Climate change increases North Korea’s hunger: implications for social resilience


 Adaption based on social resilience is proposed as effective measures to mitigate hunger and avoid disaster caused by climate change. But these have not been investigated comprehensively in climate-sensitive regions especially necessary-quantitative paths. North Korea (NK, undeveloped) and its neighbors (SK, South Korea, developed; China, developing) represent three economic levels that provide us with examples of how to examine climatic risk and quantify the contribution of social resilience to rice production. Our data-driven estimates show that climatic factors determined rice biomass changes in NK, while non-climatic factors dominated biomass changes in NK’s neighbors. If no action is taken, NK will face a higher climatic risk (with continuous high temperature heatwaves and precipitation extremes) by the 2080s with high emission scenario when rice biomass and production are expected to decrease by 20.2% and 14.4%, respectively, thereby potentially increasing hunger in NK. The contribution of social resilience to food production in the undeveloped region (15.2%) was far below the contribution observed in the developed and developing regions (83.0% and 86.1%, respectively). These findings highlight the importance of social resilience to mitigate the negative effects of climate change on food security and human hunger, and provide necessary-quantitative information.

social-economic assessments due to differences in climate vulnerability. Comparing economic 78 vulnerability based on social resilience among the three regions excepting climate risk is a valuable 79 approach, i.e., the ability to adapt to climate risk for food security resulting from climate and 80 economic vulnerability. Rice (Oryza sativa L.) is one of the most important foods in NK. Rice 81 composes more than 60% of the total grain production, and directly affects food security for NK in 82 terms of planting area and production 20 . Importantly, the adverse impacts of climate change on rice 83 systems is increasing 21 . 84 It is well-known that obtaining reliable statistics and survey data from NK is difficult due to 85 NK's politics and economics. Therefore, this study attempted to fully use remote sensing and climate 86 data with openly available statistical information to examine and assess climatic risk and food security 87 with interaction between climate and social-economic vulnerability for NK and its neighbors (Table   88  S1). In addition, the method presented in this study can be used in regions of the world that lack 89 official information to evaluate climatic risk and food security status. In this research, we need to 90 answer three key questions: (1) How has climate change (climate extremes) affected rice production 91 in NK? (2) What is the future projection of rice production loss? and (3) How have human activities 92 (adaptation based on social-economic resilience) exacerbated or ameliorated food deficits in NK and 93 its neighbors. Specifically, we focus on normal and extreme climate changes in NK over the past 18 94 years (2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016)(2017), and attribute rice biomass changes to climatic factors resulting from high-95 frequency climate extremes and increased vulnerability. Furthermore, we introduce 27 global climate 96 models (GCMs) under two future scenarios (SSP245 and SSP585 from CMIP6) to assess climate 97 changes and production losses in the future from a climatic risk perspective. Finally, the contribution 98 of social resilience based on five factors (population development, resource use, science and 99 education, economic development, and agricultural input) to rice production is explored 100 quantitatively by contrasting the differences between NK and its neighbors (SK and China). Further

104
When food famine occurs. The first challenge relative to analyzing food security is to determine the 105 time of the food crisis occurrence, i.e., when food famine occurred. To do this, we employed a general 106 method with food self-sufficiency rate and national-level statistics from FAO to identify the degree 107 of food self-sufficiency in NK and SK, and defined the years when the rice self-sufficiency rate was 108 less than 70% as famine years (see Methods). Over the period from 1984 to 2017, we obtained an 109 impressive result in which the rice self-sufficiency rate in NK was the highest in 1988 and then less than 70%). In contrast, SK's self-sufficiency rate has remained stable and close to 100% ( Figure   113 2a). From 1984 to 2017, the self-sufficiency rate for rice in NK was below 90% in fifty percent of the 114 years, but in SK the self-sufficiency rate was higher than 90% in all the years. We identified two 115 famine years in NK (2000 and2007), and used data from those two years to study the causes of the

128
The twelve climatic variables (including three average and nine extreme variables) were 129 incorporated in the analysis to determine climate attribution of famine years (Table S2). We then 130 conducted a multicollinearity analysis of all climate variables to exclude high collinearity for machine 131 learning modeling, i.e., variance inflation factor more than 10 (Table S3). Although the eLUE model 132 is driven by remote sensing and flux data, it introduces solar radiation into the simulation. We 133 therefore additionally analyzed the impact of solar radiation on attribution to climate. In the two 134 famine years (2000 and 2007), similar accuracy was found for the model with and without solar 135 radiation (Table S4). In the famine years, climate variables produced different levels of explanation 136 of GPP changes in NK, CH_1_2, and SK, with 73%, 43%, and 44%, respectively, explained in 2000, 137 and 65%, 44%, and 55% explained in 2007 (Figure 2b). This meant that climate variability was 138 responsible for nearly three quarters of rice GPP changes in NK, while approximately half of GPP 139 changes in CH_1_2 and SK in the famine years were due to climate change. 140 We next fully examined the spatial-temporal changes of each climatic factor by comparing the 141 baseline using anomalies to attribute the famine (see Methods). The gray bands in Figure 2c represent 142 the period from heading to tillering during the rice growing season (GS). This period is usually 143 sensitive to climate change, and especially sensitive to high temperatures that can cause plant death.  Figure 2c). In addition, a few long-term periods of rainfall hardly alleviated the high temperatures 152 and heat wave in local areas (Figure 2c). In terms of the spatial distribution of anomalies, the 153 frequency of abnormal increases for TXx, TR20, SU30, and FD0 (annual count of days when TN < 154 0℃) in 2000 accounted for 27%, 35%, 37%, and 54%, respectively, of the entire region ( Figure S2).

155
Specifically, the occurrence of extreme high temperatures increased abnormally in southwest NK, 156 and the anomalies of TXx, TR20, and SU30 were all more than three standard deviations, with the 157 anomalous occurrences covering the rice growing regions (Figure 2c). Significant increases for 158 precipitation and precipitation days were observed in the non-rice region of northeast NK, and this 159 also explains why precipitation did not alleviate the reduction in production caused by the high  (Figure 2c). Furthermore, the long-term and substantial 166 amount of precipitation produced conditions that made plants highly susceptible to crop root rot and 167 flood damage. TP (total precipitation during GS), R50 (annual count of days when R ≥ 50mm), and 168 R25 (annual count of days when R ≥ 25mm) in 2007 accounted for 87%, 72%, and 80%, respectively, 169 of the entire region ( Figure S2). This was similarly evidenced by the spatial distribution of rainfall 170 extremes that covered the west/southwest rice-growing region in 2007 (anomaly was more than three  (Table S5). Future climate would 176 see marked increases in temperature and precipitation in the vulnerable climatic region of NK.  Figure S3c, i, j). In general, no matter which climate scenario is 184 considered, the risk of high temperatures and extreme rainfall due to future warming will increase.

185
More specifically for NK, climatic variables explained 80% of the GPP changes observed from biomass, and the assessments of future rice losses are highly uncertain (Figure 2b). Therefore, the 196 future production losses in SK and CH_1_2 were not calculated in this study. However, it is worth 197 clarifying that robust results have shown that the rice production in CH_1_2 and SK was greater than 198 in NK in the context of past climate shocks, and was dominated by non-climatic factors.

199
Analysis of adaptability in NK, CH, and SK. Although rice production is generally subject to 200 natural-environmental change from the standpoint of a climatic risk framework, the adaptive capacity 201 at regional or national levels from social resilience is more determinant of rice production losses 23 .

202
Social resilience will be driven by population, economics, technology, and culture 12 . To quantify the 203 contribution of social resilience to rice production, we collected and used economic statistics from 204 FAO, World Bank, and an agricultural dataset of remote sensing that involved five factors: population 205 development, resource use, science and education, economic development, and agricultural inputs 206 ( Table 1) that together constituted social resilience 24 . The social resilience data was characterized as 207 one of two types to fully reveal the real-social situation: soft-adaptive and hard-adaptive 25 (Table 1). 208 Additionally, discontinuous-economic data for NK were interpolated using regression models to   have a greater need to increase the proportion of adolescents in the population to adjust for serious 250 aging 28 so that a larger agricultural labor force is available to increase rice production. Energy use 251 produced an increase in rice production for the developing country (CH), yet the opposite result was 252 observed for the developed country (SK) owing to the capacity to import resources because of 253 sufficient capital 29 . In general, social resilience likely results in various impacts of rice production 254 among NK, CH, and SK.

255
The non-linear response of rice production to agricultural inputs in the three sub-regions is 256 shown in Figure S8. The response curves for nitrogen, phosphorus, and irrigation showed an overall 257 increasing trend in SK and CH. However, in NK, the production responses of the three agricultural 258 practice inputs were seen to be expressed as hump or concave curves ( Figure S8). Based on the 259 random forest model and its out of bag error, agricultural inputs (nitrogen, phosphorus, and irrigation) 260 explained -4.8%, 51%, and 77% of rice production changes for NK, SK, and CH, respectively. The 261 explanatory degree for NK was negative, meaning that the model was overfitted, and the three 262 agricultural practices cannot support the increase in rice production in the current situation.  However, if the focus is only on increasing crop yields, farmers will still suffer severe production 290 losses due to their low perception of risk concerning the effects of extreme weather disturbances.

291
Learning reflects the ability to produce, absorb, and transform new information about climate risk, 292 adaptation, and coping with uncertainty. This ability to learn and apply new scientific information is 293 a mitigation mechanism applicable to climate change 39 . Capital investment depends on economic 294 development, and is a more direct approach. These investments include building early warning 295 systems and climate insurance, resource and energy use, international trade, and reducing poverty.

296
Insurance is a tool to mitigate climatic risk and restore livelihoods, especially in response to climate 297 extremes 40 , but if the insurance structure is not correct, it has an inhibiting effect on risk reduction 41 .

298
For regions that cannot compensate for losses through trade, these years of low productivity can still 299 be devastating. Poor countries are more vulnerable and less resilient to climate change. When the 300 poor are struck, they have less support from friends, family, and the financial system. Policies meant 301 to reduce poverty under similar climatic risk conditions can also reduce the adverse impacts of climate 302 change 42,43 . The interaction of social-economic factors from many aspects is the key to decreasing 303 economic vulnerability and increasing social resilience to mitigate climate vulnerability. In this study,   Topography may be one of the factors that limit future rice production potential in NK due to 325 the cultivation of poor-quality land that is not suitable for planting rice. NK contains a wide range of 326 slopes compared with CH_1_2, and SK. The steeper slopes (slope >10°) accounted for 64% of the 327 country's land, and the gentler slopes (5°-10°) accounted for 15% of the country's land ( Figure S9b).

328
In addition, land with DEM values greater than 500 m in NK accounted for 47% of the entire land 329 area, while those higher elevation areas in CH_1_2 and SK accounted for 23% and 16%, respectively, found that assessing the impact of climate change on yield depended on soil type because soil 345 characteristics and moisture buffer or amplify climatic impacts 3 . This study also did not consider the 346 effects of rice genotype due to a lack of cultivar data, and therefore there may be potential uncertainty 347 regarding regional production differences 52 . For GPP loss estimation, we did not introduce CO2  In the future, we plan to engage in further in-depth study of multiple climate indexes, particularly 357 extreme climate variables, combined with process-based ecological models to explain the 358 mechanisms of crop growth. However, achieving this next step will require a great deal of data 359 support.

360
Climate change is expected to increase the frequency and intensity of climate extremes that will 361 likely reduce global food production and induce famines. An accurate assessment of food insecurity, shocks in the future. Using this method, we can clarify future food risks and provide quantifiable 368 pathways and goals for efforts that will contribute to improved national risk awareness, make up for 369 weaknesses in food programs, and guide adjustments to food strategies, and thereby provide further 370 guidance for optimizing social-economic policies.    Figure 4: The contribution of vulnerable indexes for social resilience to rice production variability. 512 Twelve indexes of social resilience are from five factors, i.e., population development, resource use, 513 science and education, economic development, and agricultural input (Table 1). a, The red and green 514 bars represent significance tests for p < 0.05 and p > 0.05, respectively. The importance indicates 515 explanation degree of the non-linear model for contribution of social resilience to rice production in the 516 three regions. b, The non-linear response of rice production to standardized variables of social resilience. 517 The black lines are smoothed representations of the response, with fitted values (model predictions) for 518 the calibration data. The trend of the line, rather than the actual values, describes the nature of the 519 dependence of rice production on the social resilience. The green and orange dash lines in the partial 520 dependence plots represent rice production from the baseline and SSP585 scenarios, respectively. NK, 521 SK, and CH represent North Korea, South Korea, and China. 522

523
Data sources. We used daily reanalysis data to analyze climate variables over the years of the study.

524
The daily reanalysis data were obtained from the European Center for Medium-Range Weather 525 Forecasts (ECMWF) gridded dataset at 0.1-degree resolution, and included daily 2-m temperature 526 (24-hour maximum, minimum, and mean temperature), precipitation flux, and solar radiation flux 527 from 1979 to 2018. See Table S1 for more details. We selected the ECMWF's ERA-5 dataset for two 528 reasons: 1) ground data are not readily available; 2) by using the global dataset, we can apply our 529 methods to other areas with limited data. Our primary analysis focused on calculating normal and 530 extreme climate indexes for 2000-2017 (Table S2), and statistical downscaling of data from 1979-531 2018 was conducted to project future climate change using ERA-5.

532
For remote sensing information, we employed the MODIS gridded mosaic including 8d_surface 533 reflectance, 8d_leaf area index, and 8d_ gross primary productivity accessed from Google Earth  We used the daily net ecosystem exchange and heterotrophic respiration from the flux towers to 558 calculate the gross ecosystem CO2 exchange. We refer to gross ecosystem CO2 exchange as gross 559 ecosystem primary productivity.

560
The phenological periods of the main crops were obtained from the China Meteorological Data  Table S3 of Zhou et al., 2016 55 . The reason for choosing these sites was based on the similar 564 climate zone in the study area (Fig 1) and the lack of real data in North Korea. All data was strictly 565 examined to meet the standards for further analysis, including cross-validation and comparison with 566 existing data.

567
The statistical data were obtained from the Food and Agriculture Organization (FAO) of the School enrollment, Patent applications, and Net ODA received per capita) (Table 1) were from the "Official data" and "Aggregate" categories to ensure high quality.

582
For further analysis of non-climate attribution, we considered the effect of irrigation on rice 583 growth. We used water consumption coefficient (WCE) of rice paddy to replace irrigation because of 584 the lack of irrigation data over the study areas, and because of the large uncertainty in irrigation timing, 585 amounts, and methods. The irrigation period for these maps was from 2001-2017, and the ratio of 586 evapotranspiration (from MOD16A2, Table S1) to precipitation was used to calculate the WCE.

587
Furthermore, we averaged WCE of rice paddy in the study areas.

589
Rice paddy mapping and estimating biomass. Rice paddy maps were produced on the GEE cloud 590 platform based on vegetation index (see formula 1-3) and rice phenophase (Table S6)  We averaged daily GPP (GPPEC, g C m-2 d -1 ) and daily (MJ m -2 d -1 ) from two EC towers sufficiency is generally considered to be the degree to which a country can meet its own food needs 632 from domestic production". Generally, FSS is closely related to food security.

633
A more pragmatic understanding of FSS is that domestic food production equals or exceeds 100% 634 of a country's food consumption 65 . This concept can be reflected by the self-sufficiency ratio (SSR) 66 : where SSR is the self-sufficiency ratio, P represents food production, and D represents food demand.

637
Food demand can be further refined as food production, food exports, food imports, and fluctuations 638 in domestic food storage 67 . Therefore, the clear definition of food demand is: where I is food imports, E is food exports, ∆ represents annual change in food storage.

641
However, SSR focuses on key crops in order to give an approximate value for a country's grain 642 self-sufficiency 65 . In NK and SK, rice is the dominant food produced and affects food security 20 . We 643 assumed that the annual food storage remains constant due to difficult-to-obtain inventory 644 fluctuations for a country. Then the rice self-sufficiency rate (RSSR) formula is: to analyze GPP changes controlled by climatic factors (Table S7). The random forest regression model  (Table S7). 681 We further used function importance that was quantified as the Gini index to compute the variable 682 importance. Compared with linear regression, RF explained the non-linear response of climate 683 variables and unraveled the influence of related variables.

684
In order to attribute climatic factors in famine years (that is, to identify climate anomalies), we 685 initially calculated normal and extreme meteorological factors for 2000-2017 (for more details, see 686   Table S2). We then calculated the mean value in non-famine years as the baseline. The ratio of the 687 difference between baseline and famine years to the standard deviation of baseline was used to 688 determine the climate anomaly. Therefore, if the anomaly was >1 or < -1, we considered that the  according to the climatic index as described above for Table S2. We then assessed climate change and . is the number of years and A is the pixel area.

723
The non-linear model (RF model, as described above for GPP changes analyzed by the machine 724 learning model) was then established through historical gridded climate data, and GPP was estimated 725 by the eLUE model to assess rice GPP in the future. Future production was projected by using mean 726 harvest index and predicted GPP: where is future production. ∑ 2 represents the sum of each predicted GPP. Finally, we used 729 future and historical GPP and production to calculate relative changes as the losses.

731
The contribution of social resilience to rice production. We interpolated missing statistics from shows the geographic boundaries of the study area. BSk, Cwa, Cfa, Dwa, Dwb, Dwc, Dfa, and Dfb are de ned as arid and cold steppe, temperate regions with dry winter and hot summer, temperate regions with hot summer and without dry season, cold regions with dry winter and hot summer, cold regions with dry winter and warm summer, cold regions with dry winter and cold summer, cold regions with hot summer and without dry season, and cold regions with warm summer and without dry season19, respectively. NK, SK, CH_1, and CH_2 are North Korea, South Korea, Liaoning province of China, and Jilin province of China. The World Bank de nes North Korea, South Korea, and China as low, high, and uppermiddle income countries/regions, respectively, based on income levels.

Figure 2
Identi cation of famine years in North Korea and attribution of production losses. a shows rice selfsu ciency every year calculated from import, export, and production quantity data from FAO. The dash line is the 70% self-su ciency rate (that  Projected rice production losses under four future scenarios. a, Observed vs. predicted rice biomass from 2000 to 2017 (baseline period). The blue and red points represent the calibration and validation data sets, respectively. b, Relative importance of climatic variables from non-linear modeling (the meaning of the variable abbreviations is provided in Table S2. c, Box plots of projected rice biomass (top panel) and production (bottom panel) losses under four different future scenarios based on nonlinear modeling. Box boundaries indicate the 25th and 75th percentiles across 27 GCMs, and whiskers below and above the box indicate the 10th and 90th percentiles, respectively. The black lines within each box indicate the multi-model median. The contribution of vulnerable indexes for social resilience to rice production variability. Twelve indexes of social resilience are from ve factors, i.e., population development, resource use, science and education, economic development, and agricultural input (Table 1). a, The red and green bars represent signi cance tests for p < 0.05 and p > 0.05, respectively. The importance indicates explanation degree of the non-linear model for contribution of social resilience to rice production in the three regions. b, The non-linear response of rice production to standardized variables of social resilience. The black lines are smoothed representations of the response, with tted values (model predictions) for the calibration data.
The trend of the line, rather than the actual values, describes the nature of the dependence of rice production on the social resilience. The green and orange dash lines in the partial dependence plots represent rice production from the baseline and SSP585 scenarios, respectively. NK, SK, and CH represent North Korea, South Korea, and China.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download. FinalSupplementaryinformation.pdf