Introduction

Water is a key input for producing rice, the most important food crop in many developing countries and the staple food consumed by more than half of the world’s population (Mishra et al. 2022). Irrigated lowland rice farming plays a crucial role in global food security, providing 75% of the world's rice supply (Global Rice Science Partnership 2013). However, irrigated rice also requires more water (40% of the global irrigation water) than any other staple crops (Surendran et al. 2021). In addition to water, nitrogen (N) is another essential input limiting rice production (Johnson et al. 2019; Saito et al. 2019). Most cultivable soils contain insufficient amounts of N, and therefore, N accounts for the largest share of fertilizer consumption (Aulakh and Malhi 2005; Johnson et al. 2023b).

With increasing competition for water resources between domestic, industrial, and agricultural sectors, water scarcity has become one of the main issues faced by poorer communities, particularly in arid areas. Over 2.3 billion people live in water-scarce areas and their number will increase (UN-Water 2021) due to rapid demographic growth. In West Africa, and mainly in semi-arid and arid zones, water scarcity threatens the sustainability of irrigated rice production (Johnson et al. 2023a).

The rice crop takes up only ~ 40% of applied N fertilizers, indicating that about 60% of N applied is not assimilated and may be lost from the soil–plant system (Wopereis et al. 1999; Yu et al. 2022), making it a potential source of environmental (water and air) pollution (Tilman et al. 2002). In addition, N fertilizer represents the second-highest input cost in rice production for smallholder farmers in sub-Saharan Africa, and its inefficient use thus contributes significantly to economic losses (Chidiebere-Mark et al. 2019; Saito et al. 2023b). The sharp rise in fertilizer prices following the COVID-19 pandemic and the Russia-Ukraine war further increased concerns in the Global South and emphasized the relevance of improved nutrient management (Arndt et al. 2023). Furthermore, water scarcity may also affect nutrient use efficiency (Datta et al. 2017) and reduce crop N uptake (Lassaletta et al. 2023).

In this context, reducing water input while maintaining grain yield and nutrient use efficiency is of critical importance for sustaining the production of irrigated rice. To tackle this challenge, several water-saving technologies have been advocated. Among these, alternate wetting and drying irrigation (AWD) has become one of the most prominent and widely recommended water-saving technologies in Asia (Cheng et al. 2022; Chu et al. 2018). However, in the semi-arid zones of West Africa, AWD is largely unknown to smallholder farmers (Johnson et al. 2023a). This water-saving irrigation technology involves intentionally allowing the paddy fields to dry out until the field water level reaches a threshold, before re-irrigation, rather than keeping them continuously submerged (Lampayan et al. 2015). In practice, AWD is categorized into two groups according to either the soil water potential or the soil water table threshold: (1) alternate wetting and moderate soil drying (AWD15), which involves re-irrigation when the water level reaches 15 cm below the soil surface (or soil water potential ≥ -20 kPa); and (2) alternate wetting and severe soil drying (AWD30), where re-irrigation occurs when the field water level drops below 15 cm and up to 30 cm below the soil surface (or soil water potential < -20 kPa) (Carrijo et al. 2017). Globally, alternate wetting and moderate soil drying irrigation, considered as “safe” AWD procedure, reduced the water input by 23% without yield penalty compared to continuous submerged field practices (Carrijo et al. 2017). A recent study, comparing it to farmers’ irrigation practices, demonstrated its effectiveness as a water-saving technology in semi-arid areas of sub-Saharan Africa (SSA) (Johnson et al. 2020). Compared to farmers’ irrigation practices, it reduced water input by up to 30% without compromising yield (Johnson et al. 2020). Alternate wetting and severe soil drying has the potential to further reduce water inputs (Carrijo et al. 2017). However, extended periods of aerobic soil conditions may stimulate the germination and growth of terrestrial weeds, thereby increasing the demands for weed control management (Rodenburg et al. 2011). The extended dry periods during AWD implementation may lead to NO3--N losses due to nitrification and denitrification, further affecting the nitrogen use efficiency (De Datta et al. 1990). In addition, ammonia (NH3) volatilization in alkaline soils may enhance the N losses during drying events (Ishfaq et al. 2020; Moormann and van Breemen 1978). On the other hand, some studies reported no increase in N losses under AWD or even an enhanced N use efficiency due to reduced N leaching, being one of the primary routes of N losses in paddy fields (Ghosh and Bhat 1998; Peng et al. 2011). Furthermore, AWD may also potentially reduce the bioavailability of macronutrients such as P and K and some micronutrients (mainly Fe and Mn) (Ishfaq et al. 2020). To date, it is still unclear how AWD affects nutrient uptake and use efficiency.

Most previous studies assessing the effect of AWD on nutrient uptake and use efficiency were conducted in Asia and were obtained from on-station trials (Cheng et al. 2022). Among the limited number of studies providing insights from the Sahelian region in West Africa (de Vries et al. 2010; Djaman et al. 2018), the focus has primarily been on N use efficiency, comparing AWD to continuous submergence. However, continuous submergence is not the predominant water management practice implemented in irrigated rice schemes in West Africa where water is made available on a scheme-depended irrigation schedule (Johnson et al. 2023a). Comparisons between AWD and actual farmers’ irrigation practices are limited (Krupnik et al. 2012b). Furthermore, no previous study in the region assessed the effect of AWD on P use efficiency and the concentration of macro and micronutrients in straw and grain. There is a need to evaluate the impact of AWD in dry climatic zones in West Africa under farmers’ field conditions to assess its feasibility and sustainability. To address this knowledge gap, we quantified and compared water input, growth parameters, grain yield, water productivity, N and P use efficiency, and nutrient uptake (N, P, K, Mn, Fe, and Zn) under the farmers’ irrigation practices (FP) and alternate wetting and severe soil drying irrigation (AWD30). We hypothesized that due to increased N losses during the drying phases and reduced availability of P under aerobic soil conditions, AWD30 may reduce the N and P use efficiency compared to FP. To test this hypothesis, two-year on-farm trials were conducted in Burkina Faso, with the following objectives: (i) to compare the effects of alternate wetting and severe soil drying irrigation (AWD30) and farmers’ irrigation practices (FP) on growth parameters, weed infestation, grain yield, water productivity, nutrient uptake and use efficiency; (ii) to assess the effect of water input and anaerobic conditions on N, P and K uptake; and (iii) to evaluate the synergies and trade-offs between water productivity and N and P use efficiency.

Material and methods

Study area

Burkina Faso, a Sahelian country in West Africa, facing recurrent drought events and water scarcity in the last decades, was chosen as the focus of this present study. In this country, the availability of water for rice field irrigation is dwindling, especially during the dry season (Johnson et al. 2023a). The study was conducted in the Kou Valley irrigation scheme (11°23′7’’N, 4°24′20’’W, 320 m a.s.l), located in the southwest of Burkina Faso and approximately 30 km north of the town of Bobo-Dioulasso (Fig. S1). Commissioned in 1970, Kou Valley is one of the largest rice irrigation schemes in the country (Dembelé et al. 2005; Johnson et al. 2023a). Covering 1,260 ha within the Kou River watershed, this scheme relies on a run-of-river diversion intake from an upstream headwork (Wellens et al. 2013). Irrigation water is conveyed by gravity through either earthen or concrete-lined canals and then distributed to individual plots using bund breaks. Kou Valley irrigation scheme is a suitable case study, as it directly experiences water scarcity due to the reduced exploitable flow rates in the Kou River, resulting from a decrease in rainfall over the last two decades (Johnson et al. 2023a) and increasing water withdrawals for the city of Bobo-Dioulasso for industrial and households needs (Dembelé et al. 2005).

The region’s climate is classified as Sudanian with a unimodal rainfall pattern and an annual rainfall varying between 800 and 1,200 mm, primarily concentrated between June and September. The average annual minimum and maximum temperatures are 21.2 °C and 33.8 °C, respectively (Johnson et al. 2023a). The dominant soil types are Luvisols and Regosols (Johnson et al. 2023a) with generally sandy-clay-loam textured topsoils, progressively turning into clay in subsurface horizons (FAO and IIASA 2023). Farmers typically engage in two rice-growing seasons each year: the dry season (from January to May) and the wet season (from June to November). More information on the scheme is presented in Johnson et al. (2023a).

Description of the on-farm trials and settings

A series of on-farm trials were conducted during the wet and dry seasons of 2019 and the dry season of 2020, comparing alternate wetting and severe drying (AWD30) irrigation to farmers’ irrigation practices (FP). The purpose of the experiment was explained during cooperative group meetings, and farmers were chosen based on their willingness to participate. Initially, our goal was to involve 15 participating farmers per cropping season. However, logistic challenges and setbacks in on-farm trials (non-cooperative farmers, farmers’ non-compliance with harvesting data collection instructions, cattle grazing fields), led to a total sample size of 30 farmers’ fields (12 in the wet season of 2019, 12 in the dry season of 2019 and 6 in the dry season of 2020). No experiment was conducted twice in the same farmer's field.

In each field, two plots of about 500 m2 each with the same cropping histories were delimited by 30 cm high consolidated bunds of 50 cm width to prevent seepage. In these plots, two water irrigation regimes were applied, namely, i) farmer’s practice of irrigation (FP) and ii) AWD30 irrigation practice. The plots were irrigated using the scheme irrigation system. Farmer’s practice of irrigation differed from conventional continuous flooding irrigation due to the non-permanent availability of water in the canals. Irrigation frequency was further determined by both individual farmers’ water management decisions and the scheme-dependent water supply schedule, with water provision varying from every 4 to 5 days. Generally, in the absence of recent rainfall, whenever water is available, farmers irrigate their fields to reach a standing water level of 10 – 15 cm depth. In AWD30 plots, farmers were instructed to maintain a ponded water level of 2–5 cm during the initial 21 days after transplanting (DAT) to assist seedling recovery and ensure weed suppression. Thereafter, the AWD30 irrigation treatment was imposed. Following the installation of the perforated PVC tubes in every AWD plot to monitor the field water level (Lampayan et al. 2015), farmers received training in applying AWD to a threshold water level of 30 cm below the surface. As water deficit during the flowering stage of rice can increase spikelet sterility, farmers were instructed to halt the drying phase in AWD30 plots and to maintain ponded water levels of 2 to 5 cm during this critical stage (Bouman et al. 2007). We provided farmers with specific window dates corresponding to this period based on the phenological characteristics of the variety they planted. Thereafter, farmers applied again AWD30 until about 2 weeks before rice harvest. Except for irrigation management, plots in the same fields had identical cropping and land-use histories and were managed according to farmers’ cropping practices. Farmers selected their preferred rice variety and decided on the type, quantity, and timing of input (herbicides, pesticides, and fertilizer) use. For further analysis, which includes assessing nutrient concentration in plants and specific indices of N and P use efficiency, we selected a representative subset of farmers’ fields (5 in the wet season and 6 in the dry season). Within each irrigation treatment plot in these fields, we marked out and isolated a subplot of 10 m × 12.5 m with no application of mineral NPK, where we assessed rice biomass and grain yield to calculate agronomic nutrient use efficiencies. We used consolidated bunds that were 30 cm high and 50 cm wide for delimitation.

Soil and plant sampling and analysis

Before the onset of the trials, we collected approximately nine soil cores (at a depth of 0 – 20 cm) along the diagonals in each field and combined them to create composite samples. These samples were air-dried, sieved, and analyzed for soil texture (clay, silt, and sand contents), and chemical attributes such as pH (1:2.5 dry soil: water ratio), total organic C, total N, and available P (Bray-2) following the analytical methods described by Johnson et al. (2019).

Six randomly selected rice hills were sampled at maturity from each plot as well as from the nutrient omission subplot to determine yield components (tillers and panicles number), total shoot biomass, and nutrient uptake. Plant samples were separated into straw and grain and oven-dried at 60 ºC for 72 h to calculate the harvest index.

Dried grain and straw samples were fine-ground using a vibrating disc mill (Siebtechnik TS 250, Siebtechnik GmbH, Mülheim an der Ruhr, Germany) and stored in plastic containers for subsequent concentration analysis of N, P, K, Mn, Fe, and Zn. The total N in plant materials was determined using an elemental analyzer (EURO EA Elemental Analyzer series 3000, EuroVector, Pavia, Italy). Plant P, K, Mn, Fe, and Zn were determined after microwave digestion with concentrated nitric acid (HNO3) using an Inductively Coupled Plasma—Optical Emission Spectrometer (Thermo Fischer ICP-OES iCAP PRO, Thermo Fisher Scientific GmbH, Dreieich, Germany). N, P, K, Mn, Fe, and Zn concentrations in straw and grain were expressed on a dry weight basis. N, P, and K uptake were calculated as the sum of straw and grain dry matter weights, multiplied by their respective nutrient concentrations.

Field survey, observations, and data collection

Weather data (solar radiation, maximum and minimum air temperature, relative air humidity, and rainfall at daily time steps) were retrieved from “aWhere”, a cloud-based real-time data platform (aWhere, Inc., Broomfield, Colorado, United States).

We collected data on (i) farm management practices (source of seeds, straw and residue management, tillage method, land levelling, sowing date, planting density, number of seedlings per hill, weed management, fertilizer management, pests control, irrigation frequency), (ii) weed infestation at flowering, both below and above rice canopy (visually assessed, categorized and scored into four levels: 0 = No weed, 1 = weed cover ≤ 10%, 2 = weed cover > 10% and ≤ 30%, and 3 = weed cover > 30%) (Gongotchame et al. 2014), and (iii) farmers’ yield using a standardized protocol [see Niang et al. (2017) and Tanaka et al. (2017)] (Table S1). Three harvest areas of 12 m2 (~ 360 hills) each were demarcated within both FP and AWD30 plots for assessing farmers’ land and crop management practices and determining grain yield. The harvest areas were positioned along a diagonal to capture the variability of productivity within the fields. At maturity, grain yield was measured within each harvest area and expressed at a standardized moisture content of 14%.

Every three days, the field water status in each plot was visually assessed and scored using a three-point scale [1: ponded water (i.e., standing water table with a depth of more than 1 cm), 2: wet soil surface (i.e., fully saturated or partially saturated soil without standing water table), 3: dry soil surface (i.e., soil surface does not exhibit any visible signs of moisture)] (Haefele et al. 2006). Additionally, the frequency of irrigation (the number of irrigations during the growing cycle) for each irrigation management treatment was recorded. We estimated the water flow rates using the bucket method. The bucket method is a straightforward and efficient technique for measuring low to medium flow rates in channels or small streams. It entails placing a bucket of known capacity at the plot's inlet to capture all the flowing water through bund breaks while timing how long it takes for the container to fill using a stopwatch. The flow rate is then calculated by dividing the volume of the container by the filling time (Saito et al. 2023a; Trimmer 1994). To ensure accuracy, we repeated these steps at least three times and calculated the average flow rate. In each plot, the quantity of water applied per irrigation was estimated three to four times during the growing season. Total irrigation water input was calculated by multiplying the number of irrigation events by the average water quantity per individual irrigation event.

Soil water status, nutrient use efficiency indices, and performance indicators

Soil water status, nutrient use efficiency indices, and other performance indicators were computed for each plot. The soil flooding index (%) was calculated as the ratio between the number of days with ponded water and the total number of recorded days during the growing season (from three weeks after transplanting to maturity) (Eq. 1).

$$Soil\;flooding\;index=\frac{Number\;of\;days\;with\;a\;ponded\;water}{Total\;number\;of\;recorded\;days\;during\;the\;growing\;season}\times100$$
(1)

Similarly, the soil dryness index was calculated using the number of days with a dry soil surface (Niang et al. 2018).

$$Soil\;dryness\;index=\frac{Number\;of\;days\;with\;a\;dry\;soil\;surface}{Total\;number\;of\;recorded\;days\;during\;the\;growing\;season}\times100$$
(2)

A soil flooding index of 0 indicates a permanently dry soil surface or wet soil conditions, while a flooding index of 100 indicates a permanent presence of ponded water.

We computed water productivity (WP, in kg grain m−3 water), irrigation water productivity (IWP, in kg grain m−3 water), and partial factor productivity of applied mineral N (PFPN, in kg grain kg−1 N), P (PFPP, in kg grain kg−1 P), K (PFPK, in kg grain kg−1 K) (Fixen et al. 2015; Sharma et al. 2015) as follows:

$$WP=\frac{Yield}{Total\;volume\;of\;water\;input}$$
(3)
$$IWP=\frac{Yield}{Volume\;of\;irrigation\;water\;input}$$
(4)
$${PFP}_x=\frac{Yield}{{Rate\;of\;nutrient}_x\;applied}$$
(5)

whereby Yield represents grain yield expressed in kg ha−1. The total volume of water input comprises both irrigation and rainfall and is expressed in m3 ha−1 during the rice-growing period, and the rates of N, P, and K applied are expressed on an elemental basis (i.e., kg N, P, or K ha−1).

In addition, specific indices of N use efficiency such as agronomic efficiency (kg grain kg−1 N), recovery efficiency (%), physiological efficiency (kg grain kg−1 N), soil N dependent (%) rate were calculated according to Ladha et al. (2005) and Ye et al. (2007) and using the following equations:

$$Agronomic\;efficiency\;of\;N=\frac{{Yield}_T-{Yield}_0}{Rate\;of\;N\;applied}$$
(6)
$$Recovery\;efficiency\;of\;N=\frac{{Uptake}_T-{Uptake}_0}{Rate\;of\;N\;applied}\times100$$
(7)
$$Physiological\;efficiency\;of\;N=\frac{{Yield}_T-{Yield}_0}{{Uptake}_T-{Uptake}_0}$$
(8)
$$Soil\;N\;dependent\;rate=\frac{{Uptake}_0}{{Uptake}_T}\times100$$
(9)

where YieldT represents the grain yield (kg ha−1) in the plot with total N supply (from both the soil and fertilizer); Yield0 is the grain yield (kg ha−1) in the plot without N fertilizer or with only the soil N supply, UptakeT (kg ha−1) is the plant N uptake measured in aboveground biomass in the plot that received N fertilizer, and Uptake0 (kg ha−1) indicates the plant nutrient uptake in the plot without N fertilizer application.

In addition, for P, we calculated the partial P balance (%) (Syers et al. 2008), expressing the total P uptake as a percentage of the P applied (Eq. 10).

$$Partial\;P\;balance=\frac{P\;Uptake}{Rate\;of\;P\;applied}\times100$$
(10)

Statistical analyses and visualization

Field survey data were collected using Android mobile devices, and submitted on a server, where they were downloaded, cleaned, gathered into an Excel spreadsheet, and analyzed using R. 4.3.1 (R Core Team 2023).

Descriptive statistics and Student's t-test

Weather and soil fertility attributes, and farm management practices presented as continuous variables were analyzed using descriptive statistics (mean and standard deviation) while percentage shares were calculated on categorical variables. We computed Chi-squared (χ2) tests of homogeneity to determine whether the proportions of categorical variables were the same across wet and dry seasons (Franke et al. 2012). Continuous variables related to farm management practices were compared between seasons using either the Student's t-test when the normality of residuals and homogeneity of variance were met or the Mann–Whitney-Wilcoxon test as the non-parametric alternative.

Mixed-effects models

We visualized 30 pairwise comparisons of AWD and FP for yield, water productivity, and irrigation water productivity using scatterplots. We subsequently determined the proportion of fields where AWD30 outperformed FP.

Linear mixed-effects models [LMMs, lme4 package; Bates et al. (2015)] were applied with irrigation management as fixed factors and field as a random factor (1| Field), to compare the overall performance of AWD30 and FP on different response variables (weed infestation, grain yield, yield components, water productivity, partial factor productivity of applied nutrients, water input, N and P use efficiency indices, nutrient concentration in straw and grain, total N, P and K uptake) for both wet and dry seasons. “Field” was considered as a random effect to account for repeated measurements, with two plots within each farmer’s field. We combined the data from the two dry seasons due to the absence of significant differences in terms of yield and water productivity. This approach was also adopted to prevent a reduction in the statistical power of the tests. P-values for the fixed effect (irrigation management) were estimated using type-III analysis of variance with Satterthwaite's method, implemented using the lmerTest package (Kuznetsova et al. 2017). When assumptions for the application of LMMs were violated (i.e., heteroscedasticity and deviance of residuals from normality), robust linear mixed-effects models [with Robustlmm package; Koller (2016)] were alternatively run and p-values of the fixed effect were obtained using the emmeans package (Lenth 2022). To model count variables (e.g., tiller or panicle number and frequency of irrigation), we employed Poisson generalized linear mixed-effects models (GLMMs) with the log-link function, using the package lme4 (Bates et al. 2015). After checking overdispersion concerns in the Poisson model, we estimated the p-values of the fixed effect according to the Wald test.

Correlation analysis

Correlation analyses were conducted to explore the relationships between inputs (water, N, P, and K application rates), and outputs (N, P, and K uptakes, and grain yield), regardless of irrigation treatments (FP and AWD30) and season. Furthermore, correlations between grain yield, water productivity, and N and P use efficiency indices were computed to assess potential trade-offs among these performance indicators. The pairwise associations between variables were visualized using a correlogram plotted with the package corrplot (Wei and Simko 2021). As the standard Pearson's correlation estimate is parametric, relying on assumptions of normality and homoscedasticity, it can be sensitive to extreme values. To address these concerns, we opted to compute Spearman's rank correlation coefficient (ρ) (Wilcox 2016). The effect size ρ was interpreted following Cohen’s (1992) guidelines.

Results

Weather conditions, soil attributes, and farmers’ management practices

The weather conditions during the rice growing period differed between seasons. The relative air humidity varied between 64 and 39%, and the cumulative rainfall amounted to 258 and 60 mm in the wet and dry seasons, respectively. Cumulative solar radiation and maximum air temperature during the growing period were higher in the dry season (639 kWh m−2 and 37 °C) than in the wet season (556 kWh m−2 and 34 °C) (Table S2).

The soil clay content varied between 4 and 45% and the sand content between 40 and 68%. Soil pH (H2O) ranged from 5.2 to 6.2 with an average pH of 5.7. Total N ranged from 0.06% to 0.21% and total organic C from 0.70% to 2.0% with an average of 1.19%. Available P (Bray-2) exhibited high variability (CV = 101%) among farmers’ fields, with an average of 20 mg kg−1 (Table S3).

The mean sowing date in the nursery was August 14th (± 16 days) in the wet, and January 13th (± 20) in the dry season. On average, farmers transplanted 25-day-old seedlings. Rice straw from previous seasons was removed for multiple uses (animal feeding or building of rammed earth houses with straw bales) in a quarter of the studied fields, burned in about half of the fields, and incorporated, mulched, or used for animal feeding in the field in another quarter of the fields. These proportions were similar in both seasons (Table S4). More than 80% of the farmers used animal traction for land preparation and nearly all the fields were well-leveled. Besides, non-certified seeds were used in almost 90% of the surveyed fields. The average growth duration of varieties was 109 ± 11 days and was similar in both seasons. Transplanting was the main crop establishment method. During the rice-growing period, farmers weeded their rice fields between one and five times with an average of three weeding operations. In 95% of the cases, farmers applied herbicides at least once (Table S4). While organic manure was applied in approximately 20% of fields, mineral fertilizers were utilized by all farmers. N, P, and K were applied at average rates of 125, 14, and 27 kg ha−1, respectively (Table S4).

Water input

Irrigation frequency and applied amount of water differed between irrigation treatments. In the wet season, the irrigation frequency in FP (7 ± 2) was higher than in AWD30 (4 ± 2). Similarly, in the dry season, the irrigation frequency in FP (16 ± 4) exceeded that in AWD30 (11 ± 3) (p < 0.001). AWD30 reduced irrigation water inputs by 44% in the wet season and by 35% in the dry season. On average, irrigation water contributed to 27% and 79% of the total water input in the wet and dry seasons, respectively. Frequent rainfall events resulted in a higher soil flooding index in the wet season compared to the dry season. As a result, the soil flooding index did not differ between AWD30 and FP irrigation in the wet season (50 ± 28% vs. 60 ± 20%). However, in the dry season, the soil flooding index was lower in AWD30 than in FP (35 ± 19% vs. 51 ± 19%). Regardless of the season, fields rarely fell completely dry, as indicated by the soil dryness index remaining at 0% in the wet season and 2 ± 7% in the dry season (Table 1).

Table 1 Water input and soil water status indices under two water management practices: farmers’ irrigation practices (FP) and alternate wetting and severe soil drying irrigation (AWD30) in Kou Valley, Burkina Faso in 2019–2020
Full size table

Crop growth, weed infestation, grain yield, and water productivity

Across water management practices, grain yields ranged from 1.6 to 8.2 Mg ha−1, with a mean of 4.5 Mg ha−1, and did not differ between seasons. Total water productivity ranged from 0.28 to 4.1 kg grain m−3 of water, with a mean of 1.6 ± 0.7 kg grain m−3 of water, and was higher in the dry season (1.8 ± 0.9 kg grain m−3 of water) than in the wet season (1.4 ± 0.3 kg grain m−3 of water). Conversely, irrigation water productivity was lower in the dry season (2.6 ± 1.7 kg grain m−3 of water) than in the wet season (6.4 ± 3.5 kg grain m−3 of water) (Table S5).

AWD30 performed better than FP in 63%, 97%, and 97% of the pairwise observations for grain yield, total water productivity, and irrigation water productivity, respectively (Fig. 1a, b, and c). Grain yields did not differ between AWD30 and FP in both wet and dry seasons (Table 2). The same trend was apparent for the number of tillers and panicles (Table 3). Water irrigation management did not affect weed infestation in the fields (Table 3). On the other hand, AWD30 improved total water productivity by 20% and 50% compared to farmers’ irrigation practices in wet and dry seasons, respectively (Table 2).

Fig. 1
figure 1

Scatter plots of (a) grain yield, (b) total water productivity, and (c) irrigation water productivity of plots under farmers’ irrigation practices (FP) and alternate wetting and severe soil drying irrigation (AWD30) in Kou Valley, Burkina Faso, in 2019–2020. The rug lines (i.e., short lines along the axes) indicate the range of the data points plotted and provide insights into the distribution of the data along each axis. The red solid line is the identity line or line of equality

Full size image
Table 2 Grain yield, water productivity, and partial factor productivity of applied nutrients under two water management practices: farmers’ irrigation practices (FP) and alternate wetting and severe soil drying irrigation (AWD30) in Kou Valley, Burkina Faso in 2019–2020
Full size table
Table 3 Weed infestation and plant growth parameters under two water management practices: farmers’ irrigation practices (FP) and alternate wetting and severe soil drying irrigation (AWD30) in Kou Valley, Burkina Faso in 2019–2020
Full size table

In non-fertilized plots, grain yields ranged from 1.3 to 4.4 Mg ha−1 with a mean of 2.1 Mg ha−1 in the wet season and from 1.6 to 4.6 Mg ha−1 with a mean of 3.1 Mg ha−1 in the dry season. In these plots, grain yields, total water productivity, and irrigation water productivity did not differ between AWD30 and FP in both wet and dry seasons (Table S6).

Nutrient use efficiencies

Across seasons and irrigation management practices, the partial factor productivity of applied N (PFPN) ranged from 15 to 82 with a mean of 37 kg grain kg−1 N. The PFP of applied P ranged from 138 to 624 with a mean of 331 kg grain kg−1 P and was similar in both wet and dry seasons. Similarly, the PFP of applied K varied from 72 to 328 kg grain kg−1 K (Table S5). Furthermore, in both wet and dry seasons, PFPN, PFPP, and PFPK did not differ significantly between AWD30 and FP (Table 2).

The agronomic use efficiency (13 ± 5 kg grain kg−1 N), the recovery efficiency (23 ± 11%), the physiological efficiency (66 ± 30 kg grain kg−1 N) of N, and the soil N dependent rate (54 ± 17%) were comparable between AWD30 and FP in both seasons. Similarly, there were no significant differences in partial P balance between AWD30 and FP in both seasons with a mean value of 131 ± 63% (Table 4).

Table 4 N and P use efficiency indices under two water management practices: farmers’ irrigation practices (FP) and alternate wetting and severe soil drying irrigation (AWD30) in Kou Valley, Burkina Faso in 2019–2020
Full size table

Nutrient content and uptake

With coefficients of variation (CV) ranging from 13 to 62%, straw, and grain samples exhibited a wide range of nutrient concentrations. Across seasons and irrigation management practices, the N concentration in straw (CV of 18%) and the N, P, K, and Zn concentration in grain (CV of 13%, 18%, 14, and 14%, respectively) were less variable than the Mn and Fe concentrations in straw (CV of 62% and 58%, respectively). In both seasons, water management practices (AWD30 vs. FP) did not significantly affect nutrient concentrations in both straw and grain, with few exceptions. In the wet season, the N concentration in straw was reduced by 24% with AWD30 compared to FP (p = 0.01). Similarly, in the dry season, the P, K, and Zn concentrations in grain were reduced by about 10–20% compared to FP (Table 5).

Table 5 Nutrient concentration in straw and grain under two water management practices: farmers’ irrigation practices (FP) and alternate wetting and severe soil drying irrigation (AWD30) in Kou Valley, Burkina Faso in 2019–2020
Full size table

In the wet season, N, P, and K uptake values were 68 ± 18 kg N ha−1, 18 ± 5 kg P ha−1, and 101 ± 36 kg K ha−1 with farmers’ irrigation practices and 57 ± 15 kg N ha−1, 14 ± 3 kg P ha−1, and 84 ± 14 kg K ha−1 with AWD30. In the dry season, N, P, and K uptake values were 67 ± 21 kg N ha−1, 18 ± 8 kg P ha−1, and 103 ± 41 kg K ha−1 with farmers’ irrigation practices and 64 ± 25 kg N ha−1, 16 ± 8 kg P ha−1, and 107 ± 31 kg K ha−1 with AWD30. Differences between AWD30 and FP were not significant (p > 0.05) except for N uptake during the wet season, which was 16% lower in AWD30 compared to FP (p = 0.045) (Table 6).

Table 6 N, P, and K uptake under two water management practices: farmers' irrigation practices (FP) and alternate wetting and severe soil drying irrigation (AWD30) in Kou Valley, Burkina Faso in 2019–2020
Full size table

Productivity trade-offs and synergies

The N application rate had a large positive effect (0.5 < ρ < 0.7; p < 0.05) on K concentration in straw and K uptake. While total water input had no significant effect on grain yield and N, P, and K uptake, increasing the soil flooding index led to higher N and P uptake by rice (Fig. 2a). There was a strong positive correlation (ρ = 0.8; p < 0.05) between P concentration in grain and total water input, suggesting that reducing water input may lead to a decrease in P concentration in grains. Conversely, K concentration in straw showed a negative correlation (ρ = -0.48; p < 0.05) with total water input.

Fig. 2
figure 2

Relationships (a) between inputs (water, N, P, and K application rates), and outputs (N, P, and K uptakes, and grain yield) and (b) between grain yield, water productivity, and N- and P- use efficiencies (n = 22) in Kou Valley, Burkina Faso, in 2019–2020. The values displayed in the matrix are Spearman's rank correlation coefficients (ρ). Positive correlations are displayed in green and negative correlations in golden brown. The colour intensity (see colour bar) and the square size are proportional to the correlation coefficients. Significant correlation coefficients are indicated by asterisks (* p < 0.05, ** p < 0.01, and *** p < 0.001), while coefficients without asterisks are not statistically significant (p > 0.05)

Full size image

Significant positive correlations were observed between grain yield and the agronomic use and recovery efficiencies of N, and the partial P balance. Agronomic use and recovery efficiencies of N were not negatively correlated with the total water productivity (p > 0.05) (Fig. 2b).

Discussion

To the best of our knowledge, this is the first on-farm study conducted in the semi-arid zone of West Africa, investigating the effect of AWD irrigation on nutrient uptake in irrigated rice. It showed in locations with a shallow groundwater table, compared to the farmers’ irrigation practices (FP), alternate wetting and severe soil drying irrigation (AWD30) can reduce water input without compromising grain yield and the use efficiencies of applied N and P. However, the observed positive correlation between soil flooding index and N and P uptake raises concerns about possible trade-offs between water-savings and nutrient uptake in case of acute soil dryness. The following discussion of key findings is organized into three parts: (i) the water-saving potential of AWD irrigation and its impacts on productivity; (ii) the impact of AWD irrigation and anaerobic conditions on nutrient use; and (iii) the study's limitations and outlook.

AWD reduced water input without grain yield penalty

Alternate wetting and severe soil drying irrigation (AWD30) reduced total water input by 14% and 28% in the wet and dry seasons, respectively, without compromising the grain yield. This led to a 20% increase in water productivity during the wet season and a 50% increase during the dry season. This water-saving is attributed to the reduction in unproductive water outflows, including percolation, evaporation, and seepage. The water savings recorded in this study are lower than those reported in previous studies (30%-36%), comparing alternate wetting and severe drying regime to continuous submerged field conditions (Carrijo et al. 2017). This could be explained by the reference for the comparison, which differs from the continuous submerged field conditions. Indeed, in the present study, farmers’ irrigation practices provided a water layer only on ~ 60% of the days during the growing season. Compared to alternate wetting and moderate drying regime (AWD15) in the same scheme in the dry season (Johnson et al. 2024), AWD30 did not increase water savings. Thus, despite a reduced irrigating frequency, irrigation water inputs tended to be higher for individual irrigation events to compensate for the more extensive water losses from the soil water storage reservoir in AWD30 compared to AWD15. This raises questions about the relevance of severe drying in the dry season. Moreover, alternate wetting and severe soil drying irrigation presents higher risks and can entail a yield penalty of about 22% (Carrijo et al. 2017). In our study, AWD30 did not compromise grain yield likely because the soil did not completely dry out due to the capillary rise of water from the shallow groundwater table (Table 1). Additionally, partial soil drying also occurred in farmers' irrigation practices (FP). The assessment of the field water status was exclusively done through visual observation, limiting a comprehensive comparison of the effect of the two water management irrigation regimes (FP and AWD30) on soil dryness levels. Using appropriate instruments, such as tensiometers, could have improved the precision of measurements, resulting in more accurate and reliable insight into the soil moisture content.

AWD did not impair nitrogen and phosphorus use efficiencies

Regardless of the seasons, the partial factor productivity and the agronomic use and recovery efficiencies of applied mineral N were comparable for both irrigation management practices. However, the mean agronomic use efficiency of 13 kg grain kg−1 N and the N recovery of 23% were lower than the global average values of 22 kg grain kg−1 N, and 46%, respectively (Ladha et al. 2005). Such low N use efficiency values may well reflect poor N fertilizer management in farmers’ fields, a common issue in rice production in sub-Saharan Africa (Chivenge et al. 2021). Thus, site-specific nutrient management has been advocated for optimizing fertilizer use, enhancing the synchrony of N supply and demand, and improving N use efficiency (Saito et al. 2023b). Synergistic interaction effects between site-specific N management and AWD could well enhance grain yields while also increasing N-use efficiency (Liu et al. 2013).

There is no consensus in the scientific debate on whether AWD improves N-use efficiency or not. Thus, some studies indicated low N use efficiency under AWD irrigation due to increased NO3−-N losses (Dunn and Gaydon 2011). Other studies demonstrated improved N-use efficiency with AWD (Djaman et al. 2018; Liu et al. 2013), while yet others showed no significant difference in N-use efficiency between continuous submergence and AWD irrigation (Cabangon et al. 2004; Islam et al. 2022; Pan et al. 2017). These studies explained that AWD irrigation may decrease N losses due to the decrease in N leaching (Wang et al. 2011), and NH3 volatilization (Xu et al. 2012). On the other hand, AWD may potentially enhance NO3-–N losses through denitrification during severe soil drying and subsequent soil flooding (Yang et al. 2017), which may, in turn, negatively affect N use efficiency (Zhang et al. 2009). However, in the present study, agronomic use, recovery, and physiological efficiencies of N were not impaired by alternate wetting and severe drying. The shallow groundwater table may have mitigated the effects of severe drying, reducing the NO3--N formation and the subsequent N losses (Wang et al. 2016). This aligns with suggestions from studies in China and the Philippines conducted in environments with shallow groundwater tables (Belder et al. 2005; Cabangon et al. 2004). On the other hand, we found a positive correlation (ρ = 0.53; p < 0.05) between the soil flooding index and N uptake supporting that soil submergence fosters N uptake. This result is somewhat consistent with Atwill et al. (2018) who demonstrated that N uptake were reduced for aerobic irrigated rice compared to other irrigation treatments having higher soil flooding index. Soil submergence can buffer pH extremes. Thus, Narteh and Sahrawat (1999) surmised that the convergence of soil pH in the neutral range following soil submergences benefits the lowland rice crop through an improved plant availability of NH4+.

It is still unclear whether AWD irrigation improves P-use efficiency (Ishfaq et al. 2020). Contrary to the results of pot experiments conducted by Deng et al. (2021) and Acosta-Motos et al. (2020), we found no evidence of altered P-use efficiency with AWD30 compared to farmers’ irrigation practices. In the dry season, AWD30 even decreased grain P concentrations compared to farmers’ irrigation practices as reported before (Norton et al. 2017a; Song et al. 2018). Overall, sustaining a perched water table fosters rice P uptake (Fig. 2a). This is primarily because anaerobic reducing conditions in submerged soils may increase the concentration of plant availability P in the soil solution (Young and Ross 2001) due to the supplementary release of organic P via redox-sensitive dissociation from iron/manganese oxides. In contrast, aerobic conditions may reduce the amount of available P because of P precipitation with oxidized Fe and Mn, or by P immobilization in the soil microbial biomass (Adhikary et al. 2023). Therefore, increased P fertilizer application rates may be required under AWD, especially in acid soils (Johnson et al. 2024).

Although inconsistent across seasons, alternate wetting and severe soil drying irrigation negatively affected nutrient concentrations in rice grains. Thus, in the dry season, P and K concentrations in rice grain decreased under AWD30 as reported before by Norton et al. (2017b). In contrast to P, Zn is usually less available in anaerobic (low redox potential) than in aerobic (high redox potential) soils (Johnson-Beebout et al. 2009). Yet, we observed lower Zn concentration in rice grains under AWD30 in the present study (Table 5). This contrasts also with previous studies that showed AWD significantly increased available soil Zn but did not affect grain Zn concentration (Rubianes et al. 2018). Further investigations are needed to explore the effect of AWD on soil zinc availability and grain zinc concentration.

Limitations of the study and outlook

This on-farm-based study improved our understanding of the impact of AWD on performance indicators of irrigated rice in fields with a shallow groundwater table (Cabangon et al. 2004). We were able to demonstrate the feasibility of the AWD technology and its implementation in the local context of Burkina Faso. However, uncertainties persist due to the relatively small sample size, variabilities in farmers' implementation of the AWD technology, and site-specific conditions of water availability. Additionally, obtaining farmers' approval to make significant changes to the setting or layout of their fields was challenging, especially for this series of on-farm farmer-managed trials, where participation was voluntary and farmers did not receive any incentives. Consequently, we were unable to implement additional measures, such as those carried out in on-station trials (Krupnik et al. 2012a), to reduce underground seepage, apart from using large consolidated bunds. Therefore, the water savings and water productivity gains attributed to AWD30 may have been overestimated and should be considered with caution.

A larger study is needed to confirm the effect sizes on performance indicators (Hackshaw 2008) and to further investigate the interaction between water management and fertilizer application rates. Subsequent investigations using locally calibrated and validated models (Grotelüschen et al. 2021, 2022) can help draw insights into a wide range of climatic and hydro-edaphic conditions, as well as their interactions. In future on-farm trials, we recommend monitoring the field water level and soil water potential to assess their impact on grain yield, nutrient uptake, and use efficiency (Tuong and Bouman 2003). Furthermore, expanding the study to contrasting environments will be crucial in determining suitable niches for scaling the AWD irrigation technology to other irrigation schemes in the semi-arid zone of West Africa.

Conclusion

Compared to actual farmers’ irrigation practices, alternate wetting and severe soil drying reduced irrigation water input by 37% (mean across seasons) without yield penalty while maintaining N- and P-use efficiencies. This improved irrigation water productivity by about 39%. This study did not reveal any evident trade-offs between water productivity and agronomic use and recovery efficiencies of N. We conclude that alternate wetting and severe soil drying irrigation can efficiently increase water productivity and serve as a water-saving technology for sustainable rice production in locations with shallow groundwater tables, as observed in the case of the Kou Valley. Therefore, water management institutions and policymakers should incentivize smallholder rice farmers to adopt this water-saving technology in these locations in Burkina Faso and beyond. Furthermore, this study underscores the importance of the hydrological characterization of irrigation schemes in West Africa, especially in terms of percolation rate and groundwater table depth. Future investigations in contrasting environments should also explore the influence of alternate wetting and severe soil drying irrigation on greenhouse gas emissions and its long-term effects on soil chemical properties, as well as nitrogen and carbon storage.