Wetlands are areas of natural or man-made, permanent or temporary, static or flowing fresh, brackish, or salt water bodies (Convention on Wetlands, 2021). These are highly valued ecosystems for providing significant services to humanity, ranging from freshwater supply, food, groundwater recharge, flood control and climate change mitigation (Sieben et al., 2018). In addition, wetland ecosystems are vital in supporting floral and faunal diversity (Xu et al., 2020; Kumar et al., 2023). However, wetlands are facing degradation due to the conversion of wetland areas to croplands and human settlements to support the demands of rapidly growing population (Zekarias et al., 2021; Abdulmalik et al., 2022). In addition, infrastructural expansion and water extraction for agriculture reduce wetland areas (Prigent et al., 2012). Increases in greenhouse gases and air temperature can alter precipitation patterns, cause drought or increase water levels. These changes can either reduce water levels in wetland areas or flooding and affect the biodiversity of wetland ecosystems (Choudhury et al., 2021; Nayak, 2022).

Both anthropogenic and natural phenomena drive the land use and land cover (LULC) changes in wetland ecosystems, which significantly impacts the ecology of wetlands (Bakr and Afifi, 2019; Assefa et al., 2021). Imran et al. (2023) reported that human activity is one of the key reasons for fish decline in the Hakaluki Haor. Research conducted on Tanguar Haor, a RAMSAR site, found that between 1989 and 2017, around 33% of the vegetation cover was converted into agricultural land due to rapid population increase (Haque et al., 2021). Climate change through increasingly erratic weather patterns that include temperature rises, unpredictable rainfall and flash floods impacts fish ecology and diversity in the haor ecosystems in Bangladesh (Majumder et al., 2013; Pandit et al., 2023).

Monitoring wetlands can offer significant insights into their functioning and responses to anthropogenic and environmental shifts (Stratford, 2018). Thus, understanding and identifying these changes in wetland ecosystems is crucial for effective environmental management, resource planning and decision-making processes (Gokce, 2018). Traditional labor-intensive field-based and satellite-based monitoring strategies have been employed for monitoring wetland ecosystem. However, globally, satellite-based remote sensing has become an inevitable tool for monitoring and evaluating LULC change patterns, thereby contributing to the long-term sustainability of wetland resources (Hossain et al., 2021; Kavhu et al., 2023).

The commonly used method of classifying remotely sensed images, for example, the Maximum Likelihood approach, requires downloading and managing large datasets, which results in higher computational requirements and comparatively longer processing times. As a result, there is an increasing inclination toward using machine learning algorithms in cloud-based platforms worldwide to analyze LULC classifications and estimate long-term changes because of their effectiveness (Lary et al., 2016; Biswas et al., 2023). Cloud-based platforms, e.g., Google Earth Engine, allow researchers to use different Machine Learning algorithms, i.e., CART, Random Forest (RF) and Support Vector Machine (SVM), which are gaining popularity in this regard (Shaharum et al., 2020; Loukika et al., 2021).

Bangladesh, a low-lying deltaic country in South Asia, has a vast network of wetlands that occupy over 7–8 million hectares, making up almost 50% of its total area (Akonda, 1989; Yousuf Haroon and Kibria, 2017). Bangladesh is a signatory to the RAMSAR convention, an international treaty to sustainably conserve wetlands ecosystems from natural and anthropogenic threats (Mulligan and Griffin, 2012; DBHWD, 2016). The wetland network of Bangladesh possesses various hydrological features, including rivers, floodplains, lakes, haors (backswamp), baors (oxbow lake), beels (deepr part of floodplain), fishponds, swamps, and low-lying regions, generally classified as wetlands, with ‘haors” specifically denoting seasonal/intermittent freshwater floodplains (Khan et al., 1994; DBHWD, 2016). During the monsoon season, haors—essentially bowl-shaped depressions nestled between the natural leaves of rivers—receive surface runoff and transform into vast water bodies for about 6 months of the year (Nishat et al., 1993). According to the DBHWD (2016), the haor region comprises 373 haors spanning approximately 859,000 hectares, constituting 10% of the nation’s overall wetland area. Recognizing significance, two freshwater haors have been declared as RAMSAR sites: Tanguar Haor and Hakaluki Haor. The importance of freshwater wetlands is profound; however, a recent study conducted by Waleed et al. (2023) revealed a concerning trend of wetland degradation in the north-eastern part of Bangladesh. This study reported about an 80% reduction in some of the wetlands from this region.

Hakaluki Haor, a freshwater wetland, is a vital RAMSAR site in Bangladesh’s Northeast hydrological region (WARPO, 2001; DBHWD, 2016). This Ecologically Critical Area (ECA) supports diverse biodiversity and fisheries resources (Choudhury and Nishat, 2005). About 200,000 people living in the vicinity of the haor directly or indirectly depend on the wetland resources (Banglapedia, 2021). However, the degradation of land cover in Hakaluki Haor caused by natural events like siltation, droughts, human activities, and insufficient monitoring and management might affect the entire ecosystem. Therefore, these wetlands need constant monitoring. While satellite-based techniques are extensively utilized globally, their application in Bangladesh for evaluating change and monitoring wetlands remains limited. Noteworthy studies include those conducted by Waleed et al. (2023) on ML-based study of wetlands in the Sylhet region, Haque and Basak (2017) on Tanguar Haor, Bhattacharjee et al. (2021) on Lakshibaur-Nalair Haor, and Islam et al. (2018) on Hakaluki Haor.

Here, we conduct a comprehensive analysis of LULC dynamics of Hakaluki Haor applying a Machine Learning algorithm (CART) and Google Earth Engine. While Random Forest is considered to be a more robust ML algorithm in Google Earth Engine (GEE), many researchers have applied the decision tree-based algorithm, particularly Classification and Regression Trees (CART), for accurately mapping changes in land cover of wetland habitats (Yao et al., 2019; Chen et al., 2021; Aslam et al., 2024). CART is a simple, fast, and computationally efficient algorithm with straightforward visualization and analysis extensively used for land use change analysis (Amani et al., 2017; Sang et al., 2019; Yao et al., 2019; Arpitha et al., 2023; Iandolo et al., 2024).

Classification and Regression Trees (CART) is particularly useful for classifying satellite images to identify different land use patterns due to its intuitive and interpretable decision-making process, ability to capture complex non-linear relationships, and robustness in handling high-dimensional and noisy data. The binary tree structure of CART allows it to efficiently select relevant features and make accurate classifications based on the hierarchical decision rules, which mimic human decision-making processes. This makes CART an effective tool for distinguishing between various land use categories in satellite imagery, providing clear and actionable insights into land use patterns (Blanquero et al., 2021; Biswas et al., 2023). CART also offers built-in functionality for handling large geospatial datasets. Although Support Vector Machines (SVM) are frequently reliable, their black-box nature sometimes restricts the interpretability of decision rules. For these reasons, we utilized the CART algorithm in GEE for the study.

This study spans a temporal scope from 2000 to 2023, focusing on two distinct seasons- rainy and dry. Each season spanned 6 months and was chosen to provide a more comprehensive representation of the actual conditions. As relying solely on dry or rainy seasons, data may not accurately capture the full spectrum of LULC dynamics of the haor basin. Employing remote sensing techniques to mosaic imagery across multiple years and seasons, the study aims to detect spatio-temporal LULC change patterns focusing on differences in the two distinct seasons of different land cover types within this critical wetland. The study also aims to provide valuable insights into the complex interactions between seasonal variations and human-environment interactions in this freshwater wetland of great importance.

The findings of this study illustrate the evolving environmental conditions within this critical freshwater ecosystem through the comprehensive analysis of the land use land cover dynamics occurring in the Hakaluki Haor from 2000 to 2023. A discerning decline in water bodies has been observed in both the rainy and dry seasons, which can negatively affect aquatic ecosystems and the availability of water resources. The decrease in water bodies also poses a threat to fish diversity in Hakaluki Haor, the largest inland freshwater wetland ecosystem (Islam et al., 2011; Aziz et al., 2021) Hakaluki Haor, which provides habitat for a diverse range of fisheries resources, has declined due to the loss of permanent waterbodies and overexploitation. From a documented 107 species in 2001, the fish species reduced to only 64 species in 2018 (Centre for Natural Resource Study, 2003; Imran et al., 2023). Aziz et al. (2021) argue that increasingly erratic rainfall, temperature rises and siltation in the beel (i.e., deeper portion of haor), and fishing practices are also responsible for the decline of fisheries.

Overall, the wetlands of the northeastern part of Bangladesh are experiencing a reduction in their coverage. In the same region, a study by Bhattacharjee et al. (2021) in Lakshmibaur-Nalair Haor revealed that deep water bodies have diminished by ~38% over the last three decades. The researchers found that this significant decrease in deep waterbodies was caused by the construction of dams in the upstream rivers that led to increased siltation and shallow water areas were converted into agricultural activities. Furthermore, also reported that the deep water bodies of the haor area are becoming increasingly scarce in the regionIncreased siltation in deep water parts and the conversion of shallow water parts into settlements and agricultural lands are critical drivers of this decline. The decrease in deep and shallow waterbodies could exacerbate the severity of flash floods in the haor basin of the region (Kamruzzaman and Shaw, 2018).

A substantial reduction in vegetation coverage in the study area has also been observed in the specified time frame, indicating significant changes in ecological conditions. Changes in wetland vegetation can significantly affect ecological processes within the ecosystems (Myneni et al., 1997; Chang et al., 2023). The present study found a significant loss (71%) of overall vegetation cover in the region from 2000 to 2023 (Figure 6). Polash et al. (2023) reported a similar trend of a substantial decline in dense vegetation cover (80%) in the Hakaluki Haor region from 2000 to 2019. Decreasing patterns in vegetation cover were also observed in other haor regions in the greater Sylhet (Haque and Basak, 2017; Bhattacharjee et al., 2021; Haque et al., 2021). Hakaluki Haor, recognized as a RAMSAR site and an Ecologically Critical Area (ECA), harbors migratory waterfowls, resident birds, mammals, amphibians, and reptiles (Choudhury and Nishat, 2005; DBHWD, 2016). The ongoing decrease in vegetation coverage poses a threat to the biodiversity of wildlife in this region (Polash et al., 2023; Waleed et al., 2023). The findings highlight the need for this information to be considered and for authorities to design and implement effective conservation strategies.

The potential causes for the observed fluctuations in cropland coverage within the Hakaluki Haor region may include alterations in land management practices, environmental circumstances, or something related to agricultural methodologies. By performing a more comprehensive examination and interpretation of these patterns, substantial insights can be obtained concerning the complexities of land use dynamics and human-environment interactions within this crucial wetland ecosystem.

Consistent with prior research that has documented the effects of human activities on wetland ecosystems, including settlements (Hu et al., 2017; Berkessa et al., 2023) and agricultural expansion (Convention on Wetlands, 2022), the results presented in this study support this notion. The study revealed a substantial rise in settlement areas in Hakaluki Haor from 2000 to 2023, with a significant expansion observed in both the rainy and dry seasons. According to the Bangladesh Bureau of Statistics (Bangladesh Bureau of Statistics, 2021), the population in the Sylhet region, where Hakaluki Haor is located, has been steadily increasing. This population pressure likely necessitated the conversion of land for residential purposes, leading to the observed rise in settlement areas. Rapid population growth may also accelerate settlements in the region, leading to the conversion of barren land, vegetation, shallow water bodies, into agricultural lands negatively impacting the ecosystem. For instance, earlier, the haor regions in the study area had freshwater swamp forests that are now nearly destroyed due to overexploitation and conversion for agriculture (Bokhtiar et al., 2024).

A recent study on the Tanguar Haor, another RAMSAR site, unveiled a notable decline in fish diversity, with recorded species decreasing from 122 to 58 within 9 years (2012–2021). Anthropogenic factors, such as overfishing, illegal fishing gear, infrastructure development, and swamp forest degradation, were the primary drivers contributing to the decline (Sultana et al., 2022). Moreover, the extent of the substantial increase in settlement area in the Hakaluki Haor region is highly dynamic in nature. In some instances, settlements disappear to appear in other landforms such as barren land, agricultural lands, etc. (Figure 5). Several implications may be drawn from this. These include an increase in the frequency of flooding in such areas, shifts in economic opportunities, or the relocation of people to places close to newly constructed roads for better communications. In addition, in Figure 7, it seems some water parts have changed to settlements. In this case, probable implications could be increasing sedimentation (Polash et al., 2023), building embankments or roads, or filling up shallow water parts by dredging (as seen in ground truthing) to create habitable settlements as the population is increasing. This practice is not uncommon in regions experiencing rapid population growth and urbanization, as seen in other parts of Bangladesh (Rahaman et al., 2023). The lack of effective land use planning and management has been identified as a key issue in other studies of land cover change in Bangladesh. Nevertheless, as witnessed in Hakaluki Haor, these patterns or shifts underscore the necessity for additional multidisciplinary research approaches concerning the precise catalysts and mechanisms that govern these changes. Lastly, several natural and anthropogenic activities, such as land-use changes and agricultural techniques, sedimentation, and erosion, may be responsible for the observed variations in the covering of barren land.

The haor regions have been consistently lagging behind Bangladesh’s steady-paced national development progress (CEGIS, 2012). Bangladesh Haor and Wetland Development Board (BHWDB), established in 2000, is a coordinating and leading institution for sustainable development and enhancing resilience in the regions. According to the BHWDB Master Plan (2012–2032), the haor regions are unique ecosystems and are regulated by a substantial number of policies, including water, fisheries, jolmohal (deep water portions), environment, land use, tourism, agriculture, forestry etc. (CEGIS, 2012). The ecosystem, environment, and biodiversity of Hakaluki Haor are gradually declining due to anthropogenic factors (diverse human activities) and natural factors (siltation, unpredictable rainfall patterns, etc.) (Uddin et al., 2013; Aziz et al., 2021; Polash et al., 2023).

The present study in the LULC change analysis of Hakaluki Haor indicated a substantial decline in water bodies (~51%) and vegetation cover (71%) while a dramatic increase in the settlement areas (~353%) (Figure 6). Our findings about these significant losses of waterbodies, vegetation and increasing settlements are consistent with other LULC change analysis studies for Hakaluki Haor and other adjacent haor regions (Haque and Basak, 2017; Islam et al., 2018; Bhattacharjee et al., 2021; Haque et al., 2017; Polash et al., 2023; Waleed et al., 2023). So, to conserve the ecological integrity in the Hakaluki Haor region, a crucial habitat, more investigation and interpretation of such LULC dynamics and interdisciplinary approaches may provide insights into better management practices. These holistic approaches may illuminate the ecological and biological dynamics with socioeconomic drivers governing wetland ecosystems. Further research may include socio-ecological assessments to explore complex interactions among human activities and environmental processes. From a broader perspective, the present study contributes to monitoring, sustainable conservation, ensuring resilience, and management strategies highlighting the transformations and patterns of the critical freshwater wetland ecosystem, Hakaluki Haor.

The study introduces a Machine learning approach utilizing Landsat data in Google Earth Engine and ArcGIS 10.8 to discern Land Use and Land Cover (LULC) changes in Bangladesh’s RAMSAR site freshwater wetland, Hakaluki Haor. The study demonstrated the potential of the CART algorithm in cloud-based GEE platform with efficient computation in analyzing LULC changes in the context of a freshwater wetland in Bangladesh. An analysis spanning 2000 to 2023 via Landsat satellite data revealed consistent alterations in Water Bodies, Vegetation, Barren Land, Settlement, and Crop Land throughout both rainy and dry seasons. The drivers of observed declines in water bodies and vegetation alongside variable Crop Land are considered to arise from climate change (e.g., warming, more erratic rainfall) and human activity such as development and agricultural expansion. Notably, settlement experienced rapid growth during this period. These persistent patterns underscore potential implications for the region’s ecosystem and agriculture. Despite escalating human settlements, substantial declines were noted in water bodies, dense vegetation, barren land, and crop land. Understanding the spatial distribution and LULC change patterns is critical for effective land-use strategies for the Hakaluki Haor.

The original contributions presented in the study are included in the article/supplementary material, further inquiries can be directed to the corresponding author.

AS: Conceptualization, Formal analysis, Methodology, Project administration, Software, Visualization, Writing – original draft. ER: Formal analysis, Methodology, Software, Validation, Visualization, Writing – review & editing. FS: Formal analysis, Methodology, Writing – review & editing. ND: Formal analysis, Visualization, Writing – original draft. MF: Formal analysis, Writing – review & editing. SR: Methodology, Writing – review & editing. SS: Conceptualization, Formal analysis, Supervision, Visualization, Writing – review & editing.