and possible adaptation of suggested the of potato yield Risk Assessment and Possible Adaptation of Potato Production in Hokkaido to Climate Change Using a Large Number Ensemble Climate Dataset d4PDF

While global warming may expand suitable places for potato cultivation in cold regions, it may reduce the yield due to the increase of hot days during the tuber growth period. This study evaluated the effects of global warming on potato cultivation over Hokkaido by dynamically-downscaled ensemble experiments called d4PDF and assessed applicability of possible adaptive measures. In this study, we define the suitable area based on the accumulated temperature and deduced a relationship between the potato yield per unit area and the number of hot days (maximum temperature > 28°C) from crop statistic data. In a warming environment with 2K or 4K increase in global-mean temperature relative to the present climate (1981−2010), the accumulated tem peratures likely satisfied the criterion on potato production almost over Hokkaido. The risk of growth delay due to cold weather was projected to reduce. However, hot days in the tuber growth period would increase, reducing potato yield by 7% in a plus 2-K climate and 16% in a plus 4-K climate. This risk of yield loss would not be avoidable by moving up planting by 30 days, and the development of varieties that are tolerant to 31−33°C would be a possible way to adaptation. potato


Introduction
Potato (Solanum tuberosum L.) is one of the major staple foods and is widely harvested over the world (FAOSTAT 2020). Hokkaido, a northern main island of Japan, accounts for about 80% of the total potato production in Japan [Ministry of Agriculture, Forestry and Fisheries (MAFF), 2019]. The main production area is Okhotsk, Shiribeshi, and Tokachi subprefectures in Hokkai do (Fig. 1). As the potato growth rate is closely related to the temperature accumulation, it was modeled by the accumulated temperature as WOFOST (Diepen et al. 1989) and LINTUL (Kooman and Haverkort 1994;Haverkort et al. 2015). For example, WOFOST sets the lower limit of the daily temperature to 3°C, and the upper limit to 18°C from planting to sprouting and to 28°C after sprouting. The abundant potato yield hence requires a sufficient accumulated temperature in the production period, though a high temperature exceeding 28°C is not only useless for plant growth but harmful for potato tubers (Streck et al. 2007;Shimoda et al. 2018) because high temperatures excessively consume photosynthate by plant respiration.
Recently, global warming 1 affects the potato production in Hokkaido. For example, Hokkaido experienced a record-breaking heatwave in July and August 2010. As a result, the yield of potatoes, paddy rice, wheat, beats, and others greatly decreased (Hirota et al. 2011;Nemoto et al. 2011;Shimoda et al. 2015). As global warming progresses in the future, the environmental conditions like 2010 in Hokkaido would be increasing. Hence, it is worthwhile to consider global warming adaptation on crop management in Hokkaido. Hijmans (2003) showed a pioneering study on risk assessment and climate change adaptation on potato cultivation, based on climate change scenarios. They compared the climate between 1961 and 1990 with the future climate between 2040 and 2069, and estimated that global potato yields will decrease by 18% to 32%, depending on the region. They argued that an appropriate adaptation would limit the decline to 9% to 18%. There are some reasons why high temperatures can reduce potato yields. Tito et al. (2018) revealed that the increase in pests and diseases of potato due to global warming also reduce yields. Adavi et al. (2018) revealed that potatoes were exposed to high temperatures during the tuber grow period, which would increase the risk of yield loss. Other studies also suggested the risk of potato yield based on yield data of agriculture, forestry and fisheries statistics published by Hokkaido Regional Agricultural Administration Office (https:// www.maff.go.jp/hokkaido/toukei/kikaku/sokuho/index.html, accessed on 31 August 2020). The color shade is as per the reference in the bottom. The boundaries of the municipality are published by National Land Information Division, National Spatial Planning and Regional Policy Bureau (https://nlftp.mlit.go.jp/ksj/index.html, accessed on 31 August 2020

Current suitability for potato production
We analyzed the accumulated temperature in Obihiro from 1984 to 2019 (Fig. 3). According to an official record, the planting date is almost the beginning of May (Fig. 3d) and the accumulated temperature from planting to sprouting ranged between 200 and 350 degree-days (DD; Fig. 3a). As the accumulated temperature from planting to flowering ranged generally between 600 and 700 DD (Fig. 3b), the average date of flowering is 4 July. The average harvesting date is 19 September, when the accumulated temperature from planting almost attained 1900 DD (Fig. 3c). This is consistent with a WOFOST recommendation that the accumulated temperature from planting to harvesting is 1700−2000 DD. It is also recorded that the harvesting date is slightly earlier in the recent decade with the accumulated temperature to harvesting slightly increasing.
Next, we check the risk of yield loss due to high temperatures. Figure 4 displays the relationship between potato yield and the number of days with the maximum temperature exceeding 28°C (hereafter hot days) from flowering to harvesting from 1984 to 2018. The potato yield is anticorrelated with the number of hot days with the correlation coefficient being −0.56, which is statistically significant at a 5% level. We can then relate the yield Y [Mg ha −1 ] to the number of hot days N [day] by the regression as Y = -0.21N + 39.43.
(2) loss in sub-Saharan Africa (Pironon et al. 2019), India (Pradal et al. 2019), northern China (Zhao et al. 2016;Wang et al. 2019), and northern Europe (Pulatov et al. 2015). In Hokkaido, Hokkaido Research Organization (2011) showed that potato yields could be 10−15% lower in the 2030s than they are between 1990 and 2009, using WOFOST based on future climate data based on IS92a scenario from two of the four climate models in Yokozawa et al. (2003). Many of the above studies were limited to the discussion on changes in the mean climate state, but they suggested that global warming reduced the risk of persistent low temperature and increased the risk of extremely high temperatures. Phenomena occurring at a very low probability in the current climate, like the 2010 hot spell in Hokkaido become more likely as climate change progresses. These kinds of data samples can be now obtained from a large number of ensemble simulations at various levels of warming. However, no studies used a large number of ensemble data to assess the risk of potato yield to global warming, although some used the global climate model simulation with tens of years length (Hijmans 2003;Stöckle et al. 2010;Adavi et al. 2018).
The purpose of this study is to evaluate the impact of climate change on potato production in Hokkaido, especially focusing on Obihiro, the main potato production area. We used the Database for Policy Decision Making for Future climate change (d4PDF; Mizuta et al. 2017;Fujita et al. 2019), a large number of ensemble meteorological simulation dataset, which has been applied to risk assessment of river flood (Hoshino et al. 2020;Tanaka et al. 2018) and agriculture (Iizumi et al. 2018;Takimoto et al. 2019).
In this study, we first show that the current climate is suitable for potato production in Hokkaido. We then evaluate, as a change of probability distribution, the risk of potato yield loss due to high temperatures in the global warming climate. We also propose two possible adaptations of potato production to climate change.

Data and method
We used 2-m air temperature data produced by 20-km resolution dynamic downscaling simulation covering Japan in d4PDF. In d4PDF, there are three sets of experiments: a historical climate experiment (hereafter HIST run), a +2-K future climate experiment (2K run, equivalent of 2040 in RCP8.5 scenario), and a +4-K future climate experiment (4K run, equivalent of 2100 in RCP8.5 scenario). The ensemble members are 50 members in HIST run, 54 members in 2K run, 90 members in 4K run, respectively. Time interval of the model output is 1 h. We also used Automated Meteorological Data Acquisition System (AMeDAS) observation data in 1976−2018 at Obihiro (143°17.3′E, 42°55.3′N; Fig. 1b). Considering the effect on yield estimates in this paper, bias correction of the d4PDF original daily-mean temperature data was made by a simple linear offset, so that the climatology of the monthly-mean daily-maximum temperature in AMeDAS observation at Obihiro station and d4PDF at the nearest gridpoint to the station were consistent (Fig. 2a). The daily-maximum temperature was corrected by offsetting the mean and factoring the standard deviation (Figs. 2b and 2c) as where T  is the corrected temperature, T and T o are respectively simulated and observed temperatures, T  and T  o are each climatology, and σ denotes the standard deviation. This bias correction was done month by month.
We used an official record of planting, sprouting, flowering, and harvesting in Obihiro, based on Crop Statistics data of MAFF from 1984 to 1990, the Hokkaido Department of Agriculture data from 1991 to 2018. The data from 1991 to 2005, except for 1998, were not fully available (Fig. 3d). According to the record, we set the standard planting date on 1 May, unless otherwise noted. Here, for simplicity, the tuber growth period, exactly from flowering to mature, is approximated to the period from flowering to harvesting. We also used an official record of potato yield and cultivation Here, we used this simple linear regression because the Akaike Information Criterion for that model is the smallest compared to those for the quadratic and cubic models fitted.
It is noted that this statistical relation will be used in the analysis for risk assessment under future climates by converting the number of hot days to yield. Summarizing these results, Obihiro is currently a suitable place for potato production because the accumulated temperature to harvesting ranged between 1700 and 2000 DD. However, this place is being exposed to a risk of a decrease in the yield due to high temperatures from flowering to harvesting as reported below.

Risk assessment
First we assess the risk of potato yield loss for low temperatures. Figure 5 shows the probability of satisfying accumulated temperature conditions exceeding 1700 DD or 2000 DD for HIST, 2K, and 4K runs of d4PDF. In the HIST run, the distribution of probability satisfying the 1700 or 2000 DD criteria (Figs. 5a and 5d) is spatially correlated with that of potato yields in the current climate (Fig. 1a). Soya, Nemuro, and Kushiro subprefectures are   not suitable places due to low temperatures with accumulated temperatures that reach 1700 DD, whereas Tokachi, Shiribeshi, and Okhotsk subprefectures had a larger production of potatoes in Hokkaido mainly because accumulated temperatures are greater than 2000 DD. This analysis based on the HIST run is consistent with the analysis based on the AMeDAS observation (not shown). Most areas of Hokkaido except for high mountains in central Hokkaido satisfied the criteria in 2K run while nearly all areas satisfied the criteria in 4K run (Figs. 5b, 5c, 5e, and 5f). These results indicated that potato cultivation would become feasible in Soya, Kushiro, and Nemuro subprefectures under global warming.
Next, focusing on Obihiro again, we counted the number of days with the maximum temperature exceeding 28°C from flowering to harvesting for HIST, 2K, and 4K runs of d4PDF (Fig.  6a). The average number of hot days is 13 days with the standard deviation of 5.9 days in the HIST run. Compared with this, it increases by 15 days in the 2K run and by 25 days in the 4K run on average. The number of hot days was highest at 31 days in 2010, but it is even on the cold side of the extreme in the 4K run. We can convert the number of hot days to yield by a statistical relation (Eq. 2) (Fig. 6b). The frequency distribution of potato yield for the HIST run shows the range between 32 to 39.5 Mg ha −1 , consistent with the record of potato yield in Obihiro (Fig. 4). In the 2K run, the yield ranged from 30 to 37.5 Mg ha −1 . The average yield was 33 Mg ha −1 , which was the low yield in the current climate. Moreover, in the 4K run, the yield ranged from 28 to 35 Mg ha −1 with its extremely high yield side corresponding to the extremely low yield side in the current climate. The expected yield loss was up to 7% in the plus 2-K climate and 16% in the plus 4-K climate (Fig.  6b).

Possible adaptations
Here we will consider two possible ways of climate change adaptation, which mitigates the risk of yield loss due to high temperatures. One way is to shift the planting date earlier. Whereas the standard planting date was here set on 1 May, the planting date was shifted to 20 April, 10 April, and 1 April for adaptation. This change of planting date is reasonable because the snow-covered days in Hokkaido is reduced by about 30 days under plus 2-K climate relative to the 1990s (Katsuyama et al. 2019). However, the number of hot days from flowering to harvesting does not change much, if the planting date was shifted one month earlier in the 2K or 4K runs (Figs. 7a and 7b; Histograms for 20 April and 10 April are not shown.). Even in the plus 2-K climate, the risk of high temperatures exceeded 40 days, which is more than 50% of the tuber grow period. This is the reason why the risk of reduced yields is unavoidable with a change in planting date. The other way is to develop a new variety of potatoes that are tolerant of the daily maximum temperature above 28°C during flowering to harvesting. We give a hypothetical test to apply varieties that are tolerant of daily maximum temperatures at 29°C, 30°C, 31°C, 32°C, and 33°C, to be hopefully developed in the future, of which the threshold of hot days is replaced with the corresponding temperatures. To maintain the probability distribution of hot days as the current climate level, a new potato variety that is tolerant of 31°C should be developed in the plus 2-K climate ( Fig. 7c; Histograms of 32°C and 33°C are not shown.). Moreover, the plus 4-K climate requires a variety that is tolerant of 33°C if we desire the same risk level as the current climate (Fig. 7d).

Conclusion and discussion
We first confirmed that the current climatic conditions of Obihiro are suitable for potato production, based on an official record from 1984 to 2018. We then analyzed the effects of climate change on potato production in Hokkaido and Obihiro by using a multiple ensemble database, d4PDF, which enabled us to evaluate the phenomena with extremely low probability. Under the plus 2and 4-K climate, the risk of growth delay due to low temperatures is eliminated almost everywhere in Hokkaido. However, the risk of yield loss due to the increase in the number of hot days is unavoidable even if the planting date is shifted a month earlier.
On the other hand, switching to varieties that are tolerant to higher maximum temperatures will mitigate the risk of yield loss. Using late-maturing varieties allows for a longer growth period and potentially mitigate the warming effects. Even though heat-tolerant varieties have not yet developed, introducing Indian and Southeast Asian varieties to Hokkaido, or developing varieties that are incorporated into the genetic component of the breed, would be an effective adaptation strategy. Varietal conversion is associated with regional branding strategies in each region.
In this study, only temperature was used as a growth indicator. This index was used after examining the statistical consistency with current growth conditions and yields. However, other climatic variables such as solar radiation and carbon dioxide concentration also affect plant growth. Among them, solar radiation is quite important for potato production because it drives leaf photosynthesis and starch accumulation in tubers. The rise in an atmospheric CO 2 concentration should not be ignored for plants because it enhances photosynthetic carbon dioxide absorption and plant growth (Donohue et al. 2013;Frank et al. 2013). Extending the scope of this paper, one could estimate more realistic yield and starch quality changes, by including more climatic variables from d4PDF output.
The probabilistic discussion using a large number of meteorological data allowed us to examine more useful climate change adaptation in agriculture. Previous studies, such as Hijmans (2003) and Adavi et al (2018), have only been deterministic, discussing how yields will be reduced by global warming, or how they will be able to reduce the yield loss with adaptation. These deterministic discussions could not discuss the possibility of future climate extremes that almost seldom occur in the current climate. Using d4PDF allowed us to discuss probabilistic evaluations with the dataset including climate extremes. A more detailed discussion would be possible by a finer resolution dataset than currently available. Hoshino et al. (2020) have recently demonstrated that dynamical downscaling of d4PDF to 5 km spatial resolution and improved the reproducibility over Hokkaido. Because the yearto-year variation in yield is one of the most important concerns in agriculture, a deterministic prediction based on a single model may not be sufficient to discuss the impact of climate change on agriculture. Probabilistic data generated by using d4PDF should be a promising tool to address this issue.