Seasonal variation in maternal dietary diversity is reduced by small‐scale irrigation practices: A longitudinal study

Abstract Some agricultural practices, such as irrigation, have the potential to buffer seasonal dietary gaps and through increased production and consumption improve diets, particularly of the rural poor relying on subsistence farming but also for rural and urban households purchasing irrigated produce on local markets. The study aimed to evaluate the effect of seasonality and irrigation on women's diet in rural Ethiopia. Using a longitudinal study design, three rounds of surveys were conducted among women of reproductive age (15–49 years). Data on socioeconomic status, food consumption and haemoglobin concentration was collected. Energy and nutrient intakes were estimated using an interviewer‐administered multiple‐pass 24‐h recall. Women's dietary diversity score (WDDS), the proportion of women meeting the minimum dietary diversity for women (MDDW), haemoglobin concentration, the prevalence of anaemia and energy and nutrients intakes were compared between irrigators and nonirrigators and by season. Associations between MDDW/WDDS and irrigation status were assessed using fixed‐effect models, after adjusting for covariates. WDDS was low (3–4 out of 10 food groups) and exhibited high seasonal variability (p < 0.05). Diets were predominantly cereal‐based, with little consumption of nutrient‐dense foods like fruits and animal source foods. High seasonal variability in energy, protein, vitamin C, calcium, iron and zinc intakes were observed (p < 0.01). Irrigators were more likely to meet the MDDW than women from non‐irrigating households (p < 0.05). No cases of malaria were reported from the three rounds of screening. There is a high seasonal variation in women's diet, but this could be partly offset by irrigation practices.

in LMIC are disproportionately affected. Furthermore, the prevalence of undernutrition is often higher in rural than in urban communities.
According to the most recent Ethiopian demographic and health survey, 22.4% of women are undernourished as reflected by a low body mass index (BMI < 18.5;CSA, 2016). In rural Ethiopia, seasonal and chronic undernutrition is exacerbated by an over-reliance on rain-fed agriculture that is increasingly becoming unpredictable. In addition, poor market linkages and limited basic infrastructure (e.g., electricity) make nutrient-dense perishable foods less accessible and affordable (Hirvonen et al., 2016). The Ethiopian government has recently increased efforts to expand irrigation, which, if implemented well, could increase dietary diversity, increase micronutrient intake and prevent anaemia by filling energy and nutrient intake deficits through increased availability and accessibility of nutrient-dense foods (Keiser et al., 2005;Passarelli et al., 2018). However, irrigation water has also been linked to vector-borne diseases such as malaria, which can adversely affect nutritional status (Kibret et al., 2014).
While the seasonality of households' and children's diets is well documented, little evidence exists regarding the seasonality of women's diets in Africa (Arsenault et al., 2014;Savy et al., 2006;Waswa et al., 2021). Moreover, whether irrigation practices bridge seasonal dietary gaps and eventually improve nutrition remains largely unknown.
Recent studies suggest that irrigation can improve nutrition indirectly, through changes in production, income, water, sanitation and hygiene environment and women's empowerment (Domènech, 2015;Passarelli et al., 2018). However, the evidence remains scant and concerns over increased incidence of malaria have also been reported (Kibret et al., 2014). The lack of research on irrigation-nutrition pathways is unfortunate given the Ethiopian governments' investments in irrigation. In 2020, the Ethiopian Irrigation Development Commission announced its plan to undertake nine irrigation projects covering 125,000 hectares and with the aim of ensuring household food security. This is on top of at least 13 ongoing large-scale irrigation development projects with a combined command area of more than 400,000 hectares that were under construction as of early 2019 (Ministry of Agriculture of Ethiopia, personal communications, February 22, 2019). The 2018 National Smallholder Irrigation and Drainage Strategy of Ethiopia estimate around 7 million hectares of land to be economically viable to be irrigated by small holders in the country (GoE, 2018). However, with adequate planning the irrigation schemes could also be designed to improve nutrition. This study aims to build the evidence base to provide insight on nutrition-sensitive irrigation development, using panel data on women's diets from three rounds of surveys at different periods during the production calendar. It aims to evaluate women's diet across seasons and how it may be influenced by irrigation practice.
2 | DATA AND METHODS

| Study sites, design and study participants
The study was conducted as part of the Sustainably Intensified Production Systems Impact on Nutrition' (SIPS-IN) program funded by the USAID Feed the Future Sustainable Intensification Innovation Lab, which aims to evaluate the impact of irrigated production systems on food security, nutrition and health outcomes. The SIPS-IN program took place in the watersheds of Robit and Dangila districts, Amhara region, in northern Ethiopia. These sites were selected given their high potential for irrigation based on an ex-ante analysis of factors such as groundwater availability, distance to surface water and market access as well as interest to participate in irrigation activities by farming communities. Within these sites, households were randomly selected using the following criteria: produced crops were permanent residents of Robit or Dangila districts, and had at least one child less than 5 years of age. A longitudinal study consisting of household-level and individual-level surveys was conducted, and the same women 15-49 years of age (n = 364) were surveyed three times: in February-April 2017 (short rainy season-belg), October-November, 2017 (harvest) and July-August, 2018 (rainy season-meher). All the women self-reported that they were nonpregnant at the time of the surveys.
February-April (Season 1) represented an irrigation season, which also was a major fasting season (Lent fasting) for the Ethiopian Orthodox Church followers that accounted for 99% of our sample. The fasting season lasted 55 days preceding Easter. During this time, observers skip meals (breakfast and sometimes lunch) and adopt a vegan diet that is devoid of all animal source foods. The months of October and November (Season 2) are harvest seasons, where produces from the main rainy season are ready for household consumption, irrespective of access to irrigation. The months of July-August (Season 3) coincide with the lean season, when food availability is the lowest for nonirrigators. Considering the means and standard deviations of the women's dietary diversity score (WDDS) and energy intakes and α at 0.05, our post-hoc analyses using GPower software suggested that our sample size had a power of 99% (two-tailed test).

| Sociodemographic and household characteristics
Information on the sociodemographic characteristics of the study participants was collected using pretested questionnaires that

Key message
• Women's dietary diversity was low and exhibited high seasonal variation.
• The associations between season and diets were mediated by irrigation.
• Higher energy intakes were observed among women from households practising irrigation.
included questions about the age of household members, family size, education level, livelihood strategy and land size. The participant's flow chart is presented in Figure S1.

| Irrigation
Irrigation status was defined for each season depending on whether the household has at least one irrigated plot in the production season during the recall period. The study included areas where large irrigation systems are rarely available in the communities. Our survey data shows that irrigator households obtain water using hand buckets or hose, diesel pumps, hand or foot pumps and gravity methods.

| Dietary intake assessment
Using a quantitative 24-h recall (1-day recall), food intake was estimated using the multiple-pass technique validated for use in developing countries (Gibson & Ferguson, 2008). The first-pass was to identify all foods and drinks consumed in the last 24 h; the second pass was to identify ingredients and processing techniques applied in preparing the foods. The third pass is to make portion size estimations. To this end, whenever possible, salted-replicas of actual foods were weighed using kitchen scales (Kinlee ACS-EK01); or else, local utensils and graduated models were used. The final-pass was to check the completeness of the data gathered. All days of the week were equally represented in the final sample.

| Estimation of energy and nutrient intakes
Food intake was converted to energy and nutrient intakes using the Nutrisurvey software and linking food intake to food composition values from the Ethiopian food composition table that included nutrient composition data of local recipes (EHNRI, 1998). Missing values were completed using either published literature (Abebe et al., 2007;Umeta et al., 2005) or data from the USDA nutrient database (https://ndb.nal.usda.gov/ndb/), after adjusting for moisture difference. For mixed dishes, energy and nutrient contributions were calculated from local recipes. The detailed procedures followed are described elsewhere (Gibson & Ferguson, 2008). Intakes (per day) of energy (kcal), protein (g), vitamin A (µg RE), vitamin C (mg), calcium (mg), iron (mg) and zinc (mg) were estimated.

| Haemoglobin and malaria screening
Haemoglobin concentration was determined using a portable photometer (Hemocue HB 301; Ängelholm) from a blood sample from fingerpicks. The haemoglobin readings were adjusted for altitude (average 1900 m above sea level) and were categorized into nonanaemic and anaemic (any anaemia). Women with adjusted haemoglobin concentrations <12 g/dl were considered to be anaemic (WHO, 2011) and were referred to local health centres. Screening for malaria was performed using a rapid diagnostic test.

| Quality control
Data were collected electronically using tablets and Survey CTO software. All survey rounds were conducted by the same experienced and trained enumerators. The enumerators were professionals with at least a Bachelors' degree, fluent in Amharic and whose main occupation was survey data collection in various regions of Ethiopia. One-week training was provided before the first-round survey, and refreshers were given just before the second and third round surveys were conducted. We also run fixed effect panel models with MDDW, WDDS, as well as women's vitamin C, calcium and vitamin A intakes as dependent variables and adjusting for time-variant confounders and time-invariant individual fixed effects. The general econometric model can be described as follows:   Table S1.

| RESULTS
The most widely consumed food groups by women were cereals, followed by pulses, and 'other vegetables' (Figure 1). The  F I G U R E 1 Food groups consumed by women in the 24 h preceding the survey, by season. FV, fruits and vegetables; MFP, meat, fish and poultry subsequent rounds. Consumption of fruits was very low, while nuts and seeds were absent from the diets of the women.
nonirrigating women who met MDDW pars with women from irrigating households (p > .05). In the lean season of July and August, the proportion of women who met the MDDW was again much higher in irrigating households than nonirrigating households. Fasting was strongly and negatively associated with both MDDW and WDDS in all the specifications (with and without the interaction of seasons and irrigation status). Survey rounds February-April, and July-August fare better in MDDW and WDDS than the October-November season, only in the models without the interactions of irrigation and seasons, as most of the seasonal variation of diets comes from households with irrigation. Such differences by season and irrigation status were not observed for anaemia and haemoglobin concentrations (Table S3).

| DISCUSSION
The present study illustrates that women's dietary diversity was low and exhibited high seasonal variation. Diets were predominantly plant-based with little consumption of nutrient-dense food groups, such as animal source foods. High seasonal variability in energy, protein, vitamin C, calcium, iron and zinc intakes were observed.
Women from irrigating households had higher WDDS and were more likely to meet the MDDW in February-April and July-August seasons, than women from nonirrigating households. No cases of malaria were reported in the three rounds. The associations between season and diets were mediated by irrigation.
Multiple reasons can explain the high seasonality of the diets of the women. Primarily, subsistence farming households rely on their own harvests that, in the absence of irrigation, happen once or twice a year (Sibhatu & Qaim, 2017). In addition to the limits of relying on rain-fed production, poor market access, high price, limited food processing activities that extend the shelf-life of perishable goods (Abay & Hirvonen, 2017;Gebru et al., 2018), high price fluctuations (Bachewe et al., 2017;Gilbert et al., 2017) and religious fasting further contribute to the seasonality of diets and consequently to varying levels of energy and nutrient intakes. Such seasonal dietary patterns have been commonly reported from other parts of rural Africa (Caswell et al., 2018;Savy et al., 2006), and can have serious adverse effects, including insufficient weight gain during pregnancy and low-birthweight of babies (Rogawski McQuade et al., 2019;Toe et al., 2015).
The lowest energy intake, lowest animal-source foods consumption and most meagre overall dietary diversity were found during the first round of the survey (February-April). This is unsurprising given that this period corresponded to the longest period of lent fasting (55 days), a period when observers reduce the number of meals and adopt a vegan diet. The majority of our study subjects are orthodox Christians. Consequently, a slightly higher consumption of vegetables accompanied by higher calcium and vitamin C intake was observed, but consumption of fruits remained low. Given that this period is in the dry season, unless households practice irrigation or have access to affordable nutrient-dense foods like vegetables and fruits grown elsewhere, their diets are likely to remain poor in Women from households that practised irrigation had a higher dietary diversity than nonirrigating households. In the model adjusting for covariates, vitamin C and calcium intakes were also higher among women from households practising irrigation (Table S1). This is similar to findings on household dietary diversity in Ethiopia (Hirvonen et al., 2016). The statistically significant interaction between irrigation and season illustrates the buffering effect that irrigation can have on the seasonality of diets and energy/nutrient intakes. Multiple pathways including income, production and women's empowerment can explain these effects, but further studies are needed to decipher this.
Interestingly, no cases of malaria were reported in the three rounds of the survey, irrespective of smallholder irrigation practice.
This supports recent reports of very low levels of malaria in the Amhara region (Yalew et al., 2017), but also indicates that smallholder irrigation-unlike larger-scale irrigation systems with canal systems and higher likelihood of standing water-is not associated with high malaria incidence.
The strengths of the present study include the longitudinal study design that captured the dry and the rainy seasons, as well as the explicit evaluation of nutrition outcomes. The use of quantitative 24-h recall data to estimate energy and nutrient intake across multiple seasons is also a key strength of the study. However, a second day repeat was not available to adjust for the within-subject variance.
In addition, our study defined the use of irrigation on plots broadly; hence, whether outcomes vary by level and type of irrigation needs further investigation.
The present study illustrates the seasonal variation of women's diet in rural Ethiopia and shows that this seasonal variability can be partly offset by irrigation practice. Higher energy intakes were observed among women from households practising irrigation. Irrigation schemes that fill the gap in nutrient-dense foods (e.g., leafy vegetables) are likely to have more nutritional impact than those that aim to produce more cereals. Consequently, the Government of Ethiopia's plan to expand irrigation projects in the country could improve nutrition if oriented towards the production of nutrient-dense foods.
Future studies should investigate whether complementing irrigation interventions with nutrition-related behavioural change communication would lead to a greater impact on improving diets and nutritional outcomes of women.