Farmers’ perception and management of water scarcity in irrigated rice-based systems in dry climatic zones of West Africa

Water scarcity threatens irrigated agriculture in sub-Saharan Africa (SSA). Knowledge of farmers’ perceptions and drivers for decision-making in view of coping with water scarcity is so far lacking but needed to improve local technologies and frame policies fostering their adoption. Here, for the first time, we investigated farmers’ perception of water scarcity, key adaptation strategies, and the determinants of their adoption in irrigated rice schemes in dry climatic zones of West Africa. We surveyed 572 farming households and conducted expert interviews with key informants in four contrasting irrigated rice schemes in Burkina Faso between April 2018 and August 2019. Information was gathered on biophysical field characteristics, grain yields, agronomic and water management practices, farmers’ perception of water scarcity, their adaptive responses, and social-economic attributes of adopting households. Nearly 80% of the respondents reported having experienced water scarcity during the past 5 years. To cope with the adverse effect of water scarcity, farmers implemented seventeen different adaptation strategies that could be categorized into seven groups. Most popular among those were “water and soil conservation practices” (consisting mainly of field bunding and leveling), “no rice cultivation,” and “crop rotation.” Farmers in drier areas (Sudano-Sahelian zone) were less likely to adopt and implement several adaptation strategies to water scarcity compared to farmers in wetter areas (Sudanian zone). Belonging to farming associations increased the probability of implementing several strategies to alleviate water scarcity, while female-headed households tended to have a lower propensity to adopt and implement concomitantly several adaptation strategies in comparison with their male counterpart. The dissemination of scheme- and household-specific technology options could contribute to mitigating water scarcity in irrigated rice-based systems in the dry climatic zones of West Africa, thus contributing to rural livelihood and food security.


Introduction
In the last century, global water consumption has been growing more than twice the rate of the world population, increasing the number of regions experiencing water scarcity (UN-Water and FAO 2007). Water scarcity is defined as a gap between available supply and expressed demand for water at a given place (e.g., country, region, catchment, and irrigation scheme) and under prevailing institutional arrangements and infrastructural conditions. It also refers to situations where water demand exceeds the exploitable water resources. Unsatisfied demand, competition for water, conflicts between users, over-extraction of groundwater, and reduced water flows are the most common warning signals of water scarcity ( Fig. 1) (FAO 2012). Demographic growth, poor land and crop management, and regional economic development are exacerbating the pressure on water resources, especially in drought-prone regions. By 2025, over one fifth of the worldwide population is expected to be living with absolute water scarcity (FAO 2012). Sub-Saharan Africa (SSA) is the region with the largest number of water-stressed countries, with the Sudanian and the Sahelian climatic zones being the most vulnerable to water scarcity (UN-Water and FAO 2007).

3
Of all sectors of the economy, agriculture is the largest consumer of global freshwater, using 72% of all water withdrawals (UN-Water 2021), with irrigated rice being by far the largest consumer of freshwater (Tuong and Bouman 2003). Water scarcity is expected to be exacerbated by climate change, heat waves, and soil degradation, entailing multiple adverse impacts on crop production. Such impacts are not only restricted to upland systems but increasingly affect also irrigated rice. Being a traditional staple food in many parts of Africa, the expansion of the growing area of rice has been larger than that of any other crop (Arouna et al. 2021), and demand for rice in SSA increases faster than anywhere else in the world. Indeed, from 2008 to 2018, the annual rice consumption in SSA increased by 81% (FAO 2020). Despite only 26% of SSAs rice being irrigated (Diagne et al. 2013), irrigated lowland systems contribute most to regional rice production. However, particularly in the Sudanian and the Sahelian climatic zones, water scarcity seriously threatens rice-based livelihoods. The farm-level adoption of water-saving strategies depends on available technologies and farmers' adaptive capacity, particularly on farmers' perception of the extent of the problem and the location-specific benefits of new production strategies. Thus, there is an urgent need to understand farmers' perception of water availability and farmers' differential ability to respond adequately and promptly to water scarcity.
Most previous studies on adaptation strategies by smallholder farmers in SSA focused on climate change (Bryan et al. 2013;Zamasiya et al. 2017), with few studies analyzing the actual occurrences of droughts events (Masih et al. 2014), presenting an integrated assessment of drought risks for irrigated systems (Meza et al. 2020), or addressing issues of irrigation water management under conditions of water scarcity (Pereira et al. 2002). Assessment of the determinants of smallholder farmers' adaptation strategies to the effects of climate variability has been conducted in different SSA countries such as Benin (Gbemavo et al. 2022), Botswana (Mogomotsi et al. 2020), Ethiopia (Alemayehu and Bewket 2017;Tofu et al. 2022), Kenya and Tanzania (Shikuku et al. 2017), Uganda (Atube et al. 2021), and Zimbabwe (Jiri et al. 2017). Leal Filho et al. (2022 recently described water scarcity trends in Africa and outlined climate impacts on key water-related sectors. However, none of these studies take into account farmers' perceptions of water scarcity and household-specific willingness and capability to adopt water scarcity-mitigating strategies, and no studies are available for irrigated rice-based systems in West Africa. Filling this gap, the present paper provides insights into the spatial and temporal dynamics of water availability, the perception of water scarcity by farmers, and emerging patterns of practices aimed at mitigating water scarcity in the dry climatic zones of West Africa. We addressed the following objectives: (1) characterize rice-based farming systems and water management practices in four irrigated lowlands of Burkina Faso, (2) investigate the farmers' perceptions of the spatial and temporal dynamics of water scarcity, (3) identify farmers' strategies to control or cope with water scarcity, and (4) determine key biophysical factors and household attributes affecting the adoption of these strategies at farm level.

Description of the study area
The survey was carried out in Burkina Faso, which covers diverse dry climatic conditions (the Sudanian and Sudano-Sahelian zones) encountered in most Sahelian countries of West Africa and these being highly vulnerable to adverse impacts of climate change (The World Bank Group 2011). Burkina Faso is divided into the Sudanian zone (located above the isohyet 900 mm), the Sudano-Sahelian zone (between the isohyets 600 and 900 mm), and the Sahelian zone (below the isohyet 600 mm) (Fig. 2). Each climatic zone is characterized by a unimodal precipitation regime and experiences distinct wet and dry seasons, with the wet season extending over five months in the South and two months in the North (MAHRH 2011;CILSS 2016). The country also experiences variable rainfall and an increase in air temperatures (Fig. S1, S2, and S3). Its croplands are characterized by low soil fertility, poor nutrient availability (Hengl et al. 2021), and widespread soil degradation as reported in most Sahelian countries (Niemeijer and Mazzucato 2002).
Cereals are the main source of food calories and are produced mainly on smallholder farmland. In 2020, Burkina Faso (like most other Sahelian countries) depended on 10% of cereal imports and recorded a prevalence of undernourishment of nearly 20% (FAO 2020). While being only the fourth most cultivated cereal after sorghum (Sorghum bicolor), maize (Zea mays), and millet (Pennisetum glaucum) (FAO 2020), rice (Oryza spp.) production area and particularly its per capita consumption of 41 kg person -1 year -1 (UN 2021; USDA 2022) have been rapidly increasing. Its local production meets only 29% of the demand (USDA 2022). The

Data collection methods and data sources
The core data are derived from a household survey conducted in four irrigated rice production schemes in Burkina Faso from April 2018 to August 2019. Multiple-stage sampling techniques were employed to select the respondent farm households. We focused on the two climatic zones where irrigated rice is mostly cultivated: the Sudanian and Sudano-Sahelian zones (Akpoti et al. 2021). Two sites were selected as being representative of irrigated lowland rice-growing areas in those zones, namely, Karfiguela and Kou Valley irrigation schemes in the Sudanian zone and Zoungou and Sourou Valley irrigation schemes in the Sudano-Sahelian zone (Fig. 2). These four sites cover diverse geographic and socio-economic conditions prevailing in West Africa. Table 1 presents in detail the attributes and differences between these irrigation schemes, providing a summary of their climate, soil, irrigation, and production system characteristics. Finally, random sampling techniques were used to select smallholder rice farming households in each scheme. While assuming a confidence interval of 95% and aiming for a precision level (or sampling error) of ±8%, the sample size was determined using the formula below (Yamane 1967): where n is the sample size, N is the population of the smallholder rice farming households per irrigation scheme, and e is the level of precision. Thus, the sample size of smallholder rice farming households was 138, 141, 185, and 108 in Karfiguela, Kou, Zoungou, and Sourou irrigation schemes, respectively. We carried out the household survey using a structured questionnaire. The respondents interviewed were the household heads (i.e., principal household male or female member) as they usually have the decision-making power for managing farm resources. The information collected using the questionnaire covered seven major sections: • Section 1: Geolocation (site name, GPS coordinates) and physical characteristics (soil texture) of the rice field. In each field, the soil texture was determined by the "feel method" (or "hand method") (Defoer et al. 2009) and classified as sandy, loamy, or clayey • Section 2: Professional profile of the household head (gender, age, education level, and rice farming experience) • Section 3: Household socio-demographic (household size and farming association membership) and economic characteristics (number of farmworkers, total farmland size, rice field size, herd size, number of farm machines, main income source, and the ratio of rice production sold). The herd size was expressed in tropical livestock units (TLU) of 250 kg • Section 4: Rice productivity (rice cropping intensity, i.e., the number of times rice is grown per year), estimated grain yield typically harvested in the wet and dry seasons by indicating the number of paddy bags harvested, and the average weight of the bags over the past five years on the cultivated area) • Section 5: Current situation of perceived water scarcity and usage restriction during wet and dry seasons including the month of occurrence and the frequency of water scarcity encountered over the past five years. Respondents rated the water availability in their rice field in the wet and/or dry season following a 5-point Likert scale: 1 = "very bad," 2 = "bad," 3 = "fairly good," 4 = "good," and 5 = "very good." They also provided perceived water scarcity effects (yield reduction or complete crop failure) • Section 6: Farmers' water management practices and perception of the critical phases for irrigation including the irrigation frequency, factors influencing the irrigation frequency, and the amount paid for irrigation water per season. Additionally, farmers indicated the possible benefits of water-saving production strategies • Section 7: Inventory of farmers' adaptation strategies to perceived water scarcity issues in their rice fields Before being submitted to all 572 randomly selected rice farmers, the questionnaire was tested with voluntary farmers, adjusted as required, and converted into Open Data Kit (ODK) collect application format. The forms were uploaded to a server and downloaded to android devices (smartphones or tablets) using the ODK collect App. Well-trained extension agents or BSc students conducted the interviews in local languages (Mooré or Dioula). Finalized data forms were sent to the server housing all the survey information.
Before and after the household survey, in each scheme, discussions with 5 to 8 local experts and key resource people were conducted to get descriptive and historical information on each irrigation scheme (Table 1) and experts' opinions on some quantitative observations from the household survey. Weather data (annual rainfall, minimum and maximum temperatures) from 1981 to 2017 were retrieved for each site/scheme from NASA's POWER (Prediction Of Worldwide Energy Resource) database (Sparks 2018). Standardized Precipitation-Evapotranspiration Index (SPEI) data  were retrieved from the SPEI Global Drought Monitor website (https:// spei. csic. es/ map/)  for each site, allowing for the assessment and monitoring of aridity over timescales from 1 to 48 months.

Data analysis
After cleaning, household survey data were analyzed using R. 4.0.5 (R Core Team 2021) according to the objectives of this study.

Characterization of rice-based farming systems and farmers' water management practices
Household characteristics presented as continuous variables were analyzed using descriptive statistics (mean, median, standard deviation, and interquartile range). Percentage shares were calculated on categorical variables such as gender, age group, education level of the household head, belonging to a farming association, and main source of income. Similarly, the rice farmers' water management practices and perception of the critical phases for irrigation were analyzed using descriptive statistics. Percentage shares were calculated for categorical variables. The diversity between households was apprehended by a farm typology based on resource endowment (Falconnier et al. 2015). The farm type classification was derived from a cluster analysis based on five variables describing primary farm capital assets and potential labor availability: (i) total household size, (ii) number of full-time farmworkers (family members and hired people), (iii) total farmland size, (iv) herd size, and (v) number of farm machines (plows, rice thresher, and motor tricycles) (Falconnier et al. 2015). After standardizing the data, the cluster analysis was performed using agglomerative hierarchical clustering (AHC) (Köbrich et al. 2003). The AHC grouped households into clusters based on their similarity using Euclidian distance and Ward's minimum variance linkage method.
As all household characteristics presented as continuous variables and average rice yield did not meet the normality of residues and homogeneity of variance, the Kruskal-Wallis H tests were computed to identify significant differences across (i) irrigation schemes and (ii) farm types. Furthermore, to compare the distribution of the categorical variables between (i) irrigation schemes and (ii) farm types, the Fisher's exact test or the approximation method (Chi-squared (χ 2 ) test of independence) were computed. When more than 20% of cells had expected frequencies of <5, we used Fisher's exact test as applying the approximation method (χ 2 test) is inadequate in such cases (Crawley 2012).

Status of water scarcity
Continuous and categorical variables explaining the perceived water scarcity were analyzed using descriptive statistics and percentage shares, respectively. As the residuals of cultivated rice field size data failed to meet the normality assumption, we run robust linear mixed-effects models (robustlmm package; Koller 2016), including farmer's field as random effect and season as a fixed factor to test its effect on cultivated rice field size.
To link the farmers' perception of months when water scarcity occurs with meteorological records, we plotted long-term (1981-2017) monthly rainfall and maximum temperature (T max ) data for each site. In addition, a longterm (1981-2017) temporal analysis of the seasonal water deficit was done using SPEI_3 (SPEI over a 3-month accumulation period) for each site. Thus, we estimated the basic potential impacts of aridity on soil moisture and streamflow and the impacts of water balance on agriculture during the crop-growing season (World Meteorological Organization (WMO) and Global Water Partnership (GWP) 2016; Copernicus European Drought Observatory (EDO) 2020). Subsequently, the annual 12-month timescale accumulation SPEI (SPEI_12) was compiled, representing the precipitation deficit for the entire year and reflecting the hydrological consequences of the water deficit. The non-parametric Mann-Kendall test as described by Hirsch et al. (1982) was used to detect the possible existence of a monotonic trend of SPEI-12. The Z-value was used to test the statistical significance of the monotonic (upward/downward) trends whereas Sen's slope estimator was used to determine the magnitude of the trends (Sen 1968). We used Pettitt's test to identify the point at which the values in the data change for each site (Pettitt 1979). These analyses were done using the R package trend (Pohlert 2020). Similar to Polong et al. (2019), the threshold of annual SPEI_12 ≤ −1.0 was inferred to stand for a dry year whereas annual SPEI_12 ≥ +1.0 represented a wet year over the study domain. The frequency of occurrence of dry years (F dry years ) was calculated according to equation 2: where n is the number of dry years (annual SPEI_12 < −1.0) and N is the total of the years for a defined period.

Categorization of adaptation strategies to water scarcity
The water scarcity-adaptation strategies adopted by farmers were aggregated into seven major categories by grouping closely related strategies based on function/effects (Smit et al. 2001) and following the groups of adaptation strategies in the agricultural sector suggested by IPCC (2014). Rice farmers could use simultaneously several strategies to cope with water scarcity in their field; therefore, these strategies were coded as binary variables (yes or no) and were unordered and non-mutually exclusive. The percentage shares of "yes" were calculated to assess the predominance of the utilization of the strategies. In addition, the Kendall rank correlation analysis was also performed to assess the monotonic association between the typical grain yield in the dry season and the number of water scarcity adaptation strategies implemented.

Analysis of farmers' decisions to adopt strategies counteracting water scarcity
Our survey revealed that farmers use simultaneously several strategies to cope with water scarcity, thereby, excluding the multinomial logistic model (Greene 2011). For each adaptation strategy, farmers' decision is a binary variable, y, comprising two possible outcomes coded as 1 (or "yes") if the adoption and implementation take place or 0 (or "no") if no adoption. Therefore, given the binary nature of the response (dependent variable), we used binary logistic regression (Zhai et al. 2018) to understand factors affecting the farmers' decision to adopt and implement each major category of water-scarcity strategy. To analyze the binary response, logit and probit functions are the most commonly used. Both functions are perfectly symmetric and sigmoid except at the tails and in most cases lead to similar results (Cameron and Trivedi 2009). Therefore, we opted for the logit function partly because the coefficients can be more easily interpreted (Klieštik et al. 2015). Unlike linear regression models, the estimated regression coefficients by the logit model only provide the direction of the effect of the explanatory variables on the dependent variables but do not indicate the actual magnitude of change. Therefore, the marginal effects of the explanatory variables were also computed. Marginal effects measure how a change in response is related to a change in a covariate while all other variables are held constant. For categorical variables, the marginal effect evaluates discrete change, i.e., how predicted probabilities change as the independent variable changes from a reference modality to another one. For continuous variables, it measures the impact that an instantaneous rate of change of an independent variable has on the outcome variable (Leeper 2021). All dependent variables are related to the farmers' decision to implement an adaptation strategy. We modeled separately the seven aggregated adaptation strategies and not the single ones as the initial models failed to generate significant parameters. Moreover, for some strategies, the frequency is very low, yet, in the case of rare event data, logistic regression models can produce inaccurate estimates or fail to converge as maximum likelihood estimation is well known to suffer from small-sample bias sparse datasets (King and Zeng 2001;Allison 2004). The explanatory variables of the logistic regression models were selected according to literature and local expert knowledge. They covered five groups of factors: (i) climatic and environmental factors, (ii) institutional factors, (iii) socio-economic and demographic factors, (iv) irrigation scheme water supply, and (v) farmers' perceptions of water scarcity. Therefore, these five groups of factors were hypothesized to influence the choices of farmers' adaptation strategies to water scarcity in their irrigated rice fields. Table 2 describes all independent variables. At the outset of the fitting of the logit regression models, multi-collinearity issues of the explanatory variables were checked by computing the variance inflation factor (VIF). The smallest possible VIF value is 1, which indicates the complete absence of collinearity. On the contrary, as a rule of thumb, a VIF value that exceeds 10 indicates a serious issue of collinearity (James et al. 2014). The estimated VIF values of our models ranged from 1.00 to 2.98. Therefore, we concluded that multi-collinearity is not an issue in our models. In the next step, regression coefficients and average marginal effects (AME) of the logistic regressions were estimated. Among three different parameters of interest derived from marginal effects, we calculated average marginal effect (AME) as it has the potential to convey an important amount of information about the effect of each covariate on the outcome variable (Leeper 2021). The AMEs were computed using the package margins (Leeper 2021). To assess the goodness of fit of the models, the overall significance of the models was evaluated by computing likelihood-ratio tests. These tests consist to compare the log-likelihood of the model with the predictors (the full model) to a restricted model where only an intercept is included (the null model). In addition, Nagelkerke Pseudo R 2 was estimated for each model. We counted the number of active strategies implemented by each farmer to counteract water scarcity during the last years and assessed its key determinants by Poisson regression analysis. We removed the variable "Irrigation water availability in dry seasons" from the regression to reduce the variance inflation factor (VIF). After checking the overdispersion issue and the variance inflation factor (1.04 ≤ VIF ≤ 2.52), the Poisson regression model was fitted, and the likelihood-ratio test and Nagelkerke Pseudo R 2 were computed.

Characteristics of the farm households and farmers' water management practices
All environmental, institutional, socio-economic, and demographic characteristics of the rice farm households were significantly different across irrigation schemes and the descriptive statistics are presented in Table S1. Agriculture was the main income source for 94% of the households. On average, farmers cultivated 2.9 ha of land of which irrigated rice comprised 0.6 ha. Only about 40% of the rice production was sold on local markets, the largest share serving subsistence purposes. Some 58%, 34%, and 8% of households cultivated their irrigated rice on clayey, loamy, and sandy soils, respectively. About 31%, 56%, and 13% of the household heads were young adults (20-39 years), middle-aged adults (40-59 years), and senior adults (› 59 years), respectively. Thus, the mean household head age was 46 years while the mean farming experience was 22 years. Some 75% of the respondents were male and only 25% were female. About 66% of the respondents were illiterate whereas 26% and 8% of them attended primary school and higher schools, respectively. Illiteracy rates were much higher among female household heads (84%) in comparison to male household heads (60%). On average, the households had 14 members and 7 full-time employed farm workers. Less than 20% owned farm machines. Almost three quarters (74%) of the household belonged to farming groups or cooperatives. The hierarchical cluster analysis differentiated three farm types based on households' resource endowment. The largest group of 57% having a low resource endowed (LRE) referred to as type 1, 41% having medium resource endowed (MRE) referred to as type 2, and only <3% were classified as having high resource endowed (HRE), being referred to as type 3 (Table S2).
Type 1 (or LRE) farms are small households of 10 people on average who own a small farm size (on average 1.7 ha). Irrigated rice covers only 0.5 ha. Households are poor, owning no farm machinery and less than 2 TLUs. Type 2 (or MRE) farms own 4.3 ha of farmland of which 0.7 ha is under irrigated lowland rice cultivation. These farms are characterized by mid-size households of about 17 persons with 8 employed farm laborers. Households own at least one farm machine and 2 TLUs.
Type 3 (or HRE) farms are those with a high resource endowed and are characterized by farm sizes of 8.3 ha as well as a large household size of on average 43 people with 24 employed farm workers. The share of irrigated rice land is 1.6 ha. Farmers own more capital and productive farm assets, owning on average 28 TLUs and at least 2 farm machines.
Despite considerable and significant differences in household size, capital, and productive assets among farm types, there were no significant differences in terms of the age group, level of education, rice farming experience of the household heads, and the income source and affiliation to farming associations.
Rice grain yields varied (CV = 25%) across the irrigation scheme with an average of 4.7 Mg ha -1 and were significantly affected by farm resource endowment. Farms belonging to type 1 (low resource endowment) tended to yield less (4.6 Mg ha -1 ) than farms with medium (4.9 Mg ha -1 ) or high resource endowment (5.6 Mg ha -1 ) ( Table S2).
Farmers' management practices differed both between schemes and among households within a scheme (Table S3). While during the wet season, rice farmers rely mainly on rainfall with some supplemental irrigation (data not shown); during the dry season, most farmers irrigated their field once a week (60%) and up to 2 times per week (27% of the surveyed farmers). The frequency of irrigation was determined by the scheme-management fixed water turn, the water availability in the canal, the soil water status, soil type (texture, organic matter, and water-holding capacity), distance from the secondary/tertiary canal, and the topography. Among all these factors, more than half of the farmers explained that the water turn was the most important one (Table S3).
A water pricing mechanism is currently in force on all the studied irrigation schemes. Fixed by the local authority in charge of the water management in each scheme, the water price is a flat rate per unit of irrigated area per season that differs between schemes. On average, rice farmers paid € 33 per hectare and per season with a much higher price in Sourou Valley (€ 104 per hectare and season), a drier area, where water is provided by pumping stations, and the lowest price in Karfiguela (€ 9 per hectare and season), a less dry area, where water is supplied by river diversion through gravity.
When asked about the most drought-sensitive phenological stages of rice, and hence, the period with the highest need for reliable water supply, most farmers (70% of the respondents) named the heading and flowering stages, though answers differed between schemes (Table S4). More than 90% of the household heads considered relevant and 1 3 useful to save water in the rice fields. About 70% of them considered that the water saved could be used for growing legumes or other non-rice crops while 18% indicated that it could be used for livestock (Table S5).

Farmers' perceptions of water scarcity and analysis of its impact on rice production
The overall farmers' perceptions of water scarcity differed among schemes (Table S6). Nearly all farmers (97%) grow rice in the wet season and almost three quarters (73%) grow rice additionally in the dry season. In wet seasons, submergence and the lack of cropland represented about 60% of the causes explaining the absence of rice cropping. In contrast, in dry seasons, water shortage was the main cause (94%) ( Table S7).
Most respondents reported that water scarcity is a real issue for their farming activities. More than three quarters (78%) declared having experienced water scarcity issues in their irrigated rice field over the last 5 years. The average frequency of occurrence of water scarcity issues in the rice fields was twice over 5 years. Water scarcity occurred mostly during dry seasons (Table S6) and extended from a few days up to two months. Rice farmers perceived April and March as the critical months when water scarcity is frequently experienced (Fig. S4). Their perceptions are in line with the weather data analysis of the period 1981-2017 in these regions showing that March and April are the months with the highest maximum temperatures (Fig. S5a) and are among those with the lowest rainfall (Fig. S5b). March and April are also among the months which show the lowest SPEI_3 ( Fig. S6; Table S8). Moreover, from 1981 to 2017, at all the sites, annual SPEI_12 exhibits decreasing trends. From 2000 to 2017, the frequency of occurrence of drier years drastically increased in comparison with the period 1981-1999 ( Fig. S7; Table S9). This observation of the climate getting drier supports a large number of farmers complaining about water scarcity.
The majority of respondents (83%) perceived that their access to irrigation water is restricted in dry seasons against 17% who reported no restriction (Table S6). Thus, irrigation water availability during wet seasons was scored by 90% of the respondents as being "good" and "very good" while only 28% of the respondents indicated good water availability in the dry season. About 39% scored "bad" and "very bad" on the irrigation water availability in their rice fields in dry seasons (Fig. 3). The farmers' complaints were higher (around Fig. 3 Farmers' perception and assessment of the irrigation water availability in the rice fields in wet and dry seasons. The question was "score the irrigation water availability in your rice field during the wet and dry seasons." The coloring represents the proportion of each score ("very bad," "bad," "fairly good," "good," or "very good") for each season. The percentage on the left side is the combined proportion for "very bad" and "bad," the one on the middle for "fairly good," and the one on the right side for "good" and "very good" 90% of the respondents) where rice farmer density is high and water provided by river diversion and dam water intake (Karfiguela, Kou valley, and Zoungou) but were less where rice farmer density is low and water provided by pumping station (Sourou Valley) ( Table S6).
The majority of rice farmers (80%) claimed that water scarcity reduces their yield whereas a minority (20%) affirmed that it leads to complete crop failure (Table S6). This situation most likely led some farmers to stop rice cropping in dry seasons. Thus, the area cultivated with rice during the dry season (0.367 ha) was significantly smaller than in the wet seasons (0.534 ha) (n = 572; p < 0.0001).

Strategies adopted to counteract water scarcity
In response to the perceived water scarcity issues in their irrigated rice fields, farmers on the different sites have undertaken several adaptation strategies. From our survey, we inventoried seventeen adaptation measures that we grouped into seven main strategies differing in their aims and resource requirements (Table 3). The strategy group "water and soil conservation practices" was the most adopted and implemented (80%), followed by "crop rotation" (31%). Among "water and soil conservation practices," field bunding, and leveling predominate (Table 3), about 20% of the respondents reported using "adjustment of the agricultural calendar," "use of motor pump for supplemental irrigation," and "use of more irrigation water from the scheme." About 40% abandoned rice cultivation during the dry season, while 29% opted for the reduction of their rice field size (Fig. 4).
Among the farmers opting for "crop rotation" as an adaptation strategy to water scarcity, almost half replaced rice with maize (Zea mays), a less water-demanding crop, during the dry season. After maize, other alternative crops to dry season rice comprised vegetables (31%), sweet potato (Ipomea batatas) (12%), and groundnut (Arachis hypogaea) (8%) (Fig. S8). However, the alternative crops to irrigated rice were site-specific. Thus, in Kou Valley, the Table 3 Categorization of water scarcity adaptation strategies adopted by rice farmers in irrigated lowland fields (in Burkina Faso in 2013-2018). About the demand for each factor of production (land, labor, capital, know-how), "+" means yes; "0" means no effect most common alternative crop was sweet potato, whereas, in Zoungou, vegetables were the most common (Fig. S9).
To cope with water scarcity in their irrigated rice fields, farmers implemented up to six distinct adaptation strategies with an overage of two. There was a positive association between the typical grain yield and the number of active water scarcity adaptation strategies implemented. This means that as the number of adaptation strategies implemented on a farm increases, the typical rice yield tends to increase. Despite the small effect size (τ = 0.11), the association was statistically significant (p = 0.005). Farmers implementing the maximum (5 and 6) adaptation strategies in their fields increased the typical rice yield in the dry season by 12% in comparison to those implementing one adaptation strategy.

Analysis of the adaptation strategies
Factors increasing or reducing the likelihood of farmers' decision to implement an adaptation strategy differed by site and household socio-economic attributes. Thus, non-association in farming groups and the perceived frequency of water scarcity in the rice fields negatively influenced the choice of the strategy "use of more irrigation water from the scheme." Conversely, rice fields on light (loamy and sandy) soils were positively associated with this adaptation strategy group (Table S10).
In comparison with rice farmers from wetter areas, and having agriculture as their main income source, those from drier zones and having trade as their main income source preferably selected the "use of motor pump for supplemental irrigation" (Table S10).
Compared to farmers growing rice on clayey soil, those on sandy soil were less likely to adopt "water and soil conservation practices." Moreover, the probability of choosing and implementing this strategy group was lower for rice farmers in drier areas in comparison with those from wetter areas (Table S10).
Farmers complaining about irrigation water availability (scoring "very bad" or "bad") in the dry season were less likely to choose "adjustment of the agricultural calendar" as an adaptation strategy to water scarcity. In contrast, compared to middle-aged adults, young adult-headed households were more likely to implement this strategy. Fig. 4 Rice farmers' adaptation strategies to water scarcity. The coloring represents the proportion of each response ("no" or "yes") for each adaptation strategy. The percentage on the left side is for "no" and the one on the right side for "yes" "Crop rotation" was a strategy more likely chosen by households belonging to a farming association, and having large rice field sizes (p = 0.06). Farmers complaining about the water availability from the irrigation scheme in the dry season were 18% more eager to implement this strategy. Female-headed households were 17% less likely to implement this strategy in comparison with male-headed households (Table S11). "Crop rotation" was more likely adopted on sandy soils in comparison with clayey soils.
Rice farmers from drier areas and those complaining about the water availability from the irrigation scheme in the dry season have a higher propensity to abandon rice cultivation. Our model failed to identify factors influencing the strategy "reduction of the cultivated rice field size" (Tables S12 and S13).

Factors influencing the number of adaptation strategies implemented
The climatic zone, the soil texture, the farming association membership, the gender of the household head, and the amount paid for irrigation water were key drivers explaining the number of adaptation strategies implemented by farmers (Fig. 5).
Farmers in drier areas (Sudano-Sahelian zone) were less likely to implement more adaptation strategies to water scarcity compared to farms in wetter areas (Sudanian zone). Male-headed households and farmers growing rice on loamy or sandy soils were more likely to adopt and implement several adaptation strategies compared to female-headed and those growing rice on heavy (clayey) soils. Moreover, the farming association membership had a positive effect on the number of adaptation strategies adopted (Fig. 5).

Farm typology and effect of farm resource endowment on the yield
Smallholder farms differ in terms of capital assets, potential labor availability, land area, and investment capacity. Furthermore, we found that farm resource endowment had a significant effect on the grain yield with HRE and MRE farms obtaining higher rice yields than LRE farms (Table S4). This finding supports the conclusion of Falconnier et al. (2015) who showed that in the cotton zone of Southern Mali, LRE farms achieved lower land productivity for all crops than Empty dot symbols mean the regression is not significant (p > 0.05). Filled dot symbol means the regression is significant (p < 0.05) other farm types. In climbing bean farming systems in northern Rwanda, Franke et al. (2019) also reported the same conclusion by stating that poorer households achieved lower grain yields than wealthier households. Household resource endowment acts most likely as a proxy for the economic status of households. It affects a range of crop management factors such as access to mineral fertilizer and organic manure, ability to conduct crop management operations (land preparation and sowing/transplanting, weeding, fertilizer application, and bird scaring) on time, and access to good quality seeds. In turn, those factors determine crop productivity.

Water management practices in irrigated lowland rice fields
Farmers' management practices in the dry climatic zones of West Africa were different among the irrigation schemes. During the wet season, farmers rely mainly on rainfall while irrigation is supplemental. However, in the dry season, due to the high pressure on water, the majority of farmers (59%) irrigated their fields once a week. Hence, most farmers did not practice strict continuous flooding which is still a popular water management practice in irrigated rice systems in Asia and America (Datta et al. 2017;Singh et al. 2017). This study highlights two local perceptions which could promote the implementation and adoption of water-saving technologies such as alternate wetting and drying (AWD) in rice fields. First, about 70% of survey rice farmers identify the periods around heading and flowering as the most sensitive phenological stages to drought, and hence during these periods, they take care to irrigate thoroughly their rice fields. This endogenous perception is in agreement with the current scientific insight, which shows that the effect of drought stress on yield is most severe when drought occurs during panicle development (Boonjung and Fukai 1996) and at the flowering stage (Cruz and O'Toole 1984;Yang et al. 2019). Second, although there is no economic incentive to implement water-saving technologies as farmers pay a fixed irrigation fee that is determined by the field size (but not on the amount of water used), about 90% of the surveyed household heads considered relevant and useful to save water in their rice fields. In the Philippines and Bangladesh, in both gravity-based and pumping irrigation systems, the lack of economic incentives was a key reason for the failure of scaling up AWD, a water-saving technology (Enriquez et al. 2021).

Farmers' perceptions of water scarcity
This paper evidenced for the first time that water scarcity is currently a real issue in irrigated rice-based systems in dry (Sudanian and Sudano-Sahelian) climatic zones of West Africa. Indeed, more than three quarters declared having experienced water scarcity issues in their irrigated rice fields over the last five years. Water scarcity occurred essentially during dry seasons (April and March being the most critical months) (Fig. S4). This is congruent with the perception of the majority (83%) of respondents who felt that their access to irrigation water is restricted in dry seasons (Table S6). These perceptions are less subjective in comparison to those related to climate variability and climate change in other studies (Ayanlade et al. 2017;Ayal and Leal Filho 2017). Indeed, sometimes, the farmers' observations (e.g., rainfall trends) were inconsistent with the meteorological records (Ayal and Leal Filho 2017). But, in this study, as farmers are witnesses and above all victims of water scarcity and its effect on the availability of irrigation water and productivity, their perceptions should be considered as solid empirical evidence. Thus, this is an additional evidence confirming that farmers in the Sahel are seriously concerned with the climate change and its corollary (drought) (Mertz et al. 2009).
The current water scarcity in irrigated rice fields could be caused by several factors. The first cause could be the poor management and maintenance of the irrigation infrastructures. Indeed, the obsolescence of the irrigation infrastructure was noticeable in the oldest irrigation schemes: Karfiguela, Kou Valley, and Zoungou (Table 1). And these three irrigation schemes were those where we recorded more complaints about water scarcity (Table S6). Moreover, in these three irrigation schemes, we observed the illegal activity of water siphoning through bilge piping by farmers outside the irrigation schemes for the benefit of vegetable cropping. A second cause could be the adverse impact of climate change. Indeed, the trend analyses of the annual SPEI_12 from 1981 to 2017 showed that the climate is getting drier in our study sites. We found a decreasing pattern of the annual SPEI_12 and a drastic increase in the frequency of occurrence of dry years from 2000 to 2017 in comparison with the period 1981-1999. Thus, the severity and frequency of dryness events could explain the water scarcity situations. Therefore, adequate water-saving technologies should be tested, implemented, and adopted by farmers in those schemes. Last but not least, the degradation of shared freshwater ecosystems, the competing demands for water resources, and the increasing population are other probable causes of water scarcity in these irrigated schemes.

Farmers' adaptation strategies to water scarcity
Most of the farmers' adaptation strategies to water scarcity listed have been also inventoried as farmers' strategies to cope with climate change effects and drought in paddy fields in drought-prone areas in Bangladesh (Alauddin and Sarker 2014;Alam 2015), China (Huang et al. 2015), Iran (Esfandiari et al. 2020), and Nepal (Khanal et al. 2018). However, there are some water scarcity adaptation strategies listed in these studies that had not been explicitly mentioned by the rice farmers during our survey. Among those strategies appear: using high-yielding heat and drought-tolerant seed varieties (Alam 2015), direct-seed rice (Alauddin and Sarker 2014), crop insurance (Huang et al. 2015), increasing seed rate, seed priming, reducing tillage, increasing the number of weeding, and improving/increasing fertilizer use (Khanal et al. 2018). While in drought-prone areas of Bangladesh, rice farmers preferred to increase the use of groundwater irrigation (Alam 2015), our study revealed that the "water and soil conservation practices" group was the most adopted and implemented water scarcity adaptation strategy, followed by the "crop rotation." Similar to rice farmers in drought-prone regions in Bangladesh (Alam 2015), rice farmers surveyed in this study implemented multiple adaptation strategies.
About 40% of the respondents declared abandoning rice cultivation during the challenging period. This proportion is very high in comparison with the one (5.6%) of rice farmers in drought-prone areas in Bangladesh (Alauddin and Sarker 2014). This confirmed the previous findings showing that a large proportion of African smallholder farmers did not make any adjustments to their farming practices despite having perceived climate change adverse effects (Bryan et al. 2009). Limited knowledge and lack of information about appropriate water scarcity adaptation strategies, inadequate irrigation facilities, or financial constraints could be the main barriers to adaptation (Bryan et al. 2009;Alauddin and Sarker 2014). However, abandoning rice cultivation during dry seasons and focusing on other businesses could be a good risk adaptation strategy if income or any other key indicators of sustainability are improved.
In Nepal, Khanal et al. (2018) showed that the adoption of climate change adaptation strategies plays a significant role in enhancing rice yield. Our study supports this finding to some extent. It is also in line with Di Falco et al. (2011) and Huang et al. (2015) who showed the positive effects of adaptation on crop productivity in studies in Ethiopia and China, respectively. Moreover, we found that as the number of active adaptation strategies implemented on a farm increases, the rice yield tends to increase.

Determinants of smallholder farmers' choices of adaptation strategies
Empirical evidence showed that the decision-making drivers regarding farm-level adaptation strategies to water scarcity or climate change fall into five groups of factors: climatic and environmental factors ), institutional factors (Bryan et al. 2009;Thoai et al. 2018), socio-economic and demographic factors (Alam 2015;Khanal et al. 2018), irrigation scheme water supply (Aguilar et al. 2022), and farmers perceptions (Khanal et al. 2018).

Climatic and environmental factors
Farmers in drier areas (Sudano-Sahelian zone) were less likely to adopt and implement multiple adaptation strategies to water scarcity (especially "water and soil conservation practices") in comparison with farmers in wetter areas (Sudanian zone) ( Fig. 5 and Table S10). This could be explained by the fact that water scarcity events in drier areas are more severe, risks are higher, and farmers who are willing to continue to grow rice find these strategies less relevant and inefficient in their context. This finding could also be a bias of our study. Indeed, in pump irrigation systems such as in Sourou Valley (a scheme in the Sudano-Sahelian zone), water is available and therefore there is no incentive to adopt and implement multiple adaptation strategies and "water and soil conservation practices." Although the water price in the Sourou Valley irrigation system was five times more expensive than in the others (that rely on surface water, e.g., river diversion), using this system positively influenced the farmers' resilience during the dry season.
In comparison with farmers growing rice on clayey soil, those growing rice on sandy soil were more likely to adopt and implement concomitantly several active adaptation strategies such as "use of more irrigation water" and "crop rotation." However, they have a lower propensity to adopt and implement "water and soil conservation practices" (Table S10). This is understandable as they most likely consider this strategy group inappropriate for their soil. Indeed, bunding, the most popular strategy in this group, is challenging as the construction of the bund requires moderate soil clay content. Viewed from another perspective, farmers growing rice on clayey soils implemented fewer adaptation strategies than those growing rice on sandy or loamy soil because the fine-textured soils have a higher water storage capacity that is adequate for paddy rice cropping. In addition, farmers perceived the bunding on clayey soils to be an effective water conservation practice.

Institutional factors
In general, farming association membership has a positive and significant effect on the number of active adaptation strategies adopted (Fig. 5). More specifically, we found that it positively influenced the choice of the strategies "use of more irrigation water from the scheme," "use of motor pump for supplemental irrigation," and "crop rotation" (Table S10). In other words, "free rider" farmers are less prone to adopt and implement more active adaptation strategies. Indeed, most of the time, public and private research and development organizations use farming associations as the main canal for sharing information concerning training on new agricultural technologies. And, by participating in the social activities of these associations, farmers can access useful information, knowledge, skills, and resources. Thus, farmers belonging to farming associations have generally a better knowledge of adaptation strategies. In turn, they are more eager and skilled to implement several strategies. This confirms, inter alia, the studies of Khanal et al. (2019) and Vo et al. (2021) (in Nepal and Vietnam, respectively) affirming that involvement in community-based organizations is an effective institutional instrument fostering the adoption of adaptation strategies.

Socio-economic and demographic factors
The gender of the household head significantly influenced the number and the type of active adaptation strategies adopted and implemented. In comparison with male-headed households, the lower propensity of female-headed households to implement "crop rotation" could be explained by the fact that alternatives to rice such as vegetable crops are more labor-demanding (Brosseau et al. 2021). Male-headed households have also a higher propensity to adopt and implement concomitantly several adaptation strategies. Several studies (Thoai et al. 2018;Vo et al. 2021) in other countries (Botswana and Vietnam) demonstrated similar findings. Part of the explanation likely lies in the comparative advantage of men to get information about new technologies. In addition, female farmers in developing countries generally have lower socio-economic resources (Arora-Jonsson 2011) and are discriminated against access to farming associations, land, and farm equipment (Kinkingninhoun-Mêdagbé et al. 2010), which most likely impede their adaptation propensity. Our results support also the hypothesis that males have generally higher natural risk-taking behavior than females (Asfaw and Admassie 2004).
Contrary to Deressa et al. (2009), we did not find a significant and positive effect of the age of the household head on the probability of adapting. However, we found that in comparison with middle-aged farmers, young farmers are more likely to adjust their agricultural calendar. This could be explained by their strong motivation in adapting to water scarcity and higher natural risk-taking behavior.
Education is an important determinant in adopting adaptation strategies (Alam 2015). However, contrary to Gebrehiwot and van der Veen (2013) and Vo et al. (2021), this study did not reveal a significant and positive effect of the level of education of the household head on the adaptation propensity to water scarcity-related risks.
The group strategy "use of motor pump for supplemental irrigation" is resource-demanding as it requires holding a motor pump and buying fuel for irrigation. This explains that farmers having trade as their main income source in comparison with those having agriculture as their main income source have a higher propensity to implement it.

Irrigation scheme water supply
The water turn and availability on an irrigation scheme significantly influenced farmers' choices of adaptation strategies. This supports previous studies showing that direct access to water sources is closely associated with overall farmers' adaptation to water scarcity (Aguilar et al. 2022) and points out the importance of good and timely maintenance of irrigation infrastructures for better scheme performance. In this sense, both national and local authorities in charge of water management on the schemes have key roles in helping farmers to cope with water scarcity.

Farmers' perceptions
The farmers' perception of water scarcity or other threats like global warming side effects shape their adaptation strategies (Mertz et al. 2009;Alam 2015). Indeed, we found that farmers' perception of the frequency of water scarcity events in the previous five-year term negatively influenced their likelihood to use more irrigation water from the scheme as an adaptation strategy. This could be explained by the fact they consider this strategy less efficient to cope with high climate-related risks.

Conclusion
Farmers in dry climatic zones of West Africa must rapidly adapt to emerging water scarcity to promote the sustainability of rice-based systems. To this end, studies on farmers' perceptions and on the factors driving decisionmaking and technology adaptation are fundamental. The present survey in contrasting irrigation schemes of Burkina Faso has provided a first inventory of the strategies adopted by rice farmers to counteract water scarcity and analyzed the determinants of adaptation strategies. Farmers' management practices differed both between schemes and among households within a scheme. While during the wet season rice farmers relied mainly on rainfall with some supplemental irrigation, during the dry season, about half of the farmers irrigated their fields once a week and a quarter up to 2 times per week. Most respondents have experienced water scarcity in their irrigated rice fields and perceived access to irrigation water as a key limitation to rice production in dry seasons. Furthermore, increasing dryness trends have been visible since 2000 in some zones. To cope with and mitigate the adverse impacts of water scarcity, farmers adopted seven groups of adaptation strategies, the three most popular being "water and soil conservation practices," "no rice cultivation," and "crop rotation." However, farmers' choices of adaptation strategies varied with climatic zones, soil texture classes, farmers belonging to associations, gender, and irrigation water availability in dry seasons. Resource endowment had a significant effect on the yield. High and mediumresource endowed farms had higher rice yields. We thus advocate national policy interventions to aim at (i) improving households' livelihoods, (ii) supporting the establishment and strengthening of farmer organizations, and (iii) providing training on site-and system-specific adaptation to water scarcity and climate change, encouraging specifically the participation of female farmers. Such measures are expected to increase farmers' capacity to adapt to emerging water scarcity with positive impacts on crop productivity in the dry climatic zones of West Africa.

Acknowledgements
The staff of INERA is greatly acknowledged for their administrative support as well as all rice farmers from Kou and Di valleys, Zoungou, and Karfiguela irrigation schemes who kindly answered the questionnaire of this study. The authors extend their gratitude to Messrs. Jean P. Eric Kaboré, Gustave Bombiri, Sibiri Kabré, Soumaila Traoré, Abdoulaye Sana, Dramane Diawara, Wouramou Doye, Ismael Ouedraogo, Issa Kadri, Zakaria Tindimba, Romaric Dabiré, and Mmes. Djeneba Tagnan and Sandrine Tougri who worked as enumerators for providing reliable survey data. Mr. Justin Djagba is warmly thanked for drawing the study area map. Funding Open Access funding enabled and organized by Projekt DEAL. This work received financial support from the German Federal Ministry for Economic Cooperation and Development (BMZ) commissioned and administered through the Deutsche Gesellschaft für Internationale Zusammenarbeit (GIZ) Fund for International Agricultural Research (FIA), grant number: 81206679. This study was conducted under the framework of the "Climate-smart rice technologies to enhance the resilience of smallholder rice farmers in Burkina Faso (CSA Rice Burkina)" project led by Africa Rice Center, University of Bonn and Institut de l'Environnement et de Recherches Agricoles (INERA).

Data availability
The dataset analyzed during the current study can be obtained from the corresponding author upon reasonable request.

Code availability
The code used for data analysis can be obtained from the corresponding author upon reasonable request.

Declarations
Ethics approval Not applicable.
Consent to participate Verbal informed consent was obtained prior to the interview.

Conflict of interest
The authors have no conflicts of interest to declare that are relevant to the content of this article.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/.