Vulnerability and Resilience to Drought and Saltwater Intrusion of Rice Farming Households in the Mekong Delta, Vietnam

Adverse impacts of climate and environmental hazards are unevenly distributed between socioeconomic groups due to differences in exposure, vulnerability and resilience. This study examines the distribution of vulnerability and resilience to drought and salt intrusion impacts among rice farmers in the rural Mekong Delta in Vietnam. By defining both aspects independently, we can study potential differences in the socio-economic factors that steer them and analyse how these two aspects of adaptive capacity are related. Using fixed-effect regressions, we find that poorer communes are more vulnerable to direct environmental impacts (loss in rice yield). Several household characteristics that reflect a low socioeconomic status, such as low asset values, small plots, and limited education level, are linked with higher vulnerability to direct drought impacts. High vulnerability does, however, not necessarily translate to low resilience, which we proxy by measuring indirect impacts (loss in household income). Several household activities and characteristics help mitigate indirect impacts. Our results suggest that the least resilient household group consists of smallholder, asset-low households that are unable to diversify to non-crop agriculture or off-farm employment. Supportive policies targeting this particular socioeconomic group to enable transition to non-crop or off-farm labour would substantially improve their resilience to future environmental events. Distinguishing between resilience and vulnerability enables a broader understanding of the mechanisms influencing the distribution of direct and indirect adverse impacts, which enables drafting targeted policy measures for specific socioeconomic groups.


Introduction
The adverse effects of climate change are felt across the world and will become even more prevalent in the future (Pörtner et al. 2022).Changes in climatic and environmental conditions, in combination with more frequent extreme events, are affecting human lives, livelihoods and health across the globe (Toya and Skidmore 2007;Noy 2009;Benson and Clay 2004;Botzen et al. 2019).However, across nations and within societies, the impacts of climatic hazards are not equally distributed due to differences in exposure, vulnerability and resilience.In general, developing countries and poorer societies experience stronger impacts than wealthier nations (Tol et al. 2004;Thomas et al. 2019;Füssel 2010).Withincountry impacts are felt more strongly by poor, rural households due to a higher dependency on natural resources and less access to credit and adaptation options (Hallegatte and Rozenberg 2017;Hsiang et al . 2020).
Many studies find that climate and environmental hazards widen inequality (Hsiang et al. 2020;Arouri et al 2015;Sedova et al. 2020).A 'vicious cycle' is found whereby initial inequality causes poor and marginalised groups to suffer disproportionally from adverse environmental events, resulting in even greater subsequent inequality (Islam and Winkel 2017;Cappelli et al. 2021;Hallegatte et al. 2020).This inequality-aggravating effect is manifested through three interrelated channels; higher exposure, higher vulnerability (or susceptibility) to losses and lower resilience, i.e. the ability to cope with and recover from the suffered losses.As a result, disadvantaged groups or societies are often worse off after a climatic hazard (Islam and Winkel 2017;Hallegatte et al. 2020).This heterogeneity in impacts is not only caused by differences in exposure to the hazard but also influenced by socioeconomic conditions that determine how entities are able to respond (Birkmann et al. 2013;Adger 2006;Cutter et al. 2012).The latter is referred to as adaptive capacities and has a large influence on impacts (Simelton et al. 2009;Adger 2006).
Although the notions of resilience and vulnerability both capture the socioeconomic mechanisms that cause heterogeneous impacts under similar exposure levels, their definitions and conceptualisation differ depending on the discipline and framework used.Within the disaster risk literature, vulnerability generally entails the narrow definition of 'susceptibility to harm', while the climate research community uses a broader definition of vulnerability, including exposure, sensitivity (similar to susceptibility) and response capacity (Adger 2006;Cardona et al. 2012).Resilience originates from the ecology domain and is considered a system's ability to absorb changes with minimum adverse effects and to return to its steady state or a 'better than' pre-event state (Holling 1973).In some frameworks, resilience is considered to be an aspect of vulnerability (Gallopín 2006;Pelling 2012;Birkmann et al. 2013), while others view vulnerability as the reverse of resilience (Cannon 2008).Some consider resilience to be a more long-term structural response capacity (Moret 2014), while others see vulnerability as connected to underlying social conditions (Rose 2009).As a result, operationalisation is often confined to either resilience (Arouri et al. 2015) or vulnerability (Cappelli et al. 2021).
A growing body of literature, however, is developing that conceptually disentangles vulnerability and resilience (Miller et al. 2010;Birkmann et al. 2013;Ran et al. 2020).Quantitative studies operationalising and connecting both notions are usually done inductively, using scoring and principal component techniques to cluster composite indicators representing vulnerability and resilience (Vázquez-González et al. 2021).Econometric studies that are outcome oriented, measuring vulnerability and resilience by the actual impacts, are still scarce (Noy and Yonson 2018).Some exceptions include the work of Akter and Mallick (2013) that distinguishes between immediate impacts related to vulnerability and longer-term impacts with regard to socioeconomic resilience (lower incomes, lower welfare, unemployment).This is in line with Rose (2009), who consider vulnerability as a pre-disaster condition and resilience as the outcome of a post-disaster response.He differentiates between immediate impacts on stocks, such as property loss, and more structural impacts on economic flows (production, consumption, incomes), which he relates to economic resilience.The economic resilience framework of Hallegatte (2014) builds on this principle, considering vulnerability as the direct asset (stock) loss incurred immediately after the event, while socioeconomic resilience is concerned with the more longer-term losses of flows, such as output, income and consumption losses.According to Hallegatte, socioeconomic resilience at the microlevel is influenced by the distribution of direct damages incurred across the affected households (vulnerability) and the household's ability to maintain consumption and incomes (resilience) despite these losses (Noy and Yonson 2018).
Studies operationalising vulnerability and resilience based on their impacts are primarily focused on fast-onset events such as tropical storms and floods, as they are characterised by immediate asset damages versus indirect output and income losses (Verschuur et al. 2020;Walsh and Hallegatte 2020;Hallegatte et al. 2016;Akter and Mallick 2013).To our knowledge, there are no examples of studies focusing on slow-onset events (e.g., drought and salt intrusion) that distinguish between vulnerability and resilience in a similar way.A better understanding of the factors that influence both vulnerability and resilience is particularly interesting for slow-onset events, as their gradual nature allows an individual or community to change or modify existing practices to either reduce the direct impacts on yield or indirect impacts on income while the event is unfolding (Cutter et al. 2008).This paper aims to address this gap by operationalising the notions of vulnerability and resilience simultaneously for a drought and saltwater intrusion event in the Mekong Delta in Vietnam.Vulnerability is linked to the immediate impacts of the drought event, expressed in the relative loss of agricultural (rice) yields.Resilience in turn is connected to the indirect impacts, translated in the loss of income for farming households in the event year.A key objective of our study is to understand to what extent vulnerability and resilience are interrelated by answering the following question: Are households vulnerable to the direct impacts of environmental events also less resilient?Another key objective is to understand which socioeconomic characteristics and activities contribute to vulnerability and resilience.
The Mekong Delta in Vietnam is chosen as a case study as it is considered one of the hotspots in terms of exposure to climate and environmental change impacts.In recent decades, the region has experienced both rapid and slow-onset environmental changes, such as sea level rise, floods, saltwater intrusion and droughts (Birkmann et al. 2012;Adger 1999).Despite rapid modernisation and urbanization in recent decades, the Mekong Delta remains a predominantly rural area, in which over 75% of people's livelihoods depend on the agriculture (rice, vegetables and fruits) and aquaculture sectors (Garschagen et al. 2012).The dominant agricultural sector is rice, which is particularly vulnerable to yield damage under adverse weather conditions.This study draws on a severe drought and saltwater intrusion event in the Mekong Delta, which took place in the dry season of 2015-2016, with upstream river flows of less than 65-70% compared to average.The associated unprecedented salt intrusion caused widespread crop losses and affected nine coastal provinces, while all thirteen provinces were impacted by water shortages.The total loss of agricultural production was estimated at US$ 300 million, with 250,000 hectares of rice damaged and more than one million inhabitants suffering from a lack of clean drinking water (Nguyen 2017).The availability of both rice yields and household-level panel data enables us to conduct the analysis.
This study contributes both to the empirical and methodological aspects of vulnerability and resilience in several ways.First, it provides empirics-based insights into the distributional impacts of a slow-onset environmental event related to drought.Second, we operationalise vulnerability and resilience based on direct and indirect impacts.Third, we assess which factors explain the degree of vulnerability and resilience of households and how both are connected.Finally, we consider different scale levels, exploring vulnerability at both regional (commune and district) and household scales.Findings from our study provide useful insights for policy makers, as understanding the distributional consequences across different groups within a society helps in designing more targeted policy options focusing on vulnerability, resilience or both.
This paper is organised as follows.The second section discusses the methodological framework, its key concepts and data sources as well as the econometric details.The third section outlines the main results, and the fourth section discusses the key findings and relevant policy implications.

Methodological Framework
Vulnerability and resilience are related concepts that both refer to adaptive capacity (Engle 2011;Nelson et al. 2007); the ability to mitigate adverse impacts.The socioeconomic characteristics of a household or individual lie at the heart of adaptive capacity (Adger and Kelly 1999;Akter and Mallick 2013), influencing vulnerability and resilience in a slightly different manner.When it refers to the immediate ability to cope, adaptive capacity is generally seen as an integral part of vulnerability (Gallopín 2006;Nelson et al. 2007), while if it constitutes relatively long-term and structural changes, the term is commonly linked to resilience (Nelson et al. 2007).In our study, we build on this distinction, which was also proposed by Hallegatte (2014), and relate vulnerability to the level of direct impacts, measured in our case by agricultural rice yield losses.Resilience is associated with the ability to limit indirect impacts and is measured by changes in household income in the event year.Figure 1 visualises the different concepts, and their operationalisation is further elaborated in the next sections.

Hazard and Exposure
In this study, the hazard of concern is the severe drought in the Mekong Delta that occurred in early 2016, resulting in low river flows and high salt levels and salt intrusion near the coast (Nguyen 2017;Eslami et al. 2019).We express the degree of exposure to this hazard as the potential crop loss due to local salt levels.This loss is based on the salt response curves of the FAO1 (Maas and Hoffman 1977), indicating the rice crop tolerance to salt:

Fig. 1
Framework of the different concepts and their operationalisation applied in this study, adapted from Hallegatte et al. (2020) Salt Exposure Loss (SEL) is calculated for a base year, representing a 'normal' year (2014) and the drought event year (2016) and averaged at the commune level (smallest administrative division in Vietnam, see Fig. 2b).Spatial data on salt levels for each year are obtained from a study by Eslami et al. (2019).Salinity, S , is measured as the amount of dis- solved salts in water (in gram per litre); a is the salt threshold value, which is the salt value for 100% yield potential; b is the relative yield loss per unit increase in salinity.The b value is determined as follows: where S 0 is the threshold of S whereby rice yield is no longer possible (yield potential is 0%) and S 1 is the value of S whereby maximum yield is still possible (yield potential is 100%).The use of the SEL is preferred over the use of absolute salt levels, as rice is sensitive to a specific value range of salinity.The equation assumes a linear impact on yield until a threshold(roughly 8-10 g/litre) after which the yield potential is zero.By comparing the potential impact of salinity on rice yields with actual rice yields, we assess the degree to which farmers are vulnerable to this hazard.Farmers that are better able to cope with or adapt to higher salinity levels will be able to obtain higher yields than their more vulnerable counterparts.
In more upstream areas in the Mekong Delta, which have a SEL of zero, many farmers and communes also suffered rice yield losses as a result of shortages in fresh water for irrigation (Nguyen 2017).We include these areas in our analysis as they also reflect drought impacts but unfortunately lack spatially explicit data on freshwater shortages to explain these occurrences.

Vulnerability
We assess the heterogeneity of vulnerability to direct environmental impacts at the district, commune and household levels.The direct impacts are operationalised by calculating the relative loss of rice productivity in the event year compared to our base year (2014).Productivity is expressed as the rice yield in tons per hectare.Only dry-season rice production is considered, which is the first-season "Dong Xuan" rice or the year-round floating rice "Lua Mua" rice, as droughts and salt intrusion primarily impact the dry season.The difference in rice yields between the years is expressed in a rice yield loss ratio (YLR), which is calculated using the following equation: where Y te is the dry-season rice yield in event year 2016 and Y tb is the rice yield in the base year 2014.The ratio accounts for regional differences in productivity and focuses on the change in productivity in the drought year.Figure 2 presents the average YLR at the district and commune levels, with a high ratio expressing a severe loss and a ratio of 0 expressing no impact on productivity.At the household level, the YLR is based on the total reported dry season rice production in tons and the number of hectares used for rice cultivation in 2014 and 2016.For the household-level analysis of rice yield losses, we

Fig. 2
Mekong Delta case study site and direct impacts measured by the rice yield loss ratio at the a) district and b) commune levels employ panel survey data from the Vietnam Household Living Standard Survey (VHLSS) in 2014 and 2016 (GSO, 2014(GSO, , 2016)).Our dataset only includes households that reported cultivating rice in both 2014 and 2016, which totals 851 households.As a result, the household-level data represent a small sample of dry-season rice farmers rather than the total Mekong Delta rice farmers and thus will not be fully representative.
Commune-and district-level data are obtained from the statistical yearbooks of the Department of Agriculture and Rural Development (DARD), which collects information on total rice cultivated area, hectare and production (ton) per season and the yield per season (ton/ha).Data was available at the district level for all rice-growing districts for 2014-2016, while at the lower aggregated level, we were able to collect commune-level data in seven rice-growing provinces.

Explanatory Variables
The difference between the potential direct impacts (expressed in SEL) and the actual direct impacts (measured by YLR) indicates how vulnerable regions or households are to climate hazards.We analyse how several socioeconomic factors, at both regional and household scales, influence the heterogeneity in vulnerability.At the commune and district levels, we use the poverty level as a socioeconomic indicator, which is measured as the fraction of households living below the National Poverty Line.Poverty is widely used as a socioeconomic indicator contributing to higher vulnerability (Hallegatte et al. 2020;Akter and Mallick 2013;Narloch and Bangalore 2018).Data on poverty is obtained from the 2016 Rural Agricultural and Fishery Census, a census of all rural communes in Vietnam that is conducted every five years by the National Statistical Survey Programme (GSO 2018).Table 1 presents the key statistics for districts and communes.
At the household level, we utilize several socioeconomic characteristics as sources of vulnerability to direct impacts, as these potentially influence the sensitivity to yield losses and the adaptive capacity to prevent losses.We explore the contribution of the education level of the household head, the farm size, the household asset value in the base year and 'adaptive strategies', such as the amount of fertiliser applied in the event year and the level of agricultural diversification (HDI)2 .We also explore the relation between indebtedness and access to a bank account with vulnerability to yield losses.The socioeconomic household data stems from the VHLSS, which is conducted biannually by the General Statistics Office of Vietnam (GSO) and contains household and commune data.The survey includes data on demography, education, labour, income, asset value, housing and health and covers approximately 9,000 households in the Mekong Delta.Detailed information on the type of employment, such as self-employment in agriculture, is available, including output and costs per agricultural crop.Approximately 40% of households participate in a sequential year.

Resilience
We define resilience as the ability of a household to minimise household income loss (indirect impacts) after suffering agricultural yield losses (direct impacts).As such, resilience is measured by changes in the natural log of per capita household income between 2014 and 2016, whereby households with no change or an increase are considered resilient, as they have managed to maintain or improve their annual income despite experiencing direct losses.To assess which household characteristics and adaptation strategies contribute to resilience, we consider the same socioeconomic variables as for vulnerability3 , which are complemented by several household activities that act as income diversification strategies.For the latter we include the share of household labour hours spent on other agricultural activities besides cropping, the share of household labour hours spent outside the agricultural sector and whether the household cropped a third season of rice.We also include the ratio of working household members.Resilience is assessed for vulnerable households only, i.e. households that experienced direct impacts (yield losses) as a result of drought and salt intrusion.We set a threshold of at least 3% yield loss in the event year (a rice yield loss ratio > 0.03), which includes 483 households, or 57% of our total sample.To test the robustness of this threshold we conduct a sensitivity analysis set at 1% yield loss (535 households).A description of the key variables is presented in Table 2.
To assess household income change, agricultural diversification and labour hours per sector, we make use of the income data of the VHLSS panel data (GSO).Individual-level data on income from self-employment, wage employment, business and other incomes are calculated per household and modified using the equivalence scale of the Organisation for Economic Co-operation and Development (OECD).This scale assigns a value of one to each adult household member and 0.3 to each child below 15 years.The 2016 income data is adjusted to 2014 prices to account for inflation.We use income level due to lack of consumption data and assume income and consumption are similar.This is a common assumption that is particularly valid for the lower income brackets (Hallegatte et al. 2016;Kind et al. 2020) that have our special attention.Note: *According to the official Vietnamese grading system: 1-5: primary education, 6-9: secondary education and 10-12: tertiary education

District and Commune Level
To analyse the relationship between exposure, vulnerability and direct impacts at the district and commune levels, we apply an OLS regression model with robust standard errors, employing the following equation: where YLR i,j denotes the rice yield loss ratio observed in commune i or district j; 0 is a constant;P i,j is the fraction of poor households in commune i or district j, SEL i,j,t+2 is the salt exposure loss in commune i or district j in event year t + 2 (2016), based on average salt levels in the district or commune and its potential impact on rice yields; SEL i,j,t represents the SEL of base year t (2014); and P i,j S i,j,t+2 is an interaction between P i,j and S i,j,t+2 .To account for unobserved variables causing spatial heterogeneity, we include provincial fixed effects p .The equation includes an interaction between the 2016 SEL and fraction of poor households to assess whether poorer communes/districts are more vulnerable to yield loss compared to wealthier communes when exposed to saltwater intrusion.

Household Level
We employ a second OLS regression4 to analyse at the household level which socioeconomic variables limit vulnerability to direct impacts, i.e. rice yield losses.
where YLR hi is the yield loss ratio of household h in commune i; SEL it+2 and S it are the commune-average SEL in 2016 and 2014, respectively; Y ht is the rice yield of household h in year t; E ht is the education level of the household head in year t;L ht is the natural log of the households' land size in year t;A ht is the natural log of the asset value of the household in year t;F ht+2 is the fertiliser use of the household in year t + 2;H ht+2 is the level of agricultural diversification (HDI) of household h in year t + 2; U ht+2 and B ht+2 represent binary variables indicating whether a household had a full unpaid loan or access to a bank account in year t + 2; D jt+2 is a vector of characteristics of districts, such as poverty level; and i is a provincial or district level fixed effect to account for unobserved spatial variety5 .

Determinants of Resilience
To assess the relationship between resilience, vulnerability and household-level socioeconomic characteristics and adaptation strategies (such as income diversification), we estimate the income change of households that suffered yield loss using the following equation: where ln(I h�i ) represents the change in the natural log of per capita household income; YLR h′i denotes the direct impact indicator (dry-season yield loss ratio) of household h' in commune i that experienced a direct agricultural loss of at least 3%; X h′t is a vector of household characteristics in year t (size farmland in 2014, education level household head, asset value); A h�t+2 represents household activities undertaken in the event year (t + 2), such as cultivation of a third rice crop, the share of non-crop agricultural labour hours, the share of non-agricultural labour hours, agricultural diversification, ratio of working household members and access to a bank account; and i is a district-level fixed effect to account for unobserved spatial variety.The equation includes several interactions between household adaptation strategies and variables.This allows the effect of household activities to vary across households with different socioeconomic characteristics and, for instance, enables testing the effect of increasing off-farm employment on resilience for households from different wealth levels.

Vulnerability to Direct Loss at the District and Commune Levels
The effects of exposure and poverty on vulnerability to direct impacts at the district and commune levels are presented in Table 3.At the district level, we find a positive relationship between poverty level and yield loss, but its significance disappears when introducing the salt exposure variable (column 1b), indicating that at the district level, there is no proof that poverty levels increase vulnerability to direct impacts.Salt exposure is a more important driver of yield loss (column 1b); overall, districts exposed to saltwater intrusion experienced higher yield losses, showing an average yield loss of approximately 21% with maximum exposure to salt.In addition, when interacting the poverty fraction with the salt index, we do not find any proof of poverty levels exacerbating direct impacts.The insignificant relationship of poverty and yield loss at the district level in models 1b-1c may, however, be explained by the relatively coarse district level of aggregation in this analysis that may obscure differences between smaller communes and individual households.
At the finer commune level, a highly significant relationship is found between direct impact and poverty (fraction of poor households in commune), which remains in place when salt exposure (SEL) is introduced (column 2b).Both poorer and more exposed communes show significantly higher rice yield loss ratios in the event year.As presented in model 2c, being exposed to high levels of salt in the base year decreases rice yield loss in the event year.This confirms the adaptive nature of these areas, where farmers may have started to use crop salt-tolerant varieties or adapted their cultivation season (Nhan et al. 2012).In model 2c, commune poverty level is interacted with salt exposure, revealing that rice yield losses are higher in poorer communes under similar salt exposure levels.Therefore, vulnerability to direct environmental impacts is significantly higher for poorer communes.This finding remained invisible at the coarser district level of aggregation.
The distribution of impacts cannot be attributed to the fact that poorer households are located within more exposed communes, as indicated by the relatively low correlation between exposure to salt and poverty level.At the district level, the Pearson's correlation coefficient is 0.334, while at the commune level, it is only 0.168 (both are significant at the P < 0.001 level).This is further visualized in Fig. 3, showing the rice yield loss ratio as a function of both exposure (on the X-axis) and poverty level for the different communes (represented in the colour of the points).With increasing SEL in 2016, the YLR rises.This impact is larger for poorer communes, as indicated by the steepness of the three lines representing the response per poverty category.The slope of the high poverty category (communes with more than 22.5% poor households) is steepest, followed by the middle poverty category (between 7.5 and 22.5% poor households) and the low category (fewer than 7.5% poor households).This suggests that drought and saltwater intrusion can aggravate inequality, as the negative effects of exposure to salt are much stronger in poorer communes than in communes with a low percentage of poor households.

Fig. 3
Rice yield loss as a function of salt exposure loss in 2016 for three commune-level poverty categories (high poverty = > 22.5% poor households; medium poverty = 7.5-22.5% poor households; low poverty = < 7.5% poor households)

Household Vulnerability
The analysis of vulnerability to rice yield loss at the household level is displayed in Table 4. Robust standard errors are used in all models, and we control for unobserved variation by including regional fixed effects.The first model takes another look at the impact of poverty at the district level.As with the commune-level analysis (Vulnerability to Direct Loss at the District and Commune Levels section), we find that households living in poorer districts are confronted with higher yield losses (51% of additional yield loss if the fraction of poor households is maximal at 1).The interaction shows that this effect is even stronger in districts with high salt exposure, indicating that households living in wealthier districts have a better capacity to adapt to salt intrusion.We also find higher yield losses for households living in urban districts or districts with special economic zones, which might be explained by the fact that there are more off-farm opportunities and skills available, making households in these districts less concerned about yield loss. Models 1b and 1c show that maximum salt exposure in 2016 results in an additional yield loss of approximately 15%.SEL 2014 is not included as a variable, as the variable 'yield 2014' already reflects saline conditions in 2014.The yield of 2014 is included as a control factor to account for variations in yield prior to the event.Both models indicate that higher vulnerability to yield loss is significantly linked to lower education levels, a low asset base and a smaller farm size.Higher fertiliser use in the event year, on the other hand, has a significant relationship with lower yield loss, confirming a higher vulnerability for households with less access to technology and inputs.Having a full unpaid loan in the event year is also significantly correlated with yield loss, suggesting that indebted households are more vulnerable to drought and salt intrusion, possibly because they have fewer resources to cope and respond.The interaction between farm size and loans (column 1c) indicates slightly higher losses for indebted smallholders compared to large-scale farmers with a loan, which suggests that indebted smallholder farmers are most vulnerable to adverse direct impacts of drought and salt intrusion.These findings confirm our hypothesis that certain socioeconomic household conditions (smaller farms, lower education levels and a low asset base) contribute to higher vulnerability and exacerbate the impact of salt exposure.

Household Resilience
In our final set of regression results, we look at resilience, which we define as the ability to limit income loss.More specifically, we analyse the degree to which direct impacts (as measured in YLR) translate into a decline in (the natural log of) per capita income.In this analysis, we look for the household characteristics and adaptation responses that mitigate this impact and thus contribute to resilience.The results of the analysis are presented in Table 5.As we are especially interested in the relationship between vulnerability and resilience, we only look at households that suffered direct environmental impacts (YLR > 3%) and thus were vulnerable to drought and salt intrusion.All models include the direct environmental impact variable (YLR) and household-level socioeconomic characteristics as explanatory variables.Models 1b-1d contain separate interaction variables.The log of per capita income in the base year is added as a control factor to account for initial variation in income.To account for omitted regional effects, district-level fixed effects are added in all models.
None of the model results suggest a significant relationship between the level of direct impacts and resilience.This indicates that the severity of yield loss incurred in the dry season does not influence annual household income changes.A sensitivity analysis including all households confirms this finding, showing that once loss is incurred, the severity of loss does not affect resilience6 .Instead, key socioeconomic variables and household adaptation responses contribute to the heterogeneity of resilience.Positive contributors include a third rice cropping season in the event year (which likely enables the farmer to overcome yield losses in the first season), a large farm size, high asset value (in base year), and a high ratio of working household members in the event year.Household responses significantly increasing resilience include a larger share of labour hours spent on non-crop agriculture or on non-agricultural activities in the event year (positively affecting log household income change by 0.4 and 0.7, respectively).A significant negative effect on resilience is found for the interactions of both farm size and asset value with the share of non-agricultural labour hours (columns 1b and 1c), suggesting that spending more labour hours in non-agricultural activities has a higher positive effect on resilience for smallholder and low-asset households than for wealthier and largescale farm households.Finally, having access to a bank account has a stronger positive relationship with resilience for wealthier households, compared to low-asset households, which is likely related to a higher accessibility of wealthier households to formal bank accounts.A sensitivity analysis with a different yield loss threshold (YLR>1%) showed similar outcomes.

Discussion and Conclusion
The drought and saltwater intrusion in the Mekong delta in winter-spring of 2015-2016 imposed significant agricultural and socioeconomic damage (CGIAR 2016;Nguyen 2017;Nguyen et al. 2019a), and saltwater intrusion is projected to increase due to climate change and upstream dam developments (Eslami et al. 2021).The key objective of this paper was to enhance our understanding of the distributional implications of drought and saltwater intrusion in the Mekong Delta by disentangling and operationalizing the concepts of vulnerability and resilience.Connecting vulnerability to direct impacts (agricultural rice yield loss) and resilience to the indirect impact on overall income reveals several interesting results.We found evidence that agricultural losses due to drought and salt intrusion are unequally distributed among different socioeconomic groups and communes, with poorer communes and asset-low, low-skilled and smallholder households bearing the brunt of adverse impacts under similar exposure levels.We also find a clear correlation between yield loss and having a full unpaid loan.Our results suggest that drought and salt intrusion hazards have an inequality aggravating effect, which is consistent with findings of previous studies, of which most focus primarily on fast-onset hazards (Sedova et al. 2020;Bui et al. 2014;Akter and Mallick 2013;Salvucci and Santos 2020).
Although saltwater intrusion and drought unequivocally led to agricultural yield loss, this does not necessarily translate into lower resilience in terms of household income loss.Our findings suggest that the adaptive and response capacity of households is strong, as most can absorb and mitigate agricultural losses through household strategies, most notably non-crop agriculture and non-agricultural labour activities.In particular, the asset-low, smallholder household benefits from labour activities outside the rice and agricultural sector.Our results confirm the narrative that households are not passive victims of environmental and climatic changes but are able to generate income from additional sources and activities, in line with findings from existing studies (Poelma et al. 2021;Akter and Mallick 2013;Tran et al. 2019;Le and Ngoc 2020).At the same time, our study reveals that several socioeconomic household characteristics are linked to income loss, indicating that not all rice households have the capacity to adapt, which confirms earlier work (Hoan et al. 2019;Tran et al. 2021;Nguyen et al. 2019a;Tran et al. 2020).Based on our findings, the least resilient household is the smallholder, asset-low household that is unable to diversify to non-crop agriculture or off-farm employment.Interestingly, agricultural diversification does not impact vulnerability to yield loss or resilience, while access to a bank account is positively related to resilience only for wealthier households, which is likely related to the lower accessibility for poorer farmers.Enhancing access to microcredits and formal bank accounts may strengthen the resilience of households (Arouri et al. 2015).
Connecting and linking both direct 'asset' (yield) loss and indirect 'welfare' (income) impacts through the concepts of vulnerability and resilience enhances the assessment of effective strategies from the perspective of different socioeconomic groups (Verschuur et al. 2020).Structural 'hard' measures preventing salt intrusion focus on reducing the exposure and direct impacts on yield loss, while not necessarily enhancing the adaptive capacity that increases the socioeconomic resilience of vulnerable households.An example is the Bai Lai salt prevention project, which converted the brackish ecosystem in Ben Tre province into a freshwater system.While it has enabled large-scale rice farmers to enhance their rice productivity, the reduced waterflows have significantly increased pollution and riverbank erosion, increasing risks and input costs for smallholder farmers.(Hoang et al. 2009;Ngo et al. 2017).Increasing the socioeconomic resilience of households can be done at both the macro and micro levels, such as capacity building to switch from rice agriculture, increasing local non-agricultural employment opportunities or supportive strategies to guide the spontaneous migration of rural poor from exposed areas (Noy 2018).Smajgl et al. (2015) argue that both structural and soft measures are needed to accomplish the most effective results for people's livelihoods in the Mekong Delta.
Currently, the Vietnamese government is actively pursuing agricultural modernisation and restructuring under Resolution 120 (Prime Minister 2017).This entails shifting away from a 'rice-first'-based orientation towards sustainable and climate-resilient delta development to better cope with adverse climatic and environmental changes such as droughts, floods and saltwater intrusion (Nguyen et al. 2020).This requires widespread adoption of adaptive systems for smallholder farmers, such as shifting to salt-tolerant crops and/or sustainable brackish aquaculture practices.This calls for improved skills and capital resources, however, which many of these farmers lack (Nguyen et al. 2019a, b).Second, income generation outside the agricultural sector is the most effective resilient strategy for smallholder and asset-low households, as our results show, but non-farm employment opportunities in the Mekong Delta remain scarce.As a result, the out-migration of the agricultural workforce is increasing rapidly, especially in areas where environmental conditions are worsening (Tran 2019).This may not be desirable, as migration of a large proportion of the rural workforce puts severe strains on both receiving urban areas and sending rural communes (Entzinger and Scholten 2022;Ngo et al. 2022).Therefore, local off-farm employment is essential for a successful transformation and required to increase the resilience of the most vulnerable groups.This also entails enhancing business and entrepreneur conditions, better access to formal credits (with low interest rates) and training the right skills (Arouri et al. 2015).
Our study has some limitations, primarily related to the fact that resilience is measured by income changes over a short period of time, comparing the situation prior to the event and during the event year.This does not allow us to assess the medium to long-term indirect impacts related to recovery.Nevertheless, our study shows that during the immediate aftermath of a drought event with saltwater intrusion, the resilience of directly impacted households differs markedly, which can be attributed to their socioeconomic characteristics.It can be expected that income effects are further smoothened over a longer period as households have more time to adapt and recover.
As a follow-up to this research, it would be interesting to explore the effect of environmental changes on specific household adaptation strategies and responses, such as leaving the agricultural sector and labour migration.Furthermore, exploring the heterogeneous impacts and responses of sudden shocks versus more gradual environmental changes may shed light on the likeliness and effectiveness of different household adaptive strategies related to the pace of environmental events.

Table 1
Summary statistics of the variables used for vulnerability analysis at the district and commune levels

Table 2
Summary statistics of household and district characteristics to examine vulnerability and resilience

Table 3
Explanatory analysis results of the rice yield loss ratio (YLR) at the district and commune levels Notes: Robust standard errors in parentheses.Significance coding: *p < 0.1.**p < 0.05.***p < 0.01 1 At the coarser district level, 2014 SEL cannot be included because of its high correlation with SEL 2016 (Pearson's correlation coefficient = 0.864, significant at P > 0.01)

Table 4
Explanatory analysis results of the dry-season rice yield loss ratio at the household level

Table 5
Explanatory analysis results of resilience (measured as change in the natural log of income per capita) for vulnerable households (that have a dry-season YLR > 3%)