Potential Barriers to Adaptive Actions in Water–Rice Coupled Systems in Japan: A Framework for Predicting Soft Adaptation Limits

The changing climate makes it more difficult to manage water resources and food production sustainably. Various adaptation measures have been proposed to moderate the negative impacts of climate change; however, implementation of these measures may be hampered by other factors even if the benefits are acknowledged—a situation termed “soft adaptation limits” by the IPCC. We hypothesized that societal rules can be a potential barrier to adaptive action if they are too fixed, because such rules have coevolved with the interactions between the past climate and human activities. To test this hypothesis, we present a framework based on the assumption that associated societal rules of Japan's matured irrigation systems are potential barriers to adaptation. The framework consisted of two process‐based models, one to evaluate the water deficit risk and one to evaluate the benefits of optimizing rice yield and quality. We applied each model to an experiment in which we shifted the current transplantation date by 1 week for up to 5 weeks before and after the current date under the historical (1981–2000) and RCP 2.6 and 8.5 (2011–2030 and 2031–2050) scenarios. We revealed two contrasting development pathways in the study watershed. Soft adaptation limits imposed by water availability will occur by 2030 if farmers optimize for quality, whereas mutual benefits to farmers and river administrators will be achieved if farmers seek yield. We argue that more participatory research with stakeholder engagement, as well as policy discussions about these possible developments, is needed to ensure successful adaptation.

multiple local stakeholders such as farmers, river administrators, and domestic or industrial water users.Therefore, if we are to understand and predict the fate of agricultural societies under climate change, we need an analysis based on the region-specific rules and interrelationships among stakeholders in a human-water coupled system.
Several region-specific human-water modeling frameworks have been proposed to assess sustainable developments in human-water coupled systems.Cai et al. (2002) developed a pioneering framework to study human-water interrelationships and feedback in the Syr Darya River basin in Central Asia.Their framework successfully provides insights that can help us to find sustainable development pathways in irrigation-dominated watersheds.They mathematically solved functions to assess the sustainability of a watershed on the basis of not only short-term economic benefits but also long-term environmental risks, as well as equality within watersheds or between generations.In more recent years, studies based on such a framework have predicted the impacts of climate change on human-water coupled systems.Giuliani et al. (2016) proposed a framework for predicting the interrelationships between two stakeholders (irrigated agriculture and lake operations) in the Adda River basin, Italy.They concluded that more sustainable pathways can be achieved by cooperative adaptation of the stakeholders than by adaptation without cooperation.To simulate decision-making in irrigated agriculture, they assumed that the irrational decisions of individual farmers could be filtered out when the farmers were grouped in the form of institutions and that the behaviors of institutions are fully rational and designed to maximize profitability; this constitutes a normative meta-modeling approach.By using this approach, Li and Sivapalan (2020) assessed the trade-off relationship between environmental risks (e.g., a decrease in the groundwater table) and population growth in Beijing, China.They introduced parameters that represent the sensitivities of human societies to long-term environmental risks, leading to restrictive decisions regarding development.By using these parameters, they succeeded in representing situations in which the increasing severity of environmental risks would prompt humans to take adaptative actions.Their findings have enhanced our understanding of future interrelationships in human-water coupled systems; however, given the complexity and ambiguity associated with the introduced parameters, the study was "not aimed at predicting an accurate future of the water situation.Instead, the model outcomes are deemed as just possibilities" (Li & Sivapalan, 2020).
To evaluate the sustainability of agricultural societies under a changing climate, here we use the meta-normative modeling approach, in which the decisions of irrigation districts are deemed rational (Giuliani et al., 2016;Li & Sivapalan, 2020).However, in contrast to the studies that introduce parameters that themselves implicitly constrain socioeconomic activities (Li & Sivapalan, 2020;Li et al., 2019), we employ societal rules that explicitly govern institutional decisions.In other words, we hypothesize that rules can be a barrier to adaptive actions to changing climate if they are too fixed in societies.This is because these rules-especially those associated with agriculture-have coevolved with interactions between the past climate and human activities.To test this hypothesis, we focused on human-water interactions in a typical watershed dominated by irrigation water use in Japan, namely the Shinano River watershed (see Section 2 for more details).The evolution of irrigation systems in Japan dates to the 1700s at the earliest; therefore, there have been many implicit rules regarding irrigation water use (Satoh & Ishii, 2021).Under the modern river administrative framework established in the 20th century, these rules on irrigation water use have been documented explicitly as water laws.The water rights documented in the law define the location, volume, and period of water diversion.Water rights for irrigation are usually updated approximately every 10 years through negotiations between representatives of farmers and river administrators; however, because of the maturity of Japan's irrigation systems, changing the current water rights (e.g., the maximum volumes and periods of water diversion) is challenging.
The recent degradation of rice quality has posed further difficulties in setting the appropriate water-use periods for irrigation.This degradation is caused by high temperatures during the heading stage of rice, at the beginning of August (Ishigooka et al., 2011;Takimoto et al., 2019).In the 2000s, the problem was widely known throughout Japan and various adaptive actions were implemented.Among the options, changing the rice transplantation date is deemed relatively easy to implement (Iizumi, 2019).However, if there is a possibility that changes in the transplantation date will impinge on the water rights of other water users, then river administrators will not allow farmers to change the water-use period.In this example, adaptive options are readily available, but no action can be taken because it is hindered by other factors.This is termed a "soft adaptation limit" in the Intergovernmental Panel on Climate Change (IPCC) Sixth Assessment Report on Climate Change (IPCC, 2022).The IPCC coined this concept to illustrate the difference between soft adaptation limits and hard ones, in which adaptative options are not available.This relatively new concept could help to warn water users that factors endogenous to societies (e.g., administration, institutions, and policies) can bring about adaptation limits.Especially in the case of mature systems such as Japanese agriculture, these endogenous factors may act as potential barriers, making it hard for us to predict the soft adaptation limits.
Here, we built a framework to predict the occurrence of soft adaptation limits on the Japanese water-rice coupled system in a study watershed.The framework consisted of two process-based models, of crop science and watershed hydrology, that evaluated three components of human-water coupled systems, namely rice yield and quality, and the balance between water demand and available streamflow within the watershed.By using the crop model, we evaluated the rice yield and quality by shifting the transplantation date by 1 week for up to 5 weeks before and after the current date for changing climate scenarios.In terms of the future decisions of farmers, we assumed that the rice transplantation dates were determined at an institutional level (i.e., the same transplantation dates were applied to the entire irrigation district) to maximize either rice yield or quality.We then simulated the water demand-availability balance of the watershed for each of the shifted transplantation dates and examined whether it was acceptable from the perspective of the river administrator.Throughout this work, we asked the following questions: How does an effective measure in one sector (agriculture) influence the other (water resources)?How does climate change affect the relationship between agriculture and water resources?and Does the proposed framework help us to understand the occurrence of soft adaptation limits?

Relationship Between Water Resources and Rice Cultivation
Agriculture is the largest water user in Japan, accounting for about 70% of all water withdrawals (Ministry of Land, Infrastructure, Transport and Tourism [MLIT], 2021).Approximately 95% of the agricultural water is drawn from rivers, and more than 90% of the water is used for paddy rice cultivation (MLIT, 2021).Development of rice paddies was most advanced in the 17th century, when local governments attempted to develop as many rice paddies as possible wherever irrigation water and suitable land were available (Satoh & Ishii, 2021).Irrigation methods using open channels, which were developed at this time, are still common today.Using open channels to carry or divert water to branch channels requires farmers to keep water levels in the open channels high, and this requires extra water above the actual water requirements.Thus, the current irrigation systems using open channels have a high proportion of return flow from the irrigated area to the river (Yoshida et al., 2016), and irrigation efficiency is not high.This is also reflected in the small rate of decline in agricultural water withdrawals compared with the rate of decline in rice paddy areas in recent years.In fact, whereas the area of rice paddies decreased by 22.8% between 1975 and 2015, the rate of decrease in agricultural water withdrawal was only 5.26% (MLIT, 2021).
Currently, water rights allow maximum amounts and periods of water withdrawal.The water withdrawal fees for agricultural purposes are based on a fixed-rate system per unit area.Water rights for irrigation are determined through negotiation between representatives of farmers and river administrators.Approximately every 10 years, farmers apply to the river administrators for water rights on the basis of their rice cultivation plans (e.g., cultivation area and period) and the weather conditions (e.g., precipitation, evapotranspiration, water demand for crops).The maximum withdrawal amount is planned on the basis of the water demand of irrigation districts and the irrigation efficiency under favorable hydrological conditions.Thus, if the planned water withdrawals are secured, the paddy rice is not stressed by water deficit.However, in the planning of water withdrawals it is also assumed that water withdrawal for irrigation will be restricted during low precipitation events with an approximately 10-year probability of occurrence.
The reason for allowing a 10-year probability of water withdrawal restrictions is that farmers have historically created mechanisms to prevent water stress on paddy rice by using human power.Once low precipitation events occur, irrigation water is generally restricted by 10%-30% before restrictions are placed on other water users.In this case, the farmers adjust their branch channels to change the route of water diversion every day.As a result of these efforts by farmers, each paddy field is supplied with water at intervals of several days, even during the restriction period.These efforts have prevented water stress on paddy rice, even when water withdrawal for irrigation is restricted.This means that, although drought generally causes water stress and yield loss of paddy rice globally (e.g., Hasegawa et al., 2008;Steduto et al., 2009), these negative effects are less likely to occur in Japan.Surprisingly, according to records of an extraordinary drought that occurred throughout Japan in 1994, although water withdrawal for irrigation was restricted to between 10% and 90% in many rice paddy districts (Kawasima & Takahashi, 1995;Nakagiri et al., 1999), rice growth was not severely stressed owing to the tremendous water management efforts of farmers (Satoh, 1997).
Although the amount of water withdrawn in Japan for paddy farming has changed little historically, the withdrawal period has changed substantially over the past 70 years.From the 1950s to about 2000, transplantation dates for paddy rice shifted to about 1-5 weeks earlier nationwide.However, since 2000, transplantation dates have tended to be 1-2 weeks later in some regions.Such changes in the transplantation date imply that the periods of irrigation water withdrawal have also shifted and may continue to shift in the future.
Overall, the coupled system of water and rice in Japan is based on matured irrigation systems and an administrative framework for water resources management.Therefore, the impact of water deficit is not direct (e.g., water stress leading to decline of rice yield) but indirect (i.e., difficulties in setting appropriate water rights for rice cultivation).Changes in meteorological conditions under climate change may increase the amounts of water that need to be withdrawn, although the relatively abundant amounts available in this country for withdrawal may mean that this may not be an issue.On the other hand, changes in the water use period may lead to conflict with other water users and may disrupt the availability of streamflow, which fluctuates highly seasonally depending on the precipitation.Therefore, water-rice systems in Japan are coupled by the period of withdrawal for irrigation, rather than by the amount withdrawn, and this could result in a soft adaptation limit if water withdrawal is not available during the optimum periods.

Impacts of Climate Change on Rice Cultivation and Water Resources
Rice cultivation and water resources are both affected by climate change.Assessments of the individual impacts of climate change on these two elements have been conducted throughout Japan.In analyses of the impacts of climate change on rice cultivation in Japan, both positive and negative impacts have been observed (Iizumi, 2019).Climate warming has positive impacts on forage rice production in western Japan by increasing the annual number of harvests with ratoon-rice cropping (Nakano et al., 2009;Sakai et al., 2013).The record hot summer of 2010 in Hokkaido (2.2°C higher than the average for 1981-2010) allowed crop scientists to successfully harvest in their experimental field a rice cultivar normally grown in warmer regions (Nemoto et al., 2011).However, high temperatures during the heading period reduce the appearance quality of rice because of the occurrence of white immature grains (Ishigooka et al., 2011;Takimoto et al., 2019).In Japan, the ratio of white immature grains in a given number of grains is one criterion for determining the rice grade (i.e., first-and second-grade rice), and the price differs depending on the grade.
Adaptation measures to moderate the negative impacts on rice quality have attracted considerable attention from farmers and government.Various adaptation measures have been proposed, ranging from incremental to transformative.They include water management, fertilizer management, transplantation date shift, switching of crop type, and agricultural insurance (Iizumi, 2019).Shifting the transplantation date is relatively inexpensive and easier to implement than other adaptation measures.Therefore, it has been widely implemented in some countries (e.g., Ministry of Agriculture, Forestry and Fisheries [MAFF], 2006;Minoli et al., 2022;Urfels et al., 2021Urfels et al., , 2022)).Ishigooka et al. (2017) used a process-based rice simulation model developed by Hasegawa and Horie (1997) to evaluate the effects of shifting the transplantation date throughout Japan on rice yield and quality.They implemented the model for the period 1981-2100, and they shifted the transplantation dates at 7-day intervals from 70 to +70 days from the standard transplantation date.Their results showed that negative impacts on rice quality can be avoided by selecting the optimum transplantation date in each region.However, their analysis assumed that irrigation water would be available and the withdrawal periods could be set freely, even if the transplantation date were changed under climate change.As mentioned in Section 2.1, changes in the period of water withdrawal due to shifts in the transplantation date are dependent on river conditions.Therefore, the availability of water resources should also be considered in any assessment of the feasibility of adaptation measures for maintaining rice quality.
In analyses of the impacts of climate change on water resources, heavy snowfall areas in the temperate zone of Japan have been projected to be markedly vulnerable to temperature increases, showing a large reduction in snow water equivalent in winter (January to March) and earlier snowmelt due to climate change (Kudo et al., 2017a(Kudo et al., , 2017b)).By using a water circulation model for the whole of Japan, Kudo et al. (2017a) showed that, in northern Japan, a reduction in snow water equivalent in winter leads to a decrease in river flows during the rice puddling and heading periods, which require large amounts of irrigation water.In particular, the Hokuriku region, which Water Resources Research 10.1029/2022WR034219 TAKADA ET AL. includes our target watersheds (see Section 2.3), has heavy snowfall and relatively high temperatures in winter, and these factors could cause strong alterations in a decrease in river flows (Kudo et al., 2017b).
As a result of such changes in the seasonal variability of river flows, farmers' requests to shift the transplantation date may not be granted.Kotsuki et al. (2013) evaluated the effects of shifting the transplantation date on rice yield and the water supply-demand balance.They calculated rice yield and water supply-demand balance by using a process-based model in an experiment in which the transplantation date of paddy rice was shifted by up to 30 days before and after the reference dates throughout Japan.Their results showed that, in northern Japan, a late transplantation date had both positive and negative effects, namely an increase in rice yield and a worsening of the water supply-demand balance.However, their study did not consider the impact on rice quality.As mentioned above, the impact of climate change on paddy rice is of particular concern for quality rather than yield, and a shift in transplantation dates is being considered as an adaptation measure to avoid quality losses.Therefore, in evaluating the impact of measures for adaptation to climate change, we need to focus not only on rice yield but also on quality, and to consider the changes in the water supply demand balance.By doing this, we could predict the soft adaptation limit on the agricultural sector to climate change.

Case Study Site: The Shinano River Watershed
The Shinano River is the longest river in Japan, with a main channel length of 367 km.It has a catchment area of 11,900 km 2 -the third largest in Japan (Figure 1).The river runs through both Niigata and Nagano prefectures, included in the Hokuriku region, and flows into the Sea of Japan.The upper area of the Shinano River watershed, located in the middle of mainland Japan, is surrounded by mountains that are more than 2,000 m high.This area has precipitation lower than the average annual precipitation of approximately 1700 mm in Japan, with an annual precipitation of approximately 900 mm in the city of Nagano.Conversely, the lower watershed on the Niigata Prefecture side, where the weather is specific to areas along the Sea of Japan, has one of the heaviest snowfalls in Japan; this area includes the city of Nagaoka, which has an annual precipitation of approximately 2,300 mm, a great deal of which falls in winter.The basin of the Uono River, which joins the Shinano River in its middle reaches, is also known for heavy snowfall, with snow accumulating to depths of over 2 m.The snowmelt period coincides with the rice puddling period downstream when most of the irrigation water is required.
We targeted the Ojiya gauging station and its downstream irrigated area.The lower areas of Ojiya are among the largest rice-producing regions in Japan and include approximately 14,700 ha developed through national land improvement projects.We treated the target area as a single irrigation district because there were no major differences in weather conditions downstream from Ojiya, as shown by the difference of only 0.3°C in monthly mean temperature during the heading period (August) between the cities of Nagaoka and Niigata from 1991 to 2020 recorded at the Japan Meteorological Agency observation stations.The Ojiya gauging station is a reference point for water use in the middle and lower areas of the Shinano River watershed.A minimum flow requirement of 145 m 3 /s was defined by the river administrator during the irrigation period (from 28 April to 15 September).Because the minimum flow requirement is defined in terms of the aggregated water rights of the downstream water users (domestic, industrial, irrigation, fisheries, and environment), a hydrological drought can be evaluated as occurring when the flow rate falls below the minimum flow requirement.A MAFF data set that provides yearly statistics on rice yield and cultivation schedule summarized the dates of sowing, transplanting, heading, and harvesting at the prefectural level until 1968 and by sub-administrative regions called "sub-regions for yield statistics" (sakugara hyouji chitai in Japanese) after 1969.The peak transplantation date in our target area, included in a sub-administrative region called Chuetsu, was 9 May in 2019 (hereafter, the "current transplantation date").
As our study focused on changes in the transplantation date, we plotted the peak transplantation dates in Niigata Prefecture from 1953 to 2021 and in the Chuetsu region from 1969 to 2021 (Figure 2).The transplanting period (pink filled area) from 1982 to 2021 was defined by the start and end dates of rice transplantation.The peak transplantation date, which was 5 June in 1953, had gradually moved to 4 May by 1998.This may have been due to the adoption of mechanical transplantation, as well as to changes in rice cultivar and the higher prices at which early rice can be sold.The transplantation date tended to be delayed even further, by 1 week, in the 2000s (the latest date was 13 May, in 2012).In Japan, concerns about high-temperature injury to paddy rice began to grow in the 2000s, and studies were conducted to develop countermeasures (MAFF, 2006).The implementation of such countermeasures to maintain quality may be reflected in the data on transplantation dates.Koshihikari, a premium short-grain rice cultivar, is cultivated in the target irrigation district.Among the thousands of rice cultivars that have been developed in Japan, Koshihikari is the most preferred among Japanese consumers because of its excellent taste and texture (Kobayashi et al., 2018).In 1979 it became the cultivar with the largest cultivation area, and since then it has continuously maintained its top position in Japan.The largest cultivation area of Koshihikari is in Niigata Prefecture, and the Koshihikari produced their commands higher prices than that from other regions.In fact, the average price of Koshihikari produced in Niigata Prefecture was 15,583 yen/60 kg in 2021, whereas the average price of all rice cultivars produced in Japan was 12,804 yen/60 kg (MAFF, 2021).Furthermore, within Niigata Prefecture, Koshihikari produced in the Uonuma region (in the irrigation areas along the Uono River in Figure 1) has the highest brand power, having fetched an average price of 20,426 yen/60 kg in 2021.The price of Koshihikari is high even for the second-grade rice, with a price difference of only 600 yen/60 kg between the first and second grades.Thus, the value of Koshihikari produced in Niigata Prefecture is high in Japan, and a decline in its quality due to climate change could lead to a decline in the overall price of rice.

Framework for Predicting Potential Barriers for Adaptation Between Two Stakeholders
In our evaluation of the relationships between two stakeholders (Figure 3), we focused on rice paddy cultivation and water resources (water-rice coupled systems) in Japan.We assumed that only shifts in transplantation dates were measures of adaptation to climate change, whereas the area of rice paddies, or the cultivars used, would remain almost unchanged.First, we compiled two process-based models, one to evaluate the risk of water deficit and one to evaluate the benefits of rice production (e.g., rice yield, quality, and farmers' income) (see Sections 3.3 and 3.4, respectively).Second, we applied each model in an experiment where we shifted the current transplantation date by 1 week for up to 5 weeks before and after the current date under the same climate change scenarios.The period of water withdrawal for irrigation was also shifted in accordance with the number of days' shift in the transplantation date (see Section 3.3).Then, we plotted the results of calculation of the benefits and risks of each transplantation date under each climate scenario (Figure 3).
We evaluated the relationships between two stakeholders by examining the benefits and risks of adaptation options: namely, synergistic and trade-off relationships (Figure 3).If adaptation options are beneficial to the two stakeholders (i.e., the plots are distributed downward to the right), then the adaptation measure creates a "synergistic" relationship between the two (Figure 3a).However, if adaptation options that benefit one stakeholder result in risks to another (i.e., the plots are distributed upward to the right), then the adaptation measure creates a "trade-off" relationship (Figure 3b).The orange horizontal dashed line in Figure 3 is the predefined threshold of the risk (e.g., water deficit) at which river administrators do not allow actions that induce a further increase in risk.It is apparent that, of the two typical relationships, the trade-off relationship would be highly likely to face the occurrence of a soft adaptation limit.

Climate Change Scenarios
To apply the framework proposed in Section 3.1 and Figure 3 under different climate change scenarios, we used the general circulation model (GCM)-based historical (1981-2000; hereafter, "historical") and RCP (Representative Concentration Pathway) 2.6 and 8.5 (2011-2030 and 2031-2050) climate scenarios.There are two reasons why we chose to limit our future projections to 2050.One is that the shift in the transplantation date is an incremental adaptation measure (Iizumi, 2019), and we assumed that other transformative adaptation measures would be implemented after 2050.The other is that our projections of water risks and rice benefits could not account for the coevolutionary dynamics of system properties that could be changed by human adaptive actions and were therefore not suitable for long-term projections.
We collected the climate change scenarios from three GCMs, namely MIROC5, MRI-CGCM3, and HadGEM2-ES.These data sets were obtained from Ishizaki (2020).The GCM outputs, with spatial resolutions of approximately 100-200 km, are insufficient to describe regional climate conditions.Therefore, we spatially interpolated the outputs to 1-km grids by means of simple linear interpolation using the inverse distance weighted method.Then, we used the CDF mapping method (Ines & Hansen, 2006;Li et al., 2010) to bridge statistical gaps in climate variables between the observations and the GCM simulations.The observations  were interpolated to a 1-km grid by using daily meteorological data recorded at the Japan Meteorological Agency observation stations by means of the inverse distance weighted method.
To evaluate the risk of water deficit, the water deficit was calculated for each year (see Section 3.3); the number of data used for the evaluation per period (1981-2000, 2011-2030, or 2031-2050) was 60 (three GCMs × 20 years).To evaluate the benefits of rice production, the 20-year average of yield and quality was calculated for each of the three periods (see Section 3.4); thus, the number of data used for the evaluation per period was three (for three GCMs).
We plotted the daily mean temperature and total precipitation for each year in the Shinano River watershed during summer (June-August) (Figure 4).In the future period (2011-2090), the daily mean temperature gradually increased under both the RCP 2.6 scenario and the RCP 8.5 scenario, with a particularly high rate of increase the latter after 2050.The total precipitation showed large interannual variations under both scenarios, and no clear trend of change was observed for future periods.

Process-Based Models Used to Evaluate the Risk of Water Deficit
We used a distributed water circulation model developed by Yoshida et al. (2016) to evaluate the risk of water deficit associated with shifts in the transplantation date.The model represents a watershed as 1-km grid cells, and it calculates the river discharge for each grid cell on the basis of meteorological data such as daily precipitation, temperature, wind speed, and short-and long-wave radiation.Furthermore, the model simulates water circulation processes in paddy irrigation districts by inputting information on water-use facilities such as reservoir management, water withdrawal from rivers, and water allocation into each grid cell.The model can provide various types of information on water use for irrigation paddies, such as the amount of water released from reservoirs, the amount of water withdrawn by major water-use facilities, the amount of water supplied to paddy fields, the return flow from irrigation districts, the rice planting period and area, and the actual evapotranspiration from paddy fields.This enables us to simulate the streamflow that is highly disturbed by irrigation systems; this disturbance is typical in Japanese watersheds.Further details about the model components are provided by Yoshida et al. (2016).
To apply the model to the Shinano River watershed, we reflected the operation of 29 major reservoirs (multipurpose, municipal, hydropower, and irrigation reservoirs) and 88 irrigation districts in the watershed.Reservoir operation was modeled on the basis of predetermined operational rules and actual operational data over the past 20 years, including water release and storage data obtained from daily management reports.Water storage in each reservoir was constrained by the recorded minimum and maximum water storage on that date.In cases where the water storage exceeded the maximum constraint, water was released from the reservoir to meet the target storage level.In cases where the water storage was less than the minimum constraint, release was regulated to maintain the target storage level.The amounts and periods of water withdrawal in the irrigation districts were modeled on the basis of the recorded withdrawal rates over the past 5 years (2011)(2012)(2013)(2014)(2015).Each withdrawal amount was averaged on a daily basis, and then the amount was diverted from the simulated streamflow as long as streamflow was available.If streamflow was not available at the diversion point, the model withdrew all of the streamflow at that point.
We calibrated the model from 1981 to 2010 by using daily meteorological data recorded at Japan Meteorological Agency observation stations.We then validated it by using 5 recent years (2011-2015) at 22 stations within the watershed.We plotted hydrographs of the calculated and observed discharges at a representative station (Figure 5, for the Ojiya station in Figure 1).The 5year average Nash-Sutcliffe model efficiency coefficient at the Ojiya station was 0.68.The observed and calculated hydrographs were well fitted at all stations.The 5-year average relative error was less than 30% at 17 stations, and at the Ojiya station the model had a good fit with the observed discharges, with an average relative error of 19.3%.A mean relative error exceeding 30% occurred in the middle reaches of the Shinano River, where the effect of reservoir discharge operations was marked.The average bias in the annual water balance was less than 20% at all stations and 6% at the Ojiya station.Therefore, we judged the model to be accurate enough to withstand long-term analysis.
The risk of water deficit was quantified from the simulated river discharge and the minimum flow requirement for the irrigation period at a reference station for water resource management.The minimum flow requirement is defined at the reference station by aggregating the water demand downstream of the station.The water demand includes domestic, industrial, irrigation, fisheries, and environment, so that the minimum flow requirement can be used as a threshold to evaluate the water supply demand balance of the watershed.At the Ojiya station (the reference station for the Shinano River Watershed), a minimum required flow of 145 m 3 /s was determined for the irrigation period (from 28 April to 15 September).This corresponded to the lower 1.3% of values of past observed streamflow during the irrigation period from 1986 to 2015.However, it was problematic to use the absolute value of the minimum flow requirement as the threshold for evaluating the streamflow driven by GCM-based forcing data, because the low flow conditions calculated with the historical scenarios differed from those simulated with the observed data.As described in Section 3.2, we corrected for the bias between the observed meteorological data and the historical scenarios on a monthly basis, but the differences in the time series of precipitation could not be corrected by using this statistical method.In short, the GCM-based forcing data tended to have shorter dry spells than those of the actual data, so that the simulated low flows were higher than the observed low flows.Here, we determined the threshold values for each GCM on the basis of the 1.3rd percentile of the simulated streamflow under historical scenarios (corresponding to the probability of the minimum flow requirement not exceeding the streamflow during the irrigation period).The resulting threshold values were 163.5 m 3 /s (HadGEM2-ES), 181.8 m 3 /s (MIROC5), and 168.6 m 3 /s (MRI-CGCM3).
We calculated the cumulative amount below the threshold on an annual basis (hereafter, the "water deficit").The water deficit is a surrogate of the magnitude of the water deficit in each year.Because water deficit does not occur frequently in the watershed, a 10-year probability of the water deficit is considered in the planning of water rights (see Section 2.1).Therefore, we used the water deficit of the sixth-largest of the evaluation period (from the data of 60 years, integrating a 20-year period with the three GCMs) to depict the plots in Figure 3.An increase in the water deficit makes it more difficult to acquire water rights for adapted transplantation dates.In other words, this difficulty can be interpreted as a soft adaptation limit for selecting the optimal transplantation date in water-rice systems in Japan.

Process-Based Models to Evaluate the Benefits of Rice Production
We used a process-based rice growth model developed by Hasegawa and Horie (1997) and Ishigooka et al. (2017) to assess the rice production benefits resulting from a shift in the transplantation date.This model has three major components: phenological development, biomass production, and yield formation.The phenological development component quantified the developmental stages (emergence, panicle initiation, heading, and maturity) from the daily mean air temperature (average of daily maximum and minimum) and day length.The biomass production component estimated the daily increases in biomass and leaf area on the basis of biophysical processes.The daily biomass increase was calculated as the difference between the products assimilated by photosynthesis and consumed by respiration, accounting for the photosynthesis-enhancing effect of increasing atmospheric CO 2 concentration.Through this process, total biomass was calculated as the accumulation of daily biomass increases (dry matter).The yield formation component calculated as the brown rice yield (hereafter, "total yield") by multiplying the biomass (dry weight production of the aboveground portion) and the harvest index.The harvest index took into account three factors of yield reduction: spikelet sterility caused by low temperatures or by high temperatures, and insufficient grain filling due to delayed maturity.Further details about the model are given by Hasegawa and Horie (1997) and Ishigooka et al. (2017).
In this model, we assumed that rice growth was not affected by the availability of water resources.This is because water stress on rice growth is relatively unlikely to occur in Japan owing to the mechanisms by which human power prevents water stress on paddy rice (as mentioned in Section 2.1).
We applied this model to the target irrigation area at a spatial resolution of 1 km and assigned the cultivar as Koshihikari to set parameters of the model.The applicability of the model to the various regions of Japan has been shown by Ishigooka et al. (2017).They compared the simulated heading date (i.e., phenology) and yield with observed data for 15 cultivars.For Koshihikari, the accuracy of the model was verified on the basis of 4461 observation records-the largest number among the 15 cultivars.The mean bias and root mean square error (RMSE) of the periods from transplantation date (given) to heading date (simulated) for Koshihikari were 1.67 and 4.35 days, respectively.The values were less than 10% of the observed duration from transplanting to heading, and they were within 7 days, which was the interval of shifting transplantation date that we used here.The mean bias of the yield for Koshihikari was 0.21 t/ha, and the percentage of the observed yield was 4%.Therefore, we judged that the estimated phenology and yields corresponded well with the observed data.Further details about the validity of the model are given by Ishigooka et al. (2017).
We used three indices to evaluate the benefit of rice production: total yield, appearance quality, and farmers' income.Total yield was calculated by using the rice growth model.Appearance quality was estimated on the basis of the heat stress index for rice quality, as defined by Ishigooka et al. (2011).The heat stress index (hereafter, "HD_m26") is related to the emergence of chalky grains because of high temperatures-that is, to deterioration of the appearance quality of the rice.The index was calculated as the cumulative value of positive differences in daily average air temperature above 26°C within 20 days after the heading date.Ishigooka et al. (2011) classified yield into three classes on the basis of the degree of quality degradation risk due to high temperature during the early grain-filling period: Class A (low risk), HD_m26 < 20°C•days; Class B (moderate risk), 20°C•days ≤ HD_m26 < 40°C•days; and Class C (high risk), HD_m26 ≥ 40°C•days.Among the three classes, we used "Class A yield" as an indicator of appearance quality in the evaluation.We calculated the farmers' income by relating the appearance quality of the rice to the grade; that is, we defined Class A as firstgrade rice and Classes B and C as second-grade, and we calculated the income on the basis of the respective yields and prices.
The effect of shifting the transplantation date on yield and appearance quality (Class A yield) was described by Ishigooka et al. (2017).Later transplantation date leads to a benefit for appearance quality because the heading date and ripening period can be shifted after the peak of high temperature in midsummer.However, it leads to disadvantage in yield formation because the ripening period shifts to autumn when solar radiation energy is insufficient for adequate grain filling.On the other hand, earlier transplantation date leads to high yield because the ripening period can be completed before the peak of high temperature in midsummer and the major growing period corresponds with the period of high solar radiation (from late spring to early summer).

Relationships Between Water and Rice in the Historical Scenario
We examined the relationships between risk of water deficit and benefit of rice production (i.e., total and Class A yields, and farmers' income) in the historical scenario (1981)(1982)(1983)(1984)(1985)(1986)(1987)(1988)(1989)(1990)(1991)(1992)(1993)(1994)(1995)(1996)(1997)(1998)(1999)(2000) (Figure 6).A plot with farmers' income on the horizontal axis is not shown in Figure 6 because it was almost identical to the plots with total yield, owing to the small difference in price between the first and second grade of rice.The relationship between water deficit and total yield (Figure 6a) showed a synergistic relationship, as the water deficit decreased and the total yield increased when the current transplantation date (9 May 2019) was shifted earlier.The relationship between water deficit and Class A yield (Figure 6b) showed a trade-off relationship, as both the water deficit and the Class A yield increased when the current transplantation was shifted later.
We interpreted the relationships shown in Figure 6 on the basis of the changes in past transplantation dates in Figure 2. The current transplantation date was plotted at the location of almost minimum water deficit and maximum total yield, rather than maximum Class A yield.The peak transplantation date in the 1950s was about 4 weeks later than the current one, corresponding to the +4 weeks plot ( † in Figure 6).The current transplantation date facilitates a higher total yield and lower water deficit, although the Class A yield (i.e., appearance quality) is lower.Therefore, we inferred that the earlier transplantation date used from the 1950s to around 2000 (Figure 2) was not limited by water risk and was synergistic to both water risk and yield.
Given that the transplantation date in the 1950s corresponds to the plot of +4 weeks ( † in Figure 6) and that in 2000 corresponds to the plot of 1 week (* in Figure 6), the calculated farmers' income on each of these transplantation dates was 1,059,020 yen/ha and 1,160,650 yen/ha, respectively.Thus, from the 1950s to about 2000, the transplantation date moved in the direction of increasing income.In contrast, the late transplantation dates after 2000 have been irrational in terms of both economics and water risk.The calculated farmers' income in 2019 was 1,153,690 yen/ha-lower than in 2000.The late transplantation date could have increased the water deficit.Although the reasons for this change in the date are not clear, we consider that it reflects the effects of adaptation measures to moderate the negative impacts on rice quality, as described in Section 2.2.Therefore, the changes in transplantation date shown in Figure 2 may imply that the decision-making process changed in about 2000.

Soft Adaptation Limits in Water-Rice Coupled Systems
To predict soft adaptation limits in water-rice coupled systems under climate change by 2050, we hypothesized two possible patterns of transplantation date changes (i.e., changes in human behavior) on the basis of the past data shown in Figure 2. One was a pattern of earlier transplantation dates, as in the data from the 1950s to about 2000 (i.e., a trend toward higher rice yields and farmers' incomes).The other was a pattern of later transplanting dates, as in the data from the 2000s onward (i.e., a trend toward greater appearance quality of rice).Therefore, we selected the total and Class A yields as indicators of the benefits of rice production to predict relationships in the water-rice coupled systems.We excluded farmers' income from the indicator because it depends on the future price of rice, which is affected by socioeconomic conditions.We then created plots based on these indicators and on water deficit to examine whether the framework could predict soft adaptation limits under climate change.
The relationships between water deficit and total yield under the climate change scenarios are shown in Figure 7.In addition to the results of the RCP 2.6 and 8.5 scenarios for 2011-2030 and 2031-2050, the results under the historical (1981)(1982)(1983)(1984)(1985)(1986)(1987)(1988)(1989)(1990)(1991)(1992)(1993)(1994)(1995)(1996)(1997)(1998)(1999)(2000) scenario are shown for comparison.Under all future scenarios, the relationships between water deficit and total yield were synergistic, as in the historical scenario.With the current transplantation date, the total yields in all future scenarios were higher than that in the historical scenario.The RCP 2.6 scenario for both 2011-2030 and 2031-2050 gave higher water deficits than the historical (Figure 7a), whereas the water deficits in the RCP 8.5 scenario did not differ substantially from that in the historical scenario (Figure 7b).These results indicated that shifts in transplantation dates to increase total yield in the future may not be limited by the risk of water deficit; thus, the water rights for the optimal cultivation period could be acquired.
We also plotted the relationships between water deficit and Class A yield under climate change scenarios (Figure 8).In all future scenarios, the relationships between water deficit and Class A yield were tradeoffs, as in the historical scenario.Class A yields with the current transplantation date would decrease in the future under both scenarios, but in 2011-2030, the same yield as the historical one could be ensured by shifting the transplantation date; a 5-week delay in the transplantation date under both the RCP 2.6 and RCP 8.5 scenarios in 2011-2030 ( † in Figure 8) would ensure Class A yields of 4.4 t/ha and 4.2 t/ha, respectively, which were comparable to the 4.3 t/ha at the current transplantation date under the historical scenario.However, delaying the transplantation date by 5 weeks in 2011-2030 would result in water deficits that were approximately 2.2 and 11.8 times greater, respectively, than those with the current transplantation date for the RCP 2.6 and RCP 8.5 scenarios.Furthermore, in 2031-2050, a 5-week delay in the current transplantation date under both scenarios (* in Figure 8) would not ensure the same degree of Class A yield as with the current transplantation date under the historical scenario, and the water deficit would also increase.These results indicated that delaying the transplantation date to increase Class A yield (i.e., improve appearance quality) may be hampered by increased water deficit.
Here, we posit that there are soft adaptation limits in the water-rice coupled systems, whereby the current yield or quality of rice could be ensured by shifting the transplantation date, but the shifted transplantation date could not be accepted in terms of the water deficit.If farmers decide on the optimal transplantation date with the aim of maximizing yield, the shifts would not be restricted in terms of water resources (Figure 7).However, we argue that the farmers' decision to aim at maximizing quality would result in soft adaptation limits in 2011-2030 (Figure 8).The conditions in 2031-2050 whereby the shifted transplantation date could not ensure the same level of rice quality as currently (Figure 8) would be categorized as a hard adaptation limit, implying that more transformative adaptation measures would be required in this period (i.e., developing new cultivars that could tolerate heat stress, or switching of crop types).Our predictions suggest that the occurrence of soft adaptation limits depends on the farmers' decision on the transplantation date.In particular, we emphasize that continuation of the trend since the 2000s for a preference for later transplantation dates (Figure 2) would result in soft adaptation limits.
Soft adaptation limits are in evidence across regions and sectors and have been reported to be derived from many adaptive constraints.The major constraint to adaptation is a lack of governance, financial and technical resources, and information.Nambi et al. (2015) proposed a methodology to identify constraints that limit the implementation of adaptation measures in the agriculture and water sectors.Their methodology was based on interviews with farmers regarding adaptation measures.They analyzed the results of these interviews by using four criteria: effective awareness, economic viability, individual and institutional compatibility, and flexibility and independent benefits.Their results revealed that a lack of institutional support has limited the ability of farmers to implement adaptation, even if information about the benefits is acknowledged.Sain et al. (2017) probabilistically analyzed the cost-benefit of climate-smart agriculture practices and technologies associated with smallholder maize and beans production systems in the Dry Corridor in Guatemala.They showed that financial resources were significant constraints to implementing a relevant adaptation strategy to improve food security, resilience, and low emission development.Esteve et al. (2018) used stakeholder questionnaires to assess the impact of identified barriers on the implementation of selected adaptation measures.They revealed that lack of acceptance, lack of common understanding, and inadequate awareness were among the constraints most frequently identified across different adaptation options.As shown in these studies, it is possible to identify current soft adaptation limits and adaptive constraints on the basis of interviews with stakeholders or analyses.However, a method for predicting Our proposed framework successfully predicted the soft adaptation limits that may be faced in the implementation of adaptation measures and identified the adaptive constraints that led to these limits.The novelty of the framework is that it is based on process-based models and treats the actual adaptation measure of shifting the transplantation date as a model parameter.Moreover, societal rules that explicitly govern institutional decisions are employed in the framework as constrains on adaptive actions.We argue that the framework is effective and applicable for predicting possible future soft adaptation limits for other social problems involving multiple stakeholders, provided that process-based models that can simulate the risks and benefits of adaptive actions are available.

Understanding the Principles Behind the Human Decision-Making Process
Our results also demonstrate that the occurrence of soft adaptation limits depends on the farmers' choice of whether to maximize total yield or appearance quality.Although here we have presented possible future scenarios regarding farmers' choices, it is important to understand the decision-making process and to predict which choices humans are likely to make in the future.As discussed in Section 4.1, the turnaround in transplantation dates after about 2000 in Figure 2 was remarkable because it cannot be explained by either economic benefit or water deficit risk.We argue that this turnaround in the transplanting date will be a key to understanding the principles behind the decision-making processes in Japanese water-rice coupled systems.
We infer that the turnaround was related to the brand power of the Koshihikari rice cultivated in Niigata Prefecture.As mentioned in Section 2.3, Koshihikari is the most preferred cultivar among Japanese consumers and has the largest cultivation area in Japan.In particular, Koshihikari cultivated in Niigata Prefecture fetches a higher price than that cultivated in other regions and is highly evaluated as a high-end brand of rice.The decline of the quality of such brand rice as a result of climate change can be a concern, not only to farmers but also to the government.In fact, the Niigata Prefectural Government has issued adaptation strategies for protecting rice quality, and delaying the transplantation date was one of the incremental methods they listed to prevent degradation of the quality of Koshihikari (MAFF, 2006).Although delaying transplantation leads to a decrease in shortterm profit (i.e., yield and income), priority seems to be placed on preventing a long-term decline in the public valuation of this cultivar.In fact, the MAFF data set of yearly statistics on rice cultivation schedule reveal that only seven out of the 47 prefectures, including Niigata Prefecture, had a turnaround in the transplantation date in about 2000.This may indicate that climate and socioeconomic changes have altered the decision-making process so that protecting the high-end brand value of Koshihikari takes precedence over the short-term economic rationale of increasing yield and income.
Our results showed that the turnaround in transplantation dates could lead to soft adaptation limits.As described in Section 2.1, rice production in Japan has been based on a water-rice coupled system that has co-evolved since the 17th century.In this mature system, a turnaround due to changes in climate and socioeconomic activities may lose equilibrium and thus lead to soft adaptation limits.A model that internalizes farmers' decisions and preferences will enhance the capability of predicting soft adaptation limits in human-water coupled societies.We intend to examine the principles behind the delayed transplantation date in a future work by using information on cultivation policies, prices, cultivars, and rice production.

Conclusion
We presented a novel framework for predicting soft adaptation limits under climate change on the basis of the assumption that shifts in the transplantation date in Japanese water-rice coupled systems are related to the occurrence of the soft adaptation limits.We compiled two process-based models, one to evaluate the risk of water deficit and one to evaluate the benefits of rice production (e.g., rice yield, quality, and farmers' income).We applied each model in an experiment where we shifted the current transplantation date by 1 week for up to 5 weeks before and after the current date under the historical (1981( -2000( ) and RCP 2.6 and 8.5 (2011( -2030( and 2031( -2050) ) climate scenarios of three GCMs.The new framework was applied to the Shinano River watershed, a typical Japanese watershed where streamflow is highly disturbed by the use of irrigation systems.We found that the current transplantation date was beneficial to both water deficit and rice yield ("synergistic"), with less water deficit and higher yields than those at other transplantation dates.The synergistic relationship between water Water Resources Research 10.1029/2022WR034219 TAKADA ET AL. deficit and rice yield was also apparent under all climate change scenarios; therefore, shifts in transplantation dates to increase total yield in the future may not be limited by the risk of water deficit.On the other hand, we found that shifts in the transplantation date that benefit rice quality would result in water deficit risks under all climate change scenarios.Under the RCP 2.6 and RCP 8.5 scenarios, delaying the transplantation date by 5 weeks in 2011-2030 would ensure rice quality comparable to that with the current transplantation date under the historical scenario, but it would result in greater risk of water deficit.We posit that this situation represents soft adaptation limits.Our results revealed two contrasting development pathways in the study watershed.Soft adaptation limits imposed by water availability will occur by 2030 if farmers optimize for quality, whereas mutual benefits to farmers and river administrators will be achieved if farmers seek yield.
Our proposed framework successfully predicted the soft adaptation limits that may be faced in the implementation of adaptation measures.The novelty of the framework is that it is based on process-based models and treats the actual adaptation measure of shifting the transplantation date as a model parameter.Moreover, societal rules that explicitly govern institutional decisions are employed in the framework as constrains on adaptive actions.It is therefore possible to extend the application of our framework to a larger domain, such as the entire Japanese archipelago.Applying this framework to other regions could enhance our understanding of the occurrence of soft adaptation limits based on more robust results using different climate and regional characteristics.The framework is effective and applicable for predicting possible future soft adaptation limits for other social problems involving multiple stakeholders, provided that process-based models that can simulate the risks and benefits of adaptive actions are available.We argue that more participatory research with stakeholder engagement, as well as policy discussions about these possible developments, is needed to ensure successful adaptation.

Figure 1 .
Figure 1.Map of the Shinano River watershed and the target irrigation area.Panel at bottom left shows the monthly mean temperature (orange line) and precipitation (blue bars) and the approximate times of each step in the rice-growing cycle at Nagaoka.TP and HD denote transplanting and heading, respectively.

Figure 2 .
Figure 2. Changes in the transplantation date (solid lines) in Niigata Prefecture from 1953 to 2021 and in the Chuetsu region in the middle of Niigata Prefecture from 1969 to 2021.The transplanting period (pink filled area) from 1982 to 2021 was calculated as the difference between the start and end dates of rice transplantation in Chuetsu.

Figure 3 .
Figure3.Risk-benefit relationships of adaptation options for two stakeholders: (a) if the outcomes of an adaptation option are beneficial to both, then the adaptation measure creates a "synergistic" relationship and the risk declines as the benefit increases; (b) if the benefit of an adaptation option to one stakeholder results in a detriment to the other, then the adaptation measure creates a "trade-off" relationship whereby the risk increases as the benefit increases.The orange horizontal dashed line is the predefined threshold of the risk (e.g., water deficit) at which river administrators do not allow actions that induce a further increase in risk.

Figure 4 .
Figure 4. Changes in mean temperature and precipitation during summer (June-August) under two climate change scenarios: (a) RCP 2.6 and (b) RCP 8.5 for the period 2011-2090.The arithmetic mean and 10-year moving average of the three GCMs for each scenario are shown by solid lines, whereas the 95% confidence intervals for each element are indicated as filled areas.

Figure 5 .
Figure 5. Hydrographs of calculated and observed discharges at the Ojiya station in 2015.In that year, the relative error was 21.5%, the bias in the annual water balance was 8%, and the Nash-Sutcliffe efficiency was 0.85.

Figure 6 .
Figure 6.Relationship between (a) water deficit and total yield, and (b) water deficit and Class A yield, when the current transplantation date (square) was shifted under the historical scenario (1981-2000).The gray dashed lines indicate total and Class A yields and water deficit with the current transplantation date.The symbols of † and * indicate the peak transplantation date in the 1950s corresponds to the plot of +4 weeks and that in 2000 corresponds to the plot of 1 week, respectively.

Figure 7 .
Figure 7. Relationship between water deficit and total yield when the current transplantation date (squares) was shifted under the (a) RCP 2.6 and (b) RCP 8.5 scenarios in 2011-2030 and 2031-2050.The gray dashed lines indicate the water deficit and total yield with the current transplantation date under the historical scenario (1981-2000).

Figure 8 .
Figure 8. Relationship between water deficit and Class A yield when the current transplantation date (squares) was shifted under the (a) RCP 2.6 and (b) RCP 8.5 scenarios in 2011-2030 and 2031-2050.The dashed lines indicate the water deficit and Class A yield with the current transplantation date under the historical scenario (1981-2000).The symbols of † and * indicate a 5-week delay in the current transplantation date in 2011-2030 and 2031-2050, respectively.