1 Introduction

According to the United Nations (UN), the global population will reach 8 billion by November 2022. Nearly 55% of the world’s population (4.46 billion as of 2021) lives in urban areas [1]. By 2050, nearly 70% of people are estimated to live in urban settings. Urban sustainable development is closely linked with the future of humanity [2]. The rate of urbanization is uneven across countries because of disparities in socioeconomic development [3]. In high-income countries (HICs), urban growth is usually adaptable to population expansion because of well-planned urban infrastructures including water and electricity supply, sanitation, education, and green space or parks [4, 5]. On the contrary, in many low and middle-income countries (LMICs), the national economic growth and development are inadequate to meet the demand of the growing urban populations. Consequently, these communities face overcrowding, vector-borne diseases, and environmental pollution and degradation [5, 6]. Urbanization is linked to an increase in built-up areas and population growth impacting the loss of vegetation cover due to unplanned urban growth, which is responsible for a significant increase in land surface temperature and impervious surface [7, 8]. Triggered by population growth, urban expansion occurs at the expense of agricultural land, waterbodies and vegetation [9].

Urban sustainable development has become a significant global concern and is a crucial component in the 2030 Agenda for the UN's Sustainable Development Goals (SDGs) [10]. To achieve SDGs, it is therefore essential to track the urbanization progress around the world. The progress of urbanization is well studied in HICs, while the information is scarce for LMICs.

Bangladesh, a lower middle-income country, is one of the most densely populated (population density of 1265/km2) countries, with a population of nearly 170 million. In recent years, Bangladesh has undergone rapid urbanization. The percentage of people living in cities increased from 5.1% in 1974 to 32.8% in 2013, and by 2040, more than 50% of the population will be living in urban areas [11, 12]. The country's recent rapid economic development is considered the main driver of uncontrolled urbanization, manifested by insufficient housing, and a severe lack of infrastructure, water, and sewerage systems [13, 14].

In Bangladesh, the trends in urban growth, land use, land cover (LULC) classification, and change detection have been reported mostly for large cities, particularly Dhaka, Chittagong, and Khulna [15]. However, little is known about the progress of urbanization at the district level (65 administrative districts in Bangladesh). Based on socio-demographic factors, all district towns could be classified into three clusters where some are comparatively less developed, Jamalpur is one of them [16].

According to the Bangladesh Census, an urban area is defined as any developed area around a central location with amenities such as paved roads, electricity, gas, water supply, sewerage, and sanitation; that is densely populated; where the majority of the population is employed in non-agricultural sectors; and where a strong sense of community exists [17]. District towns are thus the most prominent urban settlement and administrative hub in a district while the rest of the areas of a district are denoted as rural. The large cities and some adjacent urban settings of districts were overpopulated in the early 2000s while the rapid urbanization process started more recently at the district level.

Understanding and quantifying the consequences of urbanization requires analyzing the Spatio-temporal aspects of land-use change to provide basic data to guide decision-making [18]. The urbanization process can be studied using remote sensing due to its multifaceted benefits. Remote sensing data have become increasingly popular because they can provide timely and spatially explicit data [19]. At the same time, conventional survey and mapping procedures are costly, time-consuming in estimating urban expansion, and lack reference datasets [20]. Temporal trajectory analysis and post-classification are two detection approaches that have been developed for urbanization monitoring. The post-classification method classifies land use/cover types at each point in time and then compares the time series classifications to examine land changes [21].

This study aimed to assess the urbanization trends (th 1991–2021) in one of the least developed districts of Bangladesh using the remote sensing approach.

2 Methods

Comparing developed countries' land cover datasets regimes, developing or least developed countries usually do not maintain central land cover database convention systems. Another significant issue to consider is the freely available satellite imagery resolution and comparison with ground representing pixel density. Although high-resolution Sentinel 10 m resolution imagery is available in the study area, its temporal resolution is limited (from 2015); therefore, Landsat is the only option for historical analysis. To study the unplanned urban growth of this country, unsupervised classification is not suitable like developed countries due to mixed class into a small chunk of an area. So supervised classification algorithm is followed in the study. The available 30 m resolution Landsat datasets are coarser in terms of classifying mixed landcover classes of this study area. Therefore we conducted field verification, accuracy assessment, post-classification refinement, and reporting approaches.

2.1 Study area with relevant demographic information

To understand the level of urbanization in the least developed districts in Bangladesh, this study was conducted in Jamalpur, one of the poorest districts in Bangladesh. Geographically it bounds by 24.9167°N to 24.7245°N latitude and 89.8139°E to 90.1991°E longitude, covering about 490 square kilometers (Fig. 1). The study sub-district is downstream of the Himalayan-Tibbet (Lusai) waterfall and under the largest Ganges–Brahmaputra-Meghna (GBM) basin in this region. Jamalpur subdistrict is near Brahmaputra and Jhinai rivers, significant sources of flash floods, river floods, and prolonged waterlogging in this region [22]. The erratic precipitation pattern in the upstream regions, along with anthropogenic mismanagement (controlling barrage, dams) in the upstream region, faces flash floods during the monsoon period from May to October, almost every year. In recent years, 35% of the landmass faced prolonged flood/waterlogging in the agricultural lands, a significant consequence of increasing seasonal job-less communities with multi-fold human health deteriorations. The Upazila is prone to seasonal flooding due to its proximity to River Brahmaputra and River Jhinai [23].

Fig. 1
figure 1

Study area map (Jamalpur Sadar Upazila)

The district ranks fourth in the country, with a poverty rate of 51% on the Upper Poverty Line [24]. With a 34 per 1000 live births rate, it ranks first in the country in infant mortality [25]. The district has a literacy rate of 38%, ranking it 60th out of 65 districts, while only 36% of the population in the district has access to household electricity [26]. Our previous work has shown that, in terms of demographic and socioeconomic factors, Jamalpur town (urban area) is classified as the least developed cluster of district towns in Bangladesh [16].

2.2 Data sources

Bangladesh is a flood-prone (typically from June to September) country, hence, four Landsat images of the years 1991, 2001, 2013, and 2021 for January and February were used for this study from different Landsat sensors based on availability to avoid seasonal variabilities like monsoon and flooding. Instead of considering continuous datasets, the study considered ten years interval imageries since typical masonry building construction takes time and onset development is not as robust as in the continuous datasets. All Landsat images were retrieved as level 1 data products (image quality >  = 8) and less cloud cover from the USGS Landsat archive. Table 1 presents detailed information about the datasets. The image data were freely accessed from the Landsat archive of the United States Geological Survey (USGS) website (https://glovis.usgs.gov/) (https://earthexplorer.usgs.gov/) District and Upazila shapefiles were retrieved from LGED GIS Portal (https://gis.lged.gov.bd/).

Table 1 Spectral characteristics of the satellite images used for LULC mapping in Jamalpur district

2.3 Image pre-processing

All composite color bands were used, excluding the thermal bands, and mosaicking of the imageries was not required. This sample selection was based on satellite imagery, GPS coordinates, and field-visited additional data [27]. ATCOR-3 was used to perform geometric, radiometric corrections and haze reduction [28]. Histogram equalization was also performed for radiometric image enhancement and found suitable for this study. HR 49 picture segments from Google Earth were analyzed to identify and include current features. This study used a Garmin GPSMAP 78 s to retrieve precise GPS locations and geotag images.

2.4 LULC methodology

The methodology included object-based image classification of Earth Observation satellite images, accuracy assessment of the images, and a GIS-based change detection study (Fig. 2). The object-based classification included image segmentation, sample selection of target LULC classes, fuzzy-based sample evaluation, standard nearest neighbor (NN) classification, field verification, and LULC map preparation. Object-based image classification was carried out in eCognition Developer (ver. 9.1), which can combine various supplementary data and produce GIS-ready output [29]. Image segmentation is the initial and most crucial step, which splits a large heterogeneous image into a finite number of homogeneous groups of pixels known as image objects. The picture objects contain numerous properties about a specific object, categorized using spectral and textural criteria. The conventional pixel-based nearest-neighbor classifier computes the Euclidean distance in n-dimensional feature space between the pixel to be categorized and the nearest training data pixel. It assigns it to that class [30]. An object-based nearest-neighbor classifier operates similarly, except instead of individual pixels, it classifies image objects way except it, classifies image objects instead of individual pixels. Various object parameters, such as the object's mean red, NIR, green, and NDVI values, and the associated standard deviation of red, NIR, green, and NDVI values and the DEM, were used to perform the standard nearest neighbor classification.

Fig. 2
figure 2

Flow chart illustrating various methods to produce LULC maps from multi-temporal satellite images

The first four features were associated with spectral measures for classification, while the last four were attributed to textural measures. The amount of object features evaluated in classification varies from image to image. Agricultural land, built-up area, bare soil, dense vegetation, light vegetation, deep waterbody, and shallow waterbody were selected for image classification as per Anderson LULC classes [31]. Training samples for each class were chosen from similar regions in all images to guarantee consistency during classification. Fuzzy-based membership functions were used to assess the quality of training samples. The sample class is more reliable if the membership value is higher. Ideal samples for that selected class are those with a membership value greater than or equal to 0.5. Finally, the feature space distance was assessed for better LULC class separation. The ideal combination of object features associated with higher Euclidean distances was chosen for classification among several combinations of object features utilized in the classification. However, object features with the greatest Euclidean distance may not always generate acceptable classification results. As a result, the visual estimation of identified image objects is further evaluated before giving it its final shape. To provide a visually pleasing classification output, these stages were repeated iteratively. Image objects categorized under several LULC classes were imported into the GIS system for map preparation and change detection analyses. The eCognition Developer can export classified image objects to a GIS-compatible data format (*.shp). Thousands of image objects of the same class were merged into a single class. The remaining unclassified picture objects (less than 1% of the entire geographical region) were categorized using visual image interpretation.

Spectral or thematic contrast was used to measure change. A thematic basis map is frequently used in such strategies to isolate better or target the change of interest. Using a change matrix, a post-classification comparison change detection approach was used. This approach compares two classed LULC maps created separately from photos from two different dates. This method for detecting LULC changes is reliable as the LULC maps were correctly created [32].

Four LULC layers from 1991, 2001, 2013, and 2021 were combined in a GIS for change detection. False change areas were corrected in the change matrix due to misclassification or spectral misunderstanding. In change detection analysis, irreversible classes such as built-up areas were also considered. The dynamics of key LULC classes in the area of interest were represented using the GIS overlay function.

The Jamalpur town land-use mapping results were expected to give information on (a) the distribution of land use categories and (b) the identification and estimation of land use changes over the past 30 years. During the field visit, the land uses in the study area were divided into seven categories: (1) agricultural land, (2) bare soil, (3) built-up, (4) deep waterbody, (5) shallow waterbody, (6) light vegetation, and (7) dense vegetation.

We also used the sprectral indices to calculate the land surface temperatures (LSTs) of the study area in different time points (1991–2021) [33].

2.5 Accuracy assessment

Accuracy assessment is used in thematic mapping using remotely sensed data to represent the degree of correctness of a classified map. ArcGIS 10.8 and Google Earth Pro were successfully utilized to process, analyze, and identify changes in those classed images. The LULC from 1991 to 2021 was estimated using area differences within a 10-year interval. For accuracy evaluation, the matrix's accuracy was examined using user accuracy, producer accuracy, overall accuracy, and the Kappa Coefficient. For ground validation, approximately 40 random locations were produced, with a minimum allowable distance of 30 m for each class.

2.6 Change detection analysis

Classification-based (map-to-map) change detection was used in this study to show the expansion of built-up area during the studied period.

3 Results

The study area’s (Jamalpur Sadar Upazila or town) LULC error matrix for 1991, 2001, 2013, and 2021 are presented in Table 2.. The overall accuracy for supervised classification images was 91%, 90.29%, 93.43%, and 91.14%, with Kappa values of 0.89. 0.89, 0.92, and 0.90, respectively. These Kappa values showed that the accuracy of the land use classification was satisfactory.

Table 2 Accuracy assessment of the LULC maps produced from Landsat data representing both confusion matrix and the Kappa statistics

3.1 Land use classification & temporal changes

The results obtained from multi-spectral imageries are illustrated in Figs. 3, 4, 5, 6. Table 3 summarizes the trend of LULC change from 1991 to 2021 based on seven classes extracted from Jamalpur Sadar Upazila with a proportionate coverage area for each, while Fig. 7 shows built-up changes and Fig. 8 represents net gain loss of each class. A brief account of these results is discussed in the following paragraph.

Fig. 3
figure 3

Land use and land cover map of Jamalpur Sadar Upazila for the year 1991

Fig. 4
figure 4

Land use and land cover map of Jamalpur Sadar Upazila for the year 2001

Fig. 5
figure 5

Land use and land cover map of Jamalpur Sadar Upazila for the year 2013

Fig. 6
figure 6

Land use and land cover map of Jamalpur Sadar Upazila for the year 2021

Table 3 Area statistics of land use classes of Jamalpur for the years 2001, 2013 and 2021
Fig. 7
figure 7

Map showing expansion of built-up area during 1991–2021

Fig. 8
figure 8

Net percent gain and loss for each LULC class during the study period (1991–2021)

Fig. 9
figure 9

Land surface temperature (LST) changes during study period (1991–2021)

In 1991, the pattern of LULC as the percentage of the total area studied was dominated by agricultural land covering 38.77% of the total studied area, followed by dense vegetation (26.4%), light vegetation (12.14%), deep waterbody (6.95%), bare soil (6.89%), shallow waterbody (5.58%), and built-up area (3.19%). Interestingly, in 2021, the observed LULC pattern was narrowly dominated by agricultural land (28.56%) followed by dense vegetation (27.26%), built-up area (27.06%), light vegetation (10.41%), shallow waterbody (5.1%), deep waterbody (1.44%), and bare soil (0.16%) (Table 3). Furthermore, Fig. 7 illustrates the pattern of LULC during the studied period (1991–2021) indicates the general increase of built-up area (748.92%) and dense vegetation (2.96%) with a decrease in bare soil (− 97.65%), deep waterbody (− 79.33%), shallow waterbody (− 8.46%), light vegetation (− 14.28%) and agricultural land (− 26.33%).

Because of increased built-ups, the land surface temperatures (LSTs) also increased over the study period. The minimun recorded LSTs were 12.84◦C, 15.89◦C, 15.8◦C and 17◦C while the maximun LSTs were 17.93◦C, 23.7◦C, 23.79◦C and 26.73◦C in the year of 1991, 2001, 2013 and 2021 respectively (Fig. 9, Supplementary Fig. 1).

4 Discussion

In our study, we found that over the last 30 years (1991–2021), the built-up area was hugely expanded while other land use classes were declined noticeably, especially deep waterbodies, shallow waterbodies, bare soil, and light vegetation at a district town level (Jamalpur Sadar Upazilla) in Bangladesh.

In this study, we have observed remarkable LULC changes in the least developed district town (this is summarized through a flow diagram, Fig. 10). Between 1991 and 2013, the area had a linear increase in built-up regions, with a linear growth pattern of along with important transportation networks such as national roads. Agricultural land dominated the LULC in 1991 while it was converted to primarily built-up areas in 2013 and 2021. The total built-up area was 1690 ha in 1991, while it was increased to ~ 3100 ha in 2013 (17.5% of increment from 1991) and 14,348 ha in 2021 (362.8% increment from 2013). Deep waterbody degraded from ~ 7% (1991) to, 1.4% in 2021. Due to siltation from seasonal flooding.

Fig. 10
figure 10

Land use conversion flow diagram in Jamalpur

deep waterbodies were mostly converted to shallow waterbodies.

On the other hand, shallow waterbodies showed an interesting trend: it increased from 5.6% (1991) to ~ 10% in 2001, while it was declined to 2.6% in 2013 and rose to ~ 5% in 2021. These changes might be due to the extraction of soil from barren ground, riverbanks, and agricultural lands to construct built-up areas. The change of dense vegetation had an unusual trajectory; decreasing from ~ 27% (1991) to ~ 11%% (2001) to ~ 41% in 2013, then dropping to ~ 27% in 2021. These changes in dense vegetation possibly occurred due to the conversion of barren soil, shallow waterbodies, agricultural fields, and landfills to orchards to reinforce the soil to develop built-up areas. Interestingly, bare soil was decreased by 50% every ten years. The bare soil bordering waterbodies was most converted to agricultural lands.

Previous studies have found that increased level of urbanization is associated with the rise of LST [34, 35]. In our study we also found similar trends of increased LST over the years.

Historically, urbanization in Bangladesh is mostly centered in big cities until late 1990s. Compared to other major cities, our study area (least developed Jamalpur Sadar Upazila) has shown rapid urban growth (approximately 8.5 folds or, nearly 750% increment of built-up area)in the last three decades. In case of Dhaka city, the built-up area growth was 14% between 2000 and 2020 [35] while it was only 5% in Chittagong Metropolitan City between 1990 and 2020 [36]. Similarly, in Khulna city, the built-up growth was 28% between 1997 and 2017 [37]. On the other hand, mid-tier district towns like Comilla experienced a net increase of 7.8% in built-up area in the last two decades [38] while it was nearly 25% in Mymenshigh [39] and ~ 69% in Barisal [40].

In other South Asian countries including India and Pakistan, a similar trend of rapid urban growth has been observed in many less developed district towns., For instance.In India, Purulia district town has grown by 122% in terms of built-up area between 1998 and 2018 [41], while Gautam Buddha Nagar district town reported an unprecedented increase of nearly 419% between 2003 and 2015 [42]. In Pakistan, the built-up area was increased by 158% in similarly sized Sialkot [43] and 146% in Sargodha [44].

Trends in urbanization differ between LMICs and HICs. The process of urbanization is relatively faster in LMICs. For instance, major cities in the United States were 40% urbanized in 1900 compared to 70% in 1960 and 75% in 1990. In contrast, major cities in Brazil were 40% urbanized in 1970, which increased to 80% in 2000. Brazil took 30 years to achieve what the United States achieved in 90 years [45].

We found that the magnitude of the progress of urbanization in our study area was unexpectedly much higher. Such a pace of urbanization is likely due to the fact that, Bangladesh has paid particular attention (Vision 2021) to developing the least developed districts to alleviate poverty in the Sixth and Seventh Five Year Plan [46, 47]. As part of this development scheme, the first-ever economic zone in the Northern part of Bangladesh was set up in Jamalpur in 2019, which now attracts agro-based industries, food processing, jute, and jute-based products, and apparel/RMG industries. In recent times, educational institutions including a medical college and a university have been established in the study area. Moreover the biggest hub of handicrafts in Bangladesh was also established along with other infrastructural development projects [48]. The overall growth in GDP, potential job opportunities, healthcare, education, and quality of life have acted as pull factors for in-migration from other Upazilas of Jamalpur and neighboring districts. Similar development schemes were also undertaken in other low-performing districts to alleviate poverty and unemployment. This suggests that other least developed districts in the country have undergone a similar trend of urbanization in the last decade. This trend might also be applicable for mid-tier districts as industries tend to relocate from major cities for cost minimization and cheap labor.

In LMICs, rapid urbanization is typically unplanned. It is important to mention that unlike the leasing system in many developed countries, Bangladeshi citizens have exclusive land ownership for indefinite periods [49]. For any developmental purpose, the state acquires lands from individual citizens. Moreover, urban development policies are not established in Bangladesh. Interestingly, Bangladesh’s 1950 State Acquisition Act and National Land Use Policy 2001 prohibits conversion of waterbodies and arable lands into non-agricultural land uses [50]. Due to lack of accountability and monitoring, these policies failed to resist unregulated land use conversions. Consequently, land-use conversion to built-up area leads to unplanned in Bangladesh. In this context, the unplanned urbanization process in Jamalpur district town could be visible from time-series images of Google Earth Pro (Supplementary Fig. 2).

Rapid and unplanned urbanization has a significant impact on socioeconomic development and sustainability. This has resulted in major issues such as a scarcity of land resources, a progressive deterioration of ecological security, the loss of traditional regional characteristics, and development pressure on historical and cultural interests. Ecosystem services have suffered due to rapid land-use change [51]. Importantly, considering impacts, local or regional environmental issues can contribute to global concerns.

5 Limitations

LULC maps derived from satellite images in this study provide an acceptable level of accuracy though they are not 100% accurate [52]. Nevertheless, these flaws were addressed by a thorough accuracy assessment conducted by ground-truthing and Google Earth Pro, resulting in reasonably accurate LULC maps.

6 Conclusion

This present study, the first of its kind in Bangladesh, found that a least developed region is experiencing a higher magnitude of unplanned urbanization. The built-up area has expanded over eight folds in three decades while other land-use classes decreased noticeably, particularly deep waterbodies, shallow waterbodies, bare soil, and light vegetation at a district town level. Our findings will serve as a baseline for future studies as well as urban planners and policymakers will get insights into the overall process of urbanization in the perspective of LMICs. Remote sensing tools can be very effective for monitoring urbanization accurately. Future studies on prediction modeling and impacts of urbanization on biodiversity and land-surface temperature are necessary to achieve urban sustainability.