Rice yields under water-saving irrigation management

Transcription

1 Department of Physical Geography Rice yields under water-saving irrigation management A meta-analysis Amanda Åberg Bachelor s thesis Geography, 15 Credits GG

2

3 Preface This Bachelor s thesis is Amanda Åberg s degree project in Geography at the Department of Physical Geography, Stockholm University. The Bachelor s thesis comprises 15 credits (a half term of full-time studies). Supervisor has been Stefano Manzoni at the Department of Physical Geography, Stockholm University. Examiner has been Steve Lyon at the Department of Physical Geography, Stockholm University. The author is responsible for the contents of this thesis. Stockholm, 16 June 2017 Steffen Holzkämper Director of studies

4

5 Abstract Water scarcity combined with an increasing world population is creating pressure to develop new methods for producing food using less water. Rice is a staple crop with a very high water demand. This study examined the success in maintaining yields under water- saving irrigation management, including alternate wetting and drying (AWD). A meta- analysis was conducted examining yields under various types of water- saving irrigation compared to control plots kept under continuous flooding. The results indicated that yields can indeed be maintained under AWD as long as the field water level during the dry cycles is not allowed to drop below - 15 cm, or the soil water potential is not allowed to drop below - 10 kpa. Yields can likewise be maintained using irrigation intervals of 2 days, but the variability increases. Midseason drainage was not found to affect yield, though non- flooded conditions when maintained throughout most of the crop season appeared to be detrimental to yields. Increasingly negative effects on yields were found when increasing the severity of AWD or the length of the drainage periods. Potential benefits and drawbacks of water- saving irrigation management with regards to greenhouse gas emissions, soil quality and nutrient losses were discussed to highlight the complexity of the challenges of saving water in rice production. Keywords: rice, yield, water scarcity, water- saving irrigation, WSI, alternate wetting and drying, AWD, soil organic matter, soil organic carbon. 1

6 2

7 Table of contents Abbreviations Introduction Research questions and problem formulation Water management in rice farming Field level water flows Field level irrigation management approaches Alternate wetting and drying, submergance- nonsubmergance and intermittent irrigation Saturated soil culture Controlled irrigation Midseason drainage Influence of water- saving irrigation management on rice yields and water savings Methodology Data collection Data compilation and evaluation Data analysis Results Yields under WSI management Regional differences in WSI yield Influence of severity of WSI management on yield Nitrogen fertilization effect on yields Discussion Rice yields under varying irrigation managements Environmental implications of water- saving irrigation Greenhouse gas emissions and soil quality Nutrient and herbicide losses Implementation of water- saving irrigation management Methodological and data weaknesses Future research Conclusions Acknowledgements References Appendix A

8 Abbreviations ASNS AWD CF FWL SOC SOM SWP WSI WUE alternate submergance- nonsubmergance alternate wetting and drying continuous flooding field water level soil organic carbon soil organic matter soil water potential water- saving irrigation water use effiency 4

9 1. Introduction Rice is a staple food for a large number of the human population and constitutes the largest food source as well as a significant income for inhabitants of developing countries (GRiSP, 2013). Worldwide rice- farming environments are oftentimes divided into four types: lowland irrigated rice, lowland rainfed rice, flood- prone rice and upland rice. Of these environments, the irrigated lowlands, covering approximately 93 million hectares of land, produce 75% of the total world rice production (Bouman et al., 2007; GRiSP, 2013). Rice receives about two to three times more water at the field level than most other crops, and these irrigated rice environments consume approximately 24-30% of the world's freshwater withdrawals. Meanwhile, decreases in water resources and declines in water quality are resulting in water scarcity. Combined with increased competition from urban and industrial sectors, this water scarcity poses a threat to the sustainability of rice production. New methods are thus required to deal with the challenges posed by water scarcity (Bouman et al., 2007). Saving water is rarely a voluntary decision made by farmers. It is more often either an imposed decision made at a higher level or a necessity dictated by physical water scarcity (Bouman et al., 2007). Due to the pivotal role of rice as a staple food for a large part of the world's population, many studies have examined the effect of different irrigation regimes on rice yields (Carrijo et al., 2017). Many of these studies have found a small decrease in yield accompanied by a significant increase in water productivity when water- saving irrigation (WSI) methods were used (see e.g. Bouman and Tuong, 2001 for a summary). Other studies found an insignificant difference in yield between continuous flooding and water- saving methods (e.g. Cabangon et al., 2001; Belder et al., 2004). The purpose of this study is to examine patterns in the relationship between rice yield and water management by systematically collating data from a number of studies. Bouman and Tuong published a meta- analysis in 2001, examining water- saving irrigation at the field level and its impact on yields. Their study provides an excellent opportunity to examine whether yield improvements under WSI management, relative to continuous flooding, have been documented in the 16 years that have passed since. A recently published meta- analysis conducted by Carrijo et al. (2017) will serve as a contemporary comparison. As has been stated by Linquist et al. (2015), though many studies have examined potential benefits of water- saving irrigation, the consequences are rarely evaluated concomitantly. Water management in rice farming has several environmental implications, aside from the challenge of water scarcity. Rice farming emits approximately four times as much greenhouse gas as wheat or maize and therefore has significant potential in terms of mitigating agricultural greenhouse gas contributions (Linquist et al., 2012). Reducing the amount of time the soil is kept under flooded anaerobic conditions has been found to decrease emissions of the strong greenhouse gas methane. However, the conversion to aerobic conditions instead leads to increased microbial activity and increased soil organic matter (SOM) decomposition and CO2 emissions (Sahrawat, 2005; Haque et al., 2016a). SOM has great importance for soil health and agricultural sustainability. The conversion to more aerobic conditions may therefore have significant implications for long- term soil fertility and rice farming sustainability. Furthermore, the implementation of water- saving irrigation has 5

10 also been found to affect nutrient availability in the soil, as well as losses of fertilizers through surface runoff and seepage (Sahrawat, 2005; Yang et al., 2015). Hence, the implementation of water- saving irrigation has many implications that should be considered in addition to the challenges of water scarcity. 2. Research questions and problem formulation The aim of this project is to examine the relationship between yields and water- saving irrigation in rice farming systems using a meta- analysis approach. This study will attempt to collate information from multiple studies to examine said relationship. Specifically, this project will attempt to answer the following questions: Is rice yield consistently higher under continuous flooding compared to alternate wetting and drying and other water- saving forms of irrigation management? Can a spatial pattern be discerned, in which water- saving irrigation has been more successful in any certain region of Asia? Through literature studies, some of the environmental implications of employing water- saving irrigation management will also be qualitatively examined and discussed. This study wishes to place water- saving irrigation in a larger context by providing a summary of both benefits and drawbacks of its implementation. Whether or not different levels of nitrogen fertilizer input affect the success of water- saving irrigation will also be briefly examined. There are many ways to save water aside from changing irrigation practices, such as proper land preparation and bund construction (Bouman et al., 2007), but these measures are largely outside the scope of this paper. No attempt will be made to quantitatively assess actual water savings, though potential water savings will be briefly discussed. The study will be limited to rice systems in East, South, and Southeast Asia. The relatively large spatial extent of field experiments included in this analysis is partly the result of the need to keep the collection of data objective and systematic. Limiting the spatial extent by using search words such as "Southeast Asia" resulted in a very limited results list. Furthermore, this approach enables an examination of whether the effects of WSI are the same over a range of different environmental conditions. Due to the necessity of being able to control the water input to implement WSI management, the focus is inevitably placed primarily on irrigated lowlands. In these environments farmers may, depending on the structure of the irrigation system, have the opportunity to influence not only drainage of water from the fields, but also the input of water (Bouman et al., 2007). Farmers can therefore to a certain degree influence the amount of water- stress experienced by plants during the growth period. A focus on these rice systems is deemed suitable for this study since irrigated lowlands are responsible for such a large part of the global rice production. 6

11 3. Water management in rice farming 3.1 Field level water flows There are various ways in which water can enter and leave a rice field. Inflow occurs through rainfall, irrigation, and capillary rise, and outflow through percolation, seepage underneath bunds, overbund flow, evaporation and transpiration. Transpiration is the only type of outflow that contributes to crop growth and is therefore termed 'productive water use'. Capillary rise is generally negated by the constant downward flow of percolation in flooded rice fields. In a series of fields, both seepage and overbund flow can contribute to adjoining farmers' fields before draining into ditches or the groundwater. Even after entering the groundwater, this water may remain reusable through pumping (Bouman et al., 2007). The high water demand for rice differs from dryland crops and is the result of the daily percolation and seepage of water that occurs in flooded rice fields, along with evaporation from exposed water surfaces. The profuse percolation rates over long periods of time have in many places served to locally raise the groundwater surface. In some locations, the groundwater table is found within 20 cm from the soil surface, and the water is therefore available for direct uptake by the rice roots (Bouman et al., 2007). When the field water level (FWL) is at or above the soil surface and the soil is saturated, such as in flooded paddies, the soil water potential (SWP) near the surface will equal 0 kpa. When the soil is saturated, most of the water is held in large pores where the molecules are not strongly bound by the soil solids and are therefore able to easily move around. As the soil dries, the remaining water is increasingly held in smaller pores closer to the soil solids, where they are more tightly bound and harder for plant roots to extract, This change is measured as an increasingly negative SWP (Brady and Weil, 2008). If there is not enough water available, the rice plant will experience drought stress, expressed, for example, in the closing of stomata and ceasing of transpiration, which can in turn result in yield declines (Bouman et al., 2007). 3.2 Field level irrigation management approaches Before rice is transplanted or seeded, the field is normally ploughed and puddled under wet conditions (Bouman et al., 2007). Puddling is a type of harrowing or rotavating that helps in controlling weeds, but also reduces soil permeability by destroying soil aggregates and creating a plough pan, usually at a depth of approximately 10 to 20 cm. The hydraulic conductivity decreases, and therefore also the loss of water through percolation (Arora et al., 2006; Bouman et al., 2007). Following puddling, fields are usually kept flooded before transplanting for a period ranging from a few days to four weeks, though it has been known to stretch as long as two months in large- scale systems. Once transplanted or seeded, the crop is traditionally kept flooded at a depth of 5 to 10 cm until one or two weeks before harvesting. Flooding following crop establishment helps to control weeds and pests (Bouman et al., 2007). Figure 1 provides an overview of the different growth stages of rice. 7

12 Figure 1. Schematic overview of the different growth stages of the rice plant. Adapted from CGIAR, n.d. For water- saving irrigation to be a feasible alternative, losses of water through seepage, percolation and evaporation must be addressed. Efforts can be made during land preparation by constructing appropriate field channels that enable the control of water levels in individual fields, maintaining good bunds, levelling the field, implementing tillage and minimising the time passing between land preparation and crop establishment (Bouman et al., 2007). Bouman et al. (2007) describe three types of water- saving irrigation; alternate wetting and drying (AWD), saturated soil culture (SSC), and aerobic rice (not covered here, predominantly used in upland environments). Which form is implemented depends on the type and severity of water scarcity, socioeconomical situation and how much control individual farmers can exercise over their irrigation. The implementation of AWD requires that a farmer can control water levels in their own field, or that a communal effort is made. With reduced water availability, saturated soil culture may be the first option, followed by AWD and then aerobic rice when faced with severe shortages (Bouman et al., 2007) Alternate wetting and drying, submergance- nonsubmergance and intermittent irrigation Alternate wetting and drying, sometimes referred to as alternate submergence- nonsubmergance (ASNS) (Belder et al., 2004) or intermittent irrigation (Lin et al., 2012), utilizes cycles of alternating flooded conditions and dry periods when the water is allowed to drop below field level. The length of the dry periods can vary from as little as one day to longer than 10 days (Bouman et al., 2007). Cabangon et al. (2001) state that AWD normally includes a midseason drainage of days in the late tillering stage, and that the dry cycles between irrigation events are normally kept at lengths of two to four days. In practice, however, the pre- designed timing and length of drainages and dry cycles can be difficult to achieve due to the variability of rainfall events. Carrijo et al. (2017) have, in their meta- analysis, chosen to define AWD as any irrigation management that contains a minimum of one single dry cycle with soil conditions below saturation. Their definition differs from those found in most other sources. AWD primarily reduces water use by lessening the amount lost through seepage and percolation. In terms of practical implementation, the use of a field water tube to monitor water levels is recommended (Bouman et al., 2007; Yang et al., 2017). The field water tube also allows farmers to detect 'hidden' groundwater sources (Lampayan et al., 2015). When the water drops to a depth of - 15 cm, the field should be re- irrigated to a ponded depth of approximately 5 cm. The AWD cycles can be implemented starting a few days after transplanting, after two to three weeks if weeds are prolific, or following panicle initiation 8

13 which occurs around 50 days after sowing (Bouman et al., 2007; Cabangon et al., 2001; CGIAR, n.d.). Bouman et al. (2001; 2007) state that the timing of dry cycles with regards to growth stages generally has little to no effect on yield, with the exception that ponded water during flowering is required to avoid yield loss. According to Yang et al. (2017), however, different thresholds should be used at different growth stages due to the variable sensitivity of rice at different points in the crop cycle. The - 15 cm field water depth is often referred to as 'safe AWD', because it keeps the root zone saturated. The water savings are generally around a modest 15%, but yield loss is avoided, and depending on local conditions farmers can experiment with longer dry cycles (Bouman et al., 2007). Though irrigation in AWD treatments is often scheduled based on FWL, other indicators are also in use, such as SWP thresholds or simply a set number of days following disappearance of previous irrigation from the soil surface Saturated soil culture In saturated soil culture (SSC), irrigation is applied to achieve a water depth of approximately 1 cm following disappearance of the previous irrigation. The goal is to keep the soil as close to saturation as possible, which requires very frequent irrigation. The practice reduces the hydraulic head, resulting in decreased seepage and percolation (Bouman et al., 2007). Though examples of similar practices can be found in the academic literature, the term 'SSC' was rarely encountered during this study Controlled irrigation The term 'controlled irrigation' is sometimes employed in the literature without a firm definition. When Yang et al. (2013, 2015) and Hou et al. (2012) employ the term, the management regime is described as including irrigation to keep the soil moist. However, graphs presented in their articles show that irrigation has been applied to reach a FWL of 1 to 4 cm in between regular dry cycles, in practice appearing to make the approach very similar to AWD Midseason drainage 'Midseason drainage' or 'intermittent drainage' are concomitantly used to describe the practice of draining the rice paddy midseason for an extended period, often lasting for about 30 days. The approach is mainly used as a means to achieve decreased methane emissions (Haque et al., 2016a, b), but has also been used as a water- saving measure (Rahman et al., 2013). 3.3 Influence of water- saving irrigation management on rice yields and water savings Based on a number of studies, Bouman et al. (2007) concluded that although AWD has in some instances been found to increase yield, it more often decreases yield. Bouman and Tuong (2001) conducted a meta- analysis based on 31 field experiments using AWD or SSC conducted under various conditions. They found that average water savings under SSC amounted to 23% with small yield reductions of approximately 6%. When SWPs in the root zone were allowed to drop to - 10 to - 30 kpa, however, yield penalties of 10-40% were recorded (Bouman and Tuong, 2001). The variability in results identified in the study is attributed to soil and hydrological conditions and the varying length of dry periods in 9

14 different experiments (Bouman et al., 2007). Carrijo et al. (2017) found a similar water use reduction of 23.4% under AWD across a selection of 56 studies, but with SWPs maintained above - 20 kpa or FWL above - 15 cm, with no yield penalty. Commenting on a several studies conducted in areas with shallow groundwater tables and fine textured soils, Bouman et al. (2007) concluded that the nearness of groundwater to the field level meant that the root zone remained saturated, supplying a hidden water source. A 15-30% lower water input could therefore be achieved without a significant penalty to rice yield. Where the groundwater table is very high and within reach of the roots, potentially negative effects of water- saving irrigation can be mitigated, and yields in relation to irrigation can therefore appear superficially high. The water savings in these environments are relatively small due to the losses already being low when using continuous flooding (CF) under such conditions. A number of studies conducted in loamier soils with deep groundwater tables presented higher water savings, exceeding 50%, but heavy yield penalties in excess of 20% (Bouman et al., 2007). Ye et al. (2013) draws on a number of studies to reason that modern rice varieties have been adapted to semi- aquatic conditions with only intermittent flooding. The aerated conditions assist in SOM mineralization and inhibition of N immobilization, promoting nutrient release and favouring good yields. Furthermore, based on recently conducted studies Yang et al. (2017) draw the conclusion that AWD within certain limits can increase yield by reducing redundant vegetative growth, elevating hormonal levels, improving canopy structure and root growth and enhancing carbon remobilization from vegetative tissues to grains. Yields have been found to increase in China under AWD, and decrease in tropical locations such as India and the Philippines; a difference that Belder et al. (2004) and Cabangon et al. (2004) reason may be the result of variable WSI practices, soil properties, groundwater depths, rice variety and crop management. AWD and other forms of WSI have been widely adopted in China where per capita fresh water availability is amongst the lowest in Asia, and is being recommended in parts of India and the Philippines (Cabangon et al., 2001; Bouman et al., 2007; Yang et al., 2013). 4. Methodology 4.1. Data collection Meta- analyses provide a tool for examining the results of studies in the context of other studies (Borenstein et al., 2009), and has been used for purposes similar to those presented in this paper by e.g. Bouman and Tuong (2001) and Carrijo et al. (2017). In this study, a search of published studies was conducted to obtain raw data on rice yield, water management method, N fertilizer input, water input, soil organic carbon or soil organic matter (SOC/SOM), crop duration, number of dry cycles, rice variety, soil texture and/or classification, some climatic variables, and whether the crop was transplanted or direct- seeded, creating a varied dataset with potential for many applications. 10

15 The article search was conducted in the Web of Science database, using the search term combinations "rice yield" AND water management AND irrigation, and "rice yield" AND water AND flood*. The abstracts of all results produced through these searches were examined, and those deemed likely to contain relevant information were obtained for more detailed study. Specific criteria considered relevant for inclusion into this study included the studies being original research based on field experiments conducted in East, Southeast or South Asia, and containing quantitative data on rice yields and information about water management methods, of which at least one had to be continuous flooding. Various WSI types have been included in the study, but in each case the irrigation approach had to be paired with a control in the form of aforementioned continuous flooding, where all other factors but water management were the same. The WSI treatment had to have a minimum of either one extended dry period, which should be more significant than the ~10- day drainage during tillering that is recommended in some locations for optimal yields under CF management (see e.g. Yang et al., 2013; 2015), or multiple shorter cycles where FWL was allowed to drop below the soil surface. The work process for the searches is visualized in figure 2. Figure 2. Flowchart describing the stages of the data collection process. The terms 'AWD' or 'alternate wetting and drying' could not be used during the data collection process, as many other terms are often employed for similar water- saving irrigation techniques that are likely to be relevant for the purpose of this meta- analysis. Examples of these terms include 'alternate submergence- nonsubmergence', 'intermittent irrigation', and 'controlled irrigation'. The environmental implications of water- saving irrigation management were qualitatively assessed based on a literature review, and the findings are summarized and discussed in section 6.2. The review is mainly based on articles encountered during the data collection for the meta- analysis and is not intended to be exhaustive. Rather, the goal is to highlight the complexity of the interactions that are affected by WSI management Data compilation and evaluation The data was compiled in Microsoft Excel, wherein all the analyses were conducted. Many of the included studies placed primary focus on issues such as greenhouse gas emissions or 11

16 identifying optimal fertilizer regimes, but were oftentimes useful for providing the quantitative data needed for this analysis. Data on irrigation management and yield was frequently presented despite any differences having been deemed to be statistically insignificant by the author(s), due to the irrigation data simply being complementary to the main focus of the study. Belder et al. (2007) promised to contain valuable data, but the focus was placed on simulation using the ORYZA2000 model. For this reason, an additional search was made to acquire the original field data, which was then used in the meta- analysis (i.e. Belder et al., 2004). A few of the articles generated by the search were found to contain data based on the same set of experiments. This was the case with Hou et al. (2012), Xu et al. (2013), Yang et al. (2013) and Yang et al. (2015). Data was primarily taken from Yang et al. (2013, 2015), and these are therefore the articles that are referred to in the henceforth. A summary of all studies included in the analysis is displayed in table 1. In some instances, only part of the data came from plots fulfilling the above stated criteria. Plots that used relevant water management methods but deviated in other management aspects, thereby invalidating any comparison with a continuous flooding control plot, were excluded. Data was digitalized in those few cases where it was only presented in graphical form. A total of 21 articles, equalling 19 original studies, were included in the analysis, covering 41 sets of comparative field trials, and 179 side- by- side comparisons of WSI with CF, in a wide range of locations (figure 3). 12

17 Figure 3. Approximate locations of the sites used for field experiments in all studies included in the meta- analysis Data analysis The collected data was analysed using simple quantitative methods. Due to the variability in field conditions between different experiments, actual yields and water input values are generally not directly comparable across studies (Bouman and Tuong, 2001). For this reason, the relative differences between WSI treatments and corresponding CF treatments have been used. For each study, yield data for every WSI- plot (YWSI) was normalized by the corresponding CF control plot (YCF). YN = YWSI / YCF (eq. 1) Due to the normalization, YN values >1 indicate that the yield was higher in the WSI plot compared to the corresponding CF plot, and values <1 indicate that WSI treatment resulted in a decreased yield. The mean normalized yield, used as 'effect size' or alternately 'treatment effect', was calculated for each study, along with the standard deviation, standard 13

18 error of the mean and 95% confidence intervals. The mean normalized yield is a simple measure of the effect of specific WSI treatments on yield. The method has weaknesses, but will be used as an indicator in this study. The treatment effect and therefore the differences in yield between WSI and CF were considered significant if the 95% confidence intervals did not overlap the value 1. A summary effect was calculated for all studies included in the meta- analysis. The summary effect is based on the mean normalized yields for all WSI/CF pairs, and not on the mean effect of each study. This approach results in the weight of each study in the summary effect being proportional to the sample size. Using effect sizes has some significant advantages over statistical significance testing. Unlike significance testing, which can only tell us whether the effect is or is not zero and which is also affected by sample size, using effect sizes allows an estimation of the magnitude of that effect (Borenstein et al., 2009). In this study, it means that we can not only tell if WSI management affects the yield, but also how large that effect is, and if certain types of WSI have a greater effect. Figure 4. Distribution of all normalized yield values from the 19 field studies. Oftentimes in meta- analyses, the log of the normalized yield is the preferred metric (see e.g. Vico et al., 2016; Carrijo et al., 2017), as the log helps make a skewed distribution of values more Gaussian, and therefore more suitable for calculating confidence intervals. The effect size used here is essentially a response ratio, as described by Borenstein et al. (2009), who also state that the log should be used for all calculations. Both the log and the exponential of the normalized yields were considered for use in this meta- analysis, but did not achieve a more Gaussian distribution of values than the normalized yields (figure 4) and were therefore dismissed. 14

19 Table 1. Summary of the 19 experiments included in the meta- analysis. 15

20 Table 1 continued. 16

21 5. Results 5.1. Yields under WSI management Of the 19 studies, eight (1, 5, 7, 9, 10, 15, 18, 19) showed a significant difference in yield between the WSI and CF treatments (not including 6 and 16 that lacked confidence intervals) (figure 5). Of these eight studies, seven displayed a significant decline in yields under WSI treatment, and only one (18) showed an increase in yields under WSI management. Of the remaining 11 studies, where the differences were not considered significant due to confidence intervals overlapping with 1, seven had a treatment effect below 1, potentially indicating a tendency toward decreased yields. Three lay above 1, and one had a treatment effect of exactly 1. The summary effect lay slightly below 1, indicating a trend of decreased yields under WSI management, and the confidence intervals indicated that this effect was significant. Figure 5. Treatment effects with 95% confidence intervals for each study, along with the summary effect size. Studies 6 and 16 only had one side- by- side comparison of WSI and CF each, and therefore do not have any confidence intervals.

22 5.2. Regional differences in WSI yield The scatter plot in figure 6 displays the relationship between WSI and CF yields for each side- by- side comparison identified in the 19 field studies. The overall distribution appears to align well with the 1:1 line, though scattering is seen both above and below the line. When the yields deviate from the 1:1 line, the deviation tends to be more pronounced in the direction of higher yields under CF. The chart indicates that though yields were oftentimes maintained under WSI, they rarely increased. The values in figure 6 have also been categorized depending on if the field experiment was conducted in East Asia (China, Taiwan, Japan, South Korea), Southeast Asia (Vietnam, Philippines) or South Asia (India). The scattering indicates no obvious pattern in terms of the ability of WSI management to maintain yields in different regions. The highest yields appear to have been achieved in East Asia, but since the various studies' yields are not directly comparable due to varying environmental conditions and management approaches, the actual yields are not reliable values for analysis. Some yields produced in experiments in South Asia appear fictitiously low, with yields below 2 t ha - 1. These low yields have been attributed to the rice variety used (Bhaduri, 2017, personal communication). Figure 6. Relationship between WSI yields and corresponding CF yields in three major regions in Asia. Figure 7 is based on the same data as figure 6, but provides the summary effects for the three regions. As expected based on figures 5 and 6, the overall effect of WSI treatment was a decrease in yield, though this effect was not significant for the experiments conducted in South Asia. The summary effect for East Asia was very similar to South Asia, but with lower variability. Southeast Asia displayed a significant decrease in yields under WSI management. 18

23 Figure 7. Summary effects for groupings of side- by- side comparison into regions; South, East and Southeast Asia. Error bars correspond to a 95% confidence interval. The summary effects contain 72 side- by- side comparisons for South Asia, 48 for Southeast Asia, and 59 for East Asia Influence of severity of WSI management on yield The different WSI treatments have been classified into a number of categories (figure 8). The mild, moderate and severe AWD treatments have been grouped with treatments using drainage periods of a maximum of 2, 4, and 7 days, respectively. Figure 8 demonstrates a very close alignment between the regression line for 'mild AWD/<=2- day drainage' and the 1:1 line, indicating that very similar yields were attained in these WSI treatments as compared to corresponding CF treatments. As the severity of the AWD management and the length of the drainage periods increased, the scattering and corresponding regression lines became increasingly displaced from the 1:1 line. The high R 2 values for all three categories indicate that the regression lines incorporate much of the variability. Figure 8. Patterns in the relationship between yield and WSI method used. Mild AWD - SWP potential >- 10kPa or FWL >- 15 cm, moderate AWD - SWP between - 10 and - 30 kpa or FWL between - 15 and - 30 cm, severe AWD - SWP <- 30 kpa. Data from studies 7, 8, 11, 13, 16, 17, 19 and parts of study 14 was excluded due to not suiting any of the designated categories. 19

24 The category 'mild AWD' corresponds to the safe AWD defined by Bouman et al. (2007), whereby the field water level should stay within - 15 cm from the surface (figure 8). The - 15 cm FWL has been paired with a SWP limit of - 10 kpa, as the SWP normally stays above - 10 kpa (measured at 15 cm depth) at a FWL of - 15 cm (Lampayan et al., 2015). Bouman and Tuong (2001) found that yield decreases often became noticeable at SWPs between - 10 and - 30 kpa, and Brady and Weil (2008) have stated that field capacity often corresponds to SWPs ranging from - 10 to - 30 kpa. As rice is classified as a semiaquatic plant (GRiSP, 2013) and in the examined lowland settings is most commonly grown under submerged conditions (Lampayan et al., 2015), SWPs at field capacity have been classified as 'moderate AWD'. When using only the data from plots that were specifically stated to have been kept under 'mild AWD' conditions, the yields corresponded almost perfectly to those achieved under CF management (figure 9). Figure 9. AWD treatments specifically stated to have been kept above a SWP of - 10kPa and field water level depth of - 15 cm. When all categories were examined individually, some additional variability was discovered. Mild AWD and midseason drainage displayed yields on par with CF plots with relatively high precision (figure 10). It is worth noting that the midseason drainage category was based on a small sample made up of seven side- by- side comparisons. Yields appeared to have increased under the <=2- day drainage treatments, and have been maintained almost on par with CF yields under 4- day drainage treatments, though neither of these treatment effects were deemed significant at the 95% level. Significant yield decreases were seen for moderate and severe AWD, as well as for 7- day drainage periods. It is clear that the yields achieved under WSI management gradually decreased from mild, to moderate, to severe AWD, as well as when drainage periods were increased from 2, to 4, to 7- day intervals. Likewise, though yields were maintained under midseason drainage treatment, they decreased when the non- flooded conditions were maintained throughout the growing season. 20

25 Figure 10. Summary effects and 95% confidence intervals for various WSI categories. Mild AWD - SWP potential >- 10kPa or FWL >- 15 cm, moderate AWD - SWP between - 10 and - 30 kpa or FWL between 15 and 30 cm, severe AWD - SWP <- 30 kpa. Data from studies 13 and 17 that lacked the necessary information to categorize the treatments were excluded. Number of side- by- side comparisons: mild AWD - 53, moderate AWD - 12, severe AWD - 21, <=2- day drainage - 36, 3 to 4- day drainage - 25, 5 to 7- day drainage - 6, midseason drainage - 7, non- flooded Nitrogen fertilization effect on yields Plotting yield against nitrogen (N) input showed an overall increase in yield with increasing N inputs, though the variability was large (figure 11). The regression lines indicate that the trend does not differ between CF and WSI management, with both irrigation treatments showing yields increasing at similar rates under increased N input. Figure 11. Relationship between nitrogen input and yield in all CF and WSI plots. Studies 3, 15 and 19 lacked data on nitrogen and are not represented. 21

26 6. Discussion 6.1. Rice yields under varying irrigation managements If the effect sizes for the studies included in a meta- analysis display consistency a summary effect size is usually calculated, and otherwise the focus is shifted to estimating the dispersion of effect sizes (Borenstein et al., 2009). In this study the effect sizes are relatively consistent with 58% of the study effect sizes in the interval between 0.9 and 1.0, and 89% between 0.8 and 1.1 (figure 5). Calculating a summary effect size was therefore deemed suitable. The summary effect size for the 19 studies was established as with the upper confidence interval limit at and the lower limit at (95% confidence level). As can be seen in figure 5 the precision of the summary effect was considerably higher than for many of the individual studies. The summary effect indicated that there was an overall decrease in yields for plots under WSI treatments and that this difference was small though significant. However, the WSI treatments used in the different studies are highly variable making it more suitable to divide the treatments into categories and looking at the summary effect for these, which was done in figures 8, 9 and 10. Figures 8, 9 and 10 demonstrate that yield was not necessarily higher under continuous flooding compared to WSI management as long as one stayed within the limits of mild or 'safe' AWD, in this case defined as a minimum field water level of - 15 cm or soil water potential of - 10 kpa. The precision for mild AWD was high, indicating that the risk of yield penalty is likely to be low. Since the category included data from various experiments the precision also indicates that this is true across a range of different locations. Carrijo et al. (2017) used fewer classes and classified SWPs of - 20 kpa as mild AWD, but when plotting data from treatments using SWP between - 10 and - 20 kpa in this study a decrease in yield was exhibited (data not shown). Carrijo et al.'s study did, however, contain a considerably larger number of side- by- side comparisons, indicating that the negative effect on yields of SWPs at - 20 kpa may be non- significant over a larger sample size. Dividing the WSI types into the number of categories used in figure 10 resulted in several categories having relatively few observations, making the summary effects somewhat weaker. The non- flooded treatment exhibited high variability, which is probably a result of it being a broad category with low precision in terms of WSI approach. In figure 10, cycles using drainages with a maximum duration of 2 days appeared to actually increase yield for unknown reasons though the difference compared to CF was non- significant. The FWL or SWP reached during these treatments was difficult to estimate. Drainage rates and soil water retention capacities are influenced by soil texture and structure with coarser fractions resulting in faster drainage (Brady and Weil, 2008). The actual SWP or FWL achieved during the drainage periods can therefore differ from site to site depending on soil texture. The mild AWD and <=2- day drainage categories each had a similar number of observations. The higher variability found in the latter likely indicates that a set number of days of drainage is a considerably more 22

27 unreliable measure compared to SWP or FWL in terms of judging the level of stress placed on the rice plants. The reliability of using set drainage periods is likely to be higher when local soil texture and water retention capacity is taken into consideration and adjustments are made accordingly. As discovered by Carrijo et al. (2017) who classified AWD treatments based on SWP and FWL, soil texture had no influence on yield. Their result is not surprising since a sandier soil using irrigation at - 20 kpa would most likely have received slightly more frequent irrigations than needed in a soil with loamier or more clayey texture, resulting in similar success in terms of yields. If, on the other hand, irrigations are scheduled based on a regular interval of dry days as seen in some studies included in this meta- analysis, the SWP or FWL reached during those intervals will likely differ between a coarser sandy soil and a finer loamy soil, and thereby also affect the perceived success of any WSI management. In figure 10, a gradual decrease in yields is seen when going from mild to moderate to severe AWD and from 2-, to 4-, to 7- day drainage. The yield decline for severe AWD was much stronger than for 4- or 7- day drainage, again indicating that a set number of days is not easily equated to a certain degree of water stress as measured by SWP and FWL. In the mild AWD category Ye et al. (2013) had dry cycles extending for as long as 12 to 15 days, but due to the soil properties and the shallow groundwater table the SWP never dropped below - 10kPa. Meanwhile, Kumar et al. (2016) reached a SWP of - 30 kpa in less than four days. In essence, one of the most important factors for the success of WSI management is not strictly the duration of dry periods but perhaps rather the level of water stress experienced as mediated by dry cycle length and soil water retention capacity. As in this study, Bouman and Tuong (2001) identified differences in yields between experiments using the same level of water stress. The authors attributed the differences to rice variety drought response and variability in the number of dry cycles used as a result of weather variability and differing seepage and percolation rates between sites. This variability in the moderating variables is likewise a factor in this meta- analysis and is at least in part demonstrated by the scattering in figure 9 and to a certain extent the width of the confidence intervals in figure 10. It is of course possible that some of the variability which points to increased yields under WSI could be attributed to the beneficial effects of aerated conditions on certain aspects of rice growth. These effects include, for example, the nutrient release and decrease in redundant vegetative growth that has been identified by Ye at al. (2013) and Yang et al. (2017). A significant decrease in yields under WSI management was displayed by experiments conducted in Southeast Asia (figure 7). This summary effect is likely skewed as a result of this category containing data from study 5 by Cabangon et al. (2011), which contains plots with some of the most severe soil drying found in any of the studies in this meta- analysis, with soil water potentials allowed to drop to - 50 and - 80 kpa. The increase in yields that Belder et al. (2004) and Cabangon et al. (2004) mention with regards to AWD 23

28 in China is not evident in figure 7. However, though most of the data in the East Asia category comes from China, some experiments were also conducted in Japan, South Korea and Taiwan, thereby possibly hiding any yield- increase trends in China. From a visual estimation, however, the yields in Japan and South Korea appear to be on par with those achieved in China with the few plots in Taiwan possibly being slightly less successful (appendix A for raw data). Based on the studies included in this analysis it does not seem as if the yields were on average higher in China under WSI management, though they were oftentimes maintained. The effect of nitrogen input on yields did not appear to differ between water managements. Yields under WSI management increased with increased N input at the same rate as under CF management. The findings are in accordance with the recommendations by Bouman et al. (2007) that N fertilizer management does not need to be altered when transitioning from continuous flooding to WSI, or more specifically, AWD management Environmental implications of water- saving irrigation Greenhouse gas emissions and soil quality Flooded rice fields are a source of methane and a sink for CO2 (Bouman et al., 2007). Under the anaerobic conditions found in flooded rice paddy environments, emissions of the strong greenhouse gas methane are inherent (Liping and Erda, 2001). More aerobic conditions instead lead to increases in CO2 emissions (Linquist et al., 2015). Haque et al. (2016a) found that though implementing a 30- day midseason drainage period during the growth season increased the CO2 emissions, the decreases in methane emissions reduced the global warming potential of the rice paddies by 17-31% whilst maintaining yields. According to Linquist et al. (2015) the contribution of CO2 from rice farming to the global warming potential of agricultural emissions in negligible. Kumar et al. (2010) have found, however, that using AWD with a soil- drying threshold of SWP - 30 kpa increased the emissions of CO2 and nitrous oxide, another strong greenhouse gas, enough to offset the beneficial effects of AWD on methane. Despite the claim that the CO2 emissions are negligible, Linquist et al. (2015) have also stated that transitioning from flooded rice monocropping where the soil carbon sequestration is high to more aerobic conditions could lead to losses of SOC (Wu et al., 2011). Sahrawat (2005) highlights the various benefits to soil fertility of growing rice under flooded conditions. The beneficial effects of flooding include neutralized soil ph, increased mineralization and availability of nutrients, and increased SOM content. These benefits could potentially be lost under aerobic soil conditions. Bouman et al. (2007) state that the beneficial effects of flooding on soil quality gradually decrease when soils are no longer kept flooded but claim that negative changes are not present when staying within the limits of safe AWD. The discussion on water- saving irrigation efforts should include careful consideration of consequences for soil fertility and the sustainability of rice production (Sahrawat, 2005; Yang et al., 2017). Lal (2010) emphasises the great importance of the SOC pool for maintaining soil quality and its role in achieving high 24

29 agricultural production. SOC plays an important role in improving soil structure and aggregation, water and nutrient retention, biomass production, mitigating non- point source pollution and decreasing erosion (Brady and Weil, 2008; Lal, 2010). Carrijo et al. (2017) found that yield decreases under both mild and severe AWD were greater in soils with SOC <1% compared to those found in soils with SOC >1%. Hence, though mild AWD might not negatively influence the SOM content of the soil, the SOM content likely has an effect on the success of AWD. Rice paddy systems have been shown to sequester organic carbon more effectively than land devoted to other uses such as arable cropping and orchards (Wu, 2011). By analysing 2700 samples taken in subtropical China, Wu found that the SOC contents were considerably higher in paddy field soils. Additionally, Wu found that the carbon stocks had increased by a factor of 1.67 between 1979 and 2003, attributing this change to the increase in rice production that occurred during the same period and concluding that rice paddy ecosystems have potential in terms of sequestering carbon. Likewise, Pan et al. (2010) investigated SOC accumulation in China's croplands, comparing soil monitoring data from rice paddy soils and dryland crop soils. Over the study period, increases in SOC stocks were found in ~70% of both soil types. Pan et al. (2010) suggest that the positive trend may in part be attributed to crop residue return and good fertilization practices. However, both the initial SOC stocks and the accumulation rates were considerably higher in the rice paddy soils, again displaying a significantly higher sequestration potential. In contrast, we have the findings of Witt et al. (2000) comparing a rice- rice cropping system with a rice- maize rotation. The rice- rice system where the crops were grown in submerged conditions experienced a 10-14% increase in SOC during a two- year period. In the rice- maize rotation, however, where one of the crops were grown under upland conditions a slight decrease was instead found due to a 33-41% increase in carbon mineralization as a result of increased microbial activity. Witt et al.'s study is certainly relevant and provides an example of the potential risks to the SOC pool of growing crops under aerobic conditions. However, the findings of Pan et al. (2010) and Wu (2011) are based on large amounts of soil monitoring data covering much longer time periods. Their results indicate that though SOC accumulate faster in rice paddies, achieving a positive SOC/SOM balance under more aerobic conditions is not impossible under the right management. According to Dawe et al. (2000) who compared 30 long- term experiments in rice- upland systems and rice monocultures, yields remained stable in the majority of experiments in both systems. The yield declines that did occur were attributed to the depletion of nutrients and the effects of prolonged wetness on soil properties. More specifically, yield declines in some rice- upland experiments were attributed to a decrease in the indigenous N supply as a result of a decline in the SOM content; an effect that was not seen in the rice monocultures. In some of the rice monocultures, however, frequent cropping cycles with very limited aerobic fallow periods in between led to accumulation of large amounts of partly decomposed lignin residues and an increased phenolic 25