Introduction

The term “land cover” describes the traits and features of the surface of the Earth, including vegetation, water bodies, metropolitan centers, and arid regions. The way in which these land cover types are utilized is termed as land use. LULC is a critical modifier of socioeconomic, biological, climatic, and hydrologic systems1. Globally, LULC has gradually changed to enable human survival; as a result, human activity has transformed 33–50% of the Earth’s surface2. Infrastructure development and the expansion of impervious surfaces have led to a transformation in agricultural land3, water bodies4, and vegetation areas5, ultimately driving urban sprawl. Conceptually, urban sprawl refers to the uncontrolled expansion of urban areas into surrounding rural or natural landscapes, including vegetation, farmland, open spaces, and water bodies6. It is typically characterized by rapid, unplanned growth at the urban fringe, resulting in fragmented and discontinuous development patterns7.

Studies have shown that urban sprawl not only leads to inefficient land use but also increases surface runoff, environmental degradation, and vulnerability to flooding in peri-urban areas and can significantly elevate flood peaks8. Given the influence of urban development on surface runoff, numerous studies have focused on quantifying the impact of LULC changes on flooding and economic losses9. For example, Tong10 observed that urbanization has led to frequent flooding in the Little Miami River Basin. Arnold & Gibbons11 have shown that urbanization and the corresponding increase in impervious surfaces can significantly alter the hydrological cycle, leading to increased peak discharge and reduced lag time between rainfall and peak flow, thus heightening flood risks. Im et al. used the Hydrologic Simulation Program-Fortran (HSPF) to evaluate the future hydrological impacts of urbanization in the Polecat Creek watershed in Caroline County, Virginia, finding that increased urbanization correlates with higher runoff volumes12. Moreover, changes in LULC, such as deforestation and agricultural expansion, can lead to soil degradation and reduced infiltration, further contributing to flood risks13. Frequent alterations in LULC can intensify this vulnerability. For instance, Barai14 observed that the South-West coastal area of Bangladesh has faced different unforeseen disaster due to the frequent LULC changes. Such frequent changes are indicative of either human-induced activities or natural events, highlighting the urgent need for strategic planning in these areas to reduce disaster risk. Collectively, these studies emphasize the critical role of continuous LULC monitoring and the identification of transformation hotspots and cold spots for effective urban and environmental planning, especially in flood-prone regions.

Bangladesh, which is located at the meeting point of the Ganges, Brahmaputra, and Meghna (GBM) river basins, is considered one of the most flood-prone areas in the world due to its distinct topography. In the northeastern region of Bangladesh, flash floods are common due to the hilly terrain and steep river slopes15. Therefore, understanding and managing LULC changes is crucial for flood risk reduction and sustainable land management in this region. The Sylhet Division comprises districts such as Sylhet, Moulvibazar, Habiganj, and Sunamganj, all of which are impacted by these flood dynamics. Sylhet district, in particular is traversed by the Surma and Kushiyara rivers, both of which are prone to overflow during the monsoon season. With some of the highest rainfall levels in Bangladesh, Sylhet frequently experiences flash floods, as the intense precipitation can overwhelm the region’s river systems and drainage capacity, leading to localized flooding. Hence, this study focuses on Sylhet district, which faces significant vulnerability to flooding due to these factors.

Satellite imagery is considered one of the most efficacious tools for measuring qualitative and quantitative LULC changes16 over a wide range since it captures images from space and continuously records global land surface17. Various spectral indices derived from satellite images are widely used to improve LULC classification18. The Normalized Difference Vegetation Index (NDVI) is a key indicator for assessing vegetation health and distinguishing vegetated areas from non-vegetated surfaces19. Similarly, the Modified Normalized Difference Water Index (MNDWI) enhances the identification of water bodies by suppressing built-up and vegetation features20, while the Enhanced Vegetation Index (EVI) improves vegetation detection in regions with dense canopy cover21. These indices help in refining LULC classification by leveraging spectral differences in vegetation, water, and impervious surfaces. However, Geographic Information System (GIS) technology offers advanced tools for analyzing these images on national, continental, and global scales22. But there are three main obstacles that traditional GIS technologies must overcome: (a) three-dimensional preprocessing and mining geographic big data; (b) effective algorithms for handling geospatial data from regional to worldwide dimensions; and (c) performing spatial-temporal analytics23. The traditional GIS systems are widely used for spatial data visualization, mapping and fundamental spatial analysis but they face significant challenges when dealing with large-scale, multidimensional and real-time data24,25. To address these challenges, Google introduced the Google Earth Engine (GEE) cloud-based computing platform in 201026. Many researchers including Haque et al.18, Nourizadeh et al.27, Becker et al.28, Ang et al.29, and Ghorbanian et al.30 have used GEE to prepare LULC maps. So, GEE is now the most frequently utilized platform for processing large geospatial datasets including LULC preparation, providing free access to a vast array of remotely sensed data31.

While previous studies32,33 have analyzed LULC in Sylhet Sadar Upazila at a smaller scale using traditional GIS technologies, there has been no comprehensive, district-wide quantitative analysis of LULC changes over an extended period using the GEE platform. This study addresses this gap by monitoring LULC changes and transitions across Sylhet district from 2000 to 2023 using GEE. The main objectives are to quantify the various LULC patterns in this district and to assess the LULC changes and to identify LULC transition hot spots and cold spots throughout the course of the study.

Materials and methods

Study area

Sylhet district is situated between latitudes 24°36’ and 25°11’ north and longitudes 91°38’ and 92°30’ east34, as shown in Fig. 1. This northeastern district of Bangladesh is bordered by India’s international boundaries to the north and east. Sylhet comprises twelve upazilas: Beani Bazar, Bishwanath, Jaintiapur, Dakshin Surma, Sylhet Sadar, Gowainghat, Kanaighat, Balaganj, Companiganj, Fenchuganj, Golabganj, and Zakiganj35. The southern portion of this district has mushy areas, whereas the northern part is made up of hilly rocks36. The main rivers flowing through this district are Surma and Kushiyara. The Sub-Recent Piedmont Basin and the Surma Kushiyara Flood Plain Ridges combine to form the majority of the district’s soil35.

Data

The required satellite images are taken from the GEE data catalog. Based on the data availability, Landsat 7 ETM + and Landsat 8 OLI satellite images are utilized in this study. To ensure consistency in temporal analysis, six specific years were selected at approximately five-year intervals: 2000, 2005, 2010, 2015, 2020, and 2023. The year 2023 was chosen as it marks the beginning of the study period. For each of these years, all available images within the calendar year were used, and cloudy pixels were masked using the pixel quality assessment (QA) band. Median composite images were generated to minimize the influence of seasonal and atmospheric variability. The descriptions of satellite images utilized in this analysis are listed in Table 1. The Landsat 7 satellite image contains 15 bands, of which 9 bands are used for data analysis. On the contrary, the Landsat 8 satellite image contains 17 bands, of which 11 are used for data analysis (Table 2). Though the imagery has different band resolutions, however, when stacking or combining bands with different resolutions in GEE, the system automatically aligns bands to the coarsest resolution using nearest-neighbor resampling. Additionally, to improve classification accuracy, three indices are added to the satellite image as distinct bands: the Enhanced Vegetation Index (EVI), Modified Normalized Difference Water Index (MNDWI), and Normalized Difference Vegetation Index (NDVI)18.

Table 1 List of Landsat satellite images used in this study, including sensor type, acquisition time, and data source.
Full size table
Table 2 Spectral bands used in the classification process, including the pixel resolution for Landsat 7 ETM + and Landsat 8 OLI.
Full size table

In this study, a supervised machine learning algorithm “Random Forest” (RF) is used to identify the satellite images using four different types of LULC: urban, barren, water body, and vegetation. For this purpose, known LULC types called ground control points (GCPs) are used to train the algorithm. Table 3 provides a summary of the data regarding the number of GCPs used in this study.

Table 3 Number of ground control points (GCPs) used for training and validating the random forest classifier across different LULC classes and years.
Full size table
Fig. 1
figure 1

Geographical location of Sylhet district, Bangladesh, showing the administrative boundaries and major river systems. Maps created in QGIS 3.28.12, https://qgis.org/.

Full size image

Methods

The methodological steps followed in this study are summarized in Fig. 2, which outlines the complete workflow from image acquisition and preprocessing to classification, accuracy assessment, change detection, and spatial hotspot analysis. The imagery datasets that are used in this study are accessed through the GEE data catalog and imported in the GEE code editor for further analysis. These datasets are preprocessed in several steps to get an accurate LULC classification map.

Image preprocessing

Landsat imagery provides a continuous time series of Earth’s land surface. For this study, the imagery was filtered based on the region of interest (ROI) and temporal requirements to obtain a suitable dataset. Cloudy pixels were then masked from each image in the collection, and the dataset was subsequently reduced to a single composite image for LULC analysis. Essential bands were selected, and three indices - NDVI, MNDWI, and EVI - were calculated and added as individual bands to the composite image. The equations for calculating NDVI, MNDWI, and EVI are presented in18 as follows:

$$\:NDVI=\frac{NIR-Red}{NIR+Red}$$
(1)
$$\:MNDWI=\frac{Green-SWIR1}{Green+SWIR1}$$
(2)
$$\:EVI=2.5\times\:\frac{NIR-Red}{NIR+6\times\:Red-7.5\times\:Blue+1}$$
(3)
Fig. 2
figure 2

Workflow diagram of the methodology used in this study, detailing the preprocessing, classification, accuracy assessment, and spatial analysis steps performed in GEE.

Full size image

Classification map

The RF classifier, a group classifier that builds multiple decision trees using a randomly selected subset of samples and covariates, is used to classify the reduced image37. The voting procedure that each tree goes through during the classification process determines the final classification38. All land use-type GCPs for a specific year were aggregated and assigned a property value for identification in order to categorize them as LULC. This process served as training data for the decision trees of the RF classifier. Of these, 40% were utilized as validation GCPs, and the residual 60% were used as training GCPs. Besides these training GCPs, six parameters are typically input to the classifier: the number of decision trees, the number of variables (or features) per split, the minimum number of leaves, the fraction of input to bag per tree, the maximum number of leaf nodes in each tree, the randomization seed, and the number of variables (or features) per split39. Taking five hundred decision trees as recommended by Belgiu et al.37 due to error stabilization and default value for other input parameters, the RF classifier classifies all other ground points with the help of training GCPs and provides a clear classification map of the study region. The classification map is then exported in QGIS for further procedures.

Evaluation of accuracy

Evaluating the accuracy of the classification map is crucial to determining its reliability. In this study, accuracy was determined using validation GCPs in the GEE code editor, employing a confusion matrix. The confusion matrix’s overall accuracy (OA) is a crucial parameter for evaluating the correctness of the map18.

Area calculation

The area of a specific LULC type is calculated using a binary image derived from the classification map. In this binary image, pixel values are either one or zero: pixels corresponding to the particular LULC type are assigned a value of one, while all other pixels are assigned a value of zero. The total area of the LULC type is then determined by summing the areas of the pixels with a value of one.

Change detection

LULC changes were analyzed using the post-classification comparison technique on the GEE platform. Each classified image comprises four distinct land use classes, where each class assigned a specific value. LULC change refers to the portion of one land use class that has changed into another over time. This process results in 16 possible outcomes of LULC transitions.

The transition image (\(\:{T}_{i}\)) is generated using the equation:

$$\:{T}_{i}={P}_{i}\times\:100+{F}_{i}$$
(4)

Where, \(\:{P}_{i}\) denotes the previous classified image and \(\:{F}_{i}\) indicates the final classified image. Using the transition image, an LULC change matrix is created, providing a comprehensive overview of the transformations among land use classes.

Land transformation hot spot analysis

To analyze land transformation hot spots, binary images were created for each LULC transition, where 0 (zero) represents “no change” and 1 (one) indicates “change.” These binary images were derived from annual LULC maps generated using classified satellite imagery (e.g., Landsat data).

Each binary image served as a weight image for subsequent hot spot analysis. A weighted overlay image was then generated by summing the binary images across all years, representing the frequency of land use changes.

Using Getis-Ord Gi* statistics40, hot spots (areas with frequent changes) and cold spots (areas with minimal changes) were identified. A threshold Z-score of ± 0.1 was applied, allowing the detection of subtle clustering patterns suitable for identifying widespread but less concentrated land transformations. While this threshold prioritizes sensitivity, it is supported by observed change dynamics in the study area.

This approach provided a spatially explicit representation of land transformation dynamics, highlighting regions requiring targeted management and further investigation.

Results and discussion

LULC changes over time

Figure 3 illustrates the LULC dynamics in Sylhet district for the years 2000, 2005, 2010, 2015, 2020, and 2023. The overall accuracy of these classification maps is 100%, 94%, 100%, 96%, 93%, and 100%, respectively. In a related study, Pande et al. analyzed LULC classifications in Akola district, Maharashtra, India, for 2008 and 2015, achieving overall accuracies of 94.1% and 88.14%, respectively41. Similarly, Seyam et al.42 evaluated LULC maps for Bhaluka in Mymensingh, Bangladesh, and reported overall accuracies of 87.2% for 2002 and 89.6% for 2022. In this study, the minimum overall accuracy achieved is 93%, which provides high confidence in the observed LULC maps.

The LULC maps reveal that vegetation was the most dominant LULC type in 2000, covering 50.51% of the district’s area (as shown in Table 4). These vegetation regions were primarily situated in the northern parts of Companiganj, Gowainghat, Jaintiapur, and Kanaighat upazilas, the southern parts of Zakiganj, Beani Bazar, and Fenchuganj upazilas, and the central region of Sylhet district, specifically in Sylhet Sadar Upazila. The second most dominant LULC type in 2000 was water bodies. Table 4 lists the areas and percentages of different categories of LULC for each year.

In 2005, the areas of vegetation and water bodies dramatically reduced from 50.51 to 24.38% and from 34.25 to 16.34%, respectively, with bare land becoming the most dominant LULC type, covering 56.66% of the area. By 2010, there was a slight reduction in the areas of vegetation, aquatic bodies, and barren land, while urban areas saw a significant increase. Urban development was particularly noticeable in the southern part of Sylhet Sadar Upazila and the eastern part of Beani Bazar Upazila (Fig. 3). The urban coverage in 2010 was approximately 67,000 hectares, accounting for about 19.5% of the total area of Sylhet district.

Fig. 3
figure 3

Land Use and Land Cover (LULC) classification maps of Sylhet district for the years 2000, 2005, 2010, 2015, 2020, and 2023, highlighting changes in urban areas, vegetation, water bodies, and bare land. Maps created in QGIS 3.28.12, https://qgis.org/.

Full size image
Table 4 Area (ha) and percentage of total land occupied by each LULC category for the years 2000, 2005, 2010, 2015, 2020, and 2023.
Full size table

In 2015, urban areas significantly decreased from 19.5 to 8.5%, while other LULC types remained relatively stable. In 2020, urban and vegetation areas slightly increased, while bare land and water bodies decreased. Finally, in 2023, urban areas, water bodies, and vegetation areas increased, covering about 18.5%, 7.5%, and 50.5%, respectively, of the total area of Sylhet district. During this year, bare land decreased from 45% to approximately 23%.

LULC change pattern

Time series analysis

Table 5 presents the LULC changes from one type to another over the study period. Approximately 156,000 hectares, or about 45% of the total land area of Sylhet district, remained unchanged over the last twenty-three years (Table 5). The study indicates that while most LULC types have increased, the area covered by water bodies has decreased (Fig. 4(a)). The most significant increase was observed in urban areas (from 4.65 to 18.46%), while the smallest increase occurred in vegetation areas (Fig. 4(b)). By 2023, around 7% of water bodies from 2000 had been converted into urban areas, 5.5% into bare land, and 15% into vegetation. Conversely, only a small portion of other LULC types converted into water bodies by 2023. This reduction in water bodies, particularly drainage areas, is a concerning trend that poses a risk to the district’s ability to manage excessive surface runoff.

Table 5 Percentage of LULC transitions between 2000 and 2023, indicating the extent of conversion from one land cover type to another.
Full size table
Fig. 4
figure 4

(a) Total area (ha) covered by different LULC classes from 2000 to 2023. (b) Percentage of total land area occupied by each LULC category over time, indicating urban expansion and waterbody decline.

Full size image
Change in urban area

In 2018, S. Gupta et al.33 analyzed urban growth in the Sylhet Sadar upazila, noting significant expansion primarily in the Sylhet City Corporation and its surrounding areas, situated between latitudes 24°52’ and 24°56’ north and longitudes 91°50’ and 91°54’ east. This trend is also evident in our study. However, Fig. 5 provides a more comprehensive overview of urban development across the entire Sylhet district from 2000 to 2023, with snapshots from 2000, 2005, 2010, 2015, 2020, and 2023.

In 2000, urbanization in Sylhet was relatively modest, with built-up areas sparsely distributed, primarily concentrated in central regions such as Sylhet Sadar and adjacent Upazilas. By 2005, a slight expansion in urban areas is observed, with notable growth in Sylhet Sadar and surrounding Upazilas, including the emergence of new urban zones in Dakshin Surma and Golapganj. The trend of urban expansion becomes more pronounced by 2010, with significant growth in central regions, particularly in Sylhet Sadar, Dakshin Surma, and Golapganj. Additionally, urbanization extends to other Upazilas like Bishwanath and Kanaighat. The momentum continues in 2015, with a marked increase in the extent and density of urban areas, especially in the central and western parts of Sylhet. New urban developments also emerge in Upazilas such as Balaganj and Zakiganj.

By 2020, the pace of urbanization accelerates further, leading to substantial expansion across the district, with Sylhet Sadar, Bishwanath, and Golapganj experiencing the most significant growth. This period suggests a clear pattern of urban spread from central to peripheral regions. In 2023, the extent of urbanization is even more pronounced, with urban areas expanding across nearly all Upazilas. Sylhet Sadar and neighboring Upazilas continue to dominate in terms of urban development, reflecting a sustained trend of urbanization.

Fig. 5
figure 5

Spatial distribution of urban area expansion in Sylhet district from 2000 to 2023, showing the concentration of built-up areas and urban sprawl. Maps created in QGIS 3.28.12, https://qgis.org/.

Full size image
Change in waterbody

Figure 6 illustrates the distribution and temporal changes in water bodies across the Sylhet district from 2000 to 2023. In 2000, a significant presence of water bodies (34.24% in Table 5) is observed, particularly concentrated in the southern Upazilas, such as Balaganj (17.82%), Fenchuganj (4.65%), and Dakshin Surma (7%). Other Upazilas, including Zakiganj (9.78%), Kanaighat (9%), and Golabganj (7.9%), also feature notable water bodies. By 2005, there is a slight reduction in the extent of these water bodies; however, their overall distribution remains largely unchanged, with prominent concentrations still found in the southern and central parts of the district, particularly in Balaganj and Fenchuganj.

The 2010 map depicts a similar distribution of water bodies as seen in 2005, with minimal changes. The southern and central regions continue to host most of the district’s water bodies. A slight increase in visibility in some areas may indicate seasonal or short-term variations. In 2015, the distribution remains relatively stable, with continued concentration in the southern Upazilas, though slight fluctuations in extent suggest potential environmental changes or seasonal shifts.

By 2020, there is a noticeable reduction in water bodies, particularly in the central and western parts of the district. However, significant water coverage persists in the southern regions, notably in Balaganj and Fenchuganj. The 2023 map indicates further reduction, especially in the central and northern parts of the district, though some water bodies remain in the southern regions, albeit in a diminished form. By 2023, about 77% waterbody has been reduced (Fig. 4).

Spatial analysis

Tables 6 and 7 present the upazila-wise LULC distribution for the years 2000 and 2023, respectively. In 2000, the largest urban and bare land areas were in Kanaighat upazila, covering approximately 3,200 and 6,900 hectares, respectively. Sylhet Sadar upazila had the second-largest urban area, with around 2,300 hectares, while Balaganj upazila had the second-largest bare land area, with about 6,400 hectares. The largest waterbody and vegetation areas were found in Balaganj and Gowainghat upazilas, covering approximately 21,000 and 29,000 hectares, respectively. Fenchuganj upazila had the smallest urban and vegetation areas, with about 125 and 5,400 hectares, respectively. The smallest bare land area was in Beani Bazar upazila, with around 250 hectares, and the smallest waterbody area was in Jaintiapur upazila, with about 5,000 hectares.

Fig. 6
figure 6

Changes in waterbody distribution across Sylhet district between 2000 and 2023, illustrating significant reductions, particularly in central and northern upazilas. Maps created in QGIS 3.28.12, https://qgis.org/.

Full size image
Table 6 Upazila-wise distribution of LULC classes in Sylhet district in the year 2000, showing the dominance of vegetation and water bodies.
Full size table

By 2023, the largest urban area, approximately 9,000 hectares, was in Gowainghat upazila. Kanaighat, Balaganj, and Gowainghat upazilas contained the largest areas of bare land (around 16,000 hectares), waterbodies (about 4,500 hectares), and vegetation (approximately 24,000 hectares), respectively. Fenchuganj upazila had the smallest urban area (about 1,500 hectares), bare land (around 800 hectares), and vegetation areas (approximately 7,500 hectares), while Bishwanath upazila had the smallest waterbody area, with around 950 hectares in 2023.

Table 7 Upazila-wise distribution of LULC classes in Sylhet district in 2023, highlighting increased urbanization and reduced water bodies.
Full size table

Table 8 displays the percentage of LULC transitions between 2000 and 2023. A rise in the LULC area is shown by positive values, whilst a fall is indicated by negative numbers. The study shows that urban areas increased in every upazila, with the most significant growth observed in Beani Bazar upazila, at approximately 1,500%. Bare land decreased only in Balaganj upazila, while Beani Bazar upazila experienced the highest percentage increase in bare land area (about 1850%) among the other upazilas. Conversely, waterbodies decreased in all upazilas, with the most substantial reduction (about 90%) occurring in Bishwanath upazila. Vegetation areas increased in some upazilas and decreased in others, with the largest increase (about 100%) in Balaganj upazila and the largest decrease (about 30%) in Kanaighat upazila.

Table 8 Percentage change in LULC classes for each Upazila between 2000 and 2023, illustrating Spatial variations in land transformation.
Full size table

LULC transformation hot and cold spots analysis

The Getis-Ord Gi* hotspot analysis was employed to identify statistically significant clusters of high (hotspots) and low (cold spots) z-scores across the Sylhet district, highlighting spatial patterns of LULC change over the study period. According to Fig. 7, the regions classified as hotspots where Z > 0.1, indicate significant clustering of high z-scores, representing areas with substantial changes in LULC classes. This LUC changes mainly occurred due to human intervention and partly by natural disaster. According to this study, notable concentrations of hotspots are observed in Sylhet Sadar (about 2.5%), Balaganj (about 5.5%), and parts of Companiganj (about 3%) and Golabganj (about 2.5%). This pattern aligns with increasing urban expansion and development activities in these regions (Fig. 5). Among all upazilas, Balaganj exhibited the highest percentage (about 5.5%) of hotspot areas relative to its total district area, confirming its significance as a zone of intensive LULC transformation (Fig. 8).

On the other hand, cold spots are identified in Fig. 7, where Z < −0.1, represent clusters of low z-scores, reflecting areas where minimal changes occurred in LULC transitions. That means in this study, the cold spots are considered as those area, where the effect of human intervention and natural disaster such as flood are minimum. Zakiganj (about 4%), Jaintiapur (about 4%), and Kanaighat (> 7%) prominently exhibit cold spots, predominantly in vegetation or water body persistence areas where urban expansion has also been minimal. Golapganj and Sylhet Sadar also display cold spots in specific zones, indicating spatial heterogeneity in LULC dynamics. Maximum cold spot coverage (more than 7%) was observed in Kanaighat upazila and then in Gowainghat upazila, which was more than 5%.

Fig. 7
figure 7

Getis-Ord Gi* (Z-Score) statistical map displaying the spatial clustering of LULC transformation hot spots and cold spots in Sylhet district from 2000 to 2023. Maps created in QGIS 3.28.12, https://qgis.org/.

Full size image
Fig. 8
figure 8

Percentage distribution of hot spots and cold spots across different upazilas in Sylhet district, highlighting areas with significant land transformation.

Full size image

In total, the study highlights significant LULC changes in the Sylhet district from 2000 to 2023. In 2005, the results show a significant reduction of vegetation (from 50.51 to 24.38%) and dramatical increment of bare land (from 10.61 to 56.66%) (Table 4). This abnormal change may occur due to the unusual climatic year 200443. In this year, Sylhet region faced an irregular early flash flood which started in the April, 2004 and continued up to end of July, 200444. This flood event damaged many vegetation areas including cropland which finally converted in bare land. By 2010, urban area has increased from 2.61% (2005) to 19.67% in Sylhet district (Table 4) and this urbanization is noticeable particularly in the southern part of Sylhet Sadar Upazila which is mainly Sylhet City Corporation (SCC). According to Rahman M45, the population of Sylhet city has increased from 263,197 (in 2001) to 663,198 (in 2010), which caused new accommodation facilities and directly increased urban area. Rahman M45 observed rural to urban migration, which is triggered by the rural unemployment, caused urbanization in SCC. Ahmed et al.46, reported that many new apartments, luxurious hotels and shopping centers were newly developed in 2010, which also caused urbanization in this area. The findings indicate that bare land was the predominant LULC type in Sylhet for most of the years examined, particularly in 2005, 2010, 2015, and 2020. However, in 2000 and 2023, areas covered by vegetation were the most extensive. Urban areas have shown a consistent upward trend in both size and number, while water bodies have steadily declined over the study period.

This study effectively captures the ongoing urban expansion (from 4.65% in 2000 to 18.46% in 2023) in Sylhet district (Table 5), correlating this growth with factors such as population increase, economic development, and infrastructure advancements. Furthermore, the research highlights a significant decline in water bodies (from 34.24% in 2000 to 7.87% in 2023) (Table 5), particularly in the central and northern parts of the district. In contrast, the southern upazilas have managed to maintain substantial water coverage over the years. The reduction of water bodies in the northern regions has impaired their capacity to store adequate water during the rainy season, leading to overflow into floodplain areas. This overflow increases the risk of flooding and eventually channels excess water toward the southern regions. This trend likely reflects the cumulative effects of land use changes, human activity, and climate change on the hydrological landscape of the area.

The hotspot analysis underscores the urgent need for sustainable urban planning in hotspot areas such as Sylhet Sadar (about 2.5%) and Balaganj (about 5.5%), where rapid urbanization could pose challenges to environmental sustainability. In contrast, cold spot areas with stable LULC clusters offer opportunities for conservation initiatives. The spatial clustering maps provide a detailed visualization of Gi* z-scores, emphasizing the spatial distribution of clusters. The spatial correlation evident in the maps suggests that factors such as proximity to urban centers, infrastructure development, and natural constraints are significant drivers of LULC changes in the district.

The trends observed in this study are primarily driven by population growth, rapid urbanization, unregulated land development, and environmental events such as flash floods. For example, early flash flooding in 2004 damaged croplands, contributing to the temporary spike in bare land in 2005. In contrast, rising demand for housing and infrastructure in urban centers, especially Sylhet Sadar, has led to widespread conversion of vegetation and water bodies into built-up areas. These transformations reflect broader patterns of rural-to-urban migration and insufficient regulatory control over land use in peri-urban areas. Theoretically, this study contributes to the LULC literature by demonstrating how the integration of high-temporal-resolution satellite imagery with spatial statistical tools like Getis-Ord Gi* can effectively identify transformation hotspots and cold spots. This approach provides a replicable framework for prioritizing areas for land use planning and flood risk mitigation, particularly in data-scarce, flood-vulnerable regions.

Conclusion

This study provides an extensive analysis of LULC transformation in Sylhet district, Bangladesh, over a 23-year period (2000–2023) using Google Earth Engine (GEE). The results highlight significant transformations in LULC, with urban areas expanding dramatically at the expense of water bodies and vegetation. Urbanization has particularly surged in upazilas such as Beani Bazar and Sylhet Sadar, with corresponding declines in natural features like water bodies, which have reduced by approximately 77% across the district. These findings underscore the critical need for sustainable urban planning and the implementation of effective drainage infrastructure to alleviate the increased risk of flash flooding in the region. Policymakers, urban planners, and water resource engineers can leverage these insights to develop strategies that balance development with environmental preservation, ensuring long-term resilience against flooding hazards in Sylhet district. The integration of advanced remote sensing technologies like GEE offers a robust framework for ongoing LULC monitoring, providing a scalable solution for other flood-prone regions globally. While this study offers comprehensive insights into LULC changes and transformation hotspots, it is constrained by the spatial resolution of Landsat imagery, which may not capture very fine-scale land cover changes. Future research may explore seasonal variations with high-resolution imagery in land cover and flood susceptibility, integrate socio-economic data to assess human vulnerability, or incorporate flash flood modeling for a more comprehensive risk assessment.