The surface reflectance products of Landsat images from each year were used for land use/ land cover classification in Google Earth Engine (GEE). The cloud mask function was applied to get cloud-free pixels. The normalized difference built-up index (NDBI), the normalized difference water index (NDWI), and the normalized difference vegetation index (NDVI) were calculated for each Landsat image. The NDBI was calculated using the NIR and SWIR bands (for Landsat 5, the formula is (Band 5Band 4) / (Band 5 + Band 4); for Landsat 8, the formula is (Band 6 – Band 5) / (Band 6 + Band 5). Gao. (1996) developed the Normalized Difference Water index in order to enhance water bodies. This index uses the NIR and SWIR bands [for Landsat 5, NDWI = (Band 4 – Band 5) / (Band 4 + Band 5), while for Landsat 8 the formula is (Band 5 – Band 6) / (Band 5 + Band 6)]. Similarly, the NDVI uses the NIR and red bands [for Landsat 5, NDVI = (Band 4 – Band 3) / (Band 4 + Band 3), while for Landsat 8 the formula is NDVI = (Band 5 – Band 4) / .(Band 5 + Band 4)].

To classify the Landsat images for the years 1990, 2000, 2010, and 2020, the random reference points were generated for four major land use categories, i.e., barren, built-up, water bodies, and agricultural land. Nearly 3500 reference points were generated for each year, of which 70% were used as training samples to create land use maps using the RF algorithm, and the remaining 30% was used for validation purposes. After obtaining LULC maps for each year, an accuracy assessment was performed for each LULC map using validation reference points, and an error matrix was employed to produce overall accuracy for each LULC map.

For the assessment of change in LULC, the classification results were used to observe the trend of land use change in Larkana and analyzed in ArcGIS 10.8. The LULC maps were categorized into four classes: built-up areas, water bodies, agricultural land, and barren land. The built-up area includes residential, commercial, and other urban areas like roads, etc., while the agricultural land includes natural vegetation, cropland, and orchards. The results were then further assessed to calculate the temporal land use change using the following equation: r = 100 × A 2 − A 1 A 1 × 1 t 2 − t 1 (1) where t1 is the initial time, t2 is the final time, A1 is the land use class cover at the initial time, and A2 is the class cover at the final time.

Furthermore, the land surface temperature (LST) for each temporal image was calculated using Google Earth Engine. The LST data were then downloaded, and the map was prepared in ArcGIS 10.8. The LST minimum and maximum were acquired from each temporal image for further interpretation. The LST was also analyzed with respect to each land use class in this study. The LST was extracted from the thermal bands of Landsat satellite imagery using the following procedure:

The results of LULC maps derived through the RF algorithm in GEE were validated using the 30% sample points. For this purpose, the kappa coefficient technique was used in this study. The 30% validation dataset points were used to compare the original land use type to the classified image. For this purpose, the value of the classified image assigned to each point was extracted using ArcGIS, and then the original image and Google Earth imagery were used as reference data. An error matrix table was generated to quantify and summarize the total accuracy of the classified images. The results showed good performance of the RF algorithm for LULC classification. The overall accuracy in 1990 was 88.3%, 89% in 2000, 90.01% in 2010, and 90.8% in 2020.

The multi-temporal satellite-based LULC mapping for the years 1990, 2000, 2010, and 2020 was categorized into barren land, built-up area, water bodies, and agricultural land. Agricultural land includes orchards and cropland. For each respective year, the Landsat satellite images were converted to color composite with bands 4,3, and 2 RGB for Landsat 5 and 5,4, and 3 for Landsat 8 satellite, using the standard false color composite band combination (Figures 2, 3). The land use for the multi-temporal years was generated from the Landsat 5 and Landsat 8 images using the RF algorithm classification method in GEE. The multi-temporal satellite-based LULC mapping results exhibited a high increase in the built-up area, which was 12.31 km² in 1990 and rose to 19.87 km² in 2000, 23.38 km² in 2010, and 43.83 km² in 2020 (Figures 2, 3, 4). In 1990, the built-up area was only 8.62%, which grew to 13.99% in 2000, 16.68% in 2010, and 30.70% of the total study area in the year 2020 (Figure 5B). During the period 1990-2010, it took two decades to double the built-up area, while during the period 2010-2020, it doubled again in just one decade. This surge in the built-up area was the result of a number of factors, including the special attention given to the city of Larkana during the People’s Party government in 1988-1990, 1993-1996, and 2008-2013. During 1988-1990, Larkana was upgraded to a division, a number of educational and health institutions were built, and housing colonies like OPF and Sindh Green Housing Society were initiated, which boosted the expansion of the city. During the 1993-1996 government, the city was given further attention, and the Board of Intermediate and Secondary Education Larkana was established in 1995. The city continued to grow to the southwest and northeast with the construction of new roads, a bus terminal, new hospitals, some new educational institutions, planned housing settlements, and small industrial areas. In the LULC map of 1990, most of the barren land is actually the land that has been cleared for different development projects in the city area. The expansion of the built-up area in 1990-2010 was almost all around the city center with a few patches in the surrounding areas, but in 2020 the city expanded along the main canal and to the west. The other factors contributing to the increased built-up area of Larkana include the natural growth of the population and people’s migration from the suburbs seeking a safe place and jobs after the massive 2010 floods. The results showed that during 1990-2010, the built-up area and agricultural land expanded over the barren land (Figures 2, 3), and the same was confirmed by the local inhabitants, who said that there was a massive amount of waterlogged and saline land up until 1990, which was later converted to agricultural land. In 1990 and 2000, barren land was the second largest LULC category, while from 1990 to 2020, agricultural land remained dominant in the study area, which substantially decreased in the last two decades, i.e., between 2010 and 2020 (Figures 4, 5).

Agricultural land accounted for 45.65% of all land in 1990, gradually increasing to 71.02 % by 2010 and decreasing again to 55% by 2020 (Figure 5A). From 1990 to 2010, barren land was the second largest LULC category in terms of area occupied, which was overtaken by built-up area in 2010, thus becoming the second largest LULC category with 16.68% of the total area in 2010 and 30.70% of the total area in 2020. The built-up area in the study area expanded in all directions during the period 2010-2020, mainly along the major communication routes (Figure 2). A huge change was observed in the built-up area between 2010 and 2020, where it increased from 16.68 % in 2010 to 30.70 % in 2020 at the expense of agricultural and barren land in the study area. The major reason for this was the mass migration from nearby areas affected by the 2010 floods, especially from Kamber-Shahdadkot, Dadu, and Shikarpur districts and other rural areas. The population spread to the western side after the 2010 floods, mainly because the area is safer from floods than the eastern side. The new expansion on the western side also included government development such as the extension of the industrial area and the building of warehouses for rice sellers. Some new housing projects were also launched and developed on the western side of the city after 2010. The barren land in the study region decreased from 39.55% in 1990 to 8.14% in 2020. This is due to the conversion of the built-up area into agricultural land. The fourth category of LULC is water bodies in Larkana city. From the analysis, an increase in water bodies was observed in the study area. This increase is due to the intensification of inland fish farming (Figure 4D).

The results revealed that the agricultural land in the study area increased from 1990 until 2010 (Figures 4, 5A). In 1990, the total area covered by agricultural land amounted to 46.65% and it increased to 65.48% in 2010, to 71.02% in 2010, and it decreased again to 55.69% in 2020. According to the field observation and discussion with the local elders, the increase in agricultural land in Larkana and its surrounding areas is the result of clearing and converting the waterlogged and saline areas into agricultural land. Most of the surrounding areas in the northern and southern parts of the city neighboring were once waterlogged and saline, and were later converted to agricultural land with support from the local community and the government. After 2010, the agricultural land decreased and the major reason is the continuous increase in the built-up area; the current study also found that the majority of agricultural land was used to expand the urban area. Therefore, we can observe that the built-up area has an incrementing trend. The relative change in the built-up area between 1990 and 2000 was 5.4%, which increased to 14% between 2010 and 2020 (Figures 4, 5B).

The overall LULC area in 1990 and 2020 and their respective changes in percentage have been shown in Table 2 and Figure 7. The results indicate that agricultural land increased by 19.24% during the study period, while the increase in the built-up area was 256.00%. The barren land in the study area decreased by almost 79.44% during the study period (Figure 7). The increase in the built-up area has grabbed the barren land and agricultural land, in the study area. Most of the changes in the built-up area were observed between 2010 and 2020 and the main reason was a mass migration from the surrounding areas after the 2010 floods. A slight increase in water bodies was observed from the study results, amounting to 5.23% (Table 2). The increase in water bodies is due to an increase in fish farming in the study region.

In this study, the change from one land use category to another was calculated (Figure 8). For this purpose, the post-classification change detection technique was applied to quantify the change in each land use category between 1990 and 2020. The change matrix technique was adopted for this purpose to analyze the LULC changes. The LULC conversion from one category to another is represented by a two-way cross-matrix, and its corresponding value was calculated from 1990 to 2020. The largest LULC change, measuring 28.71 km², went from barren land to agricultural land, followed by barren land to a built-up area (18.13 km²). Similarly, a 12.63 km² area of agricultural land was converted into a built-up area during the last three decades (Figure 8). During the same period, 3.15 km² of agricultural land was converted into barren land.

In this study, we also used the Landsat thermal band to calculate the land surface temperature (LST) of the study area. The LST map was arranged into five different classes (Figure 9A). It can be observed that in 1990, the temperature of the study area was generally under 28°C. There were only a few patches of temperature class ≥ 32.1°C because most of Larkana was covered with agricultural/open land and the built-up area was smaller (Figures 6, 8A). As the agricultural land decreased and the built-up area increased, the surface temperature started to rise and continued to do so until 2020. In the year 2020, more than 80% of the area within the city boundary experienced high surface temperatures because impervious surfaces absorb more heat, which is the main reason for high temperatures in cities (Moazzam et al., 2022). It can be observed that in 2000 and 2010, the patches of high surface temperatures grew, but in 2020, the high surface temperature expanded over the city border zone and the surrounding temperature also remained high (Figure 9A).

The Figure 9B revealed that the research area’s mean minimum and mean maximum temperatures grew steadily over the course of the investigation. In 1990, the minimum temperature was 25.1°C, while the maximum temperature was 33.4°C. By 2020, those numbers eventually grew to 26.6°C and 36.1°C, respectively (Figure 9B). Therefore, we further evaluated and compared the built-up area with the mean temperature of the urban area, and we found that as the urban area increased, so did the mean surface temperature of the urban land (Figure 8C). The mean temperature in 1990, when urban land was only 8.62% of the total area, was 29.25°C; however, as the urban area expanded, so did Larkana and the surrounding areas’ mean surface temperature, reaching a maximum of 36.1°C, and 31.35°C when the urban area increased to 30.70% in 2020 (Figure 9C).

We have extended our analysis and tried to find out the difference in the mean LST of the built-up area with barren land, agricultural land, and water bodies. It was observed that barren land has a higher surface temperature than the built-up area (Figure 10). However, the surface temperature of built-up land was higher than that of agricultural land and water bodies, and it increased consistently with time (Figure 10). In Table 3, the minimum, maximum, and mean LST and their respective standard deviations were compared for each LULC category in each year. The results of LST show a continuous rise in minimum, maximum, and mean LST for each LULC category from 1990 to 2020.

The lowest recorded mean average temperature in Larkana was 15°C in January, with a maximum temperature of 22.7°C and a minimum temperature of 7.6°C in the same month (Figure 10). It gradually started to increase and reached its highest level in May and June. The highest minimum, maximum, and mean temperatures in the study area were observed during May and June, while they started to decrease in July and August. The monsoon rains occur in the study area during July and August, which decrease the temperature to an optimal level. The annual range of mean temperature in the study area is almost 21^o C (Figure 11).

The change in LST during the study period in different land use categories was calculated as shown in Table 4. The results showed an increase in LST in all land use categories from 1990 to 2020. The highest growth in mean and maximum LST was observed in the built-up area category, while the lowest increase in mean LST was observed in the water bodies category, and the lowest increment in maximum LST was observed in the barren land category (Table 4). Similarly, the lowest increase in minimum LST was observed in the water bodies category, and the highest was in the built-up area category. The highest increase in minimum LST in the barren land category was observed between 2000 and 2010, as was the highest increase in mean LST. Similar results were revealed for other land use categories, and the highest minimum and mean LST were observed between 2000 and 2010. The main reason behind this high LST might be the highest temperature in the year 2010 during May and June in Sindh, which caused a severe heat wave in different cities of Pakistan and also created an unusually low-pressure area that attracted the monsoonal winds and became the cause of massive flooding (Rahman and Khan, 2013). In 2010, the highest temperature ever recorded in South Asia was 53.5 ^o C on 26 May at Mohenju-Daro, near the city of Larkana (Rahman and Khan, 2013). Agricultural land also showed the highest LST during 2000-2010, indicating that crops were also under high stress during this period. Similar results were observed for water bodies.

The annual mean air temperature obtained from the Larkana meteorological station for each particular year was compared with the annual mean land surface temperature derived from Landsat imagery. Both the annual mean air temperature and the mean annual LST show a gradual increase in temperature in the study area (Figure 12). In 1990, the mean annual air temperature was 26.87°C, which increased to 28.05°C in 2020, while the mean annual LST was 29.25°C in 1990, which rose to 31.35°C in 2020 (Figure 12).