Determinants of households’ livelihood diversication strategies to adapt with natural disasters: Evidence from ecologically vulnerable haor region of Bangladesh

The haor region of Bangladesh is exposed to a variety of natural disasters such as ash oods, seasonal oods, droughts, riverbank erosions, embankment breaches due to climate change, which impacts the haor people's lives and livelihoods. The study aims to assess the various livelihood strategies adopted by the haor households, as well as the factors that inuence their decision to pursue more environmentally friendly and sustainable livelihood strategies. The primary data from the 300 haor households in Kishoregonj, Netrokona and Sunamgonj districts were collected with a multi-stage stratied random sample technique taking 100 of each district. We provide inimitable insight into the analysis framework for understanding sustainable rural livelihood, as well as empirical evidence of how livelihood resources, livelihood strategies, and livelihood outcomes are strongly interrelated. The study classied households’ economic activities into ve distinct categories together with crop farming to cope with natural disasters. Among the livelihood options, crop plus livestock rearing is the most productive livelihood strategy for haor households. The ndings revealed that the higher returning livelihood diversication strategies are signicantly inuenced by the household’s head age and education, dependency ratio, land holdings, household assets value, access to credit, annual income, membership of any organization, home to road, market, and haor distances, communication during the dry season, duration of waterlogged, and agro-ecology. In order to change the local context and enable poor households to establish more protable livelihood strategies, policies should aim to promote the signicant determinants of livelihood strategies, as well as ensure livelihood assets, a strong infrastructure, and minimize natural disasters. (BDT. 6,25,324) followed by crop plus sh farming (BDT. 365645.1), crop plus poultry farming (BDT. 281090.6), crop farming (BDT.190226.70), and crop plus off-farm activities (BDT. 186207.80). The nancial capital comprises access to credit, annual income, and annual expenditure indicators of a household livelihood diversication strategy. The result shows that the percentage of households in access to credit indicator is the highest (0.57) for only crop farming and the lowest (0.37) for crop plus off-farm activities. We observed a similar pattern in the other two nancial capital indicators while annual income and annual expenditure are the least for pursuing only crop farming livelihood strategy. Membership of any organization is an important indicator of social capital which is a driver of pursuing the highest protable livelihood strategy that is crop plus livestock farming. The infrastructure indicators home to road distance (0.58 km) are the least and the home to market distance (1.49 km), home to haor distance (1.45 km) is the highest for pursuing in crop plus off-farm activities, indicating as important accessibility determinants for livelihood strategy. On the other hand, communication system such as river way or roadway or both ways play an important role in the diversication of livelihood strategy. Due to its household total land holdings (-1.426***), had a negative and signicant inuence on adopting crop plus poultry farming as a livelihood strategy. Crop plus off-farm activities are a very important part of the subsistence household, which includes a diverse range of activities. The households of that group were positively and signicantly inuenced by the HH head education (0.128**), annual income (6.29e-06***), located closer to the haor (0.552*), and high land (2.637**) type agro-ecology compared to low land for adopting such kind of livelihood strategy. Household’s total land holdings (-0.918***), and accessibility to credit (-1.169***) negatively and signicantly inuences the household's decision on adopting crop plus off-farm activities. Due to the ecological characteristics of the haor areas, infrastructural and geographical factors have had a greater impact on the adoption of livelihood diversication strategies than socioeconomic variables. with the adoption of the crop plus livestock farming indicating a unit increase of the membership facilities in an organization increases the probability of crop plus livestock farming by 7.7%. Home to road distance is positively related to the crop plus livestock farming and crop plus poultry farming, and negatively related to the crop plus off-farm activities. It indicates a 1% increase of the home to road distance results in an increase of crop plus livestock farming and crop plus poultry farming by 11.9% and 10.4% respectively. Whereas a decrease of 18% in the probability of adopting crop plus off-farm activities with the increase of 1% home to road distance. Similarly, home to market distance has a positive association with the crop plus sh farming, crop plus off-farm activities, and a negative association with the crop plus livestock farming. Home to haor distance shows positive and signicant evidence with the adoption of the crop plus off-farm activities indicating a 1% increase of the home to haor distance would lead to an increase of a 5.5% in the probability of choosing crop plus off-farm activities as a livelihood diversication strategy (Hoq et al. 2021b). Roadway communication during monsoon signicantly and negatively inuences the haor household in adopting livelihood diversication compared to its base category river way communication. The results indicate that a 1% increase in roadway communication decreases the probability of adopting crop plus sh farming by 14.5%. On the contrary, riverway and both way (road and riverways) communication during dry season signicantly and positively inuence the haor household in adopting crop plus poultry farming and crop plus livestock farming respectively. It indicates that one unit increase in the riverway communication increases the probability of choosing crop plus poultry farming by 15.2% and `increases the probability of choosing crop plus livestock farming by 7.9% as a livelihood strategy. Unexpectedly duration of waterlogged signicantly and positively inuences the crop plus livestock livelihood strategy which indicates one unit (month) increase in the duration of waterlogged is likely to increase the probability of crop plus livestock farming by 3.4% (Hoq et al. 2021b). Hence, it may be due to rearing livestock in household home yard which is physically observed during the data collection. Farmers living in different agro-ecological settings used different livelihood strategies as a measure of different adaptation strategies. Farming in medium

highly seasonal farming in the haor areas of Bangladesh (Khan and Islam 2005;Gautam and Andersen 2016;Blackmore et al. 2021). During winter, paddy is produced with minimum effort while during monsoon the same is turned into a water basin which is a breeding place for shery and a wide range of water biodiversity (Sheuli 2017). Generally, monsoon is the lean season for agriculture because all infrastructures, including road networks in the haor areas, are submerged for about 6-7 months (BHWDB 2012;CAN 2017). Therefore, this season offers a window of opportunities to the haor household to have interaction in several farms or off-farm activities i.e. shing, poultry raising (mainly duck raising), and off-farm activities like wage laborer, boating, and small scale business for sustaining their livelihood (Islam et al. 2012;Rabby et al. 2013;Gautam and Andersen 2015;Trina et al. 2015;Alam et al. 2017;Akter et al. 2020). Different livelihood diversi cation strategies require a different level of investment and offer differential earnings. The poor households are less likely to get involved in higher returning sectors as it demands higher capital (human, social or nancial capital) investment to start up (Gautam and Andersen 2016). As the majority of the haor households are living below the poverty line, it is crucial to explore and develop an in-depth understanding of the causes of poverty and vulnerability issues in the haor areas and how they manage their day to day activities as a part of the process of attempting to secure a sustainable livelihood. Previous study outside from Bangladesh consistently showed that diversi cation to the non-farm sector improves farm household incomes, enhance food security, and boosts agricultural productivity by reducing capital constraints and assisting in the management of environmental hazards (Barrett et al. 2001;Davis 2014;Loison 2015;Gebru et al. 2018). Livelihood activities of the haor household traditionally included irrigated boro rice cultivation, combined with homestead vegetable gardening, livestock rearing, shing, poultry farming, and off-farm income activities i.e. small business, boating, day laborer (agricultural and non-agricultural), and household work as sources of subsistence (Hoque et al. 2018;Islam et al. 2016a;Islam et al. 2016b). Boro rice is grown on 80% of the haor areas, and it is frequently harmed by early ash oods, hailstorms, and drought (Alam et al. 2010;Ali et al. 2019). The ood risk is a great hazard for boro rice cultivation which trends to food insecurity of haor households because boro rice is the foremost source of year-round food supply for their livelihood (Ferdushi et al. 2019;Kabir et al. 2020). Fish culture is the second most important livelihood strategy after crop farming (rice) in Bangladesh and its production contributes to the livelihoods and employment of millions of people (Islam et al. 2012;. Another lucrative livelihood strategy of haor household is livestock rearing in the haor areas during the dry season as it has large grazing land (Islam et al. 2016c). Due to ecological and geographical di culties, the monsoon in the haor area is not suitable for livestock rearing. But during monsoon, duck rearing is a pro table option for the haor household because of the ecological support (Islam et al. 2016c). Homestead duck and chicken eggs are available in the local market and they earn a considerable amount of cash income for their livelihood (Huque et al. 2011). So, adopting context-based livelihood diversi cation strategies could be able to meet the present challenges of diversi cation, thus attaining household food security and improving livelihood security (Abera et al. 2021;Gebru et al. 2018). Most previous studies focused on the different issues, such as socioeconomic status, cropping patterns, livelihood vulnerability to ooding (i.e., livelihood exposure and sensitivity), and adaptation perspectives of the haor household, which justi ed the rationale for the study to be undertaken (Basak et al.2015;Dewan 2015;Rahman et al. 2018b;2018c;Uddin et al. 2019;Hoq et al. 2021a). Only a limited number of studies were conducted on livelihood strategies and most of them concentrated on a single practice (Trina et al. 2015;Islam et al. 2016b;2016c;. Despite the signi cance of these studies, questions about how frequently and under what socioeconomic conditions diversi cation can take place remain unanswered. As a result, the focus of this research is to ll that gap by combining all possible livelihood strategies that can lead to livelihood diversi cation.
[1] A haor is a wetland ecosystem that physically is a bowl or saucer shaped shallow depression.

Description of the Study area
The research was carried out in the haor areas of Bangladesh, which are situated in the northeastern part of the country. The haor region's climate is subtropical monsoonal, with an annual rainfall of around 4,000 mm. The monsoon season, which runs from June to October, yields almost 80% of the rainfall. Pre-monsoon temperatures range from 26°C to 31°C (March to May), rainy season temperatures range from 28°C to 31°C, and winter temperatures range from 26°C to 27°C (Banglapedia 2003). The haor areas are spread over seven districts of Bangladesh; Sunamganj, Habiganj, Moulvibazar, Sylhet, Kishoreganj, Netrokona, and Brahmanbaria. Three dense haor districts (Kishoreganj, Sunamganj, and Netrokona) were purposively chosen from seven haor based districts and two administrative sub-units of each district (Upazila) were selected from each district, giving priority to the core haor area. The six administrative sub-units (Upazila) were Itna and Mithamoin in Kishoreganj, Khaliajuri and Mohonganj in Netrokona, and Derai and Salla in Sunamganj district (Fig. 1).

Determination of sample size
A total of 300 haor households with 100 from each district were surveyed from six Upazilas in three districts to collect primary data for reaching the study's objectives. Based on the household size of each Upazila, 50 potential farmers were selected from the union level (local level administrative unit) given priority for an equal distribution of six Upazila. The sample size was determined by following the given formula by Kothari (2004) Where n is the sample size and P is the estimated proportion of respondents, 0.5 was used as a p-value to get a maximum number of the respondents. Z denotes the amount of standard error in the 95% con dence interval, which is 1.96, and e denotes the margin of error that the researchers are willing to accept, which is 0.06.
Therefore, a complete of 300 households was selected from ve livelihood strategy group (Table 1). The sample household picked proportionally from each livelihood strategy group.

Data sources and data collection procedure
This study relies on farm-level primary data collected from September 2019 to March 2020, with secondary information used only to compare the research ndings. A pre-tested, semi-structured questionnaire was used to collect primary data from 300 haor households. Secondary data obtained from earlier research into the subject helped the researcher better understand the research area, shaping the research approach and identifying gaps that need to be lled by the research. Secondary information on relevant issues was gathered from various published sources, such as the Bangladesh Bureau of Statistics (BBS), the Upazila Agricultural O ce (UAO), the Department of Agricultural Extension (DAE), and many others. Face-to-face interviews were conducted with the household head by trained enumerators with Bangladeshi linguistic communication (Bengali) for collecting primary data.

Sampling technique and sampling procedure
Multi-stage strati ed sampling technique was used to select study location and sample households. In the rst stage of sampling, the study locations were selected purposively, keeping in the mind of the geographical location and core haor area. The second stage involved the identi cation of ve livelihood strategy groups, which was mostly practiced by the haor household. The numbers of sample households from each group are determined proportionately based on total household's number in the respective group. Finally, required numbers of sample households were selected randomly from each livelihood strategy group.

Analytical framework to understand sustainable rural livelihood
The sustainable livelihood concept is gradually important in the development context. The Sustainable Livelihood Framework (SLF) offers a strategy to reduce poverty through investigating the lives of poor people (Agarwala et al. 2014). The capabilities, tangible and intangible assets, and activities are needed to make a living can be de ned as a livelihood. It will be called sustainable when it can survive with and recover from shocks and stress, and uphold its capabilities and assets, including the natural resource both presently and in the future. The sustainable livelihood framework (Fig. 2) is useful for determining whether people's livelihoods are sustainable in terms of three key aspects (Khatiwada et al., 2017;Carney, 1998;Soltani et al., 2012;Babulo et al., 2008). These are livelihood assets (human, natural, physical, nancial, and social capitals), as well as mediating factors, livelihood diversi cation strategies, and outcomes that work together. However, the three aspects of the SLF are livelihood platform, livelihood strategy, and outcome (Fig. 2).
Human, natural, physical, nancial, and social capitals are the ve categories of a household's assets highlighted by the SLF. Households combine these assets together with activities and choices and construct a portfolio of activities (such as crops, livestock, poultry, sheries, etc.) to achieve their livelihood goals which can be de ned as livelihood strategies. Households' livelihood strategy choice is in uenced by its asset holding and external factors such as geographical location and infrastructure. This study also used household asset holding, geographical location (agro-ecology, waterlogging), and infrastructure (proximity to road and market center, communication system, etc.) as conditioning factors for livelihood strategy choice (Tesfaye et al., 2011;Soltani et al., 2012). Livelihood outcomes are the gains from livelihood strategies (such as income, food security, poverty reduction, environmental sustainability, etc.) 2.6 Empirical Econometric model to understand determinant of livelihood strategy When the dependent variable has more than two choices, multinomial logistic regression (MNL) is a popular and widely used model for nominal outcomes (wulff 2014). According to Green (2003) when the dependent variable has more than two alternatives from which the decision-maker must choose (i.e., unordered qualitative) the appropriate econometric model is either multinomial logit or multinomial probit regression. However, due to the estimation di culties imposed by the need to solve multiple integrations related to multivariate normal distributions, multinomial probit is rarely used in empirical investigations. Therefore, a multinomial logit model was used in the study to determine the factors that in uence a haor household's decision to participate in various livelihood strategies (Abera et al. 2021;Gebru et al. 2018;Amare and Simane 2017). This model was chosen not only for its computational simplicity, but also for its greater ability to anticipate livelihood diversi cation and differentiate between rural households' livelihood strategies (Amare and Simane 2017). The dependent variables of the multinomial logistic regression are polytomous and allow more than two discrete outcomes. It can be used to analyze socioeconomic and physical explanatory variables empirically. A logical household head selects one of ve livelihood strategy that provides the most utility. Assume for the i th respondent faced with j choices, we specify the utility choice j as: U ij is the highest of the j utility values if the respondent prefers choice j. As a result, for j selection, the statistical probability model is: Where, U ij represents the utility of the i th respondent from j th livelihood strategy and U ik represents the utility of the i th respondent from k th livelihood strategy.
If a household's utility is calculated in terms of income realizations, then its option is simply an optimal allocation of its assets endowment to select a livelihood that maximizes its utility (Brown et al. 2006).Thus, the decision of the i th household can be modeled in terms of optimizing expected utility by selecting the j th livelihood strategy from a set of J discrete livelihood strategies, thus the choice function is: If the i th household chooses the j th livelihood strategy to maximize its utility, the value could be 1 if the i th household chooses the j th livelihood strategy and 0 for otherwise. Thus, the probability of i th household with x characteristics of j livelihood strategy are: With the requirement that ∑ J j P ij = 1 for any i Where; P ij =Probability of i th respondent's chance of falling into category j, X = Predictors of response probabilities and β j = Covariate effect speci c to j th responses category with the rst reference category. The assumption that β 1 = 0 is an acceptable normalization that eliminates an indeterminacy in the model. This arises because probabilities sum to 1, so only J parameter vectors are needed to determine the J+1 probability (Galab et.al, 2002), so that exp (X i β 1 ) =1, implying that the above generalized Eq. The restriction that the J probabilities sum to 1 yields the likelihood of p i1 That is P i1 =1-∑ P ij .
We can compute J log-odds ratios, which are de ned as in the binary logit model, speci ed as: In P ij P iJ = x(βj-βJ) = xβj, ifJ = 0(7) This type of discrete outcome model can be estimated by applying the maximum likelihood method.
Marginal effects must be determined if reasonable conclusions about the direction and magnitude of the relationship between an independent and dependent variable in an MNL are to be drawn (Bowen and Wiersema, 2004

Variables used in the model and their de nition
Five livelihood strategies groups were identify from the haor household and these were used as the dependent variables. It was assumed that the livelihood strategy choice is a function of several livelihood platform variables including a household's demographic factors, assets pentagon (human, natural, physical, nancial, and social capitals) and location and geographical factors. Therefore, understanding the rural background of the haor inhabitants and reviewing the relevant literature, the explanatory variables as shown in Table 2 are selected for this study. The selected variables, their de nitions, and expected relationships to livelihood strategy choices are summarized in Table 2.

Household's income composition of different livelihood strategy
This study identi ed haor households into ve livelihood strategy groups based on the haor household's existing livelihood activities. The Majority (34%) of the haor households are engaged in only crop farming, particularly rice agriculture, which is unpro table but has to continue due to their forefathers' profession (Hoq et al., 2021a). Crop plus off-farm activities employed approximately 18% of households, followed by crop plus livestock farming (17.33%), crop plus sh farming (15.67%), and crop + poultry farming (15%) ( Table 1). The sources of income for the various livelihood strategy groups are shown in Table 3. Rice, vegetables, oilseeds, maize, livestock, sheries, poultry, wage/salary, petty business, vehicle driving, boating, power tiller operation, remittances, and land rent were all divided into fourteen categories. Crop farming is to be considered as a major component for all ve categories of livelihood strategy group because about 80% of haor households depend on rice-based agriculture. Households in group Y1 exclusively depend on crop farming. Rice is the main crop for all types of livelihood strategies which contributes 91% of total income for group Y1 followed by 70% for group Y2, 57% for group Y3, 43% for group Y4, and 64% for group Y5. Households of the livelihood strategy group Y2 are involved in a combination of crop farming and livestock rearing which is the most lucrative earning source in the haor area and livestock alone contribute 28% of total income (Table 3). Households in groups Y3 and Y4 allocate their resources to sheries and poultry farming in addition to crop farming respectively and sheries and poultry farming alone contribute 34% and 50% of total income respectively. Commercial duck rearing by income group Y4 plays an important role to achieve such a percentage. Households in livelihood strategy group Y5 are the most diversi ed as they engage in different off-farm activities that in addition to crop farming. The households of this strategy generally pursued different off-farm activities such as wage and salary (13%), petty business (15%), vehicle driving (2%), and remittances (4%) which contribute altogether 34% of total annual income. The average annual income of different livelihood strategy groups is shown in Fig. 3    geographical location, there is little difference in the value of the duration of waterlogging in the ve types of the livelihood strategy group. The value of the agro-ecology indicators is the least for only crop farming, indicating the household in low land type agro-ecology mostly engaged in crop farming. The results of MNL regression were presented in Table 5. The study used only crop farming is as a base category to assess the effect of predictor variables on the likelihood of a speci c strategy choice. The results revealed that HH family size (0.259*), HH assets value (6.88e-06***), annual income (3.88e-06*), membership of any organization (1.517**), home to road distance (1.867**), riverway (1.992**) and both way (1.548**) (river and roadway) communication during the dry season, duration of the waterlogged condition (0.600**), and comparatively high land (3.645***) type agro-ecology than low land had the positive and signi cant impact on choosing crop plus livestock farming than only crop farming. Likewise, some other variables such as total land holdings (-1.151***) and home to market distance (-1.805**) had a negative and signi cant in uence on choosing crop plus livestock farming. While comparing the crop plus sh farming with only crop farming, HH head age (0.039*), total land holdings (-1.025***), HH assets value (2.59e-06**), annual income (5.85e-06***), membership of any organization (0.887*), riverway communication during the dry season (1.290*), and comparatively high land (3.193***) type agro-ecology than low land had a signi cant in uence on adopting the livelihood strategy of the crop plus sh farming. Similarly, households with a higher annual income (8.62e-06***), closer to the road connectivity (1.231*), riverway (1.924***) communication during the dry season rather than roadway, duration of the waterlogged condition (0.379**), and generally high land (2.704***) type agro-ecology compared to low land were more likely to adopt crop plus poultry farming rather than crop farming alone. On contrary, accessibility to credit (-1.087**), household total land holdings (-1.426***), had a negative and signi cant in uence on adopting crop plus poultry farming as a livelihood strategy. Crop plus off-farm activities are a very important part of the subsistence household, which includes a diverse range of activities. The households of that group were positively and signi cantly in uenced by the HH head education (0.128**), annual income (6.29e-06***), located closer to the haor (0.552*), and high land (2.637**) type agro-ecology compared to low land for adopting such kind of livelihood strategy. Household's total land holdings (-0.918***), and accessibility to credit (-1.169***) negatively and signi cantly in uences the household's decision on adopting crop plus off-farm activities. Due to the ecological characteristics of the haor areas, infrastructural and geographical factors have had a greater impact on the adoption of livelihood diversi cation strategies than socioeconomic variables.

Determinants of livelihood diversi cation strategy via marginal effect
We may depend on another powerful interpretative technique: marginal effects after multinomial logistic regression to further make the results to our sense.
The marginal effects help us to inform the change in predicted probabilities due to a change in a particular predictor (Bowen and Wiersema 2004;Roy and Basu 2020). A negative and signi cant coe cient of the gender of HH head indicates a negative relationship between gender of HH head and probability of adoption of a speci c livelihood strategy (Table 6). For example, an increase by one male-headed HH would lead to a 20% decrease in the likelihood of adopting crop plus poultry farming. Education is an important in uencing factor in adopting new improved agricultural technologies or information, which increased agricultural productivity Elahi et al. 2015). In our study, the signi cant positive coe cient of education of the household head shows that the probability of accessing crop plus off-farm activities increase by 1% with an increase of one year of schooling compared to crop farming. A negative coe cient of family size indicates a negative relationship between family size and livelihood strategies choice (Khatun and Roy 2012), whereas a positive coe cient between dependency ratio and livelihood strategies indicates a positive relationship. Suppose, an addition by one individual in the household size is more likely to a decrease of 2.3 % in the adoption of livelihood strategy Y4 (crop plus poultry farming). On contrary, an increase by one individual in the household size is more likely to an increase of 2.2% in the adoption of livelihood strategy Y2 (crop plus livestock farming). In the case of dependency ratio, the livelihood strategy choice of Y4 (crop plus poultry farming) would increase by 7.2% with the increase of 1% dependency ratio. There is a positive association between livelihood strategy choice and skill development training indicates that a 1% increase of the skill development training would lead to a 10.7% increases probability of adopting livelihood strategy Y3 (crop plus sh farming). Total land holdings represent the total own cultivated land held by a farm household and may be taken as an indicator of wealth for a farm household. The results indicate that the total land area has negative and signi cant impacts on households involved in crop plus poultry farming. Therefore, a 1% increase in the own cultivated land is likely to decrease the probability of adopting the crop plus poultry farming as a livelihood strategy by 7.5%. The marginal effects of household assets value exhibit very different behavior depending on the outcome category (Amare and Simane 2017). An increase in the household assets value is likely to increase the probability of adopting the livelihood strategy Y2 (crop plus livestock farming), whereas it is likely to decrease the probability of adopting the livelihood strategy Y5 (   Annual income --9.33e-08 (

Discussion
The study categorizes a household's entire livelihood activities into ve livelihood strategies groups pursued in the study areas of the wetland haor ecosystem of Bangladesh. Those strategies mainly consist of various agricultural production practices and non-farm activities (e.g., petty business, wage labor, migratory work, etc.), which are consistent with the ndings of Jiao et al. (2017) in rural Cambodia. This diverse set of livelihood approaches and income composition of each livelihood strategy suggest that rural household mostly involves in multiple income-earning activities, which are essential for survival and for reducing risk and uncertainty in the production process of the unfavorable ecosystem in rural areas (Liu and Lan 2015;Jiao et al. 2017;Ellis 1998Ellis 2000. Income from the crop sector contributes the largest share to household total income, and income from rice production is still indeed the primary source of farm income for haor households, especially for the subsistence farmer. These ndings are closely related to the ndings of Rigg et al. (2016) and Gautam and Andersen (2016) in the case of East and Southeast Asia's smallholders. Previous studies reported that about 80% of haor areas are covered by boro rice cultivation and another 10% is covered by transplanted Aman rice (Huda 2004;Khan et al. 2012;Hoq et al. 2021a). The results revealed that most of the haor households engaged in agriculture-based livelihood, especially in rice crop farming (only crop farming livelihood strategy), and they earn low and sometimes negative returns from crop farming but cannot stop due to nancial barriers and nding no alternative way; these ndings are supported by Alamgir et al. (2020). The most lucrative livelihood strategy is found in pursuing crop plus livestock farming that generated higher income from crop and livestock sales. This result agrees with Khatiwada et al.'s (2017) ndings for rural household livelihood strategies in central Nepal. The study also identi ed some other pro table livelihood options such as crop plus sh farming, crop plus poultry farming, and crop plus off-farm activities in haor areas of Bangladesh which also supported by some previous studies (Islam 2011;Islam et al. 2012;Parvin and Akteruzzaman 2013;Sunny et al. 2020;Uddin et al. 2019). These livelihood strategies require a signi cant amount of capital investment to start up that is why the majority of them are unable to participate; such ndings are in line with Gautam and Andersen's ndings (2016). The asset-poor haor households are unable to overcome the entry barriers to high return sectors and are con ned to low return sectors that cannot contribute signi cantly to their well-being (Singh et al. 2018). Some innovative farmers diverse their livelihood activities into livestock, sheries, and poultry farming and off-farm activities along with crop farming to combat natural disasters in their production process. Age and education of the household head and dependency ratio are most in uencing human capital in adopting the higher returning livelihood strategies (Xu et al. 2019) (2012) and Gebru et al. (2018). One probable explanation is that when the number of members of a household increased, the likelihood of them becoming more involved in livestock farming keeps growing as well. The results of the natural capital indicated that there is a negative and signi cant relationship between total landholding and the choice of adopting all higher returning livelihood strategies. The argument is that low return from subsistence farming alone cannot sustain the smallholder's rural livelihood which compels them to generate higher income from limited farm size through diversifying their activities into more than only crop farming (Khatiwada et al., 2017). Therefore, the haor households with large landholdings do not have a surplus labor force that can be able to diversify (Rahut et al. 2017). Physical capital found a positive and signi cant relationship between household asset value (farm and non-farm) and the two most pro table livelihood diversi cation strategies: crop plus livestock farming and crop plus sh farming. It is true that households with a high initial livelihood asset indeed have greater freedom of choosing and diversifying their livelihood into higher-returning crop plus livestock and crop plus sh farming and bene ted most. These ndings are consistent with the ndings of Adepoju and Obayelu (2013), Loison (2015), and Gebru et al. (2018). The support of nancial capital is very important to start up a new livelihood strategy (Khatiwada et al. 2017;Xu et al. 2019).
Unfortunately, the credit facility is negatively related to adopting all livelihood diversi cation strategies and signi cantly in uences the crop plus poultry and crop plus off-farm livelihood strategies. The result, however, contradicts the ndings of Gebru et al. (2018) and Davies (2004). The possible explanation could be that the haor household may not be able to utilize the credit properly in the livelihood activities. Sunny et al. (2020) also found an easy source of informal credit with unfavorable interest, terms, and conditions for the haor household, which aggravate their poverty level. Household annual income from the respective livelihood strategies positively and signi cantly in uences all livelihood diversi cation strategies. Analysis reveals a positive and signi cant relationship between membership of any organization (social capital) and livelihood strategies Y2 (crop plus livestock farming) and Y3 (crop plus sh farming), which is consistent with the ndings of Khatun and Roy (2012) and Khatiwada et al. (2017). Rural infrastructure, particularly household home to road, market, and haor distance were found signi cant determinants in uencing livelihood strategy choice. The results indicate an increase in the distance from home to the road increases the probability of adopting the crop plus livestock farming and crop plus poultry farming. It could be due to the availability of land and grassland which is favorable for crop production and livestock grazing as well as poultry farming (Rabby et al. 2013). Similarly, greater distance from home to haor enables haor household to adopt the higher retuning livelihood strategy, especially signi cantly in uence in adopting the crop plus off-farm activities. On the other hand, proximity to market place is favorable for crop plus livestock farming because households being closer to market place can easily sell their livestock products at a higher price (Khatiwada et al. 2017). Riverway and both-way (river and roadways) communications are signi cantly and positively related to livelihood strategy Y2 (crop plus livestock farming) since the river and both-way communication promote livestock rearing and grazing. The livelihood strategy Y3 (crop plus sh farming) is positively and signi cantly in uenced by riverway communication during the dry season, which may be due to their ability to connect with the marketplace for sh marketing with low transportation costs (Sunny et al. 2020). In line with our expectation, riverway communication is to be positively and signi cantly in uence the livelihood strategy Y4 (crop plus poultry farming). Riverway communication is one of the most in uential factors for poultry farming, particularly duck rearing, in the rural haor areas of Bangladesh (Islam et al. 2012).
Geographical and ecological characteristics such as waterlogged duration and agro-ecology have a positive and signi cant impact on the choice of a livelihood strategy (Rahman and Hickey 2020). Opposite of our expectation, the livelihood strategy Y2 (crop plus livestock farming) is positively related to the duration of waterlogging (months), possibly due to providing more effort during monsoon as it is the lean season for crop farming. As expected, the duration of waterlogging was found to be one of the important determinants of adopting a livelihood strategy Y4 (crop plus poultry farming), because duck rearing is required of persistent water for a long time (Islam et al. 2012;Khanum and Mahadi 2015). According to the study, highland households are more likely than lowland households to adopt strategic livelihood diversi cation in the wetland ecosystem (Islam et al. 2011).

Conclusion And Recommendations
The study looked at the various livelihood strategies practiced by haor households as well as the major determinants that in uence the choice of those livelihood diversi cation strategies in the study areas. Flash oods, oods, droughts, riverbank erosion, and embankment breach are the extreme natureinduced disasters that caused increasing harm to the poor in crop production and livelihood diversi cation was the indigenous strategy to adapt to climateinduced vulnerabilities. Crop farming has been identi ed as the primary economic activity and source of income for many households, even though it is unpro table. In addition to crop farming, which is commonly done by rural haor households, the study discovered four viable livelihood diversi cation alternatives. Rice is the primary crop for all types of livelihood strategies which contributes alone a range of 43-91% of the total household income of all livelihood strategy groups. The crop plus livestock strategy group (Y2) has gained the highest annual average income followed by crop plus sh (Y3), crop plus poultry (Y4), crop plus off-farm activities (Y5), and only crop farming (Y1). Therefore, crop production supplemented by livestock farming or sh farming or poultry farming (duck rearing), or off-farm activities could be an important means of subsistence. Multinomial logistics regression results showed that household's head age and education, dependency ratio, total land holdings, household assets value, access to credit, annual income, membership of any organization, home to road, market, and haor distance, river, and both way communication during the dry season, duration of waterlogged, and agro-ecology are the major determinants that signi cantly in uence the choice of higher returning livelihood strategies. Furthermore, marginal effect analysis revealed that household's head gender and education, family size, dependency ratio, skill development training, land holdings, household assets value, access to credit, annual income, membership of any organization, home to road, market, and haor distances, roadway communication during monsoon, river and both way communication during the dry season, duration of waterlogged, and medium high land type agroecology are the signi cant determinants of choosing better-paying livelihood strategies which measure the percentage change in probability of a particular choice being made to a unit change of a particular variable. Therefore, it is recommended that such signi cant determining factors of livelihood strategy choice can be promoted through policy interventions that have a signi cant impact on sustainable livelihood. In addition, the government should put in place effective policies to make livestock, sheries, and poultry easier for the haor community. Community-owned grazing land, such as community grasslands, is a better choice for safeguarding livestock farmers in the haor areas. Fishing is a seasonal business that can only be undertaken for a few months of the year, leaving people unemployed for the rest of the year and allowing them to diversify their activities. The haor basin, on the other side, has enormous potential for commercial duck farming. The availability of natural duck feeding sources and a tradition of duck farming in the haor region are the factors that may be prompted for developing duck farming in this area. However, these sustainable strategies require enormous capital investment, which is why most of them cannot participate in these livelihood diversi cation strategies. For ensuring participation in these livelihood opportunities governments, NGOs, and other development organizations should initiate different policies and programs such as credit facilities for assets accumulations, skill development training, and training for improved sheries, livestock and poultry farming, etc. Figure 1 a) indicates the location of the study areas (three districts) in the Bangladesh map and (b-d) indicates the selected six upazilas from three districts.

Figure 2
Conceptual sustainable livelihood framework (SLF) Figure 3 Source of income by different livelihood strategies group

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download. Appendix.docx