Perceived Human-Induced Causes of Landslide in Chattogram Metropolitan Area in Bangladesh

This study investigates Land Use Land Cover changes in the Chattogram metropolitan area, the second-largest city in Bangladesh. Using a questionnaire survey of 150 local inhabitants, the study explores perceived human-induced causes of landslides. Using time series Landsat images, this study also analyzes Land Use Land Cover changes from 1990 to 2020. The analysis reveals built-up area extended rapidly during 1990–2020. In 1990, total built-up area was 82.13 km2, which in 30 years, stood at 451.34 km2. Conversely, total vegetative area decreased rapidly. In 1990, total vegetation area was 364.31 km2, which reduced to 130.44 km2 in 2020. The survey results show that most of the respondents faced landslide therefore; it is nothing new among them. Respondents were identified several reasons for landslide like extensive rainfall, hill cutting, steep hill, weak soil texture, etc. A large number of local people opined that diverse human activities are causes of landslide in their local area and it has impacted on their livelihood. Chi-square test suggests that there are statistically significant differences between local and non-local inhabitants regarding their opinion on whether excessive hill cutting is alone responsible for landslide and whether deforestation is the sole reason for landslide. This study also used four multinomial logistic regression (MLR) to examine the effects of independent variables like gender, age, level of education, income, housing pattern and experience of facing landslide on their perception of human-induced causes of landslide. Findings show that age and experience of facing landslide are two significant predictors for the first model, explaining excessive hill cutting was alone responsible for landslide. Level of education and experience of facing landslide are found statistically significant for explaining our second model that is building infrastructures solely causes landslide. Moreover, our third model claims only deforestation can be blamed for landslide which is significantly explained by three predictors, namely gender, age and income. Finally, we found our fourth model that is landslide occurs only due to excessive sand collection is significantly explained by participant's gender, level of education, and income.


Background and Introduction
Landslide is the third-most crucial natural disaster in the worldwide that takes place over a broad range of velocities (Zillman 1999). It is a poly-causal phenomena, in which it is very difficult to separate man-made causes from natural ones, but human intervention has played a key role in stimulating the natural antecedents of landslides (Alexander 1992). Almost 9% of the world's natural disasters are caused by this phenomenon, which is particularly common in mountainous regions (Galli et al. 2008). Mohan et al. (2021) found that soil rock slope lowering was the primary cause of landslides. Climate change and increased urbanization have also been mentioned as contributing factors. In addition to natural causes (Magar et al. 2021), various human activities like: road construction, deforestation, hillslope cutting, agricultural cultivation, and vibrations caused by high traffic can also be claimed for causing landslides (Shaw et al. 2013;Rabby 2021). This type of hazard causes severe damage to resources, people's and nature's resources in the world (Zumpano et al. 2018). A landslide's aftereffects are tremendous, taking into account the number of people died, the amount of money lost, and the damage to property and infrastructure that results (Cullen et al. 2016).

3
Published in partnership with CECCR at King Abdulaziz University In Bangladesh, most of the areas are floodplain on physiographic basis, with an exception of 18% of the hilly and tracked regions where a significant proportion of citizens live (Islam and Uddin 2002), particularly North-eastern, North-South and northern hilly regions are vulnerable to landslide due to lack of land use planning and weak enforcement by the local authorities (Sarker and Rashid 2013). In the last decades, devastating landslides have constantly hit the hilly areas in Bangladesh, because of climate change along with other anthropogenic influences, such as high population density, indiscriminate land use, and uncontrolled hill cutting (Sultana 2020). Chattogram, the second-largest city, contained over 200 hills in the early 1910. As the commercial and business importance of the city had substantially increased after the independence in 1971, the hill-cutting activities severely increased to accommodate excessive land demands. Since late 1990s, unlike academic studies, traditional newspaper reporting explored the problem of hill cutting as a major cause of water logging and landslide incidences in the city area of Chattogram (Alam 2017). In terms of frequency and magnitude of damage, Chattogram Metropolitan Area (CMA) is extremely vulnerable to landslide hazards, with a growing tendency of frequency and damage (Ahmed 2015). In addition to the natural cause of excessive rainfall, Chattogram Metropolitan area's landslide vulnerability is exacerbated by a variety of human-induced factors, such as rapid urbanization, increased population density, inappropriate land use, modifications in the hilly regions by illegal hill-cutting, random deforestation, weal soil structure, de-vegetation, and agricultural practices (Islam 2018;Ahmed et al. 2014). Major effects of landslide on the local communities are loss of natural scenic beauty, economic loss, destruction of lives and environmental problems (Islam 2018). In the recent time, human activities of indiscriminate hill cutting for slum expansion and residential housing development have resulted in many landslides in the Chattogram metropolitan area. Landslides that occur because of rainfall pose a severe threat in the Chattogram Hill Districts (CHD) of Bangladesh. Inhabitants living on the steep slopes are highly vulnerable to landslide disasters. A heavy rainfall in 2017 led to a major fatal landslide in Bangladesh history that caused 168 deaths and smashed around 40,000 houses in Rangamati, Chattogram, and Bandarban regions (Ahmed et al. 2018).
Human-induced landslides (HIL) refer to landslide incidents that are directly triggered or partially aggravated by anthropogenic activities. These anthropogenic causes are the modifications of topography, changes of water circulations, land use changes, and constructions of infrastructure (Jaboyedoff et al. 2016). Studies have found that the human and geomorphological factors are more significant to cause landslide hazards than geological influences (Dahal et al. 2008). Landslides cause community disruptions, and involves in both direct and indirect costs. Direct costs are the damages immediately attributable to the landslide, but the indirect costs include economic constraints and ecological effects that often exceed the direct costs (Turner 2018) which have crucial effects on the socio-economic structure (Saina et al. 2016). However, it is difficult to separate the losses from direct and indirect causes of landslide, because losses are not well documented (Kjekstad and Highland 2009). Different effects of landslides have increased in the recent times because of the rapid expansion of urbanization in the developing world and causes damaged in the many aspects of human life as well as the natural environment. Physical or socio-economic losses seriously affect populated regions (Krivoguz and Bespalova 2017). Human-caused landslides have been acknowledged by Alam (2020), Ahmed (2021), Jaboyedoff et al. (2016) and Fell (2018). Bangladesh was the focus of Alam's (2020) and Ahmed's (2021) research, who pointed to unsustainable and unplanned growth, unlawful hill cutting, settlement along hill slopes, and overpopulation as the primary causes of the problem.
Land Use and Land Cover (LULC) change can enhance or reduce susceptibility of landslide in the mountainous and hilly areas (Chen et al. 2019). Landslides are influenced by different climatic and environmental factors, such as topography, morphology, hydrology, lithology, and land use. The changing magnitudes of LULC potentially increase the quantity of unstable hillslopes (Reichenbach et al. 2014). Landslides result a major constraint on development, causing high levels of economic loss and substantial numbers of fatalities each year (Petley et al. 2007). It causes loss of life and injury to people and their domestic animals and damage to infrastructure, agricultural lands and housing (Perera et al. 2018). It is well known that stability of slopes changes based on land use land cover changes because vegetation changes may influence the mechanical and hydrological characteristics of slope (Greenway 1987). Geographical Information System (GIS) and Global Positioning System (GPS) can be used to measure the landslide catalog (Amatya et al. 2019;Moayedi et al. 2019). Moreover, the remote sensing method for landslide detection is supported by many studies including Mohan et al. (2021), Zhao and Lu (2018), Kalantar et al. (2020), Zhong et al. (2020) etc. Mostly, environmental factors can be measured using Remote Sensing (RS) images, which encompass Digital Elevation Model (DEM), aerial Imagery, LIDAR, and the Landsat8 TM image (Zhang et al. 2016;Zhao and Lu 2018;Zhu et al. 2020).
In this study, we aim to identify local people's perception regarding human-induced causes of landslide in the Chattogram metropolitan area of Bangladesh. We also tried investigating land use land cover change over the last forty years. We started this manuscript with a background and Published in partnership with CECCR at King Abdulaziz University introduction of the study. In the next section, we elaborately discussed our methodology of this research. In Sect. 3, we present our results in different subsections. Finally, we add a concluding discussion section incorporating some limitations and recommendations for further studies (Sect. 4).

Study Area
Every year, landslide occurs in the south-eastern parts of Bangladesh (Mia et al. 2015). Compare to other regions, Chattagram city has been known as one of the most susceptible cities to landslide. The most devastating case of landfall occurred on 11 June 2007 in the Chattogram, which is one of the major landslides in the history of Bangladesh (Sultana 2013). The present study explores anthropogenic causes on landslides in Chattogram metropolitan areas ( Fig. 1) of Bangladesh. The population of Chittagong city is about 5 million and is growing. This area is within 22° 14′ and 22° 24′ 30″ north latitude and between 91° 46′ and 91° 53′ east longitude (Ahmed 2015). As mentioned above, a landslide is a major geologic hazard in Bangladesh. Some specific zones of the Chattogram Metropolitan area are more landslideprone than other zones. For example, Motijharna, Baizid Bostami, Kushumbag, Batali Hill and Lichu Bagan (locally called) are mostly populated and more vulnerable to landslide. We select Lichu Bagan, Batali Hill and Motijhorna as the study area of this study.

Data Collection Tools and Techniques
This study adopts descriptive-explanatory strategy (Babbie 2004;Islam 2008), combining a survey using a selfadministered questionnaire (Mugambiwa and Dzomonda 2018) technique and GIS approach. The questionnaire was organized in a way to help achieve the aim of this study. The questionnaire had several parts highlighting the sociodemographic characteristics of respondents. This study collected data from 150 respondents about their education, age, income, occupation, breadwinners, family size, and Published in partnership with CECCR at King Abdulaziz University housing patterns. In the next part, the respondents were asked about their challenges and experiences related to landslide hazards in their area. The respondents returned their understanding about causes of landslide, including human activities: How do they consider landslide effects in their lives? Do they follow any preventive strategy to mitigate landslide hazards? and Do they experience any premanagement activities the governmental and non-governmental actors take? A categorical response was provided for respondents to check off the option according to their own choice. In the last part of questions, each respondent was asked about their more specific opinions regarding different causes of human-induced landslide, such as excessive hill cutting, infrastructural development, deforestation and excessive sand collection.
In the GIS approach, to determine LULC change, this study collected secondary data from the USGS. The study collected Landsat images for the years 1990, 2000, 2010 and 2020. Each satellite image reflects the dry season and sensor was Landsat TM and OLI/TIRS (Table 1). After pre-processing of satellites images (atmospheric and radiometric correction), it was classified into four different classes (water body, barren land, buildup area and vegetation). After images classification, accuracy assessment was conducted (Table 2) to verify the correctness of images classification.

Measurement and Data Analysis
In this study, the outcome variable is the local people's perception about human-induced causes of landslide. The questions about this variable attempt to measure whether excessive hill cutting is responsible alone for landslide. The explanatory variables of this study include: gender, age, level of education, income, housing pattern, and landslide experience. The study employed Chi-square test and multinomial logistic regression models to determine the effects of independent variables on the people's perception about human-induced causes of landslide. Researchers use these statistical techniques to analyze perceptions of local people (Manandhar et al. 2014; Brouder and Lundmark 2011).

Pre-preparation of Satellite Images
Radiometric correction In this study, only green and nearinfrared band of each image receive the radiometric and atmospheric corrections. Using radiance value, we converted the respective DN value: By aiding Erdas Imagine application, we created a model to perform the conversion.
Q cal max , Q cal min , L max , and L min values were taken from the Metadata file provided with Image file. We followed the same procedure to convert the DN values for Green and NIR bands of each image. After converting the DN value into radiance value ( L ), Eq. 2 was used to get top of atmospheric reflectance.
The value of the Earth-Sun distance, d was calculated using Julian calendar and solar zenith angle s was taken from MLT file.E sun is the mean solar exo-atmospheric irradiance in w m 2 µm, the E sun value varies with the super craft  and sensor of satellite. The E sun values for Landsat 7 and Landsat 5 were collected from Landsat 7 handbook (Irish 2000;Chander et al. 2009). Atmospheric correction Dark Object Subtraction is a simple image-based method of atmospheric correction which assumes that there are at least a few pixels within an image which should be black (% reflectance) and suck black reflectance as dark object which extracts clear water body and shadows with DN values zero (0) or close to the zero in the image (Chavez 1988).
NDVI Index Normalized Difference Vegetation Index (NDVI) is a globally accepted remote sensing index widely used to sense the vegetation, forest extension and the water bodies over the surface using red and near-infrared light. An NDVI value always ranges from − 1 to + 1 where a value of + 1 shows heavy vegetation, while -1 implies an extensive deep-water body, with 0 signifying the absence of any vegetation. NDVI equation is given below in Eq. 3 and Table 3 denotes the NDVI values which was used in this study. Table 4 demonstrates socio-demographic background of the respondents. In this study, 62.7% respondents are male and 37.3% are female. Based on age, we categorized our respondents into three different groups. 72% participants are from middle-aged group (31-60 years of age). Moreover, young-aged group comprised 26% participants who are less than 30 years of age and old-aged group consisting of participants more than 60 years representing 2% of the study population. In response to the question related to educational attainment, this study finds that 45.3% people can read and write only, 36% of the respondents have primary   Figure 2 recapitulates respondents' understanding about landslide-related issues. The results indicate landslide is a familiar hazard in the study area as most of the respondents experienced landslide (62%) in their life. However, 32% of the respondents reported that they do not have any such experience. Regarding causes of landslide, over two-quarters of the respondents spoke of extensive rainfall as the major cause compared to almost one-quarter who identified hill cutting as the principal cause. In contrast, some other participants replied steep hill (8%) and weak soil texture (8.7%) as primary reasons for landslide hazards. The findings indicate 94% respondents believe different human activities cause landslide and 95.3% pointed out it has impact on local people. A more than half of the participants (51.3%) replied that they adopt any kind of strategy that can help to avoid risk of causing landslide. In contrast, 48.7% do not follow any such strategies. Most of the participants adhered to resettlement (68.5%) for avoiding risk, whereas 24.6% respondents preferred stopping hill cutting and enhancing afforestation (6.9%) could be a better solution. Majority of the respondents opined they comply with pre-management initiatives by Government or non-governmental organizations (NGOs) to minimize the risk. Major policy actions that the respondents try to comply with are stopping illegal activities (40%), identifying landslide-prone area (24%), expanding tree plantation (19.3%), and developing water drainage system (16.7%).

Locality and Perception Regarding Human-Induced Landslide
Most of the local and non-local respondents do not agree that excessive hill cutting is solely responsible for landslide (58.1% local and 63.6% non-local). In contrast, 41.9% local and 36.4% non-local respondents believe that excessive hill cutting is a solitary cause for this hazard. The test of significance (Chi-square test) suggests that there is a statistical significant difference between local and non-local participants regarding their perception of excessive hill cutting as the only cause for landslide hazard (p < 0.05). 75.2% local and 66.7% non-local respondents said they do not agree on infrastructural development as the only cause for landslide. However, this finding is not statistically significant (p > 0.05). 24.8% local and 33.3% non-local residents opine that building infrastructure is the only cause for landslide. Table 5 also depicts that non-local respondent (78.8%) show more disagreement than local people (72.6%), when they were asked whether deforestation is the only cause for landslide. A good percentage of respondents agree that deforestation is only to blame for it (27.4% local and 21.2% nonlocal). Chi-square test result confirms a significant variation between the testimonies of local and non-local respondents (p < 0.05). Regarding excessive sand collection as the principal reason for landslide, there is no significant (p > 0.05) difference between the opinions of local and non-local respondents (94.0% local and 90.9% non-local). Only 6.0% of local and 9.1% non-local responders think excessive sand collection is the only reason for it.

Perception Regarding Human-Induced Causes of Landslide: Multinomial Logistic Regression
We try to examine the factors that are affecting the perception regarding human-induced causes of landslide using four multinomial logistic regression models. Outcome variables used for the models are statements asking whether "excessive hill cutting was alone responsible for landslide", "building infrastructures solely cause landslide", "only deforestation can be blamed for landslide", and "landslide occurs only because of excessive sand collection." The response categories of the outcome/dependent variables are coded with '0' if they agree with the statement and '1' if they do not agree with the statement. For all models, the reference category is 'disagree'. The result shows ( Table 6) that two predictors' age and experience of facing landslide have significant effects for the first model about "excessive hill cutting was alone responsible for landslide." The result illustrates that middle-and old-aged respondents compared to young respondents are less likely to agree with the statement that excessive hill cutting is alone responsible for landslide. Respondents who do not experience landslide compared to their counterpart are less likely to agree that landslide occurs mainly because of excessive hill cutting. Although the result is not significant, but the table presented below also suggests that females compared to males, respondents who earn over 10,000 Tk. per month compared to these who earn less than that and who obtained more than secondary-level study than who have below secondary-level study are more likely to agree with the statement.
We found that two predictors, such as level of education and experience of facing landslide, are statistically significant for the second model about "building infrastructures solely causes landslide." Respondents having more than secondary-level study are less likely to agree with the statement compared to their counterpart. The result also shows that people who have no experience of facing landslide in their area are 2.5 times more likely to agree with the statement "building infrastructures is the only cause for occurring landslide" compared to their counterpart. Although the result is insignificant, the tables below show that middleand old-aged people are 1.5 times more aligned with the statement than young people do. Table 6 illustrates that three predictors' gender, age, and income significantly explain the model dealing with "only deforestation can be blamed for landslide." Females compared to males, middle-and old-aged respondents compared to young and respondents whose monthly income over 10,000 Tk. compared to these who earn less than 10,000 Tk. per month are less likely to agree with the statement "deforestation is the only reason for landslide." We also found that respondents who have tin shed and earthen house compared to Published in partnership with CECCR at King Abdulaziz University these who have building or semi-building house and respondents who do not have experience of facing landslide compared to people who have experienced landslide in their home area are more likely to blame deforestation as the sole reason for landslide. The results are not statistically significant. Finally, three predictors' gender, level of education, and income significantly explain the model regarding "landslide occurs only because of excessive sand collection." Females are 4.8 times more likely to agree than males that excessive sand collection causes landslide, and that is the only cause. Respondents having more than secondary-level study compared to having below secondary-level study are 2.3 times more likely to agree with the statement. Findings also show that respondents who earn over 10,000 Tk. per month are less likely to blame excessive sand collection as a solitary cause to landslide compared to their counterpart. While the results are not statistically significant, yet middle-and oldaged people compared to young people and respondents who have not experienced landslide than respondents who have experience are more likely to agree with the statement. Table 7 and Figs. 3, 4, 5 and 6 reveals the value of NDVI analysis of CMA that shows a speedy increase in the builtup areas between 1990 and 2020. In 1990, built-up area was 82.13 km 2 which stood at 451.34 km 2 in December 2020. However, the total vegetative area decreased rapidly in the same period. In 1990, total vegetation area was 364.31 km 2 , which came down to the CMA of 130.44 km 2 in 2020. We conduct NDVI analysis with local people's perception to investigate which process further intensifies the landslide process. NDVI result reveals that in CMA built-up area increased quickly in the last 40 years and total vegetative area decreased, which further intensify the landslide susceptibility in the metropolitan area of Chattogram.

Discussion and Concluding Remarks
For the last thirty years, Bangladesh has been experiencing hill-cutting problems and consequent landslide incidence in the southeastern hilly region (Alam, 2020). The present Published in partnership with CECCR at King Abdulaziz University study finds that more than half (mention the %) of the total participants have experience of facing landslide compared to the remaining 32% who do not have this experience. This result clearly shows that landslide is not surprising among the inhabitants of the study area. Moreno and Ayala (2017) argue in their study that a better understanding of how landslide is perceived is one of the most important issues in enhancing landslide disaster risk awareness and knowledge. They suggest a knowledgeable landslide perception among people to create a resilient community.
In terms of causes of landslide, extensive rainfall (58.7%) has been identified as the most critical cause followed by hill cutting (24.7%), steep hill (8%), and weak soil texture (8.7%). This study also suggests that only 41.9% local and 36.4% non-local respondents believe that excessive hill cutting is solely responsible for landslide, and there is also a statistically significant difference found between local and non-local participants regarding their perception of excessive hill cutting as the main reason for landslide. Hassan et al. (2015) found similar results where they identified hill cutting as the principal reason for the subsequent landslide occurrence in Chattogram city area.
The findings in this study show that most of the local and non-local inhabitants opine that different human activities cause landslide in the study area, which is detrimental to their local life. As landslide is not uncommon to most of the respondents, more than half of the participants followed any kind of strategy, which helps them avoid risk of happening landslide. Alam (2020) and Burton et al. (1993) hold in their study that it is crucial to recognize how people living in the unstable environment perceive hazards and risks and to understand their indigenous knowledge and awareness regarding particular hazards. The authors argue that this kind of perception may be critical to lessen the consequential effects of natural hazards. The current study asked participants to mention what strategy particularly they follow to avoid risk of landslide. Most of the people responded they follow resettlement (68.5%) to avoid risk followed by stopping hill cutting (24.6%) and enhancing afforestation (6.9%). They also recommended some pre-management activities, Published in partnership with CECCR at King Abdulaziz University such as blocking illegal activities, identifying landslideprone area, expanding tree plantation and developing. In the risk mitigation approaches, hazard knowledge and risk discernment are vital components (Gaillard, 2008). It is also suggested that human interference has played a key role in stimulating the natural precursors of landslides occurring and risk appreciation does not promote adequate risk mitigation (Alexander, 1992).
A study conducted by Hassan and Nazem (2016) about LULC change and urban growth in Chattogram city and result of the study reveals that because of the increase of built-up areas, 56% of the land covers have undergone change. This change triggers further encroachment and degradation because of other human activities near urban areas. Roy and Saha (2016) explore the temporal pattern of land cover change in Chattogram district that shows that urban area and barren land are rising, whereas the forestland is declining at an alarming rate during 2002 and 2014. Gazi et al. (2020) conducted a study about spatio-temporal changes of land cover in Chattogram metropolitan area that reveal urban structures increased rapidly (from 20.83 to 58.93%), while vegetation area decreased from 56.54 to 20.24% in the study period. The present study also found that built-up (in 1990, the area was 82.13 km 2 . and in 2020, it was 451.34 km 2 ) area increased rapidly and vegetation (in 1990, it was 364.31 km 2 , but in 2020, it was 130.44 km 2 ) decreased in the study period.
We found that young respondents and respondents who experienced landslide in their local area are more likely to agree that excessive hill cutting is the only reason for landslide compared to their counterparts. Respondents having more than secondary-level education and who have experienced landslide in their local area are more likely to agree that building infrastructures is the only cause for occurring landslide than their counterparts. We also found that females, middle-and old-aged and these to earn over 10,000 Tk. per month are comparatively less likely to think that deforestation is the only reason for landslide. Females compared to men, respondents having more than secondary-level education compared to below secondary-level education and respondents having less than 10,000 Tk. earning per month compared to these who earn over 10,000 Tk. monthly are more likely to agree that excessive sand collection is the lone cause of landslide. A study conducted on an indigenous community living in Taiwan by Roder et al. (2016) where their results suggest that gender, age education and experience of natural hazards were significant predictors in hazards knowledge and risk perception also paying attention to the indigenous perception of a hazard and risk, can increase the effectiveness of projects implemented by government or any organization. Rieux et al. (2012) work on coping strategies and landslides in two villages of Central-Eastern Nepal and finding suggest that importance of investing in organizational skills, while building on local knowledge about landslide mitigation for reducing landslide risk.
This study explores the perception of local people about human-induced causes of landslides. The results show that human alteration influences natural causes of landslides and people's perceptions vary based on gender, age, educational attainment, experience of facing landslide and their financial condition. As local inhabitants face the effects of landslide directly, their perceptions and opinions, especially make them more resilient. This kind of information is significant for decision-makers and authorities who need to recognize and take action for effective landslide management at the local level in the hilly area of Bangladesh and beyond. Findings of this study uncover the local perceptions regarding landslide causes that may be helpful for the policy-makers and other stakeholders to find a better solution to this problem and assist the responsible bodies for taking better plan related to landslide. The present study can be an example for the future study which will be combined-topography, geology, geography, climatic data and land use and land cover change-help the decision-maker for the formulation of rules and guidelines about human-induced causes of landslide especially in hill cutting, infrastructure development in the hilly area and sand mining in the hilly region and aid in minimizing negative effects on local inhabitants. One of the major limitations of the study is that it only looks at one particular area and our sample size was 150, which is too small to generalize the overall scenarios of perception regarding this issue. A comparative study among different parts taking large samples could be an interesting work. However, these limitations will certainly pave the way for future studies to overcome these weaknesses.