Global Spatial Distributions of and Trends in Rice Exposure to High Temperature

Abstract

Due to the effects of global warming, extreme temperature events are posing a great threat to crop yields, especially to temperature-sensitive crops such as rice. In the context of disaster risk theory, exposure is central to disaster prevention and reduction. Thus, a comprehensive analysis of crop exposure is essential to better reduce disaster effects. By combining the maximum entropy model (MaxEnt) and a multiple-criteria decision analysis (MCDA), this paper analyzed the global distribution and change in rice exposure to high temperature. The results showed the future states of rice after exposure to high temperatures. Our results are: (1) the areas of potential rice distribution zones decreased within the representative concentration pathway (RCP) scenarios RCP2.6 to RCP8.5 in MaxEnt, where the long-term (2061–2080) decreases are greater than those seen in the medium term (2041–2060). (2) In the future, the number of high temperature hazards in potential rice distribution areas increased. In the RCP8.5 scenario, the intensities of global high temperature hazards on rice were reduced because the total area of potential rice distribution zones decreased. (3) Through the view of barycenter shift, the barycenter of the global potential rice and high temperature hazard distributions showed a trend of backward motion, which meant the global rice exposure to high temperature was in a downward trend. With the background of global change, this paper has great significance for the mitigation of high temperature risk in rice and its effect on the potential security of future global rice production. Future research is warranted to concentrate on discussing more socioeconomic factors and increasing rice exposure change from the temporal vision.

1. Introduction

The latest Intergovernmental Panel on Climate Change (IPCC) special report estimates that if the current warming rate continues, the world will reach a human-induced global warming of 1.5 °C by approximately 2040 [1]. This increase in temperature will increase the frequency of extreme weather events, especially extreme high temperature events. Rice is among the world’s most important food crops, and it is a primary source of food for over 50% of the world’s population [2,3]. Frequent and extreme high temperature events will increase the risk of rice yield losses [4], which is a serious threat to the global food supply. Furthermore, according to the latest IPCC report, temperatures in the mid-latitudes (the most important crop producing areas) may rise by approximately 3 °C [5]. The sensitivity of the mid-latitudes means any temperature changes will significantly affect the crops growing therein. As a temperature-sensitive crop [6], rice will clearly be affected by high temperature disasters. Therefore, studies on the changes in rice distribution resulting from high temperature exposure, and changes under different representative concentration pathway (RCP) scenarios, can directly reflect the severity of high temperature hazards that rice may face in the future. In addition, the study of this topic can help prevent and reduce the risk of rice yield losses, which is of great significance to world food security [7].

Current research on disaster exposure are focused on quantitative assessments of exposure, including statistics related to the spatial distributions of hazard-bearing bodies [8,9], estimations of hazard-factor intensities [10,11,12,13,14], etc. Wu used the crop choice decision model to calculate the sown areas of major crops based on the economic values, and the changes in crop habitat factors caused by climate change were not taken into consideration [15]. Liu used the maximum entropy model (MaxEnt) to simulate rice suitability zones during the historical period (1980–2010), and achieved the classification of suitability grade according to the appearance frequency from the model results [16]. Some other scholars used a layer overlay analysis of geographic information system(GIS), and took the habitat index of crop distribution as layers to obtain crop suitability zones by multiple constraints and then they classified suitability degree though the accuracy of habitat index [17,18,19,20,21,22,23,24]. These studies provide a useful means for the calculation of rice suitability zones. On the other hand, Shi investigated the spatio-temporal variation of rice post-heading heat stress in South China during the period 1981–2010 [25]. Zhang conducted a comprehensive analysis on maize exposure, vulnerability, and adaptation to extreme temperature to understand the effects on global maize production, especially in major production countries [18]. Wang examined the changes in the area exposed to heat stress and historical movements of the geographical centroid of rice exposure to heat stress (EHS) during the period 1980–2015 across southern China [17]. These studies provide a scientific and efficiency measure of rice high temperature hazard and its variation. Current research usually focuses on a single factor, hazard-bearing body or hazard. Hence, analyses of the comprehensive characteristics of many factors related to exposure are few. Moreover, most study areas are located in China and southeast Asia, and most studies usually consider only a single temporal period [23,24,25], which makes it difficult to fully reflect the trends and regional differences in crop exposure. Therefore, the research on rice exposure to high temperature needs to start with two aspects of disaster-bearing bodies and hazard factors, by taking into account the impact of climate and socioeconomic factors on the rice planting range. According to the features of high temperature hazards in multiple scenarios and multiple periods, we can obtain the overall spatial patterns of rice exposure to high temperature, then reveal the hotspots and analyze how the exposure features help in finding trend rules.

Moreover, the Representative Concentration Pathway (RCP), which is commonly used in future crop yield simulation research, is a greenhouse gas concentration trajectory adopted by the IPCC for its fifth Assessment Report (AR5) in 2014. Dias identified the yield and growth changes in rice under the global climate change scenario RCP 8.5 [26]; Dar focused on simulating the effect of climate change on crop yield with the Decision Support System for Agrotechnology Transfer (DSSAT) v 4.6.1 under RCP4.5 scenario [27]; Chun applied a multi-scale crop modeling approach to assess the impacts of climate change on future rice yields in southeast Asia under the RCP 4.5 and RCP 8.5 scenarios [28]; Van Oort calculated the impacts of climate change on rice production in Africa under four kinds of RCP scenarios [29]; Kim predicted potential epidemics of rice leaf blast and sheath blight in South Korea under the RCP 4.5 and RCP 8.5 scenarios [30]; Zhang investigated the spatio-temporal change in extreme temperature stress across China under the RCP (2.6, 4.5, 6.0, and 8.5) scenarios [31]. Therefore, for future rice research, RCP provides a reliable and easy way for calculations, which is the reason why this paper chooses RCP as the future scenarios.

In order to study the changes in rice habitats in different future scenarios, with the background of climate change, in this study, our research was devoted to the assessment of global rice exposure to high temperature in medium-term (2041–2060) and long-term (2061–2080) periods. The results are based on RCP scenarios and share socioeconomic pathway (SSP) scenario data [32], and we used the MaxEnt model and multiple-criteria decision analysis (MCDA) methods to set up cost-effective and spatially targeted protection measures in order to safeguard world food supplies in the future. (1) In Section 3.1, we analyze the characteristics of area changes and barycenter shift of potential rice distribution under different scenarios. (2) In Section 3.2, we analyze the characteristics of intensity change and barycenter shifts of high temperature hazard factors under different scenarios. (3) In Section 3.3, we analyze rice exposure to high temperature by comparing the barycenter shifts of potential rice and high temperature hazard distributions. Throughout the study, we include references on future rice exposure for governments, so that they can rationally allocate resources and formulate reasonable farming policies to mitigate the high temperature risk for rice and its effect on the potential security of future global rice production.

2. Materials and Methods

2.1. Basic Idea and Research Framework

From the view of disaster risk theory [33], disaster risks can be divided into three categories: hazard (H), vulnerability (V) and exposure (E). Among these three types of disaster risk, exposure directly determines the effect of a disaster, which can, in some scenarios, reverse the results of the risk assessment. Generally, high degree exposure means a large disaster risk [34]. Therefore, we can estimate the potential risk for a specific disaster in a region by analyzing the extent of exposure therein.

In this paper, we presented and analyzed exposure maps of crops from two aspects: hazard-bearing bodies and hazard factors. For a hazard-bearing body, based on the MaxEnt model, we screened the rice biotope data to map potential rice distribution, and classified the suitability criteria for each considered region. The hazard factor analyses were based on four different scenario data (RCP2.6 and SSP1, RCP4.5 and SSP2, RCP6.0 and SSP4, and RCP8.5 and SSP5) during three periods (Baseline: 1971–2000, medium term: 2041–2060, and long term: 2061–2080). We calculated the intensities of the high temperature hazards in different periods and overlaid them with potential rice distribution to obtain the high temperature exposure experienced by rice under the different scenarios.

We also analyzed crop exposure from two aspects: the first is a change in potential rice distribution, which is described by the change rate of an area and the shift of its barycenter. The second considers changes in high temperature hazard intensity, which is denoted by the accumulative extreme high temperature in a year (SDD). Through shifts in potential rice distribution areas and high temperature hazard intensities, we obtained changing global rice exposure during different periods, thus describing the potential risk of high temperature on rice globally. The specific process is as shown in Figure 1.

2.2. Data Collection

The data were divided into three categories: disaster-formative environment data, hazard-bearing body data, and scenario data. The disaster-formative environment data include global digital elevation model (DEM), global-slope and global-soil parameter data [35]. The hazard-bearing body data included actual rice cultivated area data from 2000 and 2005, which were used to test the accuracy of potential rice distribution estimates. The scenario data included global meteorological data and global land use data.

In order to mitigate errors among the different climate model datasets, we used three sets of CMIP5 (Coupled Model Inter-comparison Project phase 5) meteorological data from the Inter-Sectoral Impact Model Intercomparison Project (ISI-MIP) [36,37,38]: IPSL-CM5A-LR (IPSL(Institute Pierre Simon Laplace) Earth System Model for the 5th IPCC report with low resolution) [39], MIROC-ESM-CHEM (Earth System Model for Interdisciplinary Research on Climate, atmospheric chemistry coupled version) [40] and NorESM1-M (Norwegian Earth System with intermediate resolution) [41]. The meteorological data contained the maximum temperature, minimum temperature, solar radiation intensity, wind speed, and relative humidity. The bioclimatic data contained the monthly averaged data of 19 parameters. In this paper, we considered the metrological data and bioclimatic data as rice habitat suitability factors to rate the rice habitat suitability. Moreover, we used the land use data to build socioeconomic scenarios, and we took the annual C3 crop distribution zones as te potential rice distribution zones.

We normalized the spatial resolution of all the data to 0.5° × 0.5°.

Table 1. Datasets.

Data Category	Data Name	Temporal Resolution	Spatial Resolution	Source
Disaster-Formative Environment Theory	Global Digital Elevation Model Data GMTED2010 (DEM)	2010	1 km × 1 km	United States Geological Survey (USGS) https://topotools.cr.usgs.gov/gmted_viewer/
	Global-Slope Data	2006	10 km × 10 km	International Institute for Applied Systems Analysis Global Agro ecological Zones (GAEZ): http://www.gaez.iiasa.ac.at
	Global-Soil Parameter Data	2012	1 km × 1 km	International Soil Reference and Information Centre (ISRIC): http://www.isric.org
Hazard-Bearing Body	Rice Cultivation Range Data	2000 or 2005	0.5° × 0.5°	①Harvested Area and Yield for 175 Crops around the year 2000 (EARTHSTAT 2000): http://www.earthstat.org/harvested-area-yield-175-crops/ ②Global monthly irrigated and rainfed crop areas around the year 2000 (MIRCA 2000): http://www.unifrankfurt.de/45218031/data_download ③Spatial Production Allocation Model (SPAM) 2000/2005: http://mapspam.info/maps/
Scenarios	IPSL-CM5A-LR	1971~2099	0.5° × 0.5°	LOCEAN/IPSL: https://www.locean-ipsl.upmc.fr/smos/
	MIROC-ESM-CHEM	1971~2099	0.5° × 0.5°	Japan Agency for Marine-Earth Science and Technology, Atmosphere and Ocean Research Institute (The University of Tokyo), and National Institute for Environmental Studies: http://adsabs.harvard.edu/abs/2011GMD.....4..845W
	NorESM1-M	1971~2099	0.5° × 0.5°	Norwegian Earth System: https://portal.enes.org/models/earthsystem-models/ncc/noresm
	Land use Data (LUH2)	1971~2100	0.25° × 0.25°	Land-Use Harmonization: http://luh.umd.edu/data.shtml
	Bioclimatic Variables	1971~2000 2041~2060 2061~2080	1 km × 1 km	Global Climate Data (WorldClim): http://worldclim.org/CMIP5v1

shows the contents of the data used in this paper.

2.3. Methods

2.3.1. Potential Rice Distribution Estimation

The MaxEnt model [42,43] and MCDA were used to evaluate future rice potential distributions [22,23,44]. First, this paper inputted the habitat parameters, which included the global DEM data, global slope data, global soil data (20 parameters) and bioclimatic data (19 parameters). Then, this paper screened the 13 parameters to consider only those with contribution rates that were greater than 95% (area under curve (AUC) values > 0.75) as the necessary habitat variables for potential rice distribution estimates.

Second, according to the quantiles of the habitat variables, the rice habitat suitability was ranked into five grades from I to V, where I was the lowest suitability, and V was the highest suitability.

Table 2. Suitability classification rules.

Grade	Variable Valuation Rules
I	>99 quantiles, <1 quantile
II	97~99 quantiles, 1~3 quantiles
III	95~97 quantiles, 3~5 quantiles
IV	90~95 quantiles, 5~10 quantiles
V	<90 quantiles, >10 quantiles

shows the suitability grade rank definitions.

Third, this paper overlaid a habitat variables layer, and eliminated areas without C3 crop distributions to obtain potential rice distribution zones. Moreover, the regions of grade II to V were regarded as suitable zones for rice growth and the region of grade I was regarded as a non-suitable zone.

2.3.2. High Temperature Hazard Intensity Calculation

According to previous research results and generally accepted standards [45,46,47,48,49,50,51], this paper took the SDD as the high temperature hazard intensity index. SDD represents the annual accumulative extreme high temperature, where a high temperature event is defined as the daily average temperature Tas ≥ 30 °C. The calculation process can be divided into two steps as follows: first, this paper calculated the daily accumulative extreme high temperature according the high temperature standard (Tas ≥ 30 °C). Secondly, this paper summed the daily accumulative extreme high temperature to obtain the SDD, as shown in Equations (1) and (2):

$$D{D}_i=\{\begin{array}{ll}{T}_i-T_{base}& for T_i\ge T_{base}\\ 0& for T_i\le T_{base}\end{array}$$

(1)

$$SDD={\displaystyle \sum }_{i=1}^{n}D{D}_i$$

(2)

where T_i is the average temperature on the ith day of the year, T_base is the threshold temperature of the high temperature event (30 °C in this study) and n is the frequency of high temperature events during the year.

2.3.3. Barycenter Shift of Potential Rice and High Temperature Hazard Distributions

This paper used the barycenter shift [52,53] to measure and analyze the spatial distributions of the rice and high temperature hazards during the different periods. The barycenter of global rice habitat suitability was used to characterize the rice distribution, where the barycenter of an SDD was used to characterize the global high temperature hazard distribution, as respectively shown in Equations (3) and (4).

(1) Barycenter shift of rice habitat suitability calculation:

$$\{\begin{array}{c}LO{N}_{s, t}=\frac{{{\displaystyle \sum }}_{i=1}^n\left(S{D}_{i,t}\times lo{n}_{i,t}\right)}{{{\displaystyle \sum }}_{i=1}^{n}S{D}_{i,t}}\\ LA{T}_{s, t}=\frac{{{\displaystyle \sum }}_{i=1}^n\left(S{D}_{i,t}\times la{t}_{i,t}\right)}{{{\displaystyle \sum }}_{i=1}^{n}S{D}_{i,t}}\end{array}$$

(3)

where LONs and LATs are the longitude and latitude of the barycenter, respectively, SD_i,t is the suitability of the ith grid with a suitability grade of t, lon_i,t and lat_i,t are the respective longitude and latitude of the ith grid with a suitability grade of t, and n is the number of the suitability grade t grids.

(2) Barycenter shift of the SDD calculation:

$$\{\begin{array}{c}LO{N}_h=\frac{{{\displaystyle \sum }}_{i=1}^n\left(SD{D}_i\times lo{n}_i\right)}{{{\displaystyle \sum }}_{i=1}^{n}SD{D}_i}\\ LA{T}_h=\frac{{{\displaystyle \sum }}_{i=1}^n\left(SD{D}_i\times la{t}_i\right)}{{{\displaystyle \sum }}_{i=1}^{n}SD{D}_i}\end{array}$$

(4)

where LON_h and LAT_h are the longitude and latitude of the SDD, respectively, SDD_i is the SDD value of the ith grid, lon_i and lat_i are the longitude and latitude of the ith grid, respectively, and n is the number of grids.

Moreover, we calculated the potential rice distribution and SDD change rates to quantify the changes in high temperature exposure experienced by rice, as shown in Equations (5) and (6), respectively:

$$Dif{f}_{a,t,i,j}=\frac{Are{a}_{j,t}-Are{a}_{i,t}}{Are{a}_{i,t}}\times 100\%$$

(5)

where Diff_a,t,i,j is the change rate of the area of the t-grade zone from the ith period to the jth period, and Area_i,t is the area of the t-grade zone in the ith period.

$$Dif{f}_{h,k,i,j}=\frac{SD{D}_{j,k}-SD{D}_{i,k}}{SD{D}_{i,k}}\times 100\%$$

(6)

where Diff_h,k,i,j is the SDD change rate of the kth grid from the ith period to the jth period, and SDD_i,k is the SDD value of the kth grid.

2.3.4. Estimation of High Temperature Exposure

Using the previously calculated barycenter shift of a hazard-bearing body and the barycenter shift of a hazard, this paper expressed the characteristics of high temperature exposure and analyzed the potential risks to rice in terms of their direction and distance. This paper then compared the moving direction and distance of the barycenter in different periods, and qualitatively divided the high temperature exposure into five grades, as shown in Figure 2.

3. Results

We calculated three sets of climate data to obtain indicators maps under four scenarios (RCP2.6 and SSP1, RCP4.5 and SSP2, RCP6.0 and SSP4, and RCP8.5 and SSP4) in the medium-term and long-term future periods. Due to the similarities of our obtained results, only the results of the IPSL-CM5A-LR model are shown in the text. The results of the MIROC-ESM-CHEM and NorESM1-M models are detailed in the appendix.

3.1. Changes in the Area and Barycenter of Rice Suitability Zones

According to the estimation method provided in Section 2.3.1, we mapped the potential rice distribution, as shown in Figure A1, Figure A2 and Figure A3. On this basis, we used the continent as the basic unit to calculate the area of rice habitat suitability zones during different periods (Figure 3), and the barycenter shift from the baseline to the long-term period (Figure 4). The specific results are as follows:

• Compared with the baseline period, the areas of the rice habitat suitability zones in Europe, Asia, Africa and South America decreased over all periods. Among them, the suitable areas of all grades in Europe showed a continuous downward trend, which decreased by 40% to 65%. The areas of the grade II and V zones in Asia decreased by over 45%, while the areas of the grade IV zones first decreased and then increased. The areas of the grade IV and V zones in Africa decreased by 30% to 40%, respectively, while the areas of the grade II zone increased significantly. The areas of the grade III to V zones in South America showed continuously decreasing trends, which decreased by 30% to 50%, and the decrease was faster from RCP2.6 to RCP8.5.
• Compared with the baseline period, the areas of the rice habitat suitability zones in North America and Oceania increased. Between them, the areas of the grade IV and V zones in North America increased significantly, by 40% to 60%, while the areas of the grade II zones increased by over 50% in Oceania. However, under the RCP8.5 and SSP5 scenarios, the areas of the rice moderate suitability zones decreased by over 60%.

In terms of the direction of the barycenter shift, compared with the baseline period, we obtained the following results:

• Under the different scenarios, the rice habitat suitability zones in Africa, Asia, North America, South America and Oceania all moved to higher latitudes. The barycenters of the higher-grade suitability zones (grade IV and V zones) and the lower-grade suitability zones (grade II and III zones) in Africa moved to the southeast. The barycenters of the higher-grade suitability zones in Asia moved to the northeast, and the barycenters of the lower-grade suitability zones moved to the northwest. The barycenters of the higher-grade and the lower-grade suitability zones in South America moved to the south. The barycenters of the higher-grade suitability zones in Oceania moved to the southeast, while those of the lower-grade suitability zones moved to the north.
• Under different scenarios, the rice habitat suitability zones of all grades in Europe moved to lower latitudes, where the barycenters of both the higher-grade (IV and V) and lower-grade (II and III) suitable zones moved to the southwest.

In terms of the distance of the barycenter shift, compared with the baseline period, we obtained the following results:

• For the higher-grade suitability zones, Europe had the largest range of movement, up to approximately 10°, while Oceania had the smallest range of movement of only approximately 1°. The ranges of movement for the other continents were approximately 4° to 6°.
• For lower-grade suitability zones, South America had the largest movement range, up to approximately 7°, while the ranges of movement for the other continents were approximately 3° to 5°.

In this paper, we argue against the established view where the amplitude of a temperature increase is proportional to the barycenter northward-moving distance of a rice habitat suitability zone. Instead, we found that for all continents except Europe, the barycenter of rice suitability zones shifted the most under the RCP8.5 scenario. However, under the RCP6.0 scenario, the shifts in the barycenters of the rice habitat suitability zones were usually smaller than those under the RCP4.5 scenario, and even in Oceania, they were lower than those in the RCP2.6 scenario. Therefore, we found it reasonable to include a socioeconomic factor into the analysis.

3.2. Spatial Changes in the High temperature Hazard Distributions

(1) Spatial changes in the high temperature hazard distributions

Here, we used the SDD (T_as ≥ 30 °C) as a high temperature hazard intensity index and, through our calculations, we obtained the spatial distribution of SDD in the baseline period (Figure 5), and the medium-term and long-term periods (Figure 6) in potential rice distribution areas.

In terms of the high temperature hazard distributions in the baseline period, high-value regions were concentrated in the Indian subcontinent, Asia Minor, Sudanian Savanna, and the northern plains of the Mexican Plateau. The SDD values were generally above 100, and the SDDs in most parts of the Indian subcontinent were greater than 300. The SDDs in the southwestern parts of the Great Dividing Range, the western side of the La Plata plains, Indo-China, and the southern and eastern China regions were relatively high, generally above 10–50. The remaining regions such as Europe, north China and northeast China all contained low-value regions. Overall, Asia had the highest exposure, followed by Africa, Oceania, North America and South America, and Europe had the lowest amount of high temperature exposure.

Next, for the high temperature hazard distribution on the different continents during the medium-term and long-term periods we found: (1) the high-value regions in Africa were mainly concentrated in the Sudanian Savanna, but the high-value regions gradually decreased in the area from the RCP2.6 to RCP8.5 scenarios. In particular, under the RCP8.5 and SSP5 scenarios, the high-value regions moved to southern Africa. (2) The high-value regions in Asia were mainly concentrated in southern China, Asia Minor and the Indian subcontinent. Among them, the high-value regions of the Indian subcontinent shrunk and migrated to the north under the RCP8.5 scenario. (3) The high-value regions in North America were mainly located in the northern plains of the Mexican Plateau, and the values were high in the south and low in the north. (4) The high-value regions in South America were located in the southern parts of the Brazilian Plateau and the La Plata plains, where the region gradually expanded as the RCP2.6 to RCP8.5 scenarios were considered. (5) The high-value regions in Oceania ere located on the southwestern side of the Great Dividing Range. (6) The SDD values were low in Europe, but under the RCP8.5 and SSP5 scenarios, the Mediterranean coastal transitioned into a high-value region where the SDDs were generally above 50 and even above 100 in the north. Overall, the medium-term and long-term high temperature hazard distributions are generally consistent with those of the baseline period. In North America, South America and Europe, the intensities of the high temperature hazards increased. In Asia and Africa, the intensities of the high temperature hazards decreased.

(2) Changes in the high temperature hazards

By comparing the SDD values of the baseline period with those of the medium-term and long-term periods, and those of the medium-term period with those of the long-term period, the change rates of the SDD values under different scenarios (Figure 7) were calculated to characterize the change in the intensities of the high temperature hazards.

Except for Asia Minor, the SDD values of most regions showed a continuously increasing trend from the baseline period to the long-term period. Under the different scenarios, the high-value regions were located in Chinese monsoon regions, southeastern United States, south-central South America, southeastern Australia and the Iberian Peninsula. However, the SDD values on the Indian subcontinent and the Sudanian Savanna changed only rarely.

The regional high temperature exposure also changed under the different scenarios. In central North America, southern Africa, northeast and north China, Mediterranean coastal regions, and the east coast of South America all changed from non-exposed regions to exposed regions, while the Indian subcontinent, Indo-China, the Sudanian Savanna, the Caspian Depression, the Black Sea lowlands, the Danube River Basin, and the West Indies all changed from exposed regions to non-exposed regions.

(3) Barycenter shift of high temperature hazards to rice

Using the equations provided in Section 2.3.3, we calculated the positions of each continent’s SDD barycenter in the different periods and scenarios to characterize the barycenter shift of the SDDs from the baseline period to the long-term period, as shown in Figure 8.

In terms of the barycenter shift directions, compared with the baseline period, the SDD barycenters in Africa, North America and Oceania moved to the southeast, the SDD barycenter of Asia moved to the east, the SDD barycenter of South America moved to the south, and the SDD barycenter of Europe moved to the northeast. In terms of the barycenter shift ranges, compared with the baseline period, Europe had the largest range of SDD barycenter movement, with an average distance of 4° to 6°. Oceania had the smallest range of SDD barycenter movement, with an average distance lower than 1°. The rest of the continents had average distances of 2° to 4°. Except for Europe, the range of the barycenter shift of high temperature hazards on other continents enlarged from RCP2.6 to RCP8.5.

3.3. Comprehensive Analysis of Rice Exposure to High Temperature

According to the method presented in Section 2.3.4, the severity level of future rice exposure to high temperature was obtained (

Table 3. The degree of change in high temperature exposure in rice.

Continents	SYD	RCP2.6 and SSP1	RCP4.5 and SSP2	RCP6.0 and SSP4	RCP8.5 and SSP5	Avg
		B-M-L	B-M-L	B-M-L	B-M-L	B-M-L
AF	High	III	III	III	III	III
	Low	I	II	II	II	II
AS	High	II	II	III	I	I
	Low	I	III	I	III	II
EU	High	IV	V	IV	III	IV
	Low	III	V	IV	IV	V
NA	High	V	V	V	III	IV
	Low	IV	III	III	III	III
OC	High	III	III	III	III	IV
	Low	II	III	II	IV	III
SA	High	V	V	IV	V	V
	Low	III	V	V	V	V

Note: I–V represents the severity of exposure changes (from aggravated to relief), or the level of potential risk (from high to low): I indicates significant exacerbation; II indicates general exacerbation; III indicates no significant change; IV indicates general reduction; V indicates a significant reduction. SYD represents the suitability level: High indicates V and IV regions, and Low indicates III and II regions. B represents baseline; M represents medium term; L represents long term.

). The specific calculation process is shown in Appendix A.7. In all scenarios, the high temperature exposure of the higher-grade rice habitat suitability zones in Asia were aggravated the most, which were followed by the zones in Africa. The high temperature exposure of all grades of rice habitat suitability zones in South America and Europe were alleviated, as well as the higher-grade rice habitat suitability zones in North America. Overall, the potential risks of high temperature hazards on rice seriously increased in Asia, especially on the Indian subcontinent, Indo-China and the Chinese monsoon regions. The potential risks of high temperature hazards on rice significantly reduced in southern Europe, south-central South America, and the southeastern United States.

4. Discussion

4.1. Accuracy Tests of Potential Rice Distribution Estimates

In order to test the accuracy of the estimated results of potential rice distribution areas, we compared the estimated results with existing sets of rice harvesting and seeding data (EARTHSTAT 2000 harvest range data, SPAM 2000 and 2005 harvest range and sowing range data, and MIRCA 2000 harvest range data), as shown in

Table 4. The degree of change in high temperature exposure in rice.

Accuracy Type	Data for Verification
	EARTHSTAT 2000 Harvest Data	SPAM 2000 Harvest Data	SPAM 2000 Sowing Data	MIRCA 2000 Harvest Data	SPAM 2005 Harvest Data	SPAM 2005 Sowing Data
YY	3.6%	2.6%	2.5%	1.9%	2.8%	2.9%
NN	91.6%	92.6%	92.5%	93.0%	91.2%	91.4%
NY	2.7%	3.8%	3.9%	4.5%	4.7%	4.6%
YN	2.1%	1.0%	1.1%	0.6%	1.3%	1.1%
User Accuracy	52.8%	35.4%	36.1%	23.6%	32.6%	33.3%
Producer Accuracy	80.8%	83.8%	84.2%	80.4%	81.3%	81.8%
All Accuracy	93.9%	93.8%	94.2%	92.9%	92.4%	92.8%
Kappa Coefficient	0.61	0.47	0.48	0.35	0.43	0.44

.

The estimated potential rice distribution zones were generally larger than the actual rice distributions (Figure 9). The reason is that although North America and other places were potential areas of rice growth, due to the local cultivation histories and eating habits, rice cultivation was lower than the predictions, which reduced the accuracy of the results given in this paper. The accuracy of the producers of each set of data was greater than 80%, indicating that the rice potential distribution zones calculated in this paper contained over 80% of the actual rice distribution areas. So, the above-mentioned potential rice distribution zone estimation method had good accuracy and can reflect the actual rice distribution cultivation in the future.

4.2. Temporal Rice Exposure to High Temperature

From the view of disaster risk theory, exposure can be defined by two aspects: spatial and temporal. The spatial aspect was analyzed in the previous sections. The temporal aspect emphasizes the occurrence of hazards during the full growth period of crops or the key stages of crop growth. In terms of the effects of high temperature hazards on rice, the high temperature hazard intensity during a whole growth period is the leading factor that affects rice loss. With global climate change, the temporal distribution of extreme high temperature events during the rice growth period will also change.

Based on the rice growth season data from SAGE (University of Wisconsin-Madison Sustainability and the Global Environment) [54], we calculated the distribution of high temperature exposure and SSD changes during the growth period of rice, and the barycenter shifts of the SDDs during the whole growth period, and then analyzed the temporal characteristics of rice exposure to high temperature. The SDD barycenter calculation results were classified into three stages of rice growth periods: vegetative phase (VE), reproductive phase (RE) and ripening phase (RI). SDD barycenter distribution during the whole rice growth period can be calculated using Equation (7):

$$D=\frac{{{\displaystyle \sum }}_{i=1}^{n}SD{D}_i\times d_i}{{{\displaystyle \sum }}_{i=1}^{n}SD{D}_i}$$

(7)

where SDD_i is the high temperature hazard index of day d, n is the number of days with SDD values > 0 during the whole growth period, and D is the SDD barycenter distribution during the whole growth period.

(1)

Distribution of SDDs during the rice growth period

(i)

The spatial distributions of the SDDs during the rice growth period were roughly the same as those of the SDD values determined without considering temporal exposure. The high-value regions during the growth period were in the Indian subcontinent, Mesopotamia, southern areas of the Sahara Desert, and the northern plateau of Mexico, although their scopes shrank. In particular, southern China shrunk the most because the original data used in this region only consisted of local, early-season rice growth period data.

(ii)

Changes in the SDD values without considering temporal exposure were consistent with the changes in SDDs during the rice growth period, but the change rates of the SDDs during the growth period were larger, especially on the Indian subcontinent and southern China.

(2)

Barycenter shift of the SDDs during the growth period

By comparing the SDD barycenter shifts during the rice growth periods from the baseline to long-term periods, we found that there were no significant changes in most parts of the world: the SDDs in Europe, Africa, North America and Oceania were concentrated in the RE period, and in the VE period for the Mexican plateau. The SDDs in South America were concentrated in the VE period. In Asia, there was a regional difference, where those in southern China were concentrated in the RI period, and those in northern China were concentrated in the RE period; the rest were concentrated in the VE period. However, barycenter fluctuations occurred in some areas: the SDD barycenter in the northeastern United States showed a floating change of VE to RE to VE, while the SDD barycenter in northeast China transitioned from the RE period to the RI period. The southern area of the Brazilian plateau presented RE to VE to RE floating changes. In summary, major rice-growing areas in Asia need to focus on high temperature risk prevention in the future, and according to the periods of high temperature hazards, targeted high temperature protection mitigation measures should be carried out at key growth stages [55,56].

4.3. Comparative Analysis with Related Research

According to the results of our study, with the background of the CMIP5 climate models and socioeconomic scenarios, the global rice exposure presented the differences in spatial patterns and trends in various periods. In order to ensure the rationality of the results obtained in this paper, we conducted a comparative analysis with some related research. (1) In terms of the hazard-bearing body: Wu [15] simulated the rice-growing areas for the years 2000 and 2035 using the crop choice decision model; the primary rice production regions of the results, such as the Indian subcontinent, southeast Asia, southern China, southern United States, southeastern Brazil and the Sudanian Savanna are basically consistent with the higher-grade rice suitability zones (V and IV regions) of the baseline and medium-term periods in our study. In the studies on local regions, the results showed that the area and grade of rice suitability zones in southern China were decreasing, but they were increasing in northern China. Approximately 18.2% of the new rice-growing areas are in the newly suitable areas of the northern single-cropping rice system, and the northern limit of Chinese rice production has moved northward from 47° N to 52° N. The results of our study also presented these trends above from baseline to the long-term period. (2) In terms of the hazard factor: Gourdji [11] analyzed the regional features of global crops facing extreme high temperature, and found that during the period from the 2000s to 2050s, days with over critical temperatures in the Indian subcontinent and southern China increased by over 3–5 times, while these regions of our study showed a 4–5 times increase trend. In the studies on local regions, the results showed that the high-value regions of heat stress and its variation in southern China during the period 1980–2010 were the provinces of Jiangxi, Hunan and Guangdong, while the geographical barycenters of heat stress showed an eastward or northeastward shift. The trends in our study are also consistent with theirs. The comparative analysis above can indicate to a certain extent that the spatial patterns of and trends in rice exposure to high temperature obtained in our study are rational.

4.4. Limitations of This Study

4.4.1. Regional Indicator Selection and Suitability Level Division

Habitat factors have different effects on rice production in different regions. Hence, the principle of adapting to local conditions should be adopted in the screening process. In particular, the setting of the quantile, which is used in suitability grade ranking, should also reflect any regional characteristics. For example, in the northern part of the Indian subcontinent, the annual precipitation is higher than in other major rice-growing areas, which makes the regional suitability lower than the actual situation. Therefore, in future research, crop potential distribution estimation needs to be optimized from two levels: (1) mining regional characteristic habitat variables, and (2) introducing weighted elements (such as crop yield) to set the grade quantiles.

4.4.2. Consideration of Socioeconomic Factors

In this paper, we only considered land use as socioeconomic scenario data. However, changes in population, markets and eating habits, which limit the area and distribution of crop cultivation and hence affect crop prices, demand and planting willingness [57,58,59], were only marginally considered. Therefore, in order to make the estimated results more consistent with reality, a comprehensive multi-factor exposure estimation method is imperative. In addition, the level of fortification in the considered socioeconomic factors is often neglected, and any increase in the level of fortification can greatly reduce exposure [60], which should be considered in future research.

5. Conclusions

By combining the MaxEnt model and MCDA, in this paper, we mapped the potential rice distribution under different scenarios during different periods. Based on the above-mentioned results, we calculated the characteristic variations in rice exposure to high temperature from the perspective of disaster risk theory. The specific conclusions are as follows:

(1)

The areas of potential rice distribution zones changed:

(i)

From the perspective of area changes, under different scenarios, the rice potential distribution zones continuously declined. Europe had the largest decline (approximately 50%), followed by Asia, Africa and South America (with averages of 30% to 40%). Among them, the higher-grade suitability areas in Asia and Africa dropped significantly, with a decrease of 40%–50%. There were significant declines in the areas of all suitability grades in South America, while the areas of potential distribution zones in North America and Oceania increased significantly (by approximately 30%).

(ii)

From the perspective of barycenter shifts, the global potential rice distribution zones generally moved to higher latitudes, except in Europe, which moved to lower latitudes. The movement ranges in terms of latitude were the largest in Europe, 6–10°, while Oceania had the smallest range of movement, 1–3°.

(2)

The effects of rice exposure to high temperature, and the intensities of the high temperature hazards, had the following characteristics:

(i)

High temperature exposure was concentrated on the Indian subcontinent, Asia Minor, Sudanian Savanna, the north plains of the Mexican Plateau, La Plata plains, the southwestern parts of the Great Dividing Range in Australia, and southern China, where the SDDs were high, with averages of 100–500.

(ii)

The SDD values in potential rice distribution zones showed continuous and significant increases. The SDDs increased from RCP2.6 to RCP8.5, especially on the Indian subcontinent, the monsoon region of China, southeastern United States, south-central South America, the southeastern parts of Australia and the Iberian Peninsula, where the SDD increased by over 3 to 5 times.

(iii)

High temperature exposure on the Indian subcontinent, Indo-China and the Sudanese prairie region all decreased significantly, while the Mediterranean Sea, central and southern Africa, the east coast of South America, northeast China, and central North America all increased.

(3)

The characteristics of the comprehensive changes in high temperature exposure and their effect on rice are summarized as follows. In each scenario, the degree of exposure change on each continent showed a floating variation. From the overall characteristics of high temperature exposure, changes in Asia were significantly aggravated, and South America showed significant reductions. In Africa, the performance was generally aggravated, while the rest of the continents generally had fewer or no significant changes.

In summary, according to the abovementioned changes, the areas of rice suitability zones in the main producing areas of Asia, especially the Indian subcontinent, Indo-China and the eastern monsoon region of China, were significantly reduced, and the intensities of the high temperature hazards increased significantly. The potential high temperature risks in these regions are expected to be severe in the future. To combat the expected risks, key monitoring and disaster prevention should be carried out. The suitable areas for rice crops in North America will expand, while the degree of exposure will reduce. With this prediction, it may be prudent to try to expand the existing range of rice cultivation. In view of the obvious reduction in exposure in South America, cultivation can be appropriately modified. Finally, with the background of global change, this paper has great significance for the mitigation of high temperature risk in rice and its effect on the potential security of future global rice production. For example, this paper provides a spatial distribution of global rice exposure to high temperature under RCP and SSP scenarios, which is conducive to the rational allocation of funds and resources in the disaster management of rice cultivation. What is more, the results of this study can also provide a reference for the future regional transfer of rice cultivation at the high temperature hazard level.

Further study can take the following two aspects into consideration. (1) To improve the simulation accuracy of rice suitability zones, more socioeconomic factors should be added into the constraints in the terms of reasonable and quantitative calculation (such as market supply and demand, farmers’ planting willingness and residents’ eating habits). In addition, the occurrence frequency and intensity of natural disasters (such as extreme temperatures, droughts and floods) also affects the actual crop-cultivated distribution. Therefore, the level of crop field fortification will be among the important factors in the estimation and evaluation of rice suitability zones. On the other hand, by considering the regional differences in geographical environment and rice cultivars, the sectional method of selecting habitat variables in different regions can be adopted to improve the simulation accuracy. (2) To expand the connotation of rice exposure, this study only took rice cultivation zones as the exposure subject. In the future research, we should consider the climatic and socioeconomic changes comprehensively, and use scientific methods to estimate the rice-growing areas, yields and economic values, so as to better enrich the essential content of rice exposure and use these data in the risk assessment of high temperature hazard to rice.