermal bands (B10 and B11). In this study, we used band 10 to estimate LST (Datta et al. 2017). To acquire LST from brightness temperature maps, digital values have to be converted to radiances. Radiance and emissivity data were used to adjust brightness temperature calculations (Sobrino et al. 2004;Jimenez-Munoz et al. 2014;). When converting digital numbers (DN) to radiance ( L ), an ArcMap 10.8 addition called the Reflectance toolbox was utilized.
The utility used metadata files to extract parameters and apply them to thermal data. Equation (2), a single-channel Landsat estimate of Planck's blackbody temperature, was used to obtain T B from thermal radiance (Stathopoulou et al. 2006;Deilami et al. 2018). The Landsat surface temperature (LST) was calculated using geometrically corrected Landsat satellite images taken in 1990, 2004, and 2018. The LST was calculated using Eq. 2 (Weng et al. 2004). For each pixel, digital number (DN) was converted into the radiance ( L ) as follows: where L max and L min are the maximum and minimum radiance values, QCAL max the maximum quantized calibrated pixel value (corresponding to L max in DN (255)), QCAL min is the minimum quantized calibrated pixel value (corresponding to L min in DN (01)) respectively, and their values were available from the metadata of the Landsat images. Secondly, the L values were converted into brightness temperature ( T B ) as follows:  where K 1 and K 2 are constant and available from the United States Geological (see Table 3). From every thermal band, we identified from spectral radiance and black body the pixel-based land surface emissivity map (ε), as developed (Yang 2004) and also applied recently (Mushore et al. 2018). Ultimately, real LST was obtained using Eq.
The sign denotes the wavelength of the emitted thermal radiance (11.5 µm), and the symbol p denotes the wavelength of the emitted thermal radiance 1.438 × 10 −2mk . We collected LST data for all dates corresponding to the images shown in Table 3.
To predict LST, training models use long-term temperature fluctuations as a visual cue. Thermal data for 1990, 2004, and 2018 were used to calculate the LST (Table 1). To see if urbanization was still raising the LST, and if so, what kind of development might be on the horizon, this study was carried out in the first place.

Calculation of urban and non-urban indices
Temperature f luctuations across the season were explained using the LST. The average LST for 1990LST for , 2004, and 2018 was calculated using Table 1. This experiment was conducted to see if LST changed due to urban expansion (Xu et al. 2013). The approach (Smakhtin and Hughes 2007) estimated barren land, vegetation, built-up area, and water bodies. Fu's approach created the land cover index maps (Fu and Weng 2016). LST was computed using Landsat satellite images taken in 1990, 2004, and 2018 that had been geometrically adjusted (Tran et al. 2006). The indices of urban and non-urban indices formulas in Table 4 were used to correlate with the LST.

Temperature prediction using various parameters
The LST computation requires a strong correlation between the LST and the predictor selected variables, with no collinearity among them. It was determined by using the indices listed in Table 4. A linear regression model was used to predict LST using spectral indicators with the highest correlation. The correlation between such variables was also examined to avoid tightly clustered predictors, which could cause collinearity-related mistakes. To create a multivariate linear model, we picked indicators that were substantially connected with LST and poorly correlated with each other. We used the model to test its performance to predict 2018 observed LST. Accuracy was measured using mean absolute percentage error (MAPE) shown in Eq. (5) (Mann and Whitney 1947).
where T observed is the real ith pixel of LST reported and T predicted is the model based predicted LST from Landsat info. Error is expressed in percentages using the absolute mean percentage of the accuracy statistic. It was determined that the model's accuracy in predicting temperature could be quantified by calculating the root mean square error (RMS), Nash-Sutcliff performance, agreement index (AI), and mean bias error (MBE). LST distribution for 2032-2046 was predicted using a model evaluated for accuracy. The 14 years were chosen since the analysis demonstrated significant changes at the same points in time.

LULC changes and modeling for 2032 and 2046
The LULC has been represented using various predictive models (Triantakonstantis and Mountrakis 2012). CA-Markov   Mann and Whitney (1947) chain analysis can forecast the possible distribution of LULC and LST based on remote sensing data (Mumtaz et al. 2020; Tariq and Shu 2020) indicated in Fig. 2. It was compared to a genuine 2004 map for comparison. A 2018 state simulation was run during this study to verify that the expected LST distribution was in line with the actual distribution. Detailed explanations of the proposed actions can be found in "CA-Markov chain analysis to predict LULC changes" and "CA-Markov of indices for the prediction of LST" sections.

CA-Markov chain analysis to predict LULC changes
The Markov model permits the structure to advance from the initial state i to j during a time period T (Hashem and Balakrishnan 2015

CA-Markov of indices for the prediction of LST
"LULC changes and modeling for 2032 and 2046" section defines the urban indices (UI) as the most robust predictor variable of LST distribution.

Statistical importance of examining LULC and LST
Predicted changes in LULC and LST distribution were examined for their statistical significance. The test was applied to coded LULC values and to temperature level values derived from 300 points. Each period, the LST groups were classified 1-5, whereas the LULC levels were coded 1-4 according to Markov analysis criteria and performance. The Shapiro-Wilk statistic was employed to assess for regularity in the earliest stages of the study (Shapiro And Wilk, 1965). A Mann-Whitney test was used to determine the significance of LULC and LST deviations (Mann and Whitney 1947). We verified that the Ha hypothesis concerning LULC and LST spatial distributions is distinct from the alternative Hb theory: in 2018 and 2046, the pairs of LULC and LST were not identical ).

Spatio-temporal LULC changes with accuracy assessment
Results in Fig. 3  Our results indicate that vegetation cover decreased in Lahore and other neighboring cities. As seen in Table 5, there have been significant changes in LULC distribution between 1990 and 2018 using the MLC algorithm. Most of the changes were in agriculture/ vegetation and the urban environment. It should be noted that from 1998 to 2004, the rise in the built-up area was very high, but after that, the development in the built-up area was slow until 2018. The above data were validated by previous knowledge and Google Earth images (Mumtaz et al. 2020;

Observed LST changes from 1990 to 2018 using Landsat data
There has been a significant rise in Lahore's long-term LST since 1990. Figure 4 shows that in 1990, in contrast to succeeding years, the region was primarily covered by temperatures between 20-23 °C and 24-27 °C. Although  lower surface temperatures were found in several areas of the main research area in 2018, the 36-39 °C group was most prevalent in 2018. Temperature rises were more significant in the built-up areas of the center than in the surrounding areas of vegetation, water, and barren land. Figure 4a is satellite-derived land surface temperature. Figure 4a shows that in 1990 built-up areas were not too much, so the temperature was also low toward the boundary. In most areas, the temperature was between 20-23 °C and 24-27 °C. Figure 4b shows that in 2004, the built-up area had increased so as the temperature was also high toward the boundary. Figure 4c shows that in 2018, the builtup area had increased, so the temperature was also high toward the boundary. The temperature was between 32-35 °C and 36-39 °C in almost 1/3 area of the city. The low-temperature area had decreased over time, and the high-temperature area had been reduced due to an increase in the urban area in the middle of the city. In 2018, temperature was 36-39 °C in Lahore City. In 1990, most of the area had a temperature of between 20 and 23 °C. The temperature is 32-35 °C; the temperature has risen 12 °C because of urban areas for 28 years. Table 7 summarizes the variations in LST between 1990 and 2018. The percentage of LST in the 20-23 °C temperature range declined by roughly 37% as the city grew. A significant change and rise in the (35-42 °C) high LST coverage over this time period suggests that Lahore City is experiencing substantial warming of the earth. Table 8 shows the yearly variability in LST regarding multiple normalized satellite indices of the Lahore region for 1990, 2000, and 2018. The results show that the annual variability in LST has increased in the last three decades. The LST exhibits substantial interannual variability concerning yearly scales compared to the NDVI, NDWI, SAVI, NDBI, and BI variables. The yearly scale (between 1990 and 2004) revealed a statistically significant positive link between LST and SAVI, NDBI, and BI for the regions located in the central portion of Lahore City, while NDVI and NDWI have a negative association with LST. Negative trends have been seen in the western and the upper part of the Lahore City. A lack of vegetation in these places has contributed to significant changes in temperature. In contrast, an inverse link was detected in a small region (NDVI) of the eastern continent, where it is challenging to evaluate LST patterns due to the  lack of vegetation. There is a strong correlation between LST and the many factors assessed every year. In 2018, LST and indices have shown a 77.2% increase in favorable trends (Tariq et al. 2022). Figure 5 depicts the linear regression model for predicting the LST based on the urban indices (UI). Since surface temperature and UI had R 2 value of 0.87, the correlation was too significant. Thus, LST and UI had improved and did not affect their association due to saturation that disturb indices like NDVI, as the UI continuous to increase with abundant temperature.

Calculation of LST from UI
In 2018, the linear regression model was tested using Landsat data. This was very much in line with the known trends of temperature (Fig. 6). Tsat is the satellite-observed temperature, while Tmod is the derived temperature from the model. The UI temperature and Landsat 8 thermal data were analyzed to compare the two results based on 300 sites in the area of research studied (Fig. 1).

Accuracy assessment of cellular automata Markov chain LULC for 2032 and 2046
Analyzing the data visually revealed a correlation between the MLC classifier's estimate for 2018's LULC distribution and the CA-Markov model's prediction for 2018 (Fig. 7). A set of in situ measurements drives the model to reproduce the MLC-defined spatio-temporal dispersion of LULC. CA-KIA Markov's prediction for LULC was 0.88, while MLC's classifier predicted a dispersal of 0.85 using KIAs (Table 9). Vegetation and water classes on both maps agreed most strongly (KIA = 0.81), with the weakest (KIA = 0.79).

Prediction and distribution of LULC and LST in Lahore before 2046
Figure 8 displays in 2032 and 2046, the CA-Markov chain model predicted an increase in built-up areas associated with barren land, vegetation, and water. Buildings may eventually overtake parks if current trends continue, as they have in the past. Table 10 shows that the extent of built-up area is probably to increase from where they are now until 2046. According to the CA-Markov model, built-up areas are likely to grow from 830.22 to 955.53 km 2 , while vegetation area decreased from 478.91 to 402.83 km 2 from 2032 to 2046. Based on model predictions, expansion in built-up areas and decreases in water, vegetation, and barren land regions are the primary drivers of future development. Figure 8 illustrates the LST growing trend from 2032 to 2046. The model explained LST variations between 2032 and 2046 due to increased built-up area. The size of high LST (more than 36 °C) in Fig. 8a-b was projected to increase at the cost of lower LST classes. Thus, western and northern areas with vegetation were cooler than built-up areas. Maximum area with low LST (20-23 °C and 24-27 °C) was moved to the high LST class area (32-35 °C and 36-39 °C). Most of the areas, particularly in the northwestern and  (Table 10). From 2032 to 2046, the model indicated that the temperature ranges from 24 to 27 °C would decline by 293.80-256.22 km 2 , while the LST ranges from 36 to 39 °C are expected to rise by 290.25-320.17 km 2 . In the long run, temperatures in the high category (over 26 °C) are anticipated to grow at the expense of those in the lower category, as shown in Fig. 8a-b. There will be a significant increase in temperature in the built-up area. It is expected that most   Table 10.

Discussion
This study used cellular automata Markov chain models to predict land use, land cover, and LST in Lahore, Punjab, Pakistan. A number of land cover indices were evaluated in terms of their capacity to predict temperature changes in relation to time. It was found that the NDWI index was the best predictor of LST distribution, followed by SAVI, NDVI, NDBI, and BI, among others. Linear regression was used to predict the expansion of cities and the temperature of their surrounding areas. Multiple linear regression is appropriate when the projected variables are not connected to each other (Pauleit et al. 2005;Ahmed et al. 2013). It was determined that UI was the best predictor of temperature because all other variables were associated. We used the UI as a model for the spread of urbanization and its expected trajectory to estimate the potential effects of LST on the environment. With an absolute average inaccuracy of 1.85 °C, the model predicted the temperature using LST from the thermal band and a linear model utilizing UI. Recent studies show that the accuracy of UI's projections of heat from urban development is closely tied to a number of other indicators of urban growth. In Tokyo Bay, for example (Masek et al. 2000), UI was found to rise with building density while decreasing with NDVI. The great predictive power shown in this analysis is due to higher temperatures in places with built-up area and less vegetation, despite the fact that the link between temperature and UI has not been validated in earlier research. Sri Lanka and Colombo's water-intensive and residential areas  were likewise found to have high urban indices (Pauleit et al. 2005). According to studies, the severity of urban weather has a direct impact on water and energy consumption in the home, which explains the tight link between UI and LST (Cohen 2004). For a high-accuracy classification of urban LULC spread, the SVM was tried in 1990 and 2018. Furthermore, a recent study (Adelabu et al. 2013) demonstrates that the SVM classifier may produce high-precision maps. Use of digitalized regions rather than points as ground data for categorization results in the map's high level of precision, which surpasses the required 80% (Omran 2012). Between the years 1990 and 2018, the produced maps of LULC showed that residential areas expanded while vegetation and water cover decreased in the same region. This finding is consistent with the findings of earlier studies (Kamusoko et al. 2013). The high degree of agreement that exists between the projected map for the year 2018 and the map that was created using the supervised rating is evidence that the CA-Markov chain makes accurate predictions of the types of land use and land cover. The CA-Markov chain model predicted that based on variations in LULC between the years 1990 and 2018, coverage of built-up area will tend to grow until 2046 at the expense of land cover unless other measures such as green spaces, cropland, and water are implemented and identical trends exist. This prediction was based on variations in LULC between the years 1990 and 2018. With worldwide estimates predicting that human population expansion would continue to outpace the increase of natural areas, this finding is consistent with this conclusion (Araya and Cabral 2010;Hussain et al. 2022). According to this study, the lowest temperature level region is expected to decline, while the area protected by warmer categories like 36-42 °C and 26-32 °C is expected to rise. There will also be an increase in urbanization, which will lead to a decrease in green space and water availability as a result of the increasing tendencies. According to the land use land cover, the expected temperature rises due to urban growth are similarly consistent with past findings (Day et al. 2013). This study will help to modify the planning and management of the city of the Punjab such as in the department of land record, improvement of solid waste management through infrastructure support and capacity building, construction/ development of landfill sites, elimination of ponds and support local government or union council in rehabilitation of dangerous buildings will uplift the image of the city.

Conclusions
This research was carried out to evaluate urban growth phenomena and their impacts on vegetation and change in land surface temperature from 1990 to 2018. RS and GIS techniques were used to analyze the spatio-temporal patterns of land use. LST and LULC distributions in Lahore will be predicted using a new CA-Markov chain developed in this research. For urban areas categorized using Landsat 5, 7, and 8 imagery, we observed that the MLC method utilized obtained an overall accuracy of above 80% when applied. According to MLC classifications of LULC types, in situ measurements could drive the model's ability to reproduce the LULC types' spatio-temporal distribution accurately. It was shown that the CA-Markov model predicted a KIA of 0.88 and 0.85 between the LULC and the distribution obtained using the MLC classifier. This analysis can predict the growth of built-up area from 830.22 to 955.53 km 2 between 2032 and 2046, based on the development from 1990 to 2018. Considering the CA-Markov chain models based on UI anticipated that the temperature 20-23 °C and 24-27 °C regions would all decline in coverage from 5.30 to 4.79% and 15.79 to 13.77% in 2032 and 2046, while the temperature 36-39 °C regions would all grow in coverage from 15.60 to 17.21% of the city.
As a result of increased urbanization in Lahore, the observed values between 1990 and 2018 in the categories 20-23 °C and 24-27 °C declined by roughly 37% and 22%, respectively, indicating a severely biased trend of land heating in Lahore City. Only the effects of urban development on temperature variations are considered in the model, which is limiting. However, there has been a minor shift in LST patterns due to successful mitigation policies and innovative urban expansion methods. In general, the outcomes of this study show the utility of mediumresolution satellite data in estimating possible land surface temperatures in urbanized areas. In the absence of control measures, we conclude that urbanization will enhance warming and increase temperatures in the future. Urban planners must be warned of urban dwellers' potential temperature fluctuations and thermal comfort due to growth based on this study. Meanwhile, more research is needed to see if these methods and procedures can be applied globally and nationally. The CA model and Markov chain analysis were employed to identify spatial distribution changes and forecast time resolution. This model will be used in the future by researchers and policymakers to develop new policies to control and manage urban expansion.