More than 70% of the rural population in developing countries is engaged in agriculture¹. Nevertheless, food security is a pressing issue in the modern world due to multifaceted challenges such as rapid urbanization, climate change, and population explosion². By 2050, the world will need to produce roughly 70% more food to feed a projected 9 billion people, making the food security dilemma even more prominent³. To be precise, agriculture’s severe vulnerability to climate change is likely to exacerbate the dilemma¹. Climate change’s detrimental effects on farming are apparent through extreme weather events such as increased temperatures, weather variability, heavy precipitation, cyclones, drought, and invasive crops and pests^1,4.

The prevalence of weather events is predominantly high in the southwestern coastal areas of Bangladesh⁵. According to the Global Climate Risk Index, Bangladesh is one of the most vulnerable nations to climate change⁶ and approximately 62% of its coastal land is already affected by salinity, leading to substantial rice losses during cyclones⁷. The Sundarbans ambient coastal zone is highly affected by climatic shocks, i.e., cyclones, drought, salinity intrusion, and waterlogging. These adverse shocks scale the risks of crop vulnerability and agricultural production^8,9.

The agricultural sector in Bangladesh is primarily dominated by small and marginal farmers. Around two-thirds of the rural population is employed by agriculture and 87% of rural households are managing a portion of household income from this sector¹⁰. It employs 45.33% of the labor force in the agriculture sector and contributes roughly 13% to the GDP of Bangladesh¹¹. Rice is the primary crop, and occupies around 77% of all agricultural land¹². Despite occupying a large digit of land for agricultural activities, approximately 21% of the total population of Bangladesh faces moderate and severe food insecurity¹³. This phenomenon is likely to sustain the poverty rate which is higher than the national average in the coastal area due to frequent climate shocks and their adverse effects on crops, especially in paddy production^14,15.

Climate-smart agriculture (CSA) practices may be a solution to reduce yield losses caused by adverse climatic conditions¹⁶. It is an integrated farming technique that addresses the challenges of food security and mitigates the effect of climate change on agricultural production by increasing productivity¹⁷, developing resilience, and reducing emissions^3,18. It is argued that CSA protects poor farmers from livelihood crises and food and nutritional insecurity, especially in the face of extreme weather events^17,19,20.

Several CSA technologies are adopted in different parts of the world^21,22. Conservation tillage, agroforestry, crop rotation, livelihood diversification, drip irrigation, and precision farming are common among them^23,24,25,26. The decision to adopt CSA practices is determined by various factors such as land size, the previous household experience of confronting climate shocks, land fertility, the distance of the farm from the market, and so on^27,28. A study finds that farming experience, information sources, and multiple cropping systems positively influence CSAT adoption in Vietnam²⁹. Another study in Mali shows that socio-demographic characteristics such as age, gender, education, employment, and farming experience significantly affect the CSAT decision³⁰. Similarly, in Bangladesh, the same factors are reported to determine the CSAT adoption²⁸.

The benefit of CSAT practices on yield and food security has been studied by previous studies^{20,31,32,33,34}. However, its link with multidimensional poverty in the coastal context is rarely explored. A study finds that the households that implemented climate-smart farming practices had a 22.2% increase in agricultural output compared to the non-adopters³⁵. Evidence also suggest that CSAT adopters are better off due to the “technology impact”, compared to the non-adopters³⁶. In addition, former studies exhibit that CSA practices can be a mitigating approach to reducing carbon emissions in agriculture^16,37. For instance, a negative association between CSA practices and carbon emission intensity is documented in Tanzania³¹. Therefore, CSA practices not only help farm households adapt to erratic weather patterns but also promote carbon sequestration, and reduce soil erosion.

However, farmers face barriers such as high investment and maintenance costs to CSAT adoption³⁰. A study finds that 78% of respondents identified input costs as the primary obstacle to CSAT adoption²⁵. Moreover, most CSA practices are not widely adopted in South Asia due to weak organizational capacities and asymmetric information³⁸. Other challenges to adopting these CSA practices include input accessibility, credit restrictions, water scarcity, market uncertainty, and lack of adaptation information and extension services²⁵.

While several studies have highlighted the factors associated with CSAT adoption, its potential significance, and the barriers to adoption, there has been little investigation into the role of CSA practices in reducing multidimensional poverty, particularly in hazard-prone coastal areas of developing countries like Bangladesh. This study contributes to the pertinent literature and aligns with the first and second sustainable development goals (SDGs), which focus on poverty and hunger reduction. Furthermore, this study offers technology-specific recommendations for coastal farmers, which is a unique contribution compared to previous related studies, especially in Bangladesh. Given the significant impact of climate shocks on impoverished farming households, the aim of this study is to identify the key factors influencing the decision to adopt CSAT and evaluate how such practices affect multidimensional poverty. To achieve this objective, the study addresses the following three research questions:

1. i.

   What are the factors that influence the adoption of climate-smart agriculture among coastal farm households?

2. ii.

   Does CSA practices have any impact on multidimensional poverty among coastal farm households?

3. iii.

   Which CSA technologies can be efficient in reducing the MPI of farm households?

The results suggest that the decision to adopt CSAT is influenced by factors such as crop vulnerability, crop income, access to extension services, and training on input management. We find CSA practices significantly contribute to the reduction of MPI of coastal farmers. According to the counterfactual analysis, current CSAT adopters experience 41 percentage points lower MPI compared to the scenario where they did not adopt CSAT. Similarly, if non-adopters had chosen to adopt CSAT, their MPI would have been reduced by 15 percentage points. According to the current adoption and non-adoption effect, current adopters are experiencing a 10-percentage point lower MPI compared to the present non-adopters of CSAT. The study also highlights the impact of specific CSA technologies on MPI. We find that five CSA practices significantly reduce the MPI of farm households, relative to the non-adoption of such farming techniques. The practices are agroforestry, on-farm diversification, conservation tillage, contingent crop planning, and rainwater harvesting (drip irrigation). This information may be beneficial for coastal farmers when making informed decisions about CSAT adoption. To reduce multidimensional poverty in climate-affected coastal areas, the study recommends increasing access to extension services and providing training on input management to coastal farmers.

Data description and background statistics

Table 1 describes the general statistics of households’ sociodemographic characteristics, farm input, and output-related information. It shows that around 39% of the farm households reside near the peri-urban coastal area, and the rest big portion live in rural areas. The farm households hold approximately five members on average and they are mostly male-headed (almost 83%). The average educational attainment of the household head is almost six years, with a salient extent of standard deviation (4.26). It implies that the educational status is widely varied among the farm household heads.

The crop income, measured by the monetary value of the produced crop per season irrespective of its own consumption and sale, is approximately BDT 5217 (US$ 60.51) on average with a standard deviation of BDT 2333 (US$ 27.00) (inflation not adjusted). It indicates that there remains a disparity in terms of crop income and thus the farm households are non-homogenous in this case. On the contrary, the households earn around BDT 12617 monthly on average from the farm and off-farm sources of income together. According to BBS (2022), the household income is BDT 32,422 at the national level, which is BDT 26,163 in rural areas of Bangladesh.

Paddy is also unevenly harvested by the farmers. The average yield is 6.37 tons per hectare¹ with a minimum of 2.77 tons and a maximum of 9.40 tons yield per hectare. This pattern of yield may stem from the difference in factors of production, geographical disparity, fertility of the land, amount of land use for cultivation, and overall heterogeneous farm management strategies adopted by the different farm households. Regardless of yield distribution, the incidence of natural hazards on crops has been extreme among the households. Crop vulnerability index shows this phenomenon since the average incidence score exceeds 0.50 (0.65 out of 1.00).

Table 1 also shows that around 48% of the farm households do not have any savings account either in any bank or in NGOs, revealing that the farm households in the coastal areas are not highly empowered with financial inclusion. Regarding formal credit, only around 55% of households can manage to access it. It suggests a large portion has no access to formal credit. However, 69% of them have loan burden with an average of BDT 24681 (US$ 286.29) from formal and informal sources together.

Access to extension services is also limited to 41% of households. It might be a concern to the adoption of CSAT since access to extension services introduces farmers to the modern and updated farming technique which is more critical to the knowledge acquirement for coastal farmers. We find that the rate of CSAT adoption is actually 28% (those who adopted the CSAT intensively). The moderate adopters are 46.7%. To pursue the causal analysis, we merged the moderate (46.7%) and the intensive adopters (28.2) and designated them as CSA adopters.

Determinants of climate-smart agriculture technology adoption decision

Employing the ESR approach, the study identified potential factors of the CSAT adoption decision and the impact of the adoption on poverty status (MPI) of the farm households. Tables 2 and 3 exhibit the FIML and ML (maximum likelihood) estimates of ESR, respectively. The selection column (column 1 of Table 2) shows that crop income is positively associated with CSAT adoption. Higher crop income increases the capability to install CSA technologies that aid in protecting the crop from extreme climate events. Another important factor of CSAT adoption is access to extension services that positively correlate to CSAT adoption. It is rational that extension service gives farmers adequate knowledge about new technologies and inputs in agricultural production, thus, enhances the likelihood of CSAT adoption. Also, if farmers receive training on any agricultural input management, the chances of adopting CSAT increase, as evident in Table 2. It is also evident that the application of CSAT depends on the losses incurred by the climate shocks. The losses push farmers to use innovative technologies and CSAT has been a promis option in this case. This evidence is shown in Table 2. It suggests that as the crop vulnerability is higher, the likelihood of adopting CSAT increases among coastal farm households (p < 0.01). It indicates that farm households tend to internalize the negative effect of climate events on crops by adopting CSA techniques. According to the ML estimates, shown in Table 3, the variables that positively determine the CSAT adoption decision are crop income, access to extension service, and paddy yield which are quite consistent with the FIML estimates.

Impact of climate-smart agriculture technology adoption on multidimensional poverty

The study aimed to precisely evaluate whether CSAT adoption helps farm households reduce their poverty statuses. Table 4 demonstrates the counterfactual effect of the adoption and non-adoption of CSAT based on the FIML estimates. It shows that the current CSAT adopters are enjoying a 41-percentage point less MPI because of their decision to adopt CSAT rather than the non-adoption decision (p < 0.01). In other words, if the current adopters had not adopted the CSAT, they would have been experiencing a higher effect of poverty score (0.901) which is 41 percentage points larger than the MPI of current CSAT adopters. In the same way, if the current non-adopters adopted CSAT, they would be able to reduce their MPI incidence by 15 percentage points (p < 0.01). In other words, the current non-adopters are experiencing a comparatively higher MPI score (0.588), which could be reduced to 0.436 if they had adopted CSAT. Table 4 also represents the difference in simple treatment effect between the current adopter and the non-adopter. It shows that those who currently adopt CSAT experience 10 percentage points lower MPI than those who do not adopt CSAT.

Impact of specific climate-smart agriculture technologies on multidimensional poverty

To delve deeper into the impact of CSA technology adoption on MPI, we have separately executed ESR regression for each specific CSA technology. Table 5 reports the results of the selection model and Table 6 presents the main impact of specific technologies on MPI. Results show that five out of seven CSATs significantly reduce the MPI of households. More precisely, agroforestry and on-farm diversification adoption significantly reduce the MPI score by around 28 and 24 percentage points (p < 0.01), respectively, compared to the non-adoption of respective CSA technologies. Similarly, conservation tillage and contingent crop planning adoption also contribute to the MPI reduction by around 33 and 35 percentage points (p < 0.01), respectively relative to the non-adoption of these technologies. In the same way, rainwater harvesting/drip irrigation adoption has a significant negative effect on MPI, i.e., it reduces MPI by 35 percentage points (p < 0.01), compared to the non-adoption (Table 6). The ML estimates in Table 3 also verify this finding, showing that CSAT adoption reduces the MPI by 18 percentage points compared to non-adoption, indicating CSAT has a significant role in poverty reduction among farm households.

However, there is a concern that the adoption of crop diversification (CD) and farming decisions based on weather forecasting (FDWF) exhibit a positive relationship with MPI, as revealed by the current study. The plausible reason for such a relationship between crop diversification and MPI may be due to the lower economic productivity of small farm households. A majority of farm households in the southwestern coastal region of Bangladesh are small and marginal in terms of land size. Intensive crop diversification in small land size compromises the economies of scale of farmers⁴⁶ and may fail to address the MPI. Similarly, the positive relationship between FDWF and MPI can be described from the households’ possession of smartphones and internet access. We find that although around 75% of the households possess a smartphone, all do not have internet access (65%) (Table 1). It poses a barrier to acquiring timely and correct weather forecasting. Consequently, farm households cannot properly adopt FDWF and fail to reduce the MPI significantly among them.

Furthermore, crop vulnerability due to climate-related issues is found to be positively correlated with the adoption of all CSA technologies except for CD and FDWF (Table 5). This suggests that crop vulnerability may discourage the adoption of CD under the assumption that farmers might think—losing a single crop is better than multiple crop losses in the same land because of the high negative impact of climate issues on crops. It leads to high crop losses that hinder intensive adoption of CD and FDWF further, and thus, positively affects MPI. For FDWF adoption, we find crop vulnerability as a disincentive factor, meaning that high crop loss due to climate conditions reduces the probability of FDWF adoption among farm households. This is because climate risks frequently discourage farmers from relying on scientific forecasts, as observed in previous research⁴⁷.

Robustness check

We tested the results by running another ESR of maximum likelihood estimates. This result also shows that CSAT adoption reduces multidimensional poverty significantly, reducing the MPI of the adopters by 18 percentage points compared to the non-adopters (Table 3). We further estimate the relationship between CSA practices and MPI status using the average deprivation score, and equally weighted score of MPI to demonstrate the robustness of our findings. First, the mean score of MPI for each household is obtained by adding the deprivation (deprivation=1) in each component of MPI and dividing them by the total components (9). The ESR finding shows a similar effect of CSA practice on MPI (Supplementary Tables 1 and2). The technology-specific results also exhibit the same pattern of associations with MPI as aligned with the main FIML estimates (Supplementary Tables 3 and4). Second, assigning equal weight to each construct of MPI, we run the ESR with the same set of regressors. We discover that all approaches yield similar results, meaning that CSA practices significantly reduce MPI compared to non-adoption (Supplementary Tables 5 and6). Additionally, we conducted another endogenous switching regression, treating MPI as a binary variable where 1 represents poor households and 0 represents non-poor households. The regression also provides similar effects (i.e., the probability of being multidimensionally poor is significantly lower among the adopters than the non-adopters (Supplementary Table 7-8). This implies that CSA practices significantly reduce multidimensional poverty among coastal farm households.

The key premise of the study is to observe the factors determining CSAT adoption and its associated impact on MPI in the climate-vulnerable coastal households of the southwest zone of Bangladesh. The farm households in these areas are highly susceptible to climate change and recurrent climate shocks. These challenges make the struggle of poverty reduction even more difficult. The current study explores the causal relationship between CSAT and multidimensional poverty, helping to understand the necessity of CSAT in reducing households’ exposure to climate-induced challenges in agriculture, especially in paddy farming.

Our hypothesis posited that the adoption of CSAT would influence the multidimensional poverty experienced by coastal farm households. According to the FIML estimates of the ESR model, we find that farm households adopting CSAT significantly reduce their MPI compared to the CSAT non-adopting farm households. This finding is corroborated³⁶. Therefore, CSAT adoption should be enhanced to combat crop vulnerability posed by adverse climate events in the coastal areas. Specifically, vulnerability-specific CSAT interventions might be more fruitful. For instance, we find that crop vulnerability, measured by climate-induced issues such as cyclones, floods, drought, heavy/low rainfall, and salinity, positively predicts CSAT adoption. It suggests that climate-induced issues are compelling farm households to adopt CSAT. In this regard, the use of flood-tolerant seeds may be more fruitful to lessen the effect of flood. Similarly, salt-tolerant seeds may be useful to combat the negative effect of salinity on crops. Likewise, farm-based livelihood diversification, crop rotation, and drip irrigation are examples that contribute positively to protecting farmers from crop loss from specific climatic events^48,49,50,51. The adoption of specific CSAT technologies, thus, may be more useful for farmers to mitigate the bear of specific climate issues on paddy efficiently.

Moreover, the results of ESR models pivoting on technology-specific effects suggest that agroforestry, on-farm diversification, conservation tillage, contingent crop planning, and drip irrigation adoption significantly reduce the MPI compared to the non-adoption decision. Therefore, the study suggests scaling up the adoption of these CSA technologies to minimize the climate-induced effects on crop yield, paving the way for curbing the intensity of MPI in poor farm households. However, rural farmers might lack the knowledge and training regarding CSAT, which is a significant concern for its widespread adoption^14,38.

It is also argued that access to extension services is a promising way to enhance the knowledge of CSAT²⁹. The current study results suggest that access to extension services increases the probability of adopting CSAT significantly. Previous studies also support this finding^12,29,30,52. Besides, the data also provide strong evidence that training on input management increases the likelihood of CSAT adoption on a significant scale. By receiving such training, farmers are likely to become familiar with updated cultivation methods and use the most beneficial seed varieties and other inputs that work well against climate-induced crop loss²⁹.

We also find that crop vulnerability and extension services are negatively associated with the adoption of CD and FDWF. CD and FDWF are positively related to MPI. There are scopes to conduct more in-depth studies on this issue. More comprehensive and focused studies can explore consistent findings about the relationship between CD and MPI and FDWF and MPI in coastal environments.

Innovations and modern technologies in smart farming are crucial for eliminating poverty and ensuring global food security. The southwestern coastal areas of Bangladesh are highly susceptible to natural disasters and climatic events that make sustainable farming challenging for poor households. In such areas where the frequent climatic events pose threats to sustainable farming and hit households with poverty, advanced farming systems such as CSAT practices can be promising in such regions areas. CSAT practices help reduce the impacts of climate change on paddy production and minimize crop income loss induced by climate-related events. By combining CSAT with MPI, this study offers an interest in the subject of agricultural economics that can potentially answer the way of gaining SDGs such as eradicating poverty with sustainable farming in poverty-hit coastal regions vulnerable to climate hazards.

The study findings show that the adoption of CSAT practices significantly reduces the multidimensional poverty of marginalized farm households. Therefore, CSAT adoption among farmers in the coastal areas should get higher priority when launching any projects on sustainable agriculture by both any government and non-government schemes. In this case, we find that training farmers on input management can enhance the likelihood of adopting CSAT. Also, the study finding suggests that access to extension services should be broadened to far-flung areas of coastal zones packaging the CSAT tools and techniques in the extension services. Besides, based on the findings, agroforestry, on-farm diversification, conservation tillage, contingent crop planning, and drip irrigation should be scaled up with comprehensive intervention in the coastal areas since these specific CSA technologies significantly reduce MPI among the small and marginalized paddy farm households.

Although the study has a robust appeal in exploring the impact of CSAT adoption on multidimensional poverty in the coastal settings of Bangladesh, the external validity should be reexamined with a larger sample size in different geographical contexts. Longitudinal studies with experimental design also can portray a more detailed picture of such a study. Overall, success in gaining SDG1 CSAT adoption can be a long-term solution for developing nations where the coastal areas are highly vulnerable to climate shocks and climate-induced crop loss. Thus, promoting CSAT among poor farmers can help them graduate from poverty status to a sustained livelihood in agriculture.

Study area

The study is conducted in the Khulna district, a southwestern coastal zone edged by the Bay of Bengal and the Sundarbans. We randomly select three sub-districts (i.e., Dacope, Paikgachha, and Batiaghata). This allows for expected heterogeneity among the farm households, their poverty status, uses of CSAT, and varying household adaptive capacities to climate shocks to protect crops.

Sampling and survey design

The study considers a multi-stage sampling technique for data collection. The unit of analysis of the study is the farm households in southwestern Khulna district, Bangladesh. Firstly, three sub-districts are randomly chosen out of 10 sub-districts as the primary sampling units (PSUs). These three sub-districts are Batiaghata, Paikgachha, and Dacope, with a total of 25 Union Parishads (UPs). Among them, Batiaghata has seven UPs, while Paikgachha and Dacope each have nine UPs. Secondly, three UPs are randomly selected as the secondary sampling units (SSUs) from each sub-district. Finally, the study selects 39 farm households as the ultimate sampling units from each of the nine UPs and the total sample size is thus 351.

We collected the farmers’ list from the Sub-district Agriculture Office. The 9 UPs contain 7241 farm households producing paddy. We used Cochran’s sampling formula³⁹\(\left({n}_{0}=\frac{{Z}^{2}{pq}}{{e}^{2}}\right)\) to obtain the required sample size of 349, where, the critical value of desired confidence interval (z) is 95%, maximum variability (p) is 50%, i.e., 0.5 and, q is (1-p), and the margin of error (e) is 5.25%.

Out of 1319 households producing paddy in Batiaghata, a total of 580 households were retained based on our sample selection criteria. Two principal sample selection criteria were considered: (i) the household must have at least three consecutive years of farming experience and no less than 5 years of paddy farming experience in their respective regions, and (ii) the farm household must be conversant with either conventional or modern farming. The two criteria allowed for the heterogeneity in sample selection that reduces bias. After finalizing the list, a simple random sampling method is deployed to interview the head of the farm household. If any households refused to participate, we randomly chose another household from the list to fill up the gap.

Likewise, using the same selection criteria, the retained farm households for Paikgachha and Dacope remained at 1162 and 1452, respectively. From each sub-district, an equal sample size of 117 was selected for data collection. The sample distribution is outlined in Supplementary Table 9. Since the sample formula gives us an optimum sample of 349, we equally distributed the sample (39 households to each UP) to the 9 UPs. It allowed us to randomly select households from the retained households and limit the sample size to 351, which exceeds the targeted sample size. To reduce sampling bias, this sampling procedure adopts random selection at every stage.

The study follows a cross-sectional survey design allowing a face-to-face interview using a well-structured interview schedule. The study prepared the interview schedule focusing on the information related to the socio-economic status of the households, CSAT adoption, crop and on-farm diversification, poverty indicators, demographic characteristics, input-related variables, institutional factors, agricultural productivity, and land-specific variables. We place the description of variables in Supplementary Table 11. The questionnaire underwent a pretest phase, involving a two-week pilot survey. Subsequently, adjustments were made to the questionnaire based on the feedback received from the field before the final survey. The final data collection was carried out in May 2022.

Analytical approach: climate-smart agriculture

The three outcomes of CSA practices are increased productivity, enhanced resilience, and reduced carbon emission³. Increased Productivity refers to higher production of food, nutrition, and income. Enhanced Resilience focuses on equipping farmers to combat climatic challenges like droughts, pests, diseases, and erratic weather^3,40. Reduced Emission involves lowering the amount of carbon dioxide released per kilogram of food produced, minimizing deforestation, and implementing techniques that capture and absorb carbon.

The study analyzes seven CSA technologies: crop diversity, agroforestry, farming decision based on weather forecasting, on-farm diversification, conservation tillage, contingent crop planning, and rainwater harvesting/drip irrigation^23,24 (Table 7). Responses on the adoption of these technologies are collected in binary response (0=No, 1=Yes). Three levels of adoption are identified: non-adopters or insignificant adopters (CSAT ≤ 2), moderate adopters (2 > CSAT < 5), and intensive adopters (CSAT ≥ 5). For further analysis, we considered the moderate and intensive adopters as the main adopters while the rest were designated as non-adopters.

Crop vulnerability

Crop vulnerability, an important regressor in our analysis, has been measured based on the adverse effect of climate-induced issues on paddy crops. This assessment encapsulates seven climate-related indicators which collectively form the crop vulnerability index. These indicators are drought, flood, storm/cyclone, pest attack, heavy/less rainfall, riverbank erosion, and salinity. Responses to each indicator were recorded on a five-point Likert scale, ranging from 1 indicating no effect to 5 signifying a very high impact on crops, particularly paddy yield. Detailed information regarding the measurement of the crop vulnerability index is placed in Supplementary Table 10.

Multidimensional poverty index

This paper considers the multidimensional poverty index⁴¹ to analyze the impact of CSA practices on household poverty. This approach has widely been used in measuring multidimensional poverty, with a growing body of recent studies^36,42,43,44 for its closeness to the poverty reduction promise of sustainable development goals (SDGs). In addition, MPI measures numerous deprivations that households experience simultaneously. The index consists of three primary dimensions, i.e., education, health, and standard of living (Table 8). These three dimensions comprise 9 components. We assigned a distinguished weight to each component to weigh the MPI score. Table 8 explicitly describes the dimensions, indicators, deprivation cut-offs, and assigned weights.

The study sets two cut-off methods to identify poor households^36,43. The first cut-off is a direct technique of determining the degree of achievement required to be classified as deprived or non-deprived in each criterion. The second cut-off measures poverty based on the deprivation score of the farm households.

The counting approach of multidimensional poverty can be presented in the following way. Let\(n\) be the number of farm households and\(d\ge 2\) be the number of dimensions. Then,\(X=[{X}_{{ij}}]\) is the\(n\times d\) matrix of MPI achievements, with\({X}_{{ij}}\) representing the achievements of farm household\(i(i={{\mathrm{1,2,3}}},...,n)\) in dimension\(j(j={{\mathrm{1,2,3}}},\ldots ,d)\). Thus, the matrix can be formulated in the following.

Each column vector\({X}_{j}\) represents the distribution of dimension\(j\)'s achievements over the set of households, whereas each row vector\({X}_{j}\) represents the achievements of i^th household. Below the pre-ordained cut-off\(z\{{Z}_{i}=\left({z}_{1},{z}_{2},\ldots ,{z}_{n}\right)\}\), a household is recognized as deprived in dimensionj (\({X}_{{ij}}\le {z}_{i}\)). This is the first poverty-cut off that directly measures poverty based on the deprivation in each certain dimension. For further clarification, we assume a deprivation matrix\({\widetilde{X}}^{0}=[{\widetilde{X}}_{{ij}}^{0}]\), derived from the\(X\) matrix presented in Eq. (2):\(\forall i\) and\(j\).

Equation (2) clearly communicates that when\({\widetilde{X}}_{{ij}}^{0}=1\), the farm household will be deprived in dimension\(j\), and\({\widetilde{X}}_{{ij}}^{0}=0\) otherwise. The weighted deprivation gap for each household (\({C}_{i}\)), is obtained by horizontally summing up the deprivation score attached to each indicator. It is expressed through Eq. (3).

In Eq. (3),\({W}_{j}\) denotes the weight assigned to each of the indicators. Thus, the second poverty cut-off also segregates the poor from the non-poor by weighing deprivation. To identify the multidimensionally poor household, we consider Eq. (4). For specification, let introduce\({C}_{i}=({c}_{1}...{cn}){\prime}\) as the deprivation counts (number of deprivations) that are compared with the cut-off\(k\) to determine the poverty status. The condition of the comparison is\(0 \, < \, k \, < \, d\), meaning that the deprivation dimension of the poor households transcends the cut-off deprivation. The cut-off point (k = 3) is defined as one-third of the indicators (9).

Estimation strategy

Drivers of climate-smart agriculture technology adoption and its impact on multidimensional poverty

The Endogenous Switching Regression (ESR) model is employed to evaluate the impact of CSAT on MPI, following³⁶. The study considers MPI as the poverty status of the households. The MPI status of the CSAT adopters is identified by\({Y}_{1i}\), while that of non-adopters is identified by\({Y}_{2i}\). It is important to note that the decision to adopt CSAT is endogenous. For instance, poverty reduction and CSAT adoption are likely to be codetermined because of the potential reverse casualty between these two variables. To compensate for the endogeneity of the adoption decision, the study estimates a simultaneous equation model with ESR with the help of full information maximum likelihood (FIML). The CSAT adoption selection equation is as follows:

In Eq. (5),\({D}_{i}^{* }\) is a latent variable demonstrating the unobservable CSAT adoption decision. Contrarily,\({D}_{i}\) is the observable dependent variable that describes whether the farmer has adopted the CSAT with 1, or has not adopted the CSAT, with 0. The non-stochastic vector\({X}_{i}\) represents the farm and non-farm factors of CSAT decision, and\({\mu }_{i}\) is the stochastic disturbance term. The ESR model incorporating the poverty status (MPI) accounts for the selection bias in which farmers face two regimes: (a) adopt or (b) not to adopt. This phenomenon is illustrated in the following equation.

In both regimes (6a) and (6b),\({Y}_{i}\) refers to the poverty status (MPI) of the households and\({V}_{i}\) is the vector of exogenous regressors affecting the regressand\({Y}_{i}\). The error terms are assumed to have a trivariate normal distribution with a zero mean and a non-singular covariance matrix³⁶, which is expressed as:

Where,\({\sigma }_{\mu }^{2}\),\({\sigma }_{e1}^{2}\), and\({\sigma }_{e2}^{2}\) are the covariances of the error term in Eq. (5), Eq. (6a), and Eq. (6b), respectively. The reason for developing the covariance matrix and its analysis is to confirm the endogeneity and necessity of using ESR in this context. In Eq. (7),\({\sigma }_{e1\mu }\) and\({\sigma }_{e2\mu }\) are the obtained covariance of the error term of the CSAT adoption selection function (\({\mu }_{i}\)) and MPI function (i.e.,\({e}_{1i}\)\({{{\rm{and}}}}\)\({e}_{2i}\)). Therefore, since the covariance of\({\mu }_{i}\) is correlated with that of\({e}_{1i}\) and\({e}_{2i}\), the expected values of\({e}_{1i}\) and\({e}_{2i}\) subject to the sample selection is non-zero (biased).

In Eqs. (8a) and (8b), the\({\lambda }_{1i}\) equals to\(\frac{\varnothing \left(\delta {X}_{i}\right)}{{{{\boldsymbol{\Phi }}}}\left(\delta {X}_{i}\right)}\) and\({\lambda }_{2i}\) refers to\(\frac{\varnothing \left(\delta {X}_{i}\right)}{1{{{\boldsymbol{-}}}}{{{\boldsymbol{\Phi }}}}\left(\delta {X}_{i}\right)}\). Again,\({{{\boldsymbol{\Phi }}}}\) (.) is the normal probability density function (PDF) and\(\varnothing\) (.) is the cumulative density function (CDF). Equations (8a) and (8b) test whether the estimated covariances\(\widehat{{\sigma }_{e1\mu }}\) and\(\widehat{{\sigma }_{e2\mu }}\) are statistically significant. If they are significant, the decision to adopt CSAT and poverty status are said to be correlated. It indicates that there is evidence of endogenous switching and sample selection bias. We use the full information maximum likelihood (FIML) estimation procedure which is a consistent method of estimating ESR models⁴⁵. ESR estimates the average treatment effect of the treated (ATT) and the untreated (ATU), by associating the predicted values^36,45. The following framework clarifies the estimation process.

If adopters do not adopt CSAT (Counterfactual):

If non-adopters adopt CSAT (Counterfactual):

The actual expectations observed from the sample are represented by Eqs. (9a) and (9b). To estimate the ATT, the difference between Eqs. (9a) and (9c) has been considered in Eq. (10).

Likewise, the expected change (ATU) in the MPI of the CSAT non-adopter households, if they had adopted CSAT (treated) can be computed as the difference between Eq. (10d) and Eq. (10b).

Apart from the aggregate impact of CSAT on MPI, we run separate ESR regressions containing the same set of explanatory variables to explore the technology-specific impact of CSA on MPI.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

The data supporting the findings of this study consists of survey data on household characteristics and agricultural practices in coastal settings. The data can be accessed through this identifier:https://doi.org/10.7910/DVN/ZWVHNC.

The code can be requested through this identifier:https://doi.org/10.6084/m9.figshare.25944442.

We acknowledge the anonymous reviewers for their insightful and constructive instructions to improve the paper. We did not receive any specific funding for conducting this work.

Authors and Affiliations

1. BRAC Institute of Governance and Development (BIGD), BRAC University, Dhaka, Bangladesh

   Md. Karimul Islam

2. Economics Discipline, Social Science School, Khulna University, Khulna, Bangladesh

   Fariha Farjana

Contributions

Md. Karimul Islam: Conceptualization; Data curation; Formal analysis; Methodology; Software; Investigation; Validation; Visualization; Roles/Writing - original draft; Writing - review & editing. Fariha Farjana: Conceptualization; Project administration; Methodology; Investigation; Supervision; Resources; Writing - review & editing

Corresponding author

Correspondence toMd. Karimul Islam.

Competing interests

The authors declare no competing interests.

Informed consent

All the participants provided the informed consent to participate in the study.

Peer review information

Communications Earth & Environment thanks Md Kamrul Hasan and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Primary Handling Editors: Martina Grecequet. A peer review file is available.

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visithttp://creativecommons.org/licenses/by/4.0/.

Reprints and permissions