Determinants of Rural Households’ Intensity of Flood Adaptation in the Fogera Rice Plain, Ethiopia: Evidence from Generalised Poisson Regression

Abstract

Effective adaptation to flooding risk depends on careful identification and combinations of strategies which, in turn, depends on knowledge of the determinants of flood adaptation. The main objective of this study was to examine the determinants of rural households’ intensity of flood adaptation in the Fogera rice plain, Ethiopia. A three-stage stratified sampling technique was employed to select 337 sample household heads. Primary data was collected through a structured household survey. Data analysis was accompanied by a descriptive and generalised Poisson regression (GP) model. The descriptive analysis showed that households adopted an average of three (3) flood adaptation strategies. The generalised Poisson regression further revealed that family size, availability of off-farm income, previous flood experience, access to credit, access to extension services, and an early warning information system statistically significantly increase flood adaptation strategies’ average number (intensity). However, the age of the household head negatively and significantly influences the intensity of flood adaptation. More specifically, households with off-farm income, previous flood experience, access to credit, access to extension, and an early warning information system were 20%, 94%, 13%, 30%, and 29% more likely to adopt more flood adaptation strategies, respectively. The findings call for immediate response and coordination among stakeholders to design strategies that enhance households’ livelihood, access to credit, access to extension services, and early warning information systems for effective flood adaptation in the study area.

1. Introduction

Evidence-based research shows that the adverse impacts of floods have a far-reaching effect globally [1]. This explains why there is a fast-growing interest in understanding how communities and sectors are adapting to these changes. Flood events expose rural communities to a wide array of risks as the vast majority heavily rely on rain-fed agriculture. Floods negatively affect the agricultural sector through reduced yields, increased costs of production and high investment costs. Higher operational costs coupled with weak infrastructure weaken the rural household’s ability to cope with these risks subjecting them to flood-induced poverty [2,3]. This makes it paramount for policymakers, development agents, leaders, and the entire society to have the know-how, the processes, and the ways in which to accelerate adaptation. Achieving increased adaptation capacity for rural communities against flood risks is a function of understanding the behaviour of society. Successful implementation of perceived measures is highly determined by how households perceive and embrace them. Research shows that households are more likely to adapt by adopting on-farm and nonfarm strategies [2,3,4] to overcome the impacts of climate change while reducing the associated risks.

The impacts of climate change are being experienced at an unprecedented rate globally with the World Meteorological Centre projecting that global temperatures and carbon accumulation are still on the rise [5,6]. This explains why floods and droughts in the first half of 2023 have been prominent in different parts of the world. Ethiopia has neither been spared nor is it excepted from this phenomenon and extreme temperatures, prolonged droughts, and frequent unpredicted floods have been prominent across this region over the past decades.

Natural disasters are greatly impacting livelihoods with devastating impacts. Floods have been one of the natural disaster threats of recent times accounting for more than 40% globally and presenting itself as a threat to livelihoods and properties. These hazards have been exacerbated by fast-rising cases of land-use changes, anthropogenic changes, poor urban planning, poor water-management systems, and destruction of natural resources which have greatly impacted groundwater recharge coupled with increased rainfall downpours due to climate change. Other low-lying areas close to the seaside have been greatly impacted by short water cycles which have increased the occurrence of cyclones and typhoons coupled with continuous ocean rise as a result of melting ice caps due to rising global temperatures [7,8,9,10].

These effects have been associated with risks and hazards that have significantly impacted Ethiopia’s path towards achieving its sustainable development goals (SDGs) [11]. Extreme flooding has severe and far-reaching economic, social, and environmental impacts [12,13]. When floods occur, communities are displaced with goods and properties of greater value destroyed [14]. This directly or indirectly influences the livelihoods of the affected communities. For agricultural communities, crop and infrastructural destruction are some of the main impediments to sustainable community development [15]. Such effects and consequences of floods have been detrimental, especially to developing economies, which have reduced their ability to cope [13]. The increasing and expansive impact of floods has drawn the attention of both scholars and stakeholders on how to reduce their vulnerabilities to the impacts by either avoiding or adopting countermeasures. Avoidance, in most instances, is not optional for farmers as they are unable to relocate other than accepting the new reality. This situation necessitates communities to adapt to the prevailing conditions, creating resilience.

This, then, raises the question of how flood adaptation occurs. In 2001, the International Panel on Climate Change (IPCC) reported that communities in flood-prone areas can adapt by implementing actions which strengthen the existing socioeconomic or ecological systems to withstand the adverse effects of the hazard [16,17]. However, the success of such measures is not guaranteed but it is only achievable if the desired interventions aim at addressing the precursors [18]. This explains the importance of systemic detection and identification of the precursors to ensure they are first addressed to establish transformative adaptation. Failure to adapt is more likely to cause displacement and migration but, with high population density across Africa, migration causes community conflict; thus, the only way is to promote adaptation.

Flooding has been characterized as the leading global hazard and the most common type of natural hazard which has been associated with more than 40% of all deaths due to disasters [19]. It is a function of society to adapt to the vulnerability to these hazards. Adaptation capacity is positively correlated with the development level of the affected area. This, to a greater extent, explains why less developed countries, especially in Africa, are characterized by less adaptation capacity to floods. Additionally, sustainable growth and development of these regions could further be hindered by these natural hazards as they double the burden of underdevelopment. The impediment towards sustainable development in most of the African states is a result of a lack of disaster preparedness and adaptation strategies [20]. These predisposing factors further exacerbate the vulnerability of the communities to climate hazards. If not well addressed, it is more likely to cause hazard-induced poverty, which is a great barrier to sustainable development. Rural areas that heavily rely on agriculture for livelihood have been placed on the verge of food insecurity due to these seasonal changes [21]. However, all these claims on poverty and food production decline have been based on literature and there has been no data on the extent to which floods interfere with the production capacity of any region [22,23,24].

Adapting and coping with such hazards is a function of understanding the extent and dimensions of the impacts on sustainability. Loss of vegetation cover, crops, and other infrastructure can be considered economic hazards while the impact on people’s lives, and the health risks due to flood-related diseases and outbreaks, are social hazards [25]. From an environmental point of view, floods deplete the soil’s key essential nutrients and destroy the lives of essential microorganisms [26]. Additionally, the emission of greenhouse gases which contribute to air pollution is an environmental hazard due to floods. To sustainably create and establish resilient, adapted communities, understanding how these hazards can influence rural communities’ ability to cope with floods is fundamental [27]. With such knowledge at hand, it creates opportunities for investors, community leaders, and different actors to come up with development policies that have contingency plans for flood hazards, thus establishing a way for communities to sustainably adapt to floods [28,29,30].

Over the past decade, Ethiopia has been one of the African countries that extreme floods have negatively impacted. Ethiopia’s dominant flooding causes are river overflow and heavy rains [13,31]. For instance, [32] reported that floods caused an estimated economic loss of USD 6249 in households and USD 5329 in farmland activities in 2016 along the Abela–Abaya floodplain. An estimated 210,600 people were affected only between November 2015 to January 2016 in Ethiopia [33]. Furthermore, the flood hazard that occurred in 2006 affected 107,286 people, where 37,982 were displaced, and damaged 18,000 ha of crops in six zones of the Amhara region where the study area is located [33]. A damaging flood incident also happened in Dire Dawa in 2006 that displaced around 117,000 people, 256 died, 244 were reported as missing, and resulted in an estimated economic loss of USD 10 Million [34]. Furthermore, floods in Ethiopia are also responsible for affecting the education of children by reducing the number of completed grades [35]. It is one of the major causes of the internal displacement of Ethiopians next to internal communal conflicts [36]. To overcome the negative effects of climate-induced natural hazards, such as floods, the Ethiopian government initiated a climate-resilient green-economy strategy in 2010 [37]. Along with the strategy, the government has been facilitating the adoption and implementation of various flood mitigation, as well as adaptation, strategies including water diversion, construction of dykes, and temporary and permanent relocation away from flood-affected areas.

The Fogera rice plain, a critical economic area, has been one of Ethiopia’s most flood-prone areas along the Ribb–Gumara watersheds [11]. It has been regularly inundated by floods due to heavy rains during the summer rainy season [38].

Research shows that flooding triggered by climate change has remained a global challenge and its impact will continue in the future [39,40]. Such a borderless hazard calls for a multi-actor approach and a transglobal policy decision to address these climate-induced hazards in unison. However, when developing a countermeasure to establish resilient communities, household-level flood adaptation strategies should be taken as priorities from a performance and cost perspective. However, based on empirical evidence, most flood damage in Ethiopia occurred due to limited preparation or lack of adaptation from households and the government. For instance, [41] reported that poor urban drainage systems, land-use planning, lack of early warning systems, and unorganised flood risk mitigation measures at national and local levels aggravate flood damage. The continued hazards of floods are attributed to the failure of current coping mechanisms in developing countries [13].

Several factors are responsible for constraining effective adaptation to flooding by households. These constraints may be attributed to governance and institutional as well as household-specific limitations. Nigusse and Adhanom [42] mentioned that institutional attributes fundamentally determine the success of organizations, agencies, and individuals in their adaptation efforts to flooding by weakening their coordination and understanding. Since flood risk management is a complex process, it requires a well-functioning interconnection among human factors, infrastructures, and economic systems to reduce barriers to flood adaptation [43]. Although households may understand the damaging effect of flooding and the way to adapt to it, various socioeconomic constraints inhibit them from doing so. Alternatively, households limit themselves by taking short-term mitigation measures mainly due to financial, knowledge, and place constraints, as well as a lack of government support [44].

Carefully identifying effective coping mechanisms is inevitable to reduce flood risk [45]. For instance, [46] found that creating flood detention areas was the most economically attractive option to reduce risk. Implementing combinations of flood adaptation strategies, such as private and public, structural and nonstructural, and small scale and large scale, is a promising approach to increase preparedness against floods [47]. Furthermore, nonstructural methods, such as land use regulations and private precautionary techniques, effectively reduce flood risk [48]. Combining flood protection infrastructures, early warning systems, and nature-based solutions, and introducing risk-financing schemes should be the main components of adaptation strategies [49].

Many studies have tried to explore the determinants of a flood adaptation mechanism. For instance, [50] found that age, education, income level, gender, family size, and location determine households’ decision to adapt to flooding. In addition, age, income, marital status, and land size determine the choice of adaptation strategies [29]. Flood adaptation decision is also influenced by landholding size, age of household head, farm income, family size, education [51], and early warning information [41].

Some studies examined the determinants of short-term and long-term adaptation strategies separately [7] and ex ante and ex post strategies [51]. However, there are just a few studies that have examined the determinants of the intensity of adaptation by households. For instance, [52] studied the determinants of adaptation diversity (intensity). This study found that households used diversified adaptation strategies during the ex post period and farmers with previous cyclone experience used more adaptation strategies. Faruk and Maharjan [53] also showed that farmers with higher self-efficacy and response efficacy adopted more adaptation actions. Previous studies treated flood adaptation decisions as a binary variable and estimated their determinants using logit or probit models.

Most existing studies have addressed flood adaptation with a focus on the global level. Although it is of great importance to understand the holistic nature of a hazard, adaptation measures are effective when they stem from local initiatives and are locally led and driven [29]. Additionally, most of these studies address coastal floods leaving out a gap in riverine flooding. The few existing studies have not exhaustively addressed the determinants of rural households. These limited studies have applied parametric estimation and the assumptions upon which they are made may not be true. However, handling such a variable with binary regression models treating it as a dummy (0 = no adaptation or 1 = if adopted at least one), irrespective of the number of positive counts (more than one), may be misleading [54]. Moreover, the distribution of such count variables is either underdispersed or overdispersed and it should be captured by count data models (such as Poisson and its variants) [55] as data properties directly determine model choices [56].

Identifying the determinants of flood adaptation is indispensable for an effective capacity to respond to flood risks for riverine households [7]. At this backdrop, this article provides more insights into previous studies which have addressed the determinants of households and community adaptation across riverine communities [7,57,58] in developing countries by addressing two main questions: (a) what are the main existing adaptation measures against rural households and (b) what are the factors influencing the household’s intensity of adaptation to floods. In answering these questions, data drawn from 337 households across the Fogera rice plain, one of Ethiopia’s most frequently flood-affected areas, are analysed as explained below.

2. Data and Methods

2.1. Description of Study Area

The study was conducted in the Fogera district of the South Gondar Zone, mainly known as Amhara’s paddy field (rice plain). It is located between 11°43′ N and 11°53′ N and 37°35′ E–37°58′ E. The total area of the woreda (district) is about 1111 square kilometres, with a crude population density of 206 persons per square kilometre. The Woreda is dominated by flat land (76%), while the mountain slopes and rugged terrain account for 24%, making it flood-prone. Its altitude ranges from 1774 to 2415 m above sea level [59]. The land use pattern of the Wored is characterised by 44% cultivated land, 24% pasture land, 20% water body, and 12% others [60]. The woreda receives a total annual rainfall that ranges from 1100 to 1530 mm/year. The mean annual rainfall is 1216 mm [61] and mean monthly values vary between 0.6 mm (January) and 415.8 mm (July). The mean monthly temperature of the woreda is about 19 °C while the range runs about 10.7 °C–27.3 °C [62]. Fluvisols and vertisols are the dominant soil types of the woreda [63]. Moreover, the percentage composition of soil includes 65% black (vertisols), 20% brown, 12% red, and 3% gray soils [64]. The woreda is one of the sufficient food-producing areas of the Amhara region. It is characterized by mixed farming. Specifically, teff, corn, sorghum, sesame, and vegetables are produced. The woreda is also best known for its indigenous cattle breeding [59]. It was one of the few parts of Ethiopia where rice production was first introduced as government intervention to the food insecurity status of the woreda in the 1970s [65]. Currently, it has transformed itself from a food deficit to a food surplus woreda [66].

2.2. Sampling Technique and Sample Size

The study employed a three-stage stratified sampling technique. In the first stage, the Fogera rice plain was selected purposely, as the region is one of the most affected rural communities in Ethiopia by floods and its critical role in rice production. In the second stage, three (3) counties (kebeles) (namely Kokit, Wagetera, and Nabega) were selected randomly from 33 rural kebeles of the woreda. Finally, in the third stage, 337 sample household heads were selected randomly from 2130 households using Yemane Taro’s 1967 simplified sample size determination formula [67] as follows:

$$n=\raisebox{1ex}{$N$}\!\left/ \!\raisebox{-1ex}{$1+N{(e)}^2=\raisebox{1ex}{$2130$}\!\left/ \!\raisebox{-1ex}{$1+2130{(0.05)}^2\cong 337$}\right.$}\right.$$

2.3. Data Type and Method of Collection

The nature of the data used in this study was cross-section data. It was collected through a structured household survey from April to May 2022. Data on the main identified adaptation strategies (moving to high-elevation places, migration, dike, selling cattle, planting trees, using flood-tolerant rice varieties, drains, terrace, and emergency provision) and socioeconomic and institutional attributes of household heads were collected. Once the questionnaire was prepared in the English language, it was translated into the local language (Amharic) to enable further elicitation of better data from the households. Enumerators were assigned to interview the respondents and save the data at the same time.

2.4. Methods of Data Analysis

2.4.1. Descriptive Analysis

Descriptive analysis was used to describe the state of the intensity of adaptation by households and the average number of adaptation actions used. Furthermore, the percentage of households who adopted several adaptation actions was presented.

2.4.2. Empirical Model: Generalised Poisson Model

In the previous studies, determinants of households’ willingness to flood adaptation were captured by using limited dependent variable models such as the binary logit model treating households’ willingness as a binary variable where this variable takes just two values (1 if households reported that they adopted at least one flood adaptation strategy or 0 if no adaptation strategy at all) [7,50]. However, this method ignores the effect of the difference in the number of flood adaptation strategies adopted by adopters. Some households adopt a few strategies while others adopt several flood adaptation strategies. For instance, in our data set, 95 households did not adopt at all, 28 households adopted 3 strategies, 60 households adopted 4 strategies, 72 households adopted 5 strategies, 6 households adopted 8 strategies, and only 4 households adopted 9 strategies. The binary models assume this significant difference in the number of flood adaptation strategies used by households simply as adopters. Thus, to account for the effect of such differences in the number of adaptation strategies used by households, it is imperative to look for alternative empirical methods to examine what factors drive households to use additional adaptation strategies [52]. One way to overcome this drawback is treating some adaptation strategies as count data variables which, in this study, is used as a proxy for the intensity (diversity) of flood adaptation. One of the methods that addresses this limitation of limited dependent variable models is the count data models.

The Poisson model is one of the popular count data models [68,69]. The Poisson model is well explained in [69,70,71,72,73]. It has been used to capture relationships between a count-dependent variable and predictor variable(s), for instance, to model the number of children, scores of soccer games, career interruptions, and the number of vacations [73]. It also models the efficiency of researchers using the number of publications and citations [71], the number of typing errors per page, the number of telephone calls per hour, the number of faults in rolls of fabric, the number of automobile accidents, and home injuries [74]. The Poisson model is clever when mean and variance are equal or close to each other [69,75,76,77]. The Poisson model is good at generating robust maximum likelihood estimates that handle data distribution misspecification problems and its strength can be compared to least squares models for continuous data [73]. The probability mass function of the Poisson model is expressed as:

$$P\left(Y=y|\mu \right)=\frac{{\mu }^y}{y!}{e}^{-\mu }$$

(1)

The distribution measures the probability of observing y values of the random variable Y with a parameter value of µ which measures the mean number of occurrences of an event and it, at the same time, represents the variance of the distribution of the occurrence of the event in a given time interval. In the Poisson model, the mean is equal to variance (equi-dispersion) [55,78] and y! is read as y factorial which is further represented by:

$$y\left(y-1\right)\left(y-2\right)\left(y-3\right)\dots \dots \left(2\right)\left(1\right)$$

(2)

However, when the data under study involves differences in mean and variance, when variance is significantly greater than the mean (overdispersion) or variance is less than the mean (underdispersion), the Poisson model generates biased standard errors and is no longer applicable [69]. The Poisson model is inappropriate to model count data that exhibit under- or overdispersion [72]. Furthermore, the cost of overlooking overdispersion (underdispersion) in a data set is the generation of liberal or conservative standard error estimates [71]. Employing the Poisson model despite the under or overdispersion of a data set also produces biased parameter estimates that mislead conclusions and statistical inferences [76,79].

Therefore, it is apparent that there is a need to employ an alternative model that overcomes the limitations of the traditional Poisson model. The model obtained from the Poisson family but clever enough to handle under- or overdispersion of data is the generalised Poisson (GP) model [80]. It simply extends the usual Poisson model [81,82]. The GP is even better than the negative binomial model, another candidate for overdispersed data, when data contains dominant proportions of zeros [83]. In addition, GP also results in lower bias and is a better fit for overdispersed data caused by excess zeros as compared to the mixture Poisson model, zero-inflated Poisson, and negative binomial model [81]. On the other hand, [84] found that the negative binomial model was a better fit than the GP model while modelling cervical cancer number cases. In car-accident data analysis, the GP is as good as or better than other regression models [85]. The negative binomial model is suitable for underdispersed data [86].

Since its establishment, the GP has been frequently applied to handle both underdispersed and overdispersed data sets. For instance, to model the adoption intensity of improved soybean production technologies in Ghana [87]; to map quantitative trait loci [79], the number of dengue haemorrhagic fever sufferers [81]; to analyse the determinants of strikes between 1984 and 2017 in Turkey [88]; to identify the determinants of a number of car accidents in Alabama [85]; and to examine the predictors of the number of under-five malnourished children in Bangladesh [89].

In this study, the GP model is used to model the determinants of flood adaptation intensity where the number of adaptation strategies used by individual households are treated as count data ranging from 0 to 9. Furthermore, since the data showed that there is overdispersion within it (variance = 6.25 > mean = 3.4), the GP was employed for its flexibility to capture overdispersion as well as underdispersion [69,81,84,88,89]. Furthermore, the relative variance, which is calculated as the ratio of variance to mean, is greater than one (6.25/3.4 = 1.82), implying the presence of overdispersion in the data [82,90].

Following [79,81,87,88,91] and assuming the intensity of flood adaptation (the dependent variable, Y_i) follows overdispersed generalised Poisson distribution, the probability mass function can be specified as:

$$\mathrm{Pr}\left(Y_i=y_i\right)=f\left(y_i, {\mu }_i, \theta \right)={(\frac{{\mu }_i}{1+\theta {\mu }_i})}^{y_i}\frac{{(1+\theta {\mu }_i)}^{y_i-1}}{y_i!}\exp \left(\frac{-{\mu }_i\left(1+\theta y_i\right)}{1+\theta {\mu }_i}\right),y_i=0,1,2,\dots 9.$$

(3)

where ${\mu }_i$ is the mean of the function and expressed as E(Y_i|$x_i)=$ ${\mu }_i=f\left(x_i\right)=\exp \left(x_i^{\prime }\beta \right)$, where $x_i$ is a vector of independent variables, $\beta$ is a vector of coefficients to be estimated, and the variance of

$$Y_i \mathrm{is} \mathrm{E}(Y_i|x_i)={\mu }_i{\left(1+\theta {\mu }_i\right)}^2$$

(4)

When θ > 0, the data is said to be overdispersed or when θ < 0, the data is said to be underdispersed, and if θ = 0, no dispersion (equi-dispersed) [87,91]. The log-likelihood value can be taken as a measure of the goodness of fit of a model (a larger value means better fit to data); model adequacy can be measured through the Wald “t” as the ratio of θ to its standard error and should be greater than 0 confirming the appropriateness of GP). Similarly, AIC and BIC can also serve as referencing criteria to compare among models [87].

2.5. Description of Variables

The variables used in the study were defined and operationalised as follows (

Table 1. Description of variables.

Variables	Definition	Expected Sign
Dependent variable
Intensity of flood adaptation	Number of adaptation actions (count data)
Independent variables
Age	Number of years a household head has lived	-
Gender	Dummy (1 = Male, 0 = Female)	+
Education Status	Dummy (1 = Literate, 0 = Illiterate)	+
Marital Status	Dummy (1 = Married, 0 = Otherwise)	+
Family size	Number of people in a household	+
Off Farm Income	Dummy (1 = Yes, 0 = No)	+
Previous flood experience	Dummy (1 = Yes, 0 = No)	+
Access to credit	Dummy (1 = Yes, 0 = No)	+
Access to extension	Dummy (1 = Yes, 0 = No)	+
Early Warning Information	Dummy (1 = Yes, 0 = No)	+

Source: Author’s own compilation.

).

3. Results and Discussion

3.1. Descriptive Analysis

3.1.1. Descriptive Analysis of the Background of Respondents

In

Table 2. Descriptive Analysis of the Background of Respondents.

Variables	Observation	Mean	Min	Max
Age	337	44.13	18	72
Sex	337	0.807	0	1
Education Status	337	0.51	0	1
Marital Status	337	0.896	0	1
Family Size	337	6.14	1	12
Previous Food Experience	337	0.549	0	1
Off-Farm Income	337	0.187	0	1
Access to Credit	337	0.285	0	1
Access to Extension	337	0.587	0	1
Early Warning Information	337	0.071	0	1

Source: Author’s computation from a household survey, May 2022 [Stata15].

below, the socioeconomic and institutional background of respondents is presented using descriptive statistics methods.

The average age of the household heads is approximately 44 years. Approximately 81% of the household heads are males. Half of the household heads in the study area are literate (Table 2). The mean number of family members in a household is approximately six people. Nearly 55% reported that they had experienced flood hazards in the last 5 years. Only approximately 19% of the respondents have off-farm income sources. Furthermore, 28.5% of the households have received credit access services. This indicates that credit access to rural households in the study area is very limited. Extension services from extension offices were accessed by 59% of the households in their counties. This indicates that, although the local government believes that there is full access to extension services by farmers, only half of them accessed this service. An early warning information system is at its meagre level. Only 7% of the household heads reported that they had used an early warning information system for the potential occurrence of flood in their locality.

3.1.2. Descriptive Analysis of Intensity of Flood Adaptation by Households

The number of household adaptation strategies is presented below (

Table 3. Descriptive Analysis of the Intensity (Diversity) of Adaptation Strategies.

Intensity of Adaptation	Frequency	Percentage
0	95	28.19
1	4	1.19
2	13	3.86
3	28	8.31
4	60	17.80
5	72	21.36
6	35	10.39
7	20	5.93
8	6	1.78
9	4	1.19
Observation	337	100.00
Mean	3.4
Variance	6.25

Source: Author’s computation from a household survey, May 2022 [Stata15].

).

Of the households, 28% do not adopt any flood adaptation action. On the other hand, 17.8% and 21.36% of the households adopt four and five adaptation actions, respectively. Only four households adopt nine adaptations (Table 3). Furthermore, the average number of adaptation actions used in the study area is approximately three per household and the variance is six.

3.2. Generalized Poisson Regression Result

The intensity of flood adaptation was regressed on the socioeconomic and institutional characteristics of rural households.

Table 4. Generalised Poisson Regression Results.

Dependent Variable: Intensity of Flood Adaptation
Variables	Marginal Effect (dy/dx)	z-Value	p > |z|
Age	−0.013	−3.56	0.000 ***
Sex	0.077	0.95	0.340
Education status	−0.054	−0.89	0.372
Marital status	−0.073	−0.68	0.497
Family size	0.047	2.82	0.005 ***
Off-farm income	0.196	2.72	0.007 ***
Previous food experience	0.941	11.88	0.000 ***
Access to credit	0.126	2.02	0.044 **
Access to extension	0.296	4.14	0.000 ***
Early warning	0.291	3.17	0.002 ***
Constant	0.614	3.18	0.001 ***
Wald chi2(10)	293.45
Prob > chi-2	0.000
Pseudo R2	0.2056
AIC	1376
BIC	1422
Dispersion parameter (phi)	3.58 (4.55)
Observation	337

Source: Author’s computation from the household survey, May 2022 [Stata15], ***, ** (1%, 5%) significance levels respectively.

below presents the generalized Poisson regression results.

According to the generalised Poisson regression result above (Table 4), a household head’s age significantly limits flood adaptation intensity. A unit increase in age reduces the average number of adaptation options by 1.3%. On the other hand, family size significantly and positively influences flood adaptation diversity. An additional active family member promotes the implementation of a number of adaptation actions by 4.7%.

The availability of off-farm income for a household enables the adoption of more flood adaptation mechanisms. Households with off-farm income are approximately 20% more likely to increase their flood adaptation strategies. Similarly, previous flood experience of households also triggers the household’s preparation for flood adaptation. As presented above, households who have experienced flooding hazards in the last five (5) years are 94% more likely to look for more numbers of available flood adaptation options at any cost.

Furthermore, the availability of credit services to rural households also significantly promotes adopting more flood adaptation strategies. This result indicates that households with credit access are 12.6% more likely to increase their flood adaptation options. Agricultural extension services have a statistically significant influence on adopting additional flood adaptation strategies. In this study, extension services increase the adoption of more adaptation mechanisms to flood risks. In addition, the availability of early warning information urges households to take extra adaptation measures to potential flood hazards.

Pre- and Postestimation Tests

Before the model of interest was run, a pre-estimation test was performed to check if the independent variables were free from multicollinearity problems, as this problem is common in cross-section data. Multicollinearity occurs when two or more independent variables correlate with each other. The adverse effect of multicollinearity is that it results in biased parameter estimates and standard errors in regression analysis [92,93]. Having this in mind, a multicollinearity test was performed using the variance inflation factor (VIF). The result of the VIF test is presented in

Table 5. Multicollinearity test result (using VIF).

Variables	VIF
Age	1.21
Sex	1.23
Education status	1.11
Marital status	1.26
Family size	1.22
Off-farm income	1.12
Previous food experience	1.23
Access to credit	1.05
Access to extension	1.24
Early warning information	1.05
Mean VIF	1.17

Author’s computation from the household survey, May 2022 [Stata15].

as follows.

Based on the rule of thumb in this test, if the VIF of the variables is less than 10, and the mean VIF is not substantially greater than 1, multicollinearity is not a concern and this implies that there is no multicollinearity problem among the independent variables used in the study [93]. Furthermore, the dispersion parameter (phi) (3.55), as shown in Table 4, is greater than zero, implying that there is indeed overdispersion in the model, suggesting that the GP is a better fit to the data. The Wald chi2(10) is 293.45, which is quite greater than zero indicating the overall significance of the independent variables in the model.

3.3. Constraints to Flood Adaptation in the Study Area

Household heads in the study area reported various factors as constraints to effectively mitigate and adapt to the existing flooding problem in the study area. In this section, the constraints identified here were discussed further along with the existing scientific literature.

Figure 1 below presents these major constraints (barriers) and the corresponding percentage distribution.

One of the most important constraints identified was “lack of money”. It was reported by 96% of the households. Lack of money is attributed to limited resources and it is a critical economic factor influencing decision-making [94]. Similarly, [95] also found that credit constraint plays a central role in inhibiting households to not take flood adaptation measures. Furthermore, poor people ultimately fail to install generic and structural adaptation strategies [96]. Financial constraints, lack of an early warning system, lack of land use planning, and inadequate resources were reported as barriers to flood adaptation by households in a study on the determinants of flood risk mitigation strategies at a household level in Pakistan [50]. Lack of attention and information constraint can be overcome by improving the early warning system and use of local knowledge. Moreover, the lack of farmer associations can also be addressed by strengthening the local social capital and increasing community collaboration [97].

Households also reported that they lack technological inputs to detect potential river levels and alert systems [24]. This implies that the local government has not yet implemented such systems that can serve as early warning information-generating methods for households prone to flooding hazards; 82% of the households also uncovered that lack of information about the potential advantages of each adaptation option as well as the trend of flooding events in the region. Lack of information was reported as an important barrier to local-level climate-change adaptation by communities [94].

Although lack of attention was reported by the least number of respondents, it is very crucial in flood adaptation decisions. Lamond and Proverbs [98] found that, for flood resilience efforts to be successful, coordination and commitment from flood-prone populations is the best requirement. In this study, 58% of the households reported that their lack of attention to flood adaptation was one of the constraining factors. This is also called emotional constraint according to [98]. It also encompasses the involvement and commitment of local governments to overcome flood hazards in their respective administrative system. Such uncoordinated efforts between the government and households can be reduced by creating public–private partnership funding schemes to invest in adaptation strategies. Such a coordinated funding scheme can improve households’ behaviour and decision to take action against flooding [99]. Enabling environments can be developed for stakeholders through flood forecasting, information exchange, institutional reform, contingency planning for disaster risks, and periodic monitoring [100]. In general, the focus should be directed at removing these constraints to increase the specific adaptive capacity of individuals, communities, and organizations to flood risks [96].

4. Discussion

Age is a social factor influencing an individual’s decision to adopt flood adaptation strategies. This study found that increased age is associated with a decrease in the implementation of additional adaptation strategies. This can be explained by cognitive abilities such as problem-solving skills where the aged are more likely to rely on their experience to solve problems. This can be problematic, especially in the 21st century, where occurrences are technologically driven and require a swift change to approach and the adoption of new living models [101,102]. As adults get old, they respond more slowly to changes and this explains the decline in successful flood adaptation with the ageing population.

In contrast, young individuals seek more adaptation options, whereas older ones do not. More senior people prefer to stick to the same strategies that they believe are affordable [103]. Senior households are likelier to choose elevated ground floor and precautionary savings than younger households. This is because age indicates someone’s potential and capacity to cope with natural hazards such as flooding. Flood adaptation is not cost free. It requires money, time, and energy. Older people are less likely to save [104] which, in turn, limits their capacity to invest in adaptation strategies. Furthermore, [105] also found that increased age is associated with a decreased tendency to participate in knowledge-enhancing programs. On the other hand, the age of a household head is assumed farm experience and this implies that older farmers have significant relevant skills and experience. This would help them to understand the importance of adaptation strategies and, therefore, will be forerunners to adopt better agricultural practices [51,106].

Understanding the interplay between age and successful adaptation to climate change is fundamental. This has highly been encouraged through a generational renewal with the emphasis on replacing older farmers with young generations who are agile and adopt new strategies more easily when adversities strike them and implementing a policy driven towards the earlier retirement of older farmers is a promising pathway. This explains why the EU in the frontline advocating for farm generational renewal as one of their strategies to fight the impacts of climate change seamlessly [107,108,109,110].

The size of a family in this context is an indicator of the farm labour force available for work. This study found that family size increases the average number of flood adaptation strategies. Similarly, [51] found that family size determines the feasibility of any adaptation project in the first place. Large family size, if managed well, means a better division of labour and allocation which enables the successful and timely completion of household activities such as conserving farmland to protect from erosion and related factors. Furthermore, a larger family means a better information network with the community which enhances the knowledge of the family on effective agricultural approaches. This further implies that larger families would be highly positive to try on various flood adaptation measures to save themselves from natural hazards such as floods. In contrast, [103] found that increased family size means higher consumption, less saving, and a higher dependency ratio, limiting a household’s increased adaptive capacity. Increased annual work unit which is a measure of family labour deriving from large family sizes in times of adversity provides the scarce resource, especially in setting up temporary infrastructures and countermeasures. Additionally, recovery and rehabilitation post flooding is often a challenge due to a lack of labour. Those large families have been found to recover faster than their counterparts [1,111]. Although the issue of family size is one of the most debated issues in the quest for sustainability, future policies in the areas facing population decline and highly affected by floods could be framed towards incentivising large families as an adaptation strategy.

Off-farm income availability for households increases their capacity to purchase new technological inputs. Income level determines the flexibility to invest and consume in a given family. In this study, households with off-farm income are more likely to adopt more adaptation strategies than households relying only on agricultural outputs. Off-farm income means additional funds to be able to invest in improved techniques of production [105]. Off-farm income options are additions to the wealth of a household which enables them to cope with investment risks and benefit from adaptation strategies [112]. It is associated with higher income and savings which enable households to readily adapt multiple flood mitigation methods [113]. This finding also supports the findings of [114]. Farmers can increase their revenue streams to diversify their income through either value addition or diversification of farm activities.

Such investments could be geared towards rural insurance and risk management for both individuals and businesses to mitigate the impacts of floods. Both local and national governments should provide incentives such as subsidies or tax breaks for the rural households in these regions to increase their revenues, thus creating expendable income to invest in state-of-the-art technologies to increase adaptation [49,53]. In addition to tax breaks and subsidies, government policies in these regions must be crafted towards supporting infrastructural development, research and development, capacity building, and training to increase awareness in rural households on adapting to floods. However, such actions must be guided by stronger community engagement to ensure that the rural households are part of the decision-making through strong and active public participation to ensure their perspective and problems are considered in these policies’ development [99,115].

Previous flood experience is one of the predictor variables in this study. Households that previously experienced flood hazards are more likely to adopt diversified adaptation measures. This is because their experience informs them about the potential risk of future floods and would trigger them to prepare well using their utmost self-efficacy [53]. Osberghaus [116] also reported that the probability of increased flood adaptation increases with past flood-damage experience. Previous flood experience improves households’ understanding of flood risks and this, in turn, helps them to prepare themselves effectively by adopting the mitigation and adaptation mechanisms available [32]. Similarly, crop loss experienced by households was found to be a statistically significant positive influencer for the adoption of climate-change adaptation strategies [106]. The findings of this study are also in line with [114,117].

Access to credit is an institutional arrangement that creates opportunities for investment in productive activities. This study found that households with credit access are more likely to adopt flood adaptation options. Credit access offers the capacity to cope and adapt to natural hazards such as flooding. In addition, credit allows households to overcome their liquidity problem and financial constraints [118]. The findings of this study also support the results of studies by [119,120]. Credit access also serves as the capital for investment in adaptation strategies [121]. Credit plays a dual role for households; one is to enable households to invest in flood protective measures, which reduces their vulnerability, and to better recover from it in case it happens [33]. Access to credit can be further viewed as insurance for the farmers in times of adversity. Thus, the government must engage both local communities and financial institutions to establish a strong risk insurance scheme that can increase the adaptive capacity of farmers. Promoting financial inclusion for flood-prone communities is one of the most promising policy developments where government creates an enabling environment for financial institutions to design financial products specifically for those communities. However, these products must be tied with increased risk and insurance investment with the credit facilities made to be the easiest to access and in real-time when floods struck the communities [122,123,124,125].

Access to extension is tailored to spread information and scientific knowledge to rural households regarding agricultural input prices and supply and farming activities. In addition, extension services also play an essential role in preparing and informing households to take flood adaptation measures in flood-prone areas. This study found that households with access to extension services are more likely to adapt to floods than their counterparts. Bahinipati & Venkatachalam [119] also found a similar finding that confirms the role of extension services. Extension services also help improve the local knowledge of households for making the right adaptation decision [126]. Extension services through open-field demonstration and training create awareness in farm households about the importance of adopting improved agricultural technology [106,127]. Information is power and, with the advancement of technology in the 21st century, extension service providers must be hands on to ensure their services are integrated with the new trends of technology. Setting up an integrated knowledge and information system that sends real-time notifications and warning to farmers in the case of foreseen risk can increase the ability to adapt. Additionally, extension service providers in collaboration with the government must work with the telecommunication industry and the news agencies to ensure that there are farmer education programs that are being aired and channelled to the mainstream media for free to ensure all channels are made accessible to farmers for increased knowledge acquisition.

Early warning information influences the number of flood adaptation strategies households can take to cope with flooding damages. Therefore, information is a crucial input to making the right decision. Households who received early warning information take adaptation mechanisms toward floods. An early warning information system provides timely information about weather variability which can serve as a signal for households to implement more flood adaptation strategies [106,128]. This finding is in line with the results of [103,114,120]. To establish a state-of-the-art warning system, there is a need to upgrade the infrastructure network for electricity and wide coverage of a fast-speed network. Subsidizing access to fast internet and the availability of smartphones to farmers in the market makes it easier for real-time sharing of information from different corners of the country.

5. Conclusions and Recommendations

The study examined the determinants of rural households’ intensity flood adaptation in the Fogera rice plain, Ethiopia, employing the generalised Poisson (GP) model. This study attempted to contribute to methodological approaches in the study of determinants of flood adaptation by using a count data model rather than binary models that ignore more than one response in a dependent variable. The study found that off-farm income, previous flood experience, access to credit, access to agricultural extension services, and an early warning information system significantly increase the number of flood adaptation strategies rural households use in the Fogera rice plain. In contrast, the age of the household head significantly reduces the average number of adaptation options. The study, therefore, recommends adopting a multi-actor approach in a coordinated effort to enhance households’ livelihood and access to institutional facilities (credit services, extension services, and early warning information systems) so that an effective flood adaptation system would be possible. Furthermore, as the households reported several constraints that hold them back from adopting multiple adaptation strategies, stakeholders should not overlook the role of these barriers while working to improve the adaptive capacity of these farmers.

Additionally, there is a need for both the local and national governments to come up with policy frameworks and plans that accelerate rapid adaptation to floods. Such policies could be geared towards tax subsidies for rural households or incentives for any practice that promotes adaptation to floods. Establishing a water retention framework by the Ethiopian government to ensure the excess water from the rivers is contained while surface runoff is reduced is a promising pathway for policy development. Additionally, modifying the soil-health policy to incentivise farmers for increasing their water conservation can increase the ability of farmers to control floods and develop locally driven adaptation strategies. Another important area the Ethiopian government can improve is the management of short water cycles, which has the potential in reducing floods. This is achievable through setting up a state-of-the-art early warning system and improved infrastructural development for the farmers to ensure the impacts of unpredicted floods are averted early in advance. The success of the government policies is a function of stronger community engagement and active public participation; thus, these policies must advocate for stronger public participation in developing them.

Establishing and adopting a disaster and risk-reduction mechanism where the local and national governments and the private sector must work together to enhance policy implementation is crucial. This could be successfully established by ensuring the Ethiopian government provides an enabling environment for the backing sector to provide tailor-made financial products and insurance services for these communities. However, the insurance industry must customize a product for farmers that covers them from hazards which promotes action-based payouts and real time. Weather protection insurance can compensate farmers depending on the level of adversity of the risk. This must be guided by a framework based on index-based risk assessment and compensation of farmers.

Although the research focused on rural households, future research could as well provide insights into how policymakers, financial institutions, and nongovernmental organizations across the Fogera region perceive these determinants. From a scientific perspective, future research could evaluate the effectiveness of the desired determinants. This could be further enhanced by the national statistics bureau designing a demographic household survey for flood-prone areas to collect annual data for the long term to compare the performance of these measures for effective policy development.