Climate change shifts forward flowering and reduces crop waterlogging stress

Climate change will drive increased frequencies of extreme climatic events. Despite this, there is little scholarly information on the extent to which waterlogging caused by extreme rainfall events will impact on crop physiological behaviour. To improve the ability to reliably model crop growth and development under soil waterlogging stress, we advanced the process-basis of waterlogging in the farming systems model Agricultural Systems Production Systems sIMulator. Our new mathematical description of waterlogging adequately represented waterlogging stress effects on the development, biomass and grain yield of many commercial Australian barley genotypes. We then used the improved model to examine how optimal flowering periods (OFPs, the point at which long-term abiotic stresses are minimal) change under historical and future climates in waterlogging-prone environments, and found that climate change will reduce waterlogging stress and shift forward OFP (26 d earlier on average across locations). For the emissions scenario representative concentration pathway 8.5 at 2090, waterlogging stresses diminished but this was not enough to prevent substantial yield reduction due to increasingly severe high temperature stress (−35% average reduction in yield across locations, genotypes and sowing dates). It was shown that seasonal waterlogging stress patterns under future conditions will be similar to those occurring historically. Yield reduction caused by waterlogging stress was 6% and 4% on average across sites under historical and future climates. To adapt, both genotypic and management adaptations will be required: earlier sowing and planting waterlogging tolerant genotypes mitigate yield penalty caused by waterlogging by up to 26% and 24% under historical and future climates. We conclude that even though the prevalence of waterlogging in future will diminish, climate change and extreme climatic events will have substantial and perverse effects on the productivity and sustainability of Australian farms.


Introduction
Climate change predictions herald increased frequencies of extreme climatic events, with historical analysis by the IPCC indicating statistically significant increases in the number of high precipitation events (e.g. 95th percentile) in many regions (Seneviratne et al 2012). Increased frequencies of extreme rainfall events result in more flash flooding and seasonal waterlogging, though effects of superfluous water on crop growth and development have not been well studied (Chang-Fung-Martel et al 2017).
Waterlogging is highly variable and multifaceted, both temporally and spatially. Waterlogging may occur for many reasons, including geography, aspect, extreme rainfall events, high and prevalent seasonal rainfall, poor hydraulic conductivity, lateral groundwater flows, rising/perched water tables, improper irrigation management or combinations of these factors (Shaw et al 2013, Liu et al 2020b. In Australia, waterlogging primarily occurs in regions that have higher growing season rainfall, thin topsoils and/or low soil permeability. These factors prevent drainage and cause surface water accumulation. In both rainfed and irrigated environments, waterlogging may occur at any stage of crop development. Grain yield penalties tend to be highest when waterlogging occurs around heading. Waterlogging at this critical stage can reduce grain yield by up to 70%, with losses primarily attributed to reduced spikelet fertility and grain weight (Setter and Waters 2003, Liu et al 2020a, Ploschuk et al 2020. In Australia, the States of Western Australia and Victoria produce around 32% and 16% of the national harvest respectively; of which 1.8 Mha in Western Australia are periodically waterlogged and 4 Mha in Victoria are at high risk of yield losses due to waterlogging (Shaw et al 2013). Collectively, potential Australia crop production losses due to waterlogging are around AU$300 million per annum (Manik et al 2019). The winter crops most impacted by waterlogging include wheat and barley, Australia's most predominant winter grains. Despite this, most commercial Australian cereal genotypes are waterlogging intolerant (Liu et al 2020a). There is thus a clear need for the development of management practices coupled with greater genotypic tolerance to waterlogging to help alleviate yield losses caused by waterlogging. Here, we identify the impact of waterlogging on crop yields at several representative sites, then quantify the relative value of waterlogging tolerant genotypes at both optimal (and sub-optimal) sowing times.
In addition to effects on soil waterlogging, climate change has and will result in warmer days that accelerate crop development and bring forward flowering times (Bell et al 2013, Pembleton et al 2016. Crop yields tend to be highest when flowering occurs in a window that minimises exposure to frost, heat and drought stress during the critical period for yield determination that occurs prior to and during flowering (Flohr et al 2017, Liu et al 2019. An understanding of the duration of the optimal flowering period (OFP) and the timing in which it occurs over the long term across environments and genotypes may be used as a tactical lever to help lower the risk of abiotic stress (Harrison et al 2011a(Harrison et al , 2011b. While previous work has focussed on OFP in water limited environments, our knowledge of how OFPs may be affected in environments that are prone to waterlogging is still in its infancy. To disentangle the effects of waterlogging per se from that of sowing time on yield, we first simulated a range of sowing dates at each site to identify the OFP, then second examined the relative influence of waterlogging on yield at the OFP. Because waterlogging is a function of many environmental and social factors, our ability to reliably quantify the impacts of waterlogging on crops often suffers from poor reproducibility in field environments. This can make it difficult to identify management, genotypes or general agronomic practices that consistently alleviate effects of waterlogging. Some of these issues can be dealt with and appropriately scaled using simulation approaches (Liu et al 2020b). Indeed, process-based crop simulation models are becoming increasingly valuable frameworks in climate change adaptation of agricultural systems (Ho et al 2014, Alcock et al 2015, Harrison et al 2016a, 2016b. Suitably designed models can also be used to assess the interplay between long-term waterlogging frequencies, increasing temperatures and optimal flowering times. While contemporary farming system models are highly versatile, many models have not been designed to capture stress effects caused by superfluous moisture or extreme events ( (2019), improving the ability of APSIM-Soybean to model the effects of waterlogging on photosynthesis, phenology and nitrogen fixation. In support of the work by Pasley et al (2020), a recent review showed that APSIM was one of the most appropriate extant farming systems models for simulating waterlogging because this model is equipped with soil and water modules that can adequately simulate shallow water table dynamic and account for appropriate mechanisms (e.g. photosynthesis, nitrogen leaching and denitrification, and an aeration stress determined by the sensitivity of different growth stages; Liu et al 2020b). While the preceding work has clearly improved our understanding and ability to simulate soil waterlogging, little attention has been placed on cereal crops such as barley. This is important, because biological responses of legumes such as soybean under waterlogging will likely differ to those in barley.
Here, we conducted a genotype by environment by management analysis (G × E × M) to examine yield impositions caused by waterlogging, frost and heat in sites prone to waterlogging across Australia. Specifically, the objectives of this study were to (a) improve the capacity of APSIM-Barley to reliably model waterlogging effects on barley growth and development, (b) determine how climate change will influence OFPs when waterlogging stresses are accounted for, (c) examine how waterlogging frequencies and seasonal typologies influence yield under current and future climates.

Experimental design
Biophysical data were obtained from controlled environmental facilities at Mt Pleasant Laboratories (41 • 28 ′ S, 147 • 08 ′ E), Launceston, Tasmania in 2019 and 2020; see Liu et al (2020a) for details. Briefly, experiments were conducted with six barley genotypes Macquarie, Franklin, Planet, Westminster, a backcross line, Macquarie (T) and a double haploid line called TAMF169 (table 1). These barley differ in waterlogging tolerance conferred through aerenchyma formation under waterlogged conditions (for further background of these genotypes, see Zhang et al 2015 andZhang et al 2016).
Four waterlogging treatments of varying duration and timing with respect to phenology were implemented: waterlogging at Zadok stage (ZS) 12.5 for 1 or 2 months (WL1 and WL2, respectively), waterlogging at ZS15 for 2 months (WL3), and waterlogging beginning 1 d before heading for 15 d (WL4). Soil moisture of the controls were maintained at field capacity until grain filling. For treatment plots, waterlogging was achieved by raising an external water reservoir above the soil surface such that the water level within the plot increased to 400 mm so the soil was completely saturated. Further details of methods used to obtain experimental data used for the modelling in the current paper are described in Liu et al (2020a).

Model parameterisation and validation
Experimental data described above were used to parameterise and validate APSIM version 7.9 (Holzworth et al 2014). Genotypic phenology was calibrated by adjusting parameters of existing genotypes in the APSIM Barley XML file, or by adjusting the base cultivar where the genotype being calibrated did not exist in the default APSIM release. The APSIM-soil water infiltration and movement (SWIM) Module (Huth et al 2012) was first initialised using control plots without waterlogging, with the watertable height maintained at 200 mm until grain filling. Measured data for phenology, biomass, yield component and grain yield of the six genotypes in table 1 were then used to parameterise APSIM following procedures described by Harrison et al (2019) and Liu et al (2019). Parameterised values are shown in table 2. Bulk density (BD) and volumetric soil water content at air dry, lower limit (LL, i.e. 15 bar), drained upper limit (DUL or field capacity) and saturation (SAT) were measured in the laboratory.

Sites used for the factorial analysis
The majority of Australian grain cropping systems are located within a relatively narrow land area. The cropping zone is colloquially known as the Australian Wheatbelt, spanning from southern-Queensland, to New South Wales (NSW), Victoria, South Australia (SA) and Western Australia (WA). These regions experience temperate or Mediterranean climates with winter-dominant rainfall and hot, dry summers. Since the 1990s, the Wheatbelt has experienced increasing frequencies of heat waves and frost (CCIA 2021) in concert with declining rainfall during the critical winter cereal sowing periods. Here, we selected 12 sites across Australian Wheatbelt for factorial analysis.
At each location (table 3), sowing was simulated at five-day intervals from 1 April to 15 July for three genotypes (Macquarie, Macquarie (T) and Planet) differing in their phenology and waterlogging tolerance. Macquarie and Macquarie (T) were used to compare the relative impact of waterlogging stress through improved genetics (the latter being waterlogging tolerant). Macquarie and Planet were used to compare whether fast or slow developing genotypes would be more suitable to future climates. Waterlogging may occur for many reasons, including extreme rainfall events, intense rainfall in short periods, poor hydraulic conductivity, rising/perched water tables, or combinations of these factors. To examine the extent to which these factors influence waterlogging stress, we systematically categorised sites into three key groups (table 3): low hydraulic conductivity, high watertable (i.e. close to the soil surface) and high growing season rainfall (Viscarra et al 2014). To ensure crops were successfully established at sowing, initial plant available water was set to zero and 15 mm of irrigation was added so that initial soil water content was standardised across simulations. All simulations using historical climate data assumed a baseline atmospheric CO 2 concentration of 380 ppm.

Climate scenarios
Historical daily climate data for maximum and minimum temperature, rainfall and solar radiation for the period of 1900-2019 were obtained as SILO point data (Jeffrey et al 2001) for each location. We developed future climate scenarios for the period of 2030-2149 (hereafter referred to as future climate scenarios with a median time horizon of 2090) using representative concentration pathways (RCPs) 8.5 (IPCC 2014, Schwalm et al 2020), representing a radiative forcing of 8.5 W m −2 by the end of the century, because this scenario most closely matches the change in climate that has already be realised in Australia (Bell et al 2013, Phelan et al 2015, Chang-Fung-Martel et al 2017 and the world more generally (Schwalm et al 2020). Historical climate  data were used to produce future climate data for each region, in all cases with an atmospheric CO 2 concentration of 850 ppm following Collier et al (2011). The approach we used to produce future climate contained both monthly average trends in gradual climate change but also increased magnitudes and frequencies of extreme climatic events based on forecasts from regional climate models, such that climate data were more variable compared with historical climate observations. For each site, we used the historical climates as a template for modification, and all modifications were made on a daily time-step (see more details in Harrison et al 2016a). In this case, future climate dataset would have the same length as historical climate dataset. Averaged across locations, mean daily minimum and maximum temperature had increased by 1.6 • C and 3.8 • C respectively by 2150, while future average rainfall decreased by −35% relative to historical period (figure S1 (available online at stacks.iop.org/ERL/16/094017/mmedia)).

Optimal flowering and sowing windows
To examine how OFPs were influenced by waterlogging stress around flowering, we run a default version of APSIM simulation that included and not included waterlogging functions. OFPs were calculated for each genotype using a range of sowing dates (see methods described by Liu et al (2019)). Frost, heat and waterlogging-limiting yield (hereafter, 'FHW') for each year were calculated as the product of potential yield, cumulative frost and heat stress, and waterlogging stress. The OFP for a given location was identified as the flowering dates corresponding to ⩾95% of the maximum 15 d running mean FHW.

Improving biological processes influenced by waterlogging in APSIM-Barley
In the APSIM source code via the executable, we implemented the waterlogging processes documented by Ebrahimi-Mollabashi et al (2019) and Pasley et al (2020). Effects of waterlogging on photosynthesis and phenology in the APSIM source code were represented by the new functions oxdef_photo and oxdef_pheno, respectively. Each waterlogging function assumes 0-1 multipliers in the form of 'x/y pairs' (Holzworth and Huth 2009), where x is the independent variable (e.g. soil moisture) and y is the response variable (e.g. photosynthesis and phenology), wherein x = 0 when the soil is at field capacity (no stress) and x = 1 when the soil is saturated (full stress). These functions were invoked when 80% of the roots become waterlogged ( figure 1(a)).
Our previous studies have shown that waterlogging affects both photosynthesis and phenology to varying degrees depending on the timing and duration of waterlogging. We thus modelled the impacts of waterlogging as a function of crop stage (oxdef_photo, figure 1(b)), which is a significant advance on the majority of previous studies that assume that waterlogging stress depends only on water-filled pore space. Similarly, our experimental work has shown that waterlogging delays crop development, and in some cases induces premature senescence (Liu et al 2020a). The phenology function we added to APSIM (oxdef_pheno) was derived using information from environment-controlled experiments (Liu et al 2020a, Pasley et al 2020. We first initialised APSIM for the (non-waterlogged) control using APSIM's SWIM3 Module, then later tested the new functions using data measured in the waterlogging treatments. Parameterised values are shown in figure 1(a) and (b).
To examine the extent to which the new processes added to APSIM-Barley improved the ability to simulate crop growth and development under waterlogging, we also run a default (unimproved) version of APSIM-Barley with waterlogging. Simultaneous multi-objective optimisation (Harrison et al 2019) of oxdef_photo and oxdef_pheno for each of the six genotypes (table 2) was performed using the 2019 dataset for the four waterlogging treatments by minimising the sum of squared residuals across datasets.

Long-term seasonal waterlogging-stress typologies and frequencies
To characterise common waterlogging stress typologies, we output seasonal time courses of photosynthetic stress relative to phenology (i.e. oxdef_photo) and clustered these stresses across simulation years, sites, genotypes and management following the approach outlined by Harrison et al (2014). For each environment, oxdef_photo was averaged over 100 • Cd increments from emergence to 450 • Cd Figure 2. Simulated and observed anthesis, maturity, grain number, maturity biomass and yield for the default version of APSIM without the effects of waterlogging on photosynthesis and phenology. Ideal simulations fall on the 1:1 line. WL1: waterlogging exposed at ZS12.5 for 1 month; WL2: waterlogging exposed at ZS12.5 for 2 months; WL3: waterlogging exposed at ZS15 for 2 months; WL4: waterlogging exposed at ZS59 for 15 d. The WL4 treatment was not conducted on Franklin and Westminster. after flowering. We then characterised predominant seasonal water stress trends by applying k-means clustering to all seasonal trajectories of oxdef_photo against phenology. Clustering was applied using R statistical package (R Development Core Team 2013), with clusters being defined such that total within-cluster variation was minimised (further details of this method are described by Chenu et al 2013).

Performance of the default version of APSIM when crops were subjected to waterlogging
Without the waterlogging algorithms, the default version of APSIM Barley performed poorly in simulating waterlogging effects on grain yield, grain number maturity biomass and phenology (figure 2). Grain yield, grain number and maturity biomass were  figure 2). Ideal simulations fall on the 1:1 line. Colours and shapes represent alternative genotypes and waterlogging treatments. WL1: waterlogging exposed at ZS12.5 for 1 month; WL2: waterlogging exposed at ZS12.5 for 2 months; WL3: waterlogging exposed at ZS15 for 2 months; WL4: waterlogging exposed at ZS59 for 15 d. The WL4 treatment was not conducted on Franklin and Westminster. over-estimated across waterlogging treatments. Values of variation ratio (VR) ranged from 1.39 to 2.68. Flowering and maturity date were also oversimulated, with root mean square error (RMSE) ranged from 10.7 d in simulated anthesis to 5.6 d in simulated maturity.

APSIM performance when the algorithms for modelling waterlogging are improved
Including the effects of waterlogging on photosynthesis and phenology significantly improved the ability of the model to simulate waterlogging (figures S3 and 3). RMSE ranged from 4.9 d in simulated anthesis to 3.3 d in simulated maturity. Mean bias (MB) ranged from 1 d in simulated anthesis to −2 d in simulated maturity. Values of RMSE normalised by the mean of the observed data (RRMSE) were low and R 2 values were greater than 0.70. Overall agreement between the simulated and observed maturity biomass, grain number and grain yield for all genotypes across treatments was within the variability implicit to the measured data, indicating adequacy of the parameterisation (figure 3). The model appropriately captured delays in phenology and thus reduced biomass accumulation during waterlogging periods, as well as the final yield and biomass response to waterlogging stress at different growth stages. An example of model validation that includes the simulated control and waterlogging treatments is shown in figure S3.

OFPs under historical climates
Barley yields varied from 6977 to 11 943 kg ha −1 across locations due to differences in environmental conditions (figure 4 and table S1). In most sites, mean peak yield increased from 1900 to 1939 before declining. Genotypes Planet and Macquarie had the highest peak mean yield at the site of Carrick in Tasmania, with an average of 11 319 kg ha −1 due to a high growing season rainfall and cool temperature climate. The lowest peak mean yield was at Mt Barker in Western Australia with an average yield of 7911 kg ha −1 because of hot temperature. In general, OFP windows varied from September to November across sites and genotypes. Mt Barker (WA) had earliest OFP (9 September) whereas Carrick had the latest OFP (14 November). OFPs moved forward least at Albany (6 d) and the most (33 d) at Naracoorte. For most sites, warming climates have moved forwards OFPs due to increasing temperature and decreasing seasonal rainfall (figure S1).

OFPs under future climates
In general, future climates reduced grain yield and moved forwards OFPs (figure 4 and table S1). The sites of Hamilton and Tatyoon in Victoria experienced the largest yield loss (both approx. 56%), while Carrick in Tasmania experienced the lowest yield losses (16%). Generally, the start and end of the OFP at each site shifted forwards similarly across locations and genotypes. The start and close of the OFPs were 26 d earlier on average across locations. The largest shift was at Hamilton, Euroa and Lismore in Victoria because these sites had the greatest increases in maximum and minimum temperature (figure S1).

Relative importance of frost, heat and waterlogging stress in determining the OFPs
Sites with lower risk of aggregate stress during OFPs had higher mean yield (e.g. Carrick compared with short growing-season sites such Mt Barker in WA.). Frost and waterlogging stress were the main stress affecting OFPs at Carrick in Tasmania and Euroa and Hamilton in Victoria.
Severe waterlogging stress can delay phenology, penalise grain yield and hence impact the timing of OFPs. We found that waterlogging occurred more frequently at sites with the combination of low soil hydraulic conductivity (Ks), high watertable and high growing season rainfall (e.g. Carrick and Furner). Our results suggest that waterlogging stress needs to be considered when defining OFPs in waterlogging prone zones (Carrick, Funer and Hamilton, figure 5). . Stress indices of frost (blue line) and waterlogging (green line), heat stress (red line) was low at these sites. Black dotted lines (FHY) represent yields limited by frost and heat stress, but not waterlogging stress. Black solid lines (FHW) represent yields limited by frost, heat stress and waterlogging stress. The OFP (shaded zones) is the flowering duration at which yield is greater than or equal to 95% of the long-term average peak yield (yellow regions exclude waterlogging functions; blue regions include waterlogging; green regions represent overlapping yellow and blue regions).
OFPs were eight days earlier at Carrick in Tasmania and four days later at Tatyoon in Victoria if waterlogging effects were not considered. For other sites, waterlogging had a relatively minor effect on OFPs.

Waterlogging stress typologies and adapting to waterlogging stress through crop breeding and improved genetics
As most simulated sites suffered less waterlogging stress, here we only selected two representative sites with waterlogging risk to show waterlogging stress typologies in figure 6 (see other sites in supplementary figures). The three most prevalent waterlogging stress patterns accounted for 85% variation across all genotypes, sowing times, sites and climates ( figure 6). Under historical conditions, Carrick and Hamilton experienced early-onset low waterlogging stress (ELS). In 2090, the two sites will have lower frequencies of waterlogging. Frequencies of waterlogging stress under future climates will shift away from early-onset medium waterlogging stress (EMS) and towards ELS (increased by 5%-23%) but overall stress typologies under future conditions were similar to those occurring at under present conditions.
Our previous work (Zhang et al 2015(Zhang et al , 2016 documents the development of new genotypes with waterlogging tolerance through aerenchyma formation. In the present study, we examined the performance of these new genotypes under conditions for which crops would be expected to experience waterlogging over the long-term. We showed that waterloggingtolerant genotypes mitigated waterlogging-induced yield declines by 82-1711 kg ha −1 across sites (figure 7). Benefits derived from the use of waterlogging tolerant genotypes were greatest at Carrick (Tasmania), followed by Hamilton (Victoria) and Furner (South Australia). Planting waterlogging tolerant genotypes (e.g. Macquarie (T)) prevented 819-2800 kg ha −1 yield loss at Carrick in Tasmania, while the yield difference due to waterlogging tolerance was relatively minor (82-806 kg ha −1 ) in Under future climates, waterlogging stress generally declined such that differences between waterlogging tolerant and susceptible genotypes diminished over time due to drying trends with climate change. However, some sites continue to experience significant waterlogging (e.g. Carrick), underscoring the need for the continued genetic development of waterlogging tolerant lines. The highest yield difference between the waterlogging tolerant and susceptible genotypes was 1548 kg ha −1 at this site.
Yield reduction due to increasingly severe high temperature stress (−35% average reduction in yield across locations, genotypes and sowing dates) under the emissions scenario RCP 8.5 at 2090 (figure 6). Euroa, Hamilton and Lismore in Victoria will suffer the most yield loss (ca 45% on average) and Carrick will suffer the least yield reduction (16%).

Discussion
A central purpose of this study was to improve the capacity of APSIM-Barley to model the effects of waterlogging on barley growth and development through addition of processes affecting photosynthesis and phenology. We integrated new algorithms from APSIM-Soybean into APSIM-Barley with respect to the impacts of waterlogging on phenology and photosynthesis. We first showed that the ability of the default version of APSIM to simulate waterlogging was poor, because the model failed to adequately account for the effects of waterlogging on growth and phenology (figure 2). After addition and parameterisation of new relationships between waterlogging and photosynthesis and phenology, the validated model showed good correlation between simulated and observed phenology, maturity biomass, grain number and grain yield across genotypes and waterlogging treatments (figures S3 and 3). This work has built on the work of Pasley et al (2020) by using experimental greenhouse results to implement differential waterlogging stress impacts according to vegetative and reproductive phases of barley. We advanced their waterlogging algorithm by enabling different effects of waterlogging with phenology as suggested by our experimental data (Liu et al 2020a). It is also worth noting here that While effects of growth stress due to waterlogging early in the crop life cycle were well simulated, our results show a need to further improve the ability of APSIM to simulate waterlogging that occurs in later phenology. The relatively poor result for late waterlogging in waterlogging treatment 4 (figures S3 and 3) may be because grain number in APSIM-Barley is calculated as the product of stem dry matter on the last day of floral initiation with a cultivar-specific parameter that determines the number of grains per stem (grains_per_gram_stem). Grain number in APSIM-Barley thus remains constant from last day of floral initiation. In unstressed conditions this assumption works well, but under waterlogged conditions, this assumption breaks down. Waterlogging stress just prior to or after heading reduces grain number, sometimes causing death of entire ears due to effects on spikelet fertility and grain filling (Liu et al 2020a). The overestimation of simulated grain number under late waterlogging shown here exemplifies this process. While our work is a significant step forwards in process-based modelling of waterlogging, this observation highlights an avenue for future improvement of the model.
Given such limitations in simulating late-stage waterlogging, is it perhaps serendipitous that waterlogging of autumn-sown cereal crops in Australia generally occurs during vegetative stages in winter (figures 6(a) and (b)). Our results suggest that waterlogging stresses should be included in future crop models, particularly in studies that cross-examine the capacity of crop models for scaling, inter-and ensemble model comparison efforts of climate change and greenhouse gas emissions (Harrison et al 2016c, Farina et al 2021, because waterlogging stress has and will have significant effects on simulated yield at some sites. Although earlier flowering dates are expected under future climates in Australia (Gourdji et al 2013), few studies examine how OFPs across diverse environments will change under future climates. A notable exception is the study by Chen et al (2020), who found that OFPs of wheat will move forward by an average 29 d in the dry scenario and 11 d in the wet scenario under RCP 8.5 during 2061-2100 in Western Australia. These findings are consistent with our simulated results at sites in WA (e.g. Albany, Frankland and Mt Barker) but not at Wandering (30 d earlier). This is because relative temperature changes at Wandering are higher than other sites (figure S1) that are closer to the coast and experience more Mediterranean environments, thus advancing crop thermal time at a much higher rate.
Grain yields and OFPs in high rainfall environments are generally higher and wider, respectively, compared with environments that experience greater drought stress (Liu et al 2019, Chen et al 2020. Narrow OFPs in drought-prone environments are due to higher risk of late spring frost, and/or earlier risk of heat or water-deficit stress. Late-spring frosts are more likely to occur in drought-prone regions due to clear sky nights, coupled with high pressure systems (DPIRD 2021). Cereals such as barley and wheat are most susceptible to frost damage around flowering, with single frost events having the potential to damage stems and completely desiccate heads . In high rainfall environments, daily temperatures are generally lower, and when coupled with higher rainfall during crop reproductive stages, terminal water stress tends to be alleviated (Phelan et al 2018). Here, we observed that OFPs will become wide due to increasingly severe high temperature stress under future climates. Wide OFPs suggest a uniform low-yielding environment over a wide range of sowing dates, such that there is little variation in yield under many flowering dates (figure 5). This is evidenced by the substantial yield reduction due to increasingly severe high temperature stress (−35% average reduction across locations, genotypes and sowing dates, figure 7).
In addition, the relative importance of waterlogging stress in determining OFPs will become less under future climates, as most simulated sites will have minor waterlogging stress due to decreased rainfall. We extended the previous assessments of climate change impact on crop production by simulating the combined impacts of the occurrence of frost and heat stress event together with water stress (deficit and excess) during the reproductive stages (Flohr et al 2017, Hunt et al 2019, Chen et al 2020. Such method combined with our waterlogging algorithms can be applied into diverse environments (e.g. from too little water to too much water or high temperature to low temperature).
The significant decline in yields with climate change shown in figure 7 is consistent with results in other rainfed Mediterranean environments under high emission scenarios (Potgieter et al 2013, Yang et al 2019. Similar results were also observed by van Gool and Vernon (2006), who reported that climate change in Western Australia may result in large reductions (>30%) in barley potential yield in the northern agricultural region (around Mullewa) by 2050 due to reduced rainfall and higher maximum temperatures. To adapt to climate change, others have suggested that early-sown long-season varieties can sustain or improve wheat yield for future climatic conditions (Zheng et al 2012. Such effects were not observed in our study, although the earlysown long-season genotype Macquarie had higher yield compared with the relatively fast developing (short-season) genotype Planet in most sites (except Carrick in Tasmania, where fast maturity genotypes obtained higher yields under future climates; table S1). Similar to the observations made by Harrison et al (2014) for maize crops under future climates in Europe, these results suggest that earlier flowering in early terminal water stress environments is likely to result in higher yields, because crops have flowered and set seed before the majority of terminal water stress is realised.
In this study, we selected sites with different soil physics and growing season rainfall to examine the extent to which these factors influence waterlogging in APSIM. Our results show that sites with the combination of low soil hydraulic conductivity, low watertables and high growing season rainfall would have higher frequencies of waterlogging, evidenced by the evidence that low yields at Carrick and Furner (figure 5). Sites with high soil hydraulic conductivity, deep watertables and high growing season rainfall (e.g. Albany, Frankland and Mt Barker) would have little waterlogging stress compared with other sites. Euroa, Wandering and Millicent have similar soil characteristics with the only difference being soil hydraulic conductivity (Euroa having the lowest hydraulic conductivity). However, even in years of low rainfall, waterlogging stress occurred at Euroa, although waterlogging stress at Wandering and Millicent was often minimal. These findings suggest that soil hydraulic conductivity may be a governing factor of soil waterlogging stress in our study. Tests of the sensitivity of these factors were out of scope of current paper but would be an ideal avenue for future research.
Our results show that Tasmania would have similar waterlogging stress under both historical and future climates (figure 6). These results have important implications for barley breeding in Tasmania, for they indicate that current typologies (temporal exposure to waterlogging at certain crop stages) will be similar to those experienced in future. To concurrently increase yield potential and respond to climate change, many studies recommend improving tolerance to waterlogging stress in crops through breeding programs (Lobell et al 2015). Breeding of crop genotypes with improved tolerance to waterlogging stress appears to be an effective adaptation strategy to climate change. Indeed, we showed that the waterlogging tolerant genotype Macquarie (T) can mitigate up to 23% yield penalty caused by waterlogging under historical and future climates in Tasmania, suggesting further research and development of waterlogging tolerance genetics would be a worthwhile investment. These results suggest selection of elite barley germplasm for superior yield under present conditions using field trials should be an appropriate method for developing germplasm suitable for 2090 conditions in Tasmania.
As for any study, the present paper had some limitations. We adopted an atmospheric CO 2 concentration of 850 ppm, as this value was the median of an ensemble of global climate models (GCMs) at 2090. If the main objective of the present study had been on climate change impacts per se, another level of complexity that could be added to the study might have been to conduct a factorial simulation with a range of CO 2 values, spanning the uncertainty in GCMs projections at 2090. These simulations could also account for the uncertainty in projected temperature and rainfall under future climates. However, the main focus of our study was the relative variability in the impacts of waterlogging across sites and the ability of waterlogging tolerance genetics to alleviate waterlogging stress. As such, we believe the approach taken here is both sensible and appropriate.

Conclusions
Accounting for the waterlogging on photosynthesis and phenology in APSIM significantly improved the ability of the model to simulate growth and development in waterlogged environments. While biophysical effects of early-stage waterlogging were well simulated, our results suggest a need to further improve the ability of the model to simulate late season waterlogging. Our factorial G × E × M analysis showed that waterlogging stresses will diminish but this was not enough to prevent substantial yield reduction due to increasingly severe high temperature stress (−35% average reduction in yield) under future climates. Planting waterlogging tolerant genotypes mitigate yield penalty caused by waterlogging by up to 6% and 4% under historical and future climates across sites.

Data availability statement
The data that support the findings of this study are available upon reasonal request from the authors.
All data that support the findings of this study are included within the article (and any supplementary files).