Lowland rice is grown in bunded fields that farmers keep flooded for most of the season. The ponded layer suppresses weeds, buffers the crop against irregular supply and makes nutrients such as phosphorus more available, and for these reasons flooding has been the default for as long as irrigated rice has existed. It is also expensive in water. A flooded field loses water continuously by seepage through the bunds and percolation through the plough pan in addition to what the crop transpires, and the total input to a rice field is commonly two to three times that of other irrigated cereals (Bouman and Tuong, 2001; Tuong and Bouman, 2003).

Flooded soil is also anaerobic, and anaerobic decomposition of organic matter ends in methane. Rice cultivation appears in every global methane budget as one of the larger agricultural sources (Saunois et al., 2020). A practice that interrupts flooding therefore promises two benefits at once, and alternate wetting and drying (AWD) is the best known of them. Under AWD the field is irrigated, the water is left to recede until the soil surface is exposed and the shallow water table has dropped by a set amount, and the field is then irrigated again.

The practice has been taken up in national programmes in several Asian countries and in climate mitigation plans, yet reports from the field describe adoption that is patchy and often temporary (Lampayan et al., 2015). This review considers the evidence behind the two promised benefits and then turns to the question that the agronomic literature tends to leave until last: why a practice that appears to cost the farmer nothing is not more widely used.

This is a narrative review of the published literature, organised around three themes: water, greenhouse gases and adoption. It relies on field experiments from the main rice-growing regions, on the meta-analyses that have pooled them, on the guidance documents of the International Rice Research Institute and the Intergovernmental Panel on Climate Change, and on socio-economic studies of adoption in South and Southeast Asia. The selection is not exhaustive and no systematic search was carried out. Where sources disagree, the disagreement is reported and an explanation is offered where the literature supports one.

AWD is a family of practices that differ in how dry the soil is allowed to become. The form recommended for extension is "safe AWD", defined by a simple field measurement. A perforated tube is sunk into the soil so that the farmer can see the depth of the water table inside it. Irrigation is applied when the water level has fallen to about 15 cm below the soil surface, the field is reflooded to a shallow depth of around 5 cm, and ponding is maintained from about a week before to a week after flowering, when the crop is most sensitive to water deficit (Bouman et al., 2007; Richards and Sander, 2014). Figure 1 shows one cycle. At the 15 cm threshold the roots of the rice plant still reach saturated soil, which is why the practice is described as safe.

Researchers have also tested more severe regimes in which irrigation is delayed until the soil water potential in the root zone falls well below saturation. These variants save more water and, as discussed below, are the ones in which yield losses appear. The distinction matters because studies that report "AWD" without stating the threshold cannot be compared, a problem noted in every synthesis of the literature (Carrijo et al., 2017).

Field experiments consistently show lower irrigation input under AWD than under continuous flooding. The size of the reduction varies widely with soil and hydrology. On heavy soils with a shallow groundwater table the saving is small, because little water was being lost by percolation in the first place; in the experiments of Belder et al. (2004) at such sites in China and the Philippines, the ponded water disappeared but the soil stayed close to saturation and yields were unaffected. On lighter soils with deep groundwater the saving is larger. A meta-analysis of published comparisons concluded that AWD reduced water use substantially overall and that mild AWD did so without a significant yield change (Carrijo et al., 2017).

What is saved needs closer inspection. Evapotranspiration changes little under safe AWD, since the canopy still transpires at the potential rate and the soil surface is wet for much of the time. The reduction comes mainly from seepage and percolation, which fall when there is no standing head of water. At the scale of the field these are losses. At the scale of an irrigation system they are often not, since percolated water recharges an aquifer from which other farmers pump, and seepage reaches drains that are reused downstream (Bouman and Tuong, 2001; Tuong and Bouman, 2003). AWD certainly saves pumping energy and canal deliveries for the farmer who practises it. Whether it releases water for other uses depends on where the "lost" water was going, and few studies have followed it.

Two further qualifications apply. Much of the water used by a transplanted rice crop goes into land preparation, when the field is soaked and puddled before planting, and AWD does nothing to reduce that share. The comparison also depends on how the flooded field was managed. A plot kept carefully at a shallow depth loses less water than a field held deep as insurance against an unreliable supply, so the saving to be expected in practice depends on the depth of water that farmers habitually keep (Bouman et al., 2007).

The meta-analysis by Carrijo et al. (2017) distinguished mild from severe AWD and found that yields under the mild form did not differ from those under continuous flooding, whereas severe drying reduced yield markedly. Soil properties modified the response, with losses more likely on soils of low organic carbon and high pH. Individual studies range more widely. Some report yield gains under moderate AWD, attributed to deeper rooting, fewer unproductive tillers and better grain filling (Yang et al., 2017), while trials with poor water control report losses. In direct-seeded systems in California, drying cycles were introduced without loss of yield (LaHue et al., 2016). The practical reading is that safe AWD is yield-neutral where the farmer can reflood on time, and that the risk lies in not being able to.

Methane in a rice field is produced by methanogenic archaea in reduced soil, consumed in part by methane-oxidising bacteria in the thin oxic zones at the soil surface and around roots, and carried to the atmosphere chiefly through the aerenchyma of the rice plant (Le Mer and Roger, 2001; Conrad, 2007). Drying the field admits oxygen, raises the redox potential, halts methanogenesis and regenerates oxidised iron and sulphate that suppress methane production for a time after reflooding. The effect of drainage on emissions was shown by continuous field measurement in Japan in the 1990s (Yagi et al., 1996) and has since been confirmed across Asia, where water regime and organic amendments emerge as the dominant controls on emission (Wassmann et al., 2000; Yan et al., 2005).

National greenhouse gas inventories reflect this. The IPCC method scales a baseline emission factor for continuously flooded rice by a factor for water regime, and the factors for single and multiple drainage are well below one (IPCC, 2006; IPCC, 2019). A global meta-analysis of water management found that non-continuous flooding roughly halved methane emissions relative to continuous flooding (Jiang et al., 2019).

The qualification is nitrous oxide. Alternating aerobic and anaerobic conditions favour nitrification followed by denitrification, and nitrous oxide emissions rise under AWD, particularly when drying follows a nitrogen application (Lagomarsino et al., 2016). Most syntheses conclude that the increase offsets only a small part of the methane reduction when the two gases are combined on a global-warming-potential basis (Jiang et al., 2019; Linquist et al., 2015). Measurements on intermittently flooded farms in India, however, recorded nitrous oxide fluxes far above those assumed in inventories, and the authors argued that water management must be judged on both gases together (Kritee et al., 2018). The matter is not settled, and it argues for mild drying and careful timing of nitrogen in place of the assumption that drier is better.

Drying also changes the chemistry of trace elements. Aerobic periods reduce the mobility of arsenic and lower its concentration in grain (Linquist et al., 2015), a benefit where groundwater carries arsenic. Cadmium behaves in the opposite way and becomes more available as the soil oxidises, so that on cadmium-affected soils AWD may trade one contaminant for another.

If safe AWD maintains yield, saves water and reduces emissions, slow adoption needs explaining. The socio-economic literature points to the conditions under which farmers irrigate more than to the practice. Table 1 groups the barriers most often reported.

Four gaps deserve priority. The first is water accounting above the field. Claims of water saved should be tested with basin or command-area balances that distinguish reductions in consumptive use from reductions in recoverable flows. The second is the combined greenhouse gas balance under farmers' conditions. Most flux data come from research stations with good water control, and measurements on working farms, including nitrous oxide at high temporal resolution, remain scarce. The third is institutional. Comparative studies are needed of the arrangements under which AWD has persisted, such as volumetric charging, prepaid pump cards, scheduled rotations and aggregated carbon projects, so that programmes can be designed around the constraint that binds in each place. Long-term effects on soil organic matter under repeated drying also await study.

The fourth gap is verification. Carbon projects and national inventories need evidence that fields were in fact dried, and neither farmer diaries nor occasional visits provide it at scale. Satellite detection of surface water and low-cost water-level sensors have been proposed for the purpose, but their reliability beneath a closed rice canopy and their cost per farm are not yet well documented.

Safe AWD is among the better-evidenced practices in irrigated agriculture. It lowers irrigation input at the field, keeps yield where water can be reapplied on time, and reduces methane emissions by a margin large enough to survive the nitrous oxide penalty in most assessments. The same body of evidence shows that field-level water savings should not be equated with basin-level savings, and that the practice spreads only where farmers gain from saving water and can rely on getting it back. Efforts to scale AWD are therefore as much a matter of irrigation pricing, scheduling and field infrastructure as of agronomic extension.

To be declared by the author(s) before publication.

To be declared by the author(s) before publication.