Heterogeneous impacts of rainwater harvesting systems on income and food outcomes in northern Ghana

1 Introduction

Rain-fed agriculture remains the main source of livelihoods for smallholders in northern Ghana (Acheampong et al., 2018), yet, the region experiences pronounced rainfall variability, frequent dry spells, and increasing climatic instability (Abdallah, 2025a), leading to low crop yields, poor food self-sufficiency (hereinafter, FSS), volatile farm income (hereinafter, FI), and heightened food insecurity (Abdallah, 2025a; Williams et al., 2019). These challenges also intersect with a broader policy debate on whether Ghana should prioritize FSS—expanding domestic production of staples (MoFA, 2024b)—or emphasize food security (Dzanku and Sarpong, 2011). But while FSS is valued for reducing import dependence, and enhancing food sovereignty (Barker and Hayami, 1976; Clapp, 2017), food security allows for trade to fill supply gaps, but increase reliance on imports that can expose households to global price volatility (Barker and Hayami, 1976; Lele et al., 2016; Siderius et al., 2024). In northern Ghana, where both production deficits and access constraints are prevalent, a balanced strategy is needed—one that strengthens domestic production while improving household-level food access.

Agricultural water management technologies can allow for additional irrigated production in the absence of rain and increase the resilience in various seasons (Acheampong et al., 2018; IWMI, 2024). Among a range of water management technologies, ex-situ Rainwater Harvesting Systems (hereinafter, RWHS)—where water runoff is collected and stored outside the source—play an important role in production (Barnes, 2009; García-Ávila et al., 2023). These systems are the only techniques that directly addresses the water scarcity and the prolong drought from the inadequate rainfall distribution—mostly concentrated in one season, making water scarcity an issue for much of the year (Andoh et al., 2018; FAO, 2017; Mekuria et al., 2020). Ex-situ RWHS can help increase agricultural production, yields, FSS, and food security in Africa (Rockstrom et al., 2002; Siderius et al., 2024). As a result, ex-situ RWHS is therefore seen as a viable means of stabilizing yields, FSS and food security (Barry et al., 2008; FAO, 2014, 2025; Siderius et al., 2024).

In Ghana, government and NGOs have been promoting the establishment of ex-situ RWHS for harvesting of the rainwater for production and household use (Abdallah, 2025a; Namara et al., 2013). For instance, the Ghanaian government, through the National Rainwater Harvesting Strategy, provided a road-map to enhance planning, development and management of RWHS and to serve as a guide for greater use of rainwater harvesting in Ghana (MWRWH, 2011b). Further, a massive rainwater harvesting project has also been initiated by the government to enhance the availability and accessibility of water for human use in the country (Koranteng et al., 2024; Owusu and Obour, 2025). Collectively, the ex-situ RWHS include: (i) the conventional RWHS that have been globally accepted and utilized for many years and include dugouts/ponds, hand-dug wells, and community earth dams; and (ii) the emerging RWHS—a newly introduced rainwater harvesting facilities such as ferro-cement tanks, rooftop harvesting facilities and ponds that have been recently encouraged on the grounds that they have often proved effective for human use (Andoh et al., 2018; Koranteng et al., 2024; Resource Center Network Ghana, 2015). These infrastructural facilities have remained globally accepted and utilized on the grounds that they are relatively cheaper and require little technical knowledge for their implementation (Shikuku et al., 2020). However, the newly introduced rainwater harvesting facilities have often remained attractive on the grounds that they have the advantages of reducing seepage/evaporation losses or conveying high-quality water that can be utilized for dry-season production within a homestead setting. This heterogeneity in costs, water management, seasonal reliability, and potential for sustaining agricultural production could have some implications for use of RWHS, FI, FSS, and food security, yet the use, FI, FSS, and food security implications of each type remain insufficiently understood.

Although prior research demonstrates positive effects of RWHS on households in Africa (e.g., Abdallah, 2025a; Mekuria et al., 2020; Terêncio et al., 2018), most of these studies treat RWHS as a uniform intervention. This approach obscures important differences in performance across the diverse systems used by farmers. Ex-situ RWHS vary considerably in design, storage volume, maintenance, evaporation and seepage, and dependability during critical crop growth stages (FAO, 2014, 2017; Mekuria et al., 2020). These variations imply that livelihood outcomes associated with use of RWHS are unlikely to be homogeneous. Yet, empirical evidence comparing the relative effectiveness of different ex-situ RWH systems in Ghana remains limited. Therefore, an explicit comparison of conventional and emerging types of ex-situ RWHS is essential to inform policy and investment decisions. Moreover, various government organizations and NGOs have been promoting different types of RWHS. Some of these organizations include CWSA (2023), MWRWH (2011a), Resource Center Network Ghana (2015), Water Resource Commission (2012), and WaterAid-Ghana (2023). But in areas where persistent increased resource limitations and rising climatic stress are an ever-increasing concern, identifying and promoting the most beneficial RWHS is vital. This is especially true for northern Ghana where poor rainfall reliability and irrigational facilities are limiting small-scale farming populations. Nonetheless, an answer to the question of which RWHS is more beneficial is lacking in empirical literature. The lack of empirical evidence on the comparative benefits of different RWHS may constrain evidence-based policymaking, making it difficult for policymakers to identify and prioritize the most effective and context-appropriate systems for improving productivity, FI, FSS, food security, and rural livelihoods. Consequently, investments may be directed toward less effective technologies, resulting in inefficient allocation of scarce resources and reduced policy effectiveness.

These gaps will be addressed by the current research, focusing on an evaluation of the impact of alternative forms of ex-situ RWHS on FI, FSS, and food security of smallholder farmers through an empirical study of Ghana's north sector. Hence, the novelty and contribution of the current research stem from its stated purpose to compare alternatives of ex-situ RWHS, including traditional and innovative ones, as well as to empirically evaluate the varying impact of such alternatives on FSS, FI, and food security. By transcending the most common categorization of either use or non-use of RWHS, as conducted in most research, the current paper will shed more light on the varying implications of distinctly categorized system technologies on FI, FSS and food security. Moreover, the research includes the SRMFS as an instrument for crosschecking the consistency of most stated food security measures. This specificity aims to show the contribution, from an applied perspective to government agencies, as well as non-profit organizations, seeking guidelines on which water-harvesting techniques are most efficient, defined as those having the largest potential for supporting dry-season crops.

To achieve these objectives, the switching regression for multinomial treatment (SRMT) is combined with the correlated random effect (also known as the Mundlak's fixed effect). However, this approach is limited to only continuous outcomes and cannot be extended to the binary outcomes. As a result, impact of RWHS on FSS and SRMFS is analyzed by combining the correlated random effects with the multinomial endogenous treatment (MET) model. For proper identification, both methods require instrumental variables for identification in multiple treatment equations. However, this approach is also a weakness as the results may be sensitive to the instruments. Thus, with respect to the SRMT, the study checked for robustness of results using the multinomial propensity score (MNPS) methodology proposed by McCaffrey et al. (2013). For the MET, the study checked for robustness using the augmented inverse-probability weighting (AIPW) estimator (Cattaneo, 2010). Unlike the SRMT and MET, the MNPS and AIPW do not require instruments for identification. Even though these approaches still rely on the conditional independent assumption (CIA) and assumption of sufficient overlap or positivity, strong arguments have been advanced in literature to support their application (see for example, Imbens and Wooldridge, 2009; Jalan and Ravallion, 2003, etc.).

Apart from the introduction, the next section of the paper includes overview of RWH and systems in Ghana, conceptual framework, methods, results, and conclusions.

2 Overview of RWH and systems in Ghana

Rainwater harvesting (RWH) encompasses the collection, conveyance, and storing of rainfall runoff for productive purposes (Abdallah, 2025a). RWH can increase the reliability of water supply in seasonally dry environments and therefore constitutes an important adaptation option for semi-arid farming systems (Barnes, 2009; UN-HABITAT, 2005). The system for rainwater harvesting (i.e., RWHS) is commonly classified into in-situ RWHS—which enhance infiltration and soil water storage on the plot—and ex-situ RWHS—which capture runoff in structures outside the water source for later use (Rao et al., 2024). Of particular interest in this study is ex-situ RWHS where groundwater recharge is low or where stored surface water can be used for dry-season irrigation and homestead needs (Rao et al., 2024; Terêncio et al., 2018). According to the Development Technology Unit (DTU) (1999) of the University of Warwick, the basic elements of rainwater harvesting include catchment zone where rainwater can be collected, conveyance channels that help transport rainwater, first-flush diversion systems that can remove contaminants from collected rainwater, a storage structure where collected rainwater can be stored, and the delivery channel that can deliver the stored rainwater to its destination.

In Ghana, both government policy and development practice have promoted ex-situ RWHS as part of broader water-security and climate-resilience agendas. For example, national policies—including the Ghanaian policy for water (MWRWH, 2007), the strategy for harvesting of rainwater (MWRWH, 2011b) and the management plan for integrating water resources (Water Resource Commission, 2012)—lay out institutional support and scaling objectives, while multiple projects implemented by government agencies and NGOs have installed a variety of ex-situ RWHS in rural and peri-urban areas (Barnes, 2009; CWSA, 2023; Koranteng et al., 2024; WaterAid-Ghana, 2023). Despite these efforts, research have either emphasized in-situ measures (e.g., Issahaku and Abdulai, 2020a,b) or large community projects (e.g., Amoatey and Siddiqi, 2021; Koranteng et al., 2024; Owusu and Obour, 2025), leaving the diversity and performance of use of ex-situ systems less well-documented in the Ghanaian context.

Ex-situ RWHS observed and promoted in Ghana vary from low-technology, locally constructed options to more engineered and durable storage (Abdallah, 2025a; Andoh et al., 2018; Koranteng et al., 2024; Opare, 2012). Conventional and widely practiced systems include dugouts and small farm ponds, hand-dug wells, and community earth dams or reservoirs (Koranteng et al., 2024; MoFA, 2024a; Owusu and Obour, 2025). These systems are favored for their relatively low construction cost and the use of local materials and labor, but they can suffer large evaporation and seepage losses (EL-Molla and Kilit, 2025; FAO, 1986) and may require periodic desilting and maintenance (Barnes, 2009; Opare, 2012). Emerging ex-situ options include ferro-cement tanks, lined ponds, modular plastic tanks and rooftop collection systems; these systems generally reduce seepage and contamination risks and are better for homestead use and kitchen-garden irrigation, although they typically require higher upfront investment and technical support (Gyau-Boakye, 2001; Owusu and Asante, 2020).

Like any agricultural technology, adoption patterns across these systems are shaped by climatic, socioeconomic and institutional factors (Abdallah, 2025a; Abdallah et al., 2020). In the semi-arid north, dugouts and ponds remain the most common household/community options because they are affordable and can be built using local labor, whereas ferro-cement tanks and rooftop collection systems are more frequently seen where NGO support, subsidies, or project interventions reduce capital barriers (Andoh et al., 2018; Barnes, 2009; Gedamu et al., 2025; Owusu and Asante, 2020). Barriers to wider uptake of more durable systems include limited access to credit, technical capacity constraints, high initial costs, and concerns about long-term maintenance; these constraints mean that the choice of system often reflects trade-offs between cost, reliability, water quality and the intended use (Mbilinyi et al., 2005; Pachpute, 2010; Wekesa et al., 2022).

Empirical studies from elsewhere in Africa and Asia show that RWH can help improve yields and household welfare, but findings did not focused on system type, and scale or consider local conditions (Mekuria et al., 2020; Namara et al., 2013; Shikuku et al., 2020). In Ghana, relatively few studies (e.g., Abdallah and Ayamga, 2023; Abdallah, 2025a; etc.) examined the impacts of ex-situ RWH but not the system; most of these studies also treat RWH as a single category. Abdallah and Ayamga (2023), for example, examined the effect of RWH on livelihood outcomes in northern Ghana, using a dataset from the Ghana Agricultural Production Survey. The findings were mixed as RWH had positive and significant influence on yields, but negative influence on net returns and FSS in northern Ghana. Using survey data from 713 farm households in northern Ghana, Abdallah (2025a) also examined the impact of ex-situ RWH on FI and food security outcomes, revealing that ex-situ RWH had a 3.99% lower food insecurity access score, a 7.13% higher food consumption score, and a 47.8% higher FI. However, these studies only aggregated the impacts of ex-situ RWHS which can risk masking important differences in effectiveness and cost-effectiveness across RWHS. Consequently, policy and investment decisions can be made with limited evidence about which ex-situ RWHS delivers the greatest gains in FI and food security under Ghana's variable rainfall regimes. This evidence gap motivates comparative analyses that explicitly evaluate the performance of alternative ex-situ RWHS across smallholder settings.

3 Conceptual framework

Smallholders in northern Ghana operate under conditions of high rainfall variability, short rainy seasons, and limited reliable water sources (Madaki et al., 2025). The conceptual framework for this study, as shown in Figure 1, is grounded in the idea that different types of RWHS—conventional (dugouts, ponds, wells, and earth dams) and emerging (ferro-cement tanks, rooftop systems, and lined ponds)—affects household FI, FSS, and food security through several interconnected pathways. First, use of RWHS may increase the volume and reliability of water accessible to farmers throughout the production season (Alam et al., 2022). According to Atubiga et al. (2025), improved water availability reduces exposure to intra-seasonal dry spells and rainfall variability, which are major constraints in the semi-arid environment of northern Ghana. Reliable access to harvested water forms the foundation on which additional livelihood improvements occur (Lade and Oloke, 2020; Lepcha et al., 2024). This reduces crop failure risk, stabilizes yields, and encourages farmers to cultivate larger areas or engage in dry-season farming (Biazin et al., 2012). Second, improved water access enhances take-up of productivity-enhancing techniques and the production of high-value vegetables (Lepcha et al., 2024). More reliable water supplies help stabilize yields, support timely planting, reduce moisture stress, and improve overall productivity (Alam et al., 2022). At the same time, households with greater control over water can expand their production activities into the dry season, increase cropping intensity, and diversify into additional food and cash crops. These shifts in production practices contribute to higher total output and enable FSS (Tsegaye et al., 2025). Higher production and diversification also enhance FI, both through sales of surplus produce and through expanded opportunities for dry-season agriculture (Adams et al., 2025). Increased income, in turn, improves households' capacity to acquire food from the market, when necessary, invest in productivity-enhancing inputs, and smooth consumption across seasons (Lele et al., 2016; Maxwell et al., 2014). Together, the combined effects of improved water availability, higher production, increased cropping opportunities, income gains, and strengthened FSS enhance overall household food security (Abdallah, 2025b; Manyanga et al., 2025).

However, the expected magnitude of these effects can vary across RWHS. Conventional systems may offer larger but less efficient water storage because of high evaporation, seepage, seasonal unreliability, and water-quality issues (MoFA, 2024a; Zingiro et al., 2014). Emerging systems, by contrast, provide more efficient and reliable storage with lower losses (García-Ávila et al., 2023), although they often have smaller volumes and require higher capital investment (Liu et al., 2024). This implies that households using emerging systems may experience more consistent water supply, while those using conventional systems may face greater seasonal fluctuations. The framework, therefore, recognizes heterogeneous treatment effects; whereby different systems generate distinct levels of impact on income and food security. Specifically, this study conceptualizes heterogeneity in RWHS impacts along technological lines by distinguishing between conventional and emerging systems. The distinction is important because RWHS technologies differ substantially in design characteristics, storage capacity, reliability, investment requirements, and irrigation potential. Conventional RWHS generally involve relatively simple and low-cost structures with limited storage and seasonal functionality, whereas emerging RWHS comprises improved systems with greater water retention capacity, enhanced durability, and stronger potential to support supplemental and dry-season irrigation. These technological differences imply that the welfare impacts of RWHS are unlikely to be homogeneous across adopters. Rather than assuming a uniform treatment effect, the study therefore explicitly estimates subgroup-specific impacts associated with different RWHS categories.

Furthermore, RWHS use, much like the use of other water conserving technologies, is shaped by household characteristics such as resource endowment, labor availability, risk preferences, credit, and delivery of extension or NGO interventions (Foltz, 2003). These factors not only shape decisions but welfare outcomes and may create selection bias (Abdallah et al., 2020). In empirical specifications and analysis, these issues are considered.

Thus, overall, the framework posits that the type of ex-situ RWH system adopted influences water availability, production decisions, livelihood diversification, and ultimately FI, FSS, and food security. This study empirically tests these relationships by comparing the outcomes associated with alternative RWHS while accounting for potential selection effects and heterogeneity across households. To begin with, it is assumed that FI, FSS and food security are both related to RWHS and other covariates (C) expressed in Equation 1 as:

Mklt=αCklt+ρRWHSklt+eklt(1)

Where M_*kl* is an index of FI, FSS, or food security of householdin plot l a time t; C stands for other factors influencing M; RWHS_*kl* is system of rainwater harvesting; α and ρ are the respective coefficients; and is a random error term consisting of unobserved heterogeneity for household and the usual error terms of plot l in household k. Further, the effect of RWHS on M_*kl* is assumed to be ρ if RWHS is exogenously determined. However, a central empirical issue in estimating the effects of use of RWHS is the potential endogeneity of RWHS choice. Decision to use RWHS is unlikely to be random because households self-select into different systems based on both observable and unobservable characteristics. For example, unobserved household and plot attributes affecting the use of RWHS may also affect outcome M. Specifically, households with greater risk and managerial attributes, better plot quality, stronger social networks, or greater motivation may be more likely to use RWHS and may simultaneously achieve better income and food outcomes even in the absence of RWHS use. Similarly, agroecological conditions and institutional exposure may jointly influence both use decisions and welfare outcomes. Failure to account for these selection processes could therefore bias estimates of the impacts of RWHS.

To avert such problems in estimating the impact of RWHS, the SRMT, and MET models are each combined with the correlated random effect (CRE). The CRE specification controls for time-invariant unobserved heterogeneity correlated with explanatory variables, while the SRMT and MET frameworks explicitly model endogenous selection into alternative RWHS categories. Identification is strengthened using exclusion restrictions—distance to RWHS, perception of RWHS ease of use, and membership in water-user associations—which affect decisions to use RWHS but are not expected to directly affect household income and food outcomes conditional on status of RWHS use and other controls. Although other methods have been proposed for analyzing impacts of multiple technology (Binam et al., 2015; Heckman et al., 2001; Imbens and Wooldridge, 2009; McCaffrey et al., 2013; StataCorp, 2013), combination of the SRMT and MET with the CRE is more efficient in impact estimations as it accounts for both observe and unobserved heterogeneities (Abdallah et al., 2020; Kassie et al., 2008; Manda et al., 2016). Both models estimate choice of alternative RWHS (with special attention to the interdependencies of RWHS) as the first step, and the impacts of each choice on outcomes as the second step. The details of these methods are presented in the following sections. Further, the MNPS and AIPW are used to check for robustness of the results.

4 Methodology

4.1 Study area, sampling design, and data collection

Although the successful evaluation of program impacts depends heavily on the empirical methods employed, it also requires rich data on agricultural production, livelihood outcomes, and strategies to minimize potential spillover effects (Abdallah, 2025a). Thus, aside from the econometric used, two rounds panel data from household- and or plot-level surveys were employed for the study. Specifically, the panel data involved of 713 households (1,466 plots observations) in 2019/2020 and 2023/2024 cropping seasons. These households come from northern Ghana including the Northern, Upper East, and Upper West regions. This part of Ghana represents one of the most climate-vulnerable agricultural zones in the country and is characterized by high dependence on rainfed agriculture, recurrent droughts, erratic rainfall patterns, and limited irrigation infrastructure (Ahiamadia et al., 2024). Agriculture serves as the primary livelihood source for the majority of rural households, with crop production dominated by cereals such as maize, rice, millet, and sorghum, alongside legumes and groundnuts (File and Nhamo, 2023). The area falls largely within the Guinea and Sudan savannah agroecological zones, which are characterized by a unimodal rainfall regime (Owusu, 2018). Annual rainfall generally ranges between 800 mm and 1,200 mm but exhibits substantial temporal and spatial variability. The rainy season typically lasts from May to October, followed by a prolonged dry season from November to April (Davies et al., 2026). Increasing rainfall variability, delayed onset of rains, early cessation of rainfall, and more frequent dry spells have heightened production risks and contributed to unstable agricultural yields (Tahiru et al., 2026). Water scarcity constitutes a major constraint to agricultural production and rural livelihoods in northern Ghana (Alhassan et al., 2024). Most smallholder farmers rely mostly on rainfall for crop production, while access to formal irrigation infrastructure remains limited (Ahiamadia et al., 2024). During the long dry season, many communities experience severe water shortages that constrain agricultural activities, livestock production, and household welfare (Abdallah, 2025a). These challenges have increased interest in alternative water management strategies, including RWHS, which provide opportunities for supplementary irrigation, dry-season farming, and improved resilience to climatic shocks. These regions were selected because they are characterized by high rainfall variability, recurrent drought exposure, low irrigation coverage, and widespread use of rainfed agriculture, making them highly relevant for studying RWHS. The socio-economic environment is characterized by relatively high levels of rural poverty, limited access to agricultural credit, low levels of mechanization, and small farm sizes. Consequently, farm households remain highly vulnerable to weather-related shocks and fluctuations in agricultural production. In this context, RWHS have emerged as a potentially important strategy for improving agricultural productivity, stabilizing farm income, enhancing food security, and strengthening climate resilience among smallholder farmers (Ahiamadia et al., 2024; Tahiru et al., 2026).

A multistage sampling procedure was employed to select the study households. In the first stage, districts with significant agricultural activity and observable use of RWHS were identified within the study regions. In the second stage, farming communities were selected from the sampled districts. For each selected district, lists of communities with RWHS were obtained from the respective Food and Agricultural Ministry of Ghana. Using these lists, three (3) treated villages were chosen from each region relying on the number and concentration of existing RWHS. Selection was undertaken to ensure adequate representation of communities with varying levels of RWHS use and differing agroecological conditions. For each treated community, a corresponding non-treated community was also selected to serve as a control and to facilitate the identification of non-users. The control communities were chosen based on their geographical proximity to the treated communities, ensuring similarity in agroecological and socioeconomic conditions. In total, eighteen (18) communities—six (6) from each of the three regions—were included in the study. The distribution of these communities is presented in Figure 2.

In the final stage, treated and non-treated farm households were randomly selected from community household lists compiled with the assistance of officials of Ghana Statistical Service, local authorities, farmer groups, and agricultural extension officers. Although the sampled communities in stage two of the sampling process were enough for selection of control and treated households from the treated and non-treated communities, propensity scores of 3,418 households—across the sampled communities—were further generated from production data extracted from the 2010 census for population and housing for all households in the selected communities¹. The scores for the treated and untreated are shown in Figure 3. Comparing the scores obtained, non-users with similar scores to the users of RWHS were selected to serve as control and treated households. Specifically, a total sample of 720 households—consisting of 20 households from each village, and 360 RWHS users and 360 none RWHS users—were chosen randomly from the 3,418 households for the study. The sampling approach ensured representation of households that did not use RWHS, households using conventional RWHS, and households using emerging RWHS.

After sampling, primary data was collected through face-to-face interviews using a structured and pre-tested questionnaire administered by trained enumerators. The survey collected detailed information on household demographics, farm characteristics, agricultural production, income sources, climate-related shocks, cultivation, and institutional characteristics; village and community characteristics; asset ownership; use of RWHS; and food security. The information came from heads of the 720 sampled households (1,483 plot observations) during the 2019/2020 cropping season with the assistance of trained research enumerators. In 2023/2024, the same households were tracked and re-interviewed. To facilitate this process, extensive contact information, including secondary contacts and identifiers for household members, was collected during the initial survey. GPS coordinates recorded in 2019/2020 were also used to physically locate households, while community leaders and headmen supported the identification of households that had relocated within the study area. Multiple follow-up visits were undertaken for households that were temporarily unavailable during the main survey period. Despite these rigorous tracking efforts, the team was unable to interview seven heads who were either reported dead or migrated to Accra and Kumasi for trade. Thus, only 713 households (1,466 plot observations)—consisting of 353 households (715 plot observations) using none of the RWHS and 360 households (751 plot observations) using RWHS—were re-interviewed. The two rounds of data collection were then linked using unique household identifiers to construct a balanced panel dataset. To address concerns about representativeness, an attrition analysis was conducted in accordance with Baulch and Quisumbing (2011). The analysis revealed no statistically significant differences across key baseline indicators, suggesting that the panel remains representative of the original sample. The two-round panel design provides a strong basis for analyzing longitudinal changes in outcomes associated with RWHS use. The 2022–2025 timeline was intentionally selected to capture medium-term impacts and to enable robust estimation of trends while controlling for unobserved heterogeneity.

4.2 Variable measurement

To improve transparency and replicability, all variables used in the analysis are explicitly defined in Table 1, including measurement units and data sources where applicable. The study additionally combines household survey information with externally derived geospatial and market-level datasets, including WorldClim climate data, road network data, district-level fertilizer prices, and district-level wage rates. These variables are also presented in the following sections.

Table 1. Variable description/measurement and full sample descriptive statistics.

GH¢ is Ghana cedis. ^†The reference for fertility of the soil is poor soil fertility; ^§The reference for location of the farmer is Northern region. ^ΨLivestock conversion factor for cattle = 0.7, sheep = 0.1, goats = 0.1, pigs = 0.2, chicken = 0.01.

4.2.1 Rainwater harvesting system (RWHS)

The treatment variable is defined as a three-state multinomial indicator with mutually exclusive categories: the first category consists of households that did not use any ex-situ RWHS (T0) during the study period. The second category includes households using conventional RWHS (T₁), defined as relatively traditional and low-capacity systems primarily intended for short-term water storage and limited supplemental irrigation. The third category comprises households using emerging RWHS (T₂), which includes improved and relatively advanced systems characterized by larger storage capacity, enhanced durability, and stronger potential to support dry-season irrigation and intensified agricultural production. Both conventional RWHS and emerging RWHS systems are constructed outside the crop root zone and function by capturing water during rainy season. In Ghana's northern sector—where seasonal rainfall is high and intra-seasonal dry spells are frequent—most households use RWHS for enhancing availability of water, and stabilizing agricultural production (Barnes, 2009; CWSA, 2023; WaterAid-Ghana, 2023). This three-category specification is employed for two reasons: empirical prevalence and econometric identification. Empirically, many households in the sample reported either no RWHS (T_o), a conventional system only (T₁), or an emerging system only (T₂). Thus, this categorization captures both the presence and the technological form of RWHS. It also reflects the heterogeneity of water storage options available to farmers and allows the study to examine differences in livelihood outcomes across system types rather than treating water harvesting as a uniform intervention. Dual use (the simultaneous use of both a conventional and an emerging system) was rare in the study population; where it occurs it typically reflects a transition phase (e.g., a household that retains an old dugout while newly investing in a ferro-cement tank) rather than a long-term, distinct strategy. To preserve the multinomial modeling assumptions, households that reported more than one system were assigned to a single treatment state according to the system that is primarily used for agricultural production (the dominant production-use system) during the reference period. Econometrically, the three-state treatment structure satisfies the requirement that the treatment indicator be mutually exclusive and collectively exhaustive, a precondition for the SRMT and the MET models (Deb and Trivedi 2006; Bourguignon et al., 2007). Both SRMT and MET explicitly accommodate multinomial treatment choices and allow for selection into each treatment state to depend on observed and unobserved characteristics. SRMT permits the estimation of separate outcome equations conditional on each treatment state while accounting for endogeneity in the multinomial selection; MET likewise identifies average treatment effects for discrete (including binary) outcomes under a multinomial treatment regime. Thus, the chosen treatment definition aligns naturally with these econometric frameworks and enables estimation of distinct counterfactual outcomes for non-adopters, conventional-only users, and emerging-only users. With regards to use of RWHS, about 19% of the households produce using the emerging RWHS (T₂) while 32% produce using the conventional RWHS (T₁). On the other hand, about 50% do not use any of the systems (T_o) as shown in Table 2.

Table 2. RWHS applied by households.

RWHS include conventional RWHS and emerging RWHS.

4.2.2 Outcome variables

Farm income (FI) was used as the primary measure of household agricultural earnings. This variable was constructed as the total value of all farm outputs, minus the total cost of farm production. Because rural households derive income from multiple agricultural activities and often diversify to manage production risks (Abdallah, 2025a; Kato et al., 2011), the paper captured the monetary value of crops, livestock and livestock products, and earnings from agricultural labor provided to other farms. The corresponding production costs deducted comprised expenditures on purchased seeds, fertilizers, agrochemicals, feed, hired labor, machinery, and equipment services, transport, and other variable production-related expenses. All values were expressed in Ghanaian cedis (GH¢). To ensure comparability across households and communities, output values were estimated using the 2019 projected prices from the Food and Agriculture Ministry of Ghana. Although the use of fixed prices may introduce potential correlation with factors such as market proximity or input price variation, employing standardized prices avoids distortions arising from spatial and temporal price differences. The resulting net values were standardized on a per-hectare basis to improve comparability across households with different farm sizes. Higher FI values therefore indicate better agricultural economic performance and household welfare. The resulting net FI variable provides a consistent and economically meaningful indicator of agricultural profitability. Food security is a multidimensional concept encompassing food availability, access, utilization, and stability over time (Lele et al., 2016; Maxwell et al., 2014). Because no single indicator fully captures all dimensions of household food security and welfare, this study employs multiple complementary indicators following established food security literature (Magrini and Vigani, 2016). Specifically, the analysis uses food self-sufficiency (FSS), Food Consumption Score (FCS), Household Food Insecurity Access Scale (HFIAS), and self-reported measures of food security (SRMFS). Together, these indicators provide a broader assessment of household welfare by capturing objective production-based food availability, dietary diversity and food quality, experiential food insecurity, and households' own perceptions of food security conditions. The use of multiple indicators is particularly important in rural agricultural settings where households may simultaneously experience variations in food production, food access, diet quality, and subjective food adequacy. Accordingly, combining these indicators improves the robustness and comprehensiveness of the welfare analysis. Further, household food insecurity access score (HFIAS), and the food consumption score (FCS) are employed to capture the multidimensionality of food security. For HFIAS and FCS, the details can be found in Abdallah (2025b). The HFIAS measures households' experiences of food insecurity related to anxiety about food supply, insufficient food quality, and inadequate food quantity during a specified recall period. The index reflects the access dimension of food security and captures experiential aspects of deprivation. Higher HFIAS values indicate more severe food insecurity, while lower values indicate improved food access and welfare conditions. The FCS is a composite indicator developed by the World Food Program (WFP) that captures dietary diversity, food frequency, and the relative nutritional importance of different food groups consumed by the household over a reference period. Higher FCS values indicate better dietary quality and more diversified food consumption patterns, which are generally associated with improved household welfare and nutritional adequacy (Coates et al., 2007; WFP, 2008, 2009). The study also used a self-reported measure of food security (SRMFS), where household heads directly classify themselves as food secured or food insecure at varying levels. The self-reported measure of food security (SRMFS) captures households' subjective assessment of their overall food security status. Unlike objective consumption- or production-based indicators, SRMFS reflects households' own perceptions regarding the adequacy and stability of their food situation. This measure complements the other indicators by incorporating experiential and perception-based dimensions of welfare that may not be fully captured by quantitative food consumption or production measures alone. The SRMFS captures dimensions often missed by standard indicators—such as food preferences, perceived adequacy, and personal satisfaction with food access. Although self-reported measures can be prone to bias, the study ensured neutral phrasing of questions and avoided cues that might lead respondents to overstate or understate their level of food insecurity. Previous studies highlight the importance of such safeguards when using subjective food security assessments (e.g., Fawole and Ozkan, 2017; Magaña-Lemus et al., 2016; Ravallion and Lokshin, 2002). For household FSS, the study followed Jolly and Gadbois (1996) who defined FSS as:

FSS=TACPFU(2)

Where PFU in Equation 2 is population of family unit; TAC = CR + CMBC; TAC is total cereal available; CR is total cereal in a corn equivalent defined as:

CR=[(M×90100×97100)+(S×90100×97100) +(R×65100×99100) +(C×85100)] with M, S, R, and C representing produced millet, sorghum, rice and corn; 90100, and 85100 are the milling ratios respectively, 97100 and 99100 are the corn equivalence of processed M, S, and R. Further, CMBC is the cereal in corn equivalence bought from cotton, groundnuts, cowpea, and other legumes and defined as: CMBC=[(CT×PCT×85PM×100)+(GR×PGRPM+CW×PCWPM+OL×POLPM)×85100]; where CT, GR, CW, and OL are the kilogram of cotton, groundnuts, cowpea and other legumes produced in the 2018/2019 cropping season; PCT, PGR, PCW, and POL are the corresponding prices; and PM is the price per maize equivalent. Using the FAO-recommended annual requirement of 200 kg of processed cereal per capita as the benchmark for self-sufficiency (Jolly and Gadbois, 1996), a respondent is group as self-sufficient if processed cereal in corn equivalence is at least 200 kilogram and deficient if below 200 kilogram. FSS captures the household's capacity to meet minimum cereal requirements from its own agricultural production and related exchange activities (Abdallah et al., 2021). Thus, the indicator primarily reflects the food availability dimension of food security. A household is classified as food self-sufficient if annual processed cereal availability in maize-equivalent terms is at least 200 kilograms per capita, following FAO recommendations (Jolly and Gadbois, 1996). Higher values of the indicator therefore imply stronger household capacity to satisfy staple food requirements from own production sources.

4.2.3 Control variables

Following literature (e.g., Abdallah, 2025b), demographic attributes—such as age, gender, education level, household size, engagement in off-farm activities, receipt of remittances, asset holdings, livestock ownership, and access to credit, fertilizer use, labor input, farm size, soil fertility status, and prior exposure to drought on the plot—were captured during the household survey and included in the analysis to account for differences in resource endowments, labor availability, managerial capacity, liquidity, and wealth. These factors shape farm investment decisions and are therefore expected to affect both RWHS use and subsequently, income, SSF, and food security. To reflect broader economic and input market conditions, district-level fertilizer prices and agricultural wage rates were included. The wage rates are district-level agricultural wage rates constructed using the most recent nationally representative labor data from GLSS Rounds 6 (2012/13) and 7 (2016/17), the only comprehensive sources of agricultural wage information for Ghana. Hourly agricultural wages were derived from reported earnings and work hours of individuals engaged in agricultural wage labor and then averaged at the district level. These district-level rates were spatially matched to households based on district identifiers. Although the GLSS year do not coincide exactly with the 2022–2025 panel period, the wage data serve as valid proxies for local labor market conditions because agricultural wages in rural Ghana adjust slowly and exhibit persistent spatial variation over time (Abdulai and Delgado, 2000). Similar proxy approaches are widely applied in Ghanaian agricultural household models where contemporaneous wage information is missing (Abdulai and Delgado, 1999, 2000). Thus, the GLSS-based wage variable can provide a consistent and econometrically defensible measure of district-level labor opportunity costs. District-level fertilizer prices were constructed using retail price data obtained from the AfricaFertilizer price monitoring platform for Ghana (https://viz.africafertilizer.org). The platform compiles fertilizer price information from regular market surveys conducted in collaboration with national partners, reporting prices by fertilizer type, location, and period. For this study, price observations for NPK 20–10–10 (GH¢/kg)—the most widely used compound fertilizer in northern Ghana (Adzawla et al., 2021)—were extracted. Monthly retail prices reported for markets within each district were first converted to a common unit (GH¢per kilogram where necessary). These observations were then averaged across markets within the same district to obtain a district-level mean fertilizer price. Where multiple monthly observations were available within a year, prices were averaged over the relevant period to smooth short-term fluctuations and capture prevailing local input market conditions. These district-level prices serve as a credible proxy for local input market conditions, consistent with standard practice in agricultural household models where contemporaneous, farm-level price data are unavailable (Abdallah, 2025b).

Additionally, GIS coordinates were collected for each household, allowing the survey information to be supplemented with geospatial variables derived from external spatial datasets. For instance, long-term climatic conditions were controlled for using spatially referenced temperature and precipitation data. Specifically, the study employed average minimum and maximum temperatures and long-term mean precipitation derived from the WorldClim version 2 dataset, which provides high-resolution global climate surfaces at a 5 km resolution (Fick and Hijmans, 2017). These climatic variables represent multi-decadal averages based on records from 1950 to 2024 and, therefore, capture the underlying agroecological and biophysical environment within which households operate. The long-term climate variables were preferred over short-term weather observations because RWHS use responds to persistent climatic conditions rather than year-specific weather shocks (Abdallah, 2025a; Roba et al., 2022). The multi-decadal WorldClim averages (1950–2024) can capture the structural agroecological setting more reliably than annual weather data for 2019/2020 and 2023/2024. Using long-term climatic surfaces may also avoid the endogeneity that arises when short-term weather shocks simultaneously influence both RWHS use and livelihood outcomes. The climate data for each sampled community was extracted using QGIS by overlaying household geographic coordinates on the WorldClim raster layers, ensuring precise matching of local climatic conditions to each observation. Geographical accessibility was accounted for using measures of distance to major roads and the proximity of plots to existing RWHS. These variables proxy for market access, transport constraints, and ease of using water from nearby storage facilities. Institutional and informational factors were captured through household knowledge of RWHS, frequency of extension contacts, farmer perceptions of RWHS ease of use, and membership in water-user associations. These factors influence awareness, technical support, collective action dynamics, and the perceived feasibility of RWHS use. Regional dummy variables for the Northern, Upper East and West regions were included to capture unobserved spatial heterogeneity in agroecological conditions, infrastructure, and institutional environments. Table 1 presents the descriptive statistics of all variables used in the analysis. For each variable, we report the number of observations, minimum and maximum values, mean, and standard deviation. The sample consists of 713 farm households. Farm income averaged GH¢499.90 per hectare, while approximately 34% of households were food self-sufficient and 29% reported being food secure. The average household head was 49 years old and had approximately 4 years of formal education. The mean household size was about six members, and 58% of households were males. These statistics indicate substantial variation across households in terms of socioeconomic characteristics, production conditions, and food-security outcomes.

Table 3 presents treatment-specific descriptive statistics and mean-difference tests between non-users of RWHS, users of conventional RWHS, and users of emerging RWHS. The results reveal substantial heterogeneity across the three groups, suggesting that use of RWHS is not randomly distributed among farm households. Compared with non-users, households using conventional RWHS earn, on average, GH¢31.79 more FI per hectare, while users of emerging RWHS earn approximately GH¢298.69 more per hectare. The latter difference is nearly ten times larger than that observed for conventional RWHS users, suggesting that emerging RWHS generate considerably greater economic benefits. FI among emerging RWHS users is approximately 79% higher than that of non-users (GH¢677.89 vs. GH¢379.20 per hectare). Similarly, food self-sufficiency is significantly higher among adopters. While only 31% of non-users are food self-sufficient, the proportion increases to 38% among conventional RWHS users and 48% among emerging RWHS users. This represents increases of 7 and 17 percentage points, respectively. The larger improvement among emerging RWHS adopters suggests that more advanced harvesting systems may provide more reliable water supplies for crop production and household food provisioning. Food insecurity indicators also reveal important differences. The HFIAS declines from 4.64 among non-users to 4.03 among conventional RWHS users and 3.67 among emerging RWHS users. Since lower HFIAS scores indicate better food security, these results suggest that use of RWHS is associated with reduced food insecurity. Likewise, SRMFS improves substantially, increasing from 29% among non-users to 38% and 48% among conventional and emerging RWHS users, respectively.

Table 3. Treatment-specific descriptive statistics and mean difference tests between RWHS users and non-users.

T0 represents non-users of any of the RWHS, T1 represents users of conventional RWHS only and T2 represents users of emerging RWHS only; GH¢ is Ghana cedis; ^†The reference for fertility of the soil is poor soil fertility; ^§The reference for location of the farmer is Northern region. ^ΨLivestock conversion factor for cattle = 0.7, sheep = 0.1, goats = 0.1, pigs = 0.2, chicken = 0.01; Significance levels: ^***p < 0.01, ^**p < 0.05, ^*p < 0.10.

The results also reveal that conventional RWHS users are substantially older than non-users, with an average age of approximately 59 years compared with 39 years among non-users. Emerging RWHS users are also older than non-users, although the difference is smaller (approximately 11 years). Older farmers may possess greater farming experience, stronger social networks, and accumulated resources that facilitate investment in water harvesting technologies. Educational attainment also differs markedly across groups. Non-users have, on average, only 3.18 years of formal education, compared with 5.61 years among conventional RWHS adopters and 6.50 years among emerging RWHS adopters. The differences of 2.43 and 3.32 years, respectively, are highly significant. These findings suggest that education may facilitate access to information, enhance understanding of technology benefits, and improve the ability to implement and manage RWHS effectively. Further, RWHS users generally possess larger landholdings than non-users. Average farm size increases from 2.67 ha among non-users to 3.69 ha among conventional RWHS users and 3.62 ha among emerging RWHS users. The differences of approximately one hectare are statistically significant and economically meaningful, representing increases of approximately 38% and 36%, respectively. The larger farm sizes observed among users suggest that households with greater productive resources may be more capable of investing in water harvesting technologies. Such differences provide preliminary evidence of potential selection effects, whereby relatively better-endowed households are more likely to adopt RWHS. Credit access also differs substantially. While only 14% of non-users report access to credit, the corresponding figure reaches 25% among emerging RWHS users. The 11-percentage-point difference is statistically significant and suggests that liquidity constraints may influence the use of more advanced RWHS.

Membership in WUAs increases from only 9% among non-users to 23% among conventional RWHS users and 37% among emerging RWHS users. The differences of 14 and 28 percentage points, respectively, are highly significant. This pattern suggests that participation in local institutions may facilitate information sharing, collective action, and access to water-related infrastructure. Similarly, perceptions regarding the ease of implementing RWHS are significantly more favorable among users of emerging RWHS adopters. While only 10% of non-users perceive RWHS as easy to implement, the proportion rises to 21% among emerging RWHS users. Positive perceptions may therefore play an important role in influencing technology use decisions.

Non-users are located, on average, 55.61 meters from major roads, compared with 45.59 meters for conventional RWHS users and 28.89 meters for emerging RWHS users. Similarly, users are located significantly closer to RWHS infrastructure. The average distance to RWHS facilities decreases from 6.18 meters among non-users to 2.08 meters among conventional adopters and 3.09 meters among emerging adopters. These findings suggest that physical access to infrastructure may reduce transaction costs associated with technology use and improve opportunities for agricultural intensification.

Further, conventional RWHS users are more likely to cultivate plots with good soil fertility (84%) than non-users (66%). Regional differences are also pronounced. Conventional RWHS use is concentrated in the Upper West Region, whereas emerging RWHS use is more prevalent in the Upper East Region. These spatial differences may reflect variations in climatic conditions, water availability, infrastructure development, and institutional support across regions. Differences in rainfall and temperature variables further suggest that agroecological conditions influence both the feasibility and attractiveness of RWHS use. Taken together, the descriptive statistics reveal systematic differences between users and non-users across demographic, economic, institutional, locational, and agroecological characteristics. Such differences indicate that adoption decisions are unlikely to be random. Consequently, simple comparisons of outcomes between users and non-users may confound the effects of RWHS use with pre-existing household characteristics. These findings therefore provide justification for employing econometric approaches that address endogenous technology choice when estimating the impacts of conventional and emerging RWHS.

4.3 Estimation strategies

4.3.1 Identification strategy and assumptions

The empirical identification strategy seeks to estimate the causal effects of alternative RWHS on farm income, food self-sufficiency, and food security while addressing both observable and unobservable selection bias. Because households voluntarily choose whether to use conventional or emerging RWHS, simple comparisons between users and non-users may produce biased estimates. In particular, unobserved factors such as managerial ability, entrepreneurial orientation, risk preferences, soil characteristics, and social connectedness may simultaneously influence both adoption decisions and welfare outcomes. To address these concerns, the study employs SRMT and MET models combined with correlated random effects (CRE). These models jointly estimate the selection equation and the outcome equations, thereby accounting for endogenous sorting into alternative RWHS categories (Deb and Trivedi, 2006; Bourguignon et al., 2007; Dubin and Mcfadden, 1984). The CRE specification additionally controls for time-invariant unobserved heterogeneity by allowing unobserved effects to correlate with the time averages of explanatory variables. Identification of the SRMT and MET models relies on two main conditions. First, the exclusion restrictions used in the treatment equations—distance to RWHS, perception of RWHS ease of use, and membership in water-user associations—must affect adoption decisions but not directly influence the welfare outcomes after conditioning on observed covariates. Second, the models assume sufficient variation and overlap in observable characteristics across treatment groups to allow meaningful counterfactual comparisons.

For robustness, the study additionally applies MNPS weighting and AIPW estimators. These approaches rely on the conditional independence assumption (CIA), which states that conditional on observed covariates, treatment assignment is independent of potential outcomes. The estimators also require the overlap (positivity) assumption, implying that households with similar observed characteristics have a positive probability of belonging to each treatment category. The rich household, plot, institutional, climatic, market, and geospatial covariates included in the analysis are intended to make these assumptions more plausible. To further assess the credibility of the exclusion restrictions, falsification tests were implemented using the subsample of non-adopters. Since these households do not use RWHS, valid instruments should not directly predict their welfare outcomes. The results show that the instruments are jointly insignificant in the non-adopter outcome regressions, supporting the validity of the exclusion restrictions (see Supplementary Table S1 of the Supplementary material for this paper).

4.3.2 Switching regression for multinomial treatment (SRMT)

As pointed out already, calculating alternative RWHS impact on farmer income and food security indicators (i.e., FCS and HFIAS) through the SRMT is two stages. First, the head's use of RWHS is determined. Second, the farmer income and food security impact of heads' choice of RWHS on then calculated. To begin with the first stage, it is assumed that household's decision regarding the use of RWHS follows the random utility model (RUM). According to the RUM, when a household i is presented with several packages of RWHS, the household will prefer package k over l if V_*kl*(pi)>V_*km*(pi) and m ≠ 1 in all cases. This is conditioned on assumption that the household is risk neutral. However, the net utility RWHSkl*(π) derived from choosing package k is a latent variable linked to observable household/plot and choice attributes and means of household/plot attributes (A) stipulated as:

RWHSkl*=γlA+εkl(3)

Assuming RWHS is the household's choice of RWHS, Equation 3 translates into the below equation for the observed the choice of the head:

RWHS={and 1 if Vk1∗(π)>maxm≠l[Vkm∗(π)] or ηk1<0and:: :: :: ::andL if Vkl∗(π)>max[Vkm∗(πm≠l)] or ηkl<0 m#l for all(4)

In Equation 4, ηkl=maxm≠l[Vkm(π)-Vkl(π)]<0 (Bourguignon et al., 2007) and this means a household will select l to maximize expected utility if ηkl=maxm≠l[Vkl(π)-Vkm(π)]>0.

If ε_*kl* is identical and independently Gumbel distributed in Equation 3, the likelihood of a household to adopt choice l is indicated by the MNL:

Pr(ηkl<0|Ak )=exp(γlAk) ∑i=1Lexp(γiAm)(5)

The parameters of the latent variable model are then determined through maximum likelihood estimation. For stage two of the SRMT, the FI, and food security (i.e., FCS and HFIAS) determinants among users of each RWHS is estimated using a set of exogenous variables H (Bourguignon et al., 2007; Dubin and Mcfadden, 1984). The choices are non-usage of any of the RWHS (reference group), usage of only conventional RWHS and usage of only emerging RWHS, that is, l = 0, 1, and 2, respectively. The FI and food security indicators (I) for each category (i.e., l = 0, 1, 2) is:

{and Category 1: Ik1=θ1Hk+μk1 if RWHS=1and:: :: :: ::andCategory L: IkL=θLHk+μkL if RWHS=L (6)

where Ikl′s are the FI and food security (i.e., FCS and HFIAS) for k^*th* head in category l, and the errors μkl′s with E(μ_*kl*|A, H) = 0 and var(μkl|A,H )=σl2. Ikl′s are observed if a possible choice is adopted (i.e., Vkl>max(Vkm)m≠l. If a correlation exists between εkl′s and μkl′s, the coefficients θk′s from Equation 6 will be inconsistent if estimated with OLS. According to Bourguignon et al. (2007), estimating the following SRMT models yields better θk′s:

{andRegime 1: Ik1=θ1Hk+τ1ω^1+υk1 if RWHS=1and:: :: :: ::andRegime L: IkL=θLXk+τLω^L+υkL if RWHS=J (7)

In Equations 6, 7, H include households/plot and means of household/plot varying explanatory variables; τ_*l* represents covariation between εkl′s and μkl′s; υkl′s are zero-mean error terms; and ω^j is the inverse Mills ratio deduced via probabilities (R) in Equation 5 as using the following equation:

ωl=∑m≠lLρl[ln(R^km) 1-R^km+ln(R^kl)](8)

where ρ in Equation 8 stands for the coefficient of correlation between εij′s and uij′s. The inclusion of the means of household/plot varying explanatory variables in Equation 7 is the CRE or the Mundlak's fixed effect. Mundlak's fixed effect assumes that time-invariant unobserved heterogeneity as captured in A of Equation 3, and H of Equations 6, 7 is a linear function of the averages of the time and plot varying explanatory variables (Mundlak, 1978). To manage the heteroscedasticity due to the two-step method of estimation associated with SRMT, standard errors in Equation 7 are obtained via bootstrapping. Equation 7 was identified by including distance of farm to RWHS, perception of RWHS and membership to WUAs in the MNL model but excluding them from Equation 7. Even though these instruments are intuitively strong, simple falsification test was performed following literature (e.g., Abdallah, 2025a). The logic of this test is that valid instruments should affect the outcomes only through RWHS use and should have no direct effect on outcomes among households that did not adopt any RWHS. Accordingly, distance to RWHS, perception of RWHS, and WUA membership, were regressed on the FI, FSS, FCS, HFIAS, and SRMFS using a subsample of non-adopters of conventional and emerging RWHS only. Since RWHS use does not occur within this group, any statistically significant relationship between the instruments and the outcomes would cast doubt on the exclusion restriction. The falsification results (Supplementary Table S1 of the supplementary information to this paper) show that none of the instruments are statistically significant predictors of the outcome variables among non-users, supporting the idea that the instruments influence FI, self-sufficiency and food security outcomes only through their effect on RWHS use, lending further credibility to their exogeneity. Other studies (e.g., Abdallah, 2025a) used similar variables as instruments. Following Carter and Milon (2005), the outcome of users and non-users are presented as:

User with use of RWHS (actual):

E[Ikl|RWHS=l,Hkl, ω^kl] =θlHkl+τlω^kl(9)

Non-user without the use of RWHS (actual):

E[Ik1|RWHS=1,Hk1, ω^k1] =θ1Hk1+τ1ω^k1(10)

User had he/she decided to be non-user of RWHS (counterfactual):

E[Ik1|RWHS=l,Hkl, ω^kl] =θlHkl+τ1ω^ij(11)

Non-user had he/she decided to be user of RWHS (counterfactual):

E[Ikl|RWHS=1,Hk1, ω^k1] =θlHk1+τlω^k1(12)

From the above expressions, Equation 9 less Equation 11 is the average treatment effect on the treated (ATT) while Equation 12 less Equation 10 is the average treatment effect on the untreated (ATU).

4.2.3 Multinomial endogenous treatment (MET) model

In line with Deb and Trivedi (2006), this study examined the effect endogenous multinomial treatment in two steps. To start with, let IUkl* be the indirect utility obtained l^*th*system, where, and IUkl* is expressed as:

IUkl*=ϖlxk′+ψlfkl+vkl(13)

where x_*k* in Equation 13 is exogenous variables; f_*k* is latent variable indexing unobserved attributes related to choice of household and the FI and food security indicators but unrelated to and error term ν_*kl* which is identical and independently distributed; ϖ_*l* and ψ_*l* are the parameters.

Denote β_*i* as a bundle of RWHS available to household represented as with b_*l* as dichotomous covariate indexing the observable choice of RWHS. Also, letF_*k* be a latent factor incorporating unobserved attributes with household k′s choice of RWHS and outcome, such that and is the unobserved attribute for a system. Then the first stage probability estimates for choosing any of the systems is defined in Equation 14 as:

Pr(βk|xk,Fk)=h(ϖ1xk+ψ1fk1+ϖ2xk+ψ2fk2,⋯,ϖLxk+ψLfkL)(14)

Where x_*k* includes household/plot and choice attributes and as well as he Mundlak's fixed effect (i.e., means of household/plot); ϖ_*l* and are as defined; and ν_*kl* is independent and identically distributed errors assumed to have no relationship with f_*kl*; and l = 0 is the reference group (non-users of RWHS). Additionally, h is a function with multinomial probability distribution with a MNL structure (Deb and Trivedi, 2006) which is expressed in Equation 15 as:

Pr(Bk|zk,Fk)=exp(ϖlxk+δllkl)1+∑k=1Lexp(ϖkxl+ψkfkl)(15)

The second stage evaluates food security impact of RWHS as:

E(IK|Bk,Xk,Fk)=αZk+∑l=1Lϑlbkl+∑l=1Lφlfkl(16)

In Equation 16, Z_*k* stands for the exogenous variables and means of household/plot attributes with α as the associated parameter; b_*kl* and f_*kl* are defined already with ϑ_*l* denoting a system's effect relative to the control; ϵ(I_*k*|B_*k*, Z_*k*, F_*k*) is a function of the latent factor f_*kl* indicating that unobserved attributes affecting selection into RWHS also affect outcomes (i.e., FSS and SRMFS). Per Deb and Trivedi (2006), when the parameter, is positive (negative), choice of RWHS and outcome are positively (negatively) related via unobserved attributes, and this means positive (negative) selection with the parameters and φ_*l*, respectively. For the estimation, it is worth assuming a function for the outcomes. In this study, the outcomes' function is logistic for the dependent variables (i.e., FSS and SRMFS) since they are binary variables (Deb and Trivedi, 2006). In addition, the Maximum Simulated Likelihood (MSL) approach was utilized for the estimation. For identification in the estimation of the outcome equations, the variables used as instruments in the SRMT model (i.e., distance to RWHS, perception of RWHS, and WUA membership) are the same variables used in the MET model as instruments. Although the model is already identified, addition of these variable as instruments in x_*k* and excluding them in is preferable (Deb and Trivedi, 2006).

5 Results and discussions

This section presents only the results of ex-situ RWHS impact on FI, FSS and food security. Section 5.1 presents the impact of the ex-situ RWHS on FI; Section 5.2 presents the impact of the ex-situ RWHS on FSS; and section 5.3 presents the impact of the ex-situ RWHS on food security. For want of space, the full model results involving determinants of use of RWHS, FI, FSS and food security indicators are only presented in Supplementary Tables S2, S3 of the Supplementary material to this study. The empirical analysis in each section focuses on heterogeneous treatment effects across alternative RWHS categories rather than a single average treatment effect. Accordingly, the estimated impacts should be interpreted as subgroup-specific effects comparing users of conventional RWHS and users of emerging RWHS relative to non-users. This distinction is important because the two technology groups differ in technical sophistication, storage capacity, reliability of water supply, and capacity to support intensified agricultural production. Consequently, differences in estimated impacts across the two categories provide insights into how technology characteristics shape livelihood outcomes.

5.1 Farm income effect

Table 4 presents the SRMT estimates of the impacts of RWHS use on FI, accounting for both observed and unobserved factors influencing RWHS choices. The results indicate that use of both conventional and emerging RWHS significantly increases FI relative to non-use, although the magnitude of the gains differs substantially across RWHS. Households using conventional RWHS earn GH¢360.45 per hectare compared with a counterfactual FI of GH¢310.20 per hectare in the absence of use, yielding an average treatment effect on the treated (ATT) of GH¢50.25 per hectare. This represents an FI increase of approximately 16.2% compared to the counterfactual FI level. In contrast, households using emerging RWHS obtain an income gain of GH¢160.70 per hectare, representing an increase of about 44.6% relative to the counterfactual income of GH¢360.10 per hectare.

Table 4. SRMT based effects of RWHS on FI.

All models control for household demographics, farm characteristics, institutional variables, climate variables, soil fertility indicators, and regional fixed effects. l represents rainwater harvesting systems defined in Table 2. Standard errors in parenthesis; ^***P < 0.01.

The substantially larger income gains associated with emerging RWHS suggest that technological improvements in water harvesting systems generate important productivity advantages beyond those achieved through conventional systems. Emerging RWHS typically provides greater water storage capacity, improved water retention, and more reliable access to supplementary irrigation during critical crop growth stages (Gyau-Boakye, 2001; Owusu and Asante, 2020). These features are particularly important in northern Ghana, where agricultural production is highly dependent on increasingly erratic rainfall patterns and recurrent dry spells (Abdallah, 2025a). By reducing moisture stress and stabilizing crop production, emerging RWHS enable farmers to achieve higher yields and potentially cultivate higher-value crops, thereby translating water security into improved economic outcomes. The magnitude of the estimated impacts is economically meaningful given the relatively low-income levels observed among non-adopters. An additional GH¢160.70 per hectare represents a substantial increase in farm earnings for smallholder households whose livelihoods largely depend on agriculture. Such gains can improve households' capacity to purchase agricultural inputs, smooth consumption during lean periods, and invest in productive assets, thereby strengthening resilience to climatic and economic shocks.

The robustness estimates reported in Table 5 further reinforce these findings. Although the MNPS estimates are larger than those obtained from the SRMT model, both approaches consistently indicate positive and statistically significant effects of RWHS use on farm income. The larger MNPS estimates likely reflect the fact that the method accounts only for observable selection, whereas the SRMT framework additionally controls unobserved factors affecting both adoption decisions and income outcomes. The consistency in the direction and significance of the effects across the two identification strategies strengthens confidence in the causal interpretation of the results. These findings are consistent with previous studies that document positive income effects of rainwater harvesting technologies and supplementary irrigation systems among smallholder farmers (Amani and Rezaee, 2025; Andualem et al., 2020; Shokati et al., 2019). However, the present study extends the literature by demonstrating that the economic returns vary considerably across RWHS types. The results therefore suggest that treating all rainwater harvesting technologies as a homogeneous intervention may conceal important differences in effectiveness. From a policy perspective, investments that promote the use and scaling-up of emerging RWHS may generate substantially larger livelihood benefits than interventions focusing exclusively on conventional systems. Nonetheless, the results generally concur with past studies (e.g., Amani and Rezaee, 2025; Andualem et al., 2020; Shokati et al., 2019) that found RWHS to be beneficial to users.

Table 5. MNPS model estimates of the impacts of RWHS on FI.

All models control for household demographics, farm characteristics, institutional variables, climate variables, soil fertility indicators, and regional fixed effects. LSE is Linearized Standard Error; and ^***P < 0.01.

5.2 Food self-sufficiency effect

Tables 6, 7 present the estimated effects of RWHS use on FSS using the MET and AIPW approaches, respectively. The results consistently indicate that use of RWHS significantly improves households' ability to meet their staple food requirements from their own production. According to the MET estimates, conventional RWHS increases the probability of being FSS by approximately two percentage points, while emerging RWHS increases the probability by about 8 percentage points. Relative to the average FSS rate of 31% among non-users, these effects represent improvements of approximately 6.5% and 25.8%, respectively. The larger effect associated with emerging RWHS highlights the importance of technology quality and performance in shaping food security outcomes. By providing more reliable access to water during periods of rainfall shortages, emerging RWHS help stabilize crop production and reduce the likelihood of harvest shortfalls. This is particularly relevant in Ghana, where household food availability is closely tied to the performance of rainfed agriculture and where recurrent rainfall variability frequently undermines food production (Acheampong et al., 2018). The negative and statistically significant selection correction terms indicate that households using RWHS are negatively selected with respect to unobserved factors associated with FSS. This suggests that users may face constraints that would otherwise reduce their FSS prospects. Consequently, failure to account for endogenous adoption decisions would underestimate the true contribution of RWHS to household food availability.

Table 6. MET estimates of the impacts of RWHS on self-sufficiency.

All models control for household demographics, farm characteristics, institutional variables, climate variables, soil fertility indicators, and regional fixed effects. Robust standard errors in parentheses; ^***p < 0.01, and ^**p < 0.05.

Table 7. AIPW-based treatment effect of m (type of RWHS) relative to treatment l (non-users).

All models control for household demographics, farm characteristics, institutional variables, climate variables, soil fertility indicators, and regional fixed effects. Robust standard errors in parentheses; ^***p < 0.01, and ^**p < 0.05.

The AIPW estimates reported in Table 7 closely mirror the MET results, showing increases of 1.5 and 6.1 percentage points for conventional and emerging RWHS, respectively. The consistency of findings across the two estimation approaches provides strong evidence that the positive effects are not driven by model-specific assumptions. Overall, the results suggest that RWHS constitutes an important climate adaptation strategy capable of enhancing household food availability and reducing vulnerability to food shortages. The findings also support previous evidence showing that improved water management technologies contribute significantly to FSS among smallholder farmers (e.g., Hume et al., 2022).

5.3 Food security effects

The food security indicators used in this study capture distinct but complementary welfare dimensions. Improvements in FCS indicate enhanced dietary diversity and food quality (Abdallah, 2025c), reductions in HFIAS reflect improved access to food and lower food-related deprivation (Lele et al., 2016; Maxwell et al., 2014), while increases in SRMFS indicate improved household perceptions of food adequacy and security assessments (Fawole and Ozkan, 2017). Interpreting these indicators jointly therefore provides a more comprehensive assessment of the welfare implications of RWHS use than reliance on a single measure alone. The results presented in Tables 8–11 demonstrate that use of RWHS improves household food security across multiple dimensions, including dietary quality and diversity, food access, and households' own perceptions of food adequacy. Considering these indicators jointly provides a more comprehensive understanding of welfare impacts than reliance on a single food security measure. The SRMT estimates in Table 8 indicate that use of RWHS significantly improves food consumption outcomes and reduces food insecurity. For FCS, for example, households using conventional RWHS record an average score of about 50.1, compared with 44.8 of non-users, implying a statistically significant gain of 5.3 points attributable to use of RWHS. The effect is even stronger for emerging RWHS, where users achieve an average FCS of 52.4 relative to 45.9 for non-users, corresponding to a gain of 6.5 points. These magnitudes suggest that RWHS use enhances dietary diversity, caloric adequacy, and overall diet quality. The larger effect associated with emerging RWHS is consistent with their higher storage capacity and greater reliability, which likely allow households to smooth production across seasons and allocate more resources to diverse food consumption. A similar pattern emerges for the HFIAS, where lower values indicate better food security. Conventional RWHS reduces the HFIAS by about 1.4 points relative to non-users, while emerging RWHS reduces it by approximately 1.85 points. These reductions imply fewer experiences of food-related anxiety, compromised food quality, or reduced food quantity among adopting households. Notably, emerging RWHS again outperform conventional systems, reinforcing the idea that technology characteristics matter for food security outcomes (e.g., Abdallah et al., 2020, 2021).

Table 8. SRMT based treatment effects of use of RWHS on food security.

All models control for household demographics, farm characteristics, institutional variables, climate variables, soil fertility indicators, and regional fixed effects. j represents rainwater harvesting systems defined in Table 2. Standard errors in parenthesis; ^***P < 0.01.

Table 9 presents MNPS-based estimates for FCS and HFIAS as a robustness check, and the results closely align with the SRMT findings. Conventional RWHS increases FCS by about 4.85 points and reduces HFIAS by 1.18 points, while emerging RWHS increases FCS by 6.92 points and reduces HFIAS by about 1.74 points. The consistency in sign, relative magnitude, and statistical significance across SRMT and MNPS estimates strengthens confidence that the observed food security gains are not driven by model-specific assumptions. Instead, they reflect a robust association between RWHS and improved food consumption and access outcomes.

Table 9. MNPS model estimates of the impacts of RWHS on food security.

All models control for household demographics, farm characteristics, institutional variables, climate variables, soil fertility indicators, and regional fixed effects. LSE is Linearized Standard Error; and ^***P < 0.01.

Beyond consumption-based indicators, Tables 10, 11 extend the analysis to SRMFS, capturing households' own assessments of their food situation. The MET estimates in Table 9 show that conventional RWHS use increases the probability of reporting food security by about three percentage points, while emerging RWHS increases it by about eight percentage points. The significance of the selection correction term for emerging RWHS suggests non-random selection into this technology, implying that failing to address endogeneity would underestimate its true effect on perceived food security.

Table 10. MET model estimates of the impacts of RWHS on food security.

All models control for household demographics, farm characteristics, institutional variables, climate variables, soil fertility indicators, and regional fixed effects. Robust standard errors in parentheses; ^***p < 0.01, and ^*p < 0.1.

Table 11. AIPW-based treatment effect of m (type of RWHS) relative to treatment l (non-users).

All models control for household demographics, farm characteristics, institutional variables, climate variables, soil fertility indicators, and regional fixed effects. Robust standard errors in parentheses; ^***p < 0.01, and ^*p < 0.1.

The AIPW estimates in Table 11 confirm these results, indicating increases of about 2.1 percentage points for conventional RWHS and 6.7 percentage points for emerging RWHS relative to non-users. Although slightly smaller in magnitude than the MET estimates, the AIPW results preserve the same ranking and statistical significance, providing strong robustness support. Thus, across all four tables, the results demonstrate that ex-situ RWHS use significantly improves household food security along multiple dimensions—diet quality and diversity (FCS), experiential food insecurity (HFIAS), and subjective food security (SRMFS). This is consistent with a recent study by Abdallah (2025a) in which ex-situ rain water harvesting was found to improve food security. The consistently larger impacts associated with emerging RWHS across all food security indicators suggest that improvements in water storage and water availability translate directly into enhanced household welfare. In northern Ghana, where food insecurity is often driven by rainfall uncertainty, crop failures, and seasonal food shortages, improved access to harvested water can help stabilize production and strengthen household resilience (Abdallah, 2025a). The findings therefore support the growing body of evidence that water management interventions play a critical role in improving food security among smallholder farmers (Abdallah et al., 2020, 2021; Abdallah, 2025a). More importantly, the results reveal substantial heterogeneity across RWHS technologies, indicating that emerging systems generate considerably larger welfare gains than conventional alternatives. This highlights the need for policy interventions that not only promote RWHS use but also encourage the diffusion of more efficient and reliable water harvesting technologies.

6 Potential pathways linking RWHS use to welfare

Although the empirical analysis focuses on estimating the welfare impacts of RWHS, several plausible mechanisms may explain the observed improvements in farm income and food security. First, RWHS increases the availability of water during periods of rainfall deficits and short-term dry spells, thereby reducing moisture stress during critical crop growth stages (Barnes, 2009; UN-HABITAT, 2005). Improved water availability can enhance crop productivity, stabilize yields, and reduce the likelihood of crop failure (Koranteng et al., 2024; Owusu and Obour, 2025), particularly in northern Ghana where agricultural production remains highly dependent on erratic rainfall (Abdallah, 2025a). Second, RWHS may facilitate dry-season and supplementary irrigation activities, enabling households to extend production beyond the main rainy season. Increased opportunities for dry-season cultivation can raise cropping intensity, increase annual agricultural output, and generate additional income (Rao et al., 2024; Terêncio et al., 2018). The larger welfare gains associated with emerging RWHS are consistent with this explanation, as these systems typically possess greater storage capacity and provide more reliable water supplies throughout the production cycle. Third, improved access to harvested water may encourage diversification into higher-value crops, including vegetables and other market-oriented enterprises that require relatively reliable water supplies (Andoh et al., 2018; Barnes, 2009; Gedamu et al., 2025; Owusu and Asante, 2020). Such diversification can increase farm income while simultaneously improving household access to diverse and nutritious foods. This mechanism may partly explain the positive effects observed on food consumption scores. Fourth, RWHS may reduce production risk associated with rainfall variability (Lade and Oloke, 2020; Lepcha et al., 2024). By lowering the probability of crop losses during droughts or prolonged dry spells, households may experience greater income stability and improved food availability throughout the year (Abdallah, 2025a). Reduced production risk may also encourage farmers to invest more confidently in complementary inputs such as fertilizer, improved seed varieties, and hired labor, further enhancing productivity and welfare outcomes (Andoh et al., 2018; Barnes, 2009). Finally, increased agricultural production and income generated through RWHS use may improve households' purchasing power, enabling them to acquire food during periods of seasonal scarcity and invest in other welfare-enhancing activities (Abdallah, 2025a). Through these interconnected pathways, RWHS can contribute simultaneously to improved food availability, food access, dietary quality, and farm income (Owusu and Asante, 2020).

While the present study does not directly estimate these mechanisms because of data limitations, the observed impacts are consistent with these theoretical pathways and with evidence from previous studies on rainwater harvesting, supplementary irrigation, and climate-smart agricultural technologies in Sub-Saharan Africa (e.g., Abdallah, 2025a; Andoh et al., 2018; Barnes, 2009; Gedamu et al., 2025; Owusu and Asante, 2020). Future research should explicitly examine these channels using mediation or pathway analysis to identify the specific mechanisms through which different RWHS technologies influence household welfare outcomes.

7 Conclusions

This study examined the impacts of alternative ex-situ rainwater harvesting systems (RWHS) on FI, FSS, and food security among smallholder farmers in northern Ghana. By explicitly distinguishing between non-adopters, users of conventional RWHS, and users of emerging RWHS, the analysis moved beyond the common binary treatment of rainwater harvesting adoption and provided new evidence on the heterogeneity of impacts across system types. The study further strengthened credibility by employing multiple econometric approaches—SRMT, MET, MNPS, and AIPW—to address selection bias and test the robustness of the estimated effects.

The results consistently show that use of ex-situ RWHS generates substantial welfare gains for smallholder households. RWHS use significantly increases FI, improves FSS, and enhances food security measured by food consumption scores, reductions in food insecurity experiences, and households' own assessments of their food security status. Across all outcomes, emerging RWHS outperforms conventional systems, delivering larger and more robust impacts. These findings suggest that improvements in water availability and reliability translate into higher agricultural productivity, better capacity to meet household cereal requirements, and more stable and diverse food consumption patterns. The consistency of results across different indicators and estimation strategies underscores the robustness of the conclusions and confirms that the observed benefits are not driven by methodological choices. The heterogeneous effects identified carry important policy implications. The consistently larger welfare gains associated with emerging RWHS suggest that policy interventions should move beyond simply promoting rainwater harvesting adoption in general and instead prioritize the scaling of higher-performing systems. Investments in improved storage infrastructure, durable lining materials, water distribution mechanisms, and farmer training may generate substantially larger returns in terms of income growth and food security improvements than support for traditional systems alone. At the same time, the results suggest that conventional RWHS still provide positive and statistically significant benefits, implying that they may remain relevant for poorer households facing liquidity and capital constraints. Accordingly, a differentiated policy approach may be appropriate, where conventional RWHS serve as entry-level adaptation options while emerging systems are promoted through targeted subsidies, credit support, extension services, and community-based water management programs. Such targeting could improve the efficiency of climate adaptation investments in northern Ghana and similar semi-arid regions.

The evidence also highlights the importance of accounting for endogenous technology choice in evaluating agricultural water management interventions. Significant selection effects indicate that households adopting RWHS—particularly emerging systems—differ systematically from non-users, and that failure to address this selection would bias impact estimates. By correcting these issues, the study provides more credible estimates of the true welfare gains associated with RWHS adoption. Moreover, the stronger impacts observed for emerging systems point to meaningful differences in storage capacity, water retention, and seasonal reliability that matter for livelihoods and food outcomes.

From a policy perspective, the findings suggest several clear recommendations. First, scaling up investments in ex-situ RWHS remains a viable and effective strategy for enhancing resilience, income generation, and food security in semi-arid regions of Ghana. Second, greater emphasis should be placed on promoting emerging RWHS, as these systems deliver larger welfare gains than conventional structures. However, their higher upfront costs and technical requirements imply a need for complementary support, including access to credit, targeted subsidies, technical training, and extension services to lower adoption barriers. Third, policies should recognize that RWHS are not only productivity-enhancing technologies but also critical instruments for improving FSS and stabilizing household food access, particularly under increasing climate variability.

Finally, the study points to important directions for future research. Further work could examine the long-term sustainability and maintenance of different RWHS, assess cost-effectiveness across system types, and explore distributional impacts across gender and resource groups. Integrating biophysical data on water availability and crop responses with household-level outcomes could also deepen understanding of the mechanisms through which RWHS affect livelihoods. Overall, the evidence from this study supports the strategic role of ex-situ rainwater harvesting—especially emerging systems—in advancing climate-resilient agriculture and food security in northern Ghana and similar semi-arid contexts.

8 Policy implications

Although the present analysis focuses primarily on heterogeneity across RWHS, the descriptive evidence suggests that welfare impacts may also vary across household and farm characteristics. These patterns suggest that the welfare benefits of RWHS may not be uniform across all farmers and that institutional, socioeconomic, and locational characteristics could shape both adoption decisions and realized impacts. Future research should therefore examine treatment-effect heterogeneity across gender, farm-size categories, income groups, institutional participation, and agroecological zones to identify which groups benefit most from specific RWHS technologies and to better inform targeted policy interventions. The findings provide several important implications for agricultural water management and climate adaptation policy in Ghana. First, the positive effects of RWHS on farm income, food self-sufficiency, and food security suggest that rainwater harvesting should be considered a central component of climate-resilient agricultural development strategies in northern Ghana. Given increasing rainfall variability and recurrent drought events, investments that improve water availability at the farm level can strengthen household resilience while contributing to national food security objectives. Second, the results demonstrate substantial heterogeneity in impacts across RWHS types. Emerging RWHS consistently generate larger improvements in income and food security than conventional systems. This suggests that policy interventions should move beyond simply promoting RWHS use and instead focus on supporting technologies with greater storage capacity, reliability, and irrigation potential. Public investments in improved reservoirs, lined storage structures, small-scale irrigation systems, and associated water distribution infrastructure may therefore yield higher welfare returns than investments limited to traditional systems. Third, financial constraints are likely to remain a major barrier to use of emerging RWHS. Consequently, scaling up these systems may require targeted support mechanisms, including subsidized credit, matching grants, cost-sharing arrangements, and integration into existing agricultural development programs. Such interventions could reduce initial investment costs and encourage wider use among smallholder farmers. Fourth, agricultural extension services should play a more active role in disseminating technical knowledge on RWHS design, operation, and maintenance. Strengthening collaboration among extension agents, water-user associations, local governments, and development partners could improve adoption rates and enhance the long-term sustainability of RWHS investments. Finally, the findings highlight the importance of integrating water harvesting interventions into broader rural development and food security policies. Because RWHS influence both agricultural productivity and household welfare outcomes, coordinated investments across the agriculture, water resources, and rural development sectors may generate larger and more sustainable impacts than isolated interventions.

The policy implications of these findings extend beyond the promotion of rainwater harvesting per se. The evidence suggests that the type of RWHS adopted is critically important for determining welfare outcomes. While conventional RWHS generate positive benefits, emerging systems consistently produce substantially larger gains in farm income, food self-sufficiency, dietary quality, and reductions in food insecurity. Consequently, policy efforts should prioritize the expansion of higher-performing RWHS through targeted investments, improved access to finance, strengthened extension support, and enhanced community-based water management institutions. The findings are particularly relevant for northern Ghana, where agriculture remains predominantly rainfed and highly vulnerable to climatic shocks. However, the broader lessons are likely applicable to other semi-arid and drought-prone regions of Ghana, including transitional agroecological zones experiencing increasing rainfall variability. More generally, the results may inform climate adaptation and agricultural water management strategies across Sub-Saharan Africa, where improving water availability for smallholder farmers remains a critical development challenge. From a scalability perspective, the results suggest that RWHS can serve as a viable pathway for enhancing agricultural productivity and food security among smallholders. Scaling up adoption will likely require complementary investments in rural infrastructure, farmer training, institutional support, and affordable financing mechanisms. Such integrated approaches could improve the effectiveness of water management interventions and contribute to broader goals of poverty reduction, food security, and climate resilience.

Generative AI statement

The author(s) declared that Generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher's note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Supplementary material

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/frevc.2026.1808592/full#supplementary-material

Supplementary material Supplementary material at the source

Footnotes

Edited by: Jaime Moll de Alba, United Nations Industrial Development Organization, Austria

Reviewed by: Inoussa Zagre, UNFCCC, Germany

Mustapha Mohammed Suraj, CSIR-Savanna Agricultural Research Institute, Ghana

Footnotes

The 2010 PHC were made available to us by the Ghana Statistical Service.