Impacts of Climate Change on Wildﬁres in Central Asia

: This study analyzed ﬁre weather and ﬁre regimes in Central Asia from 2001–2015 and projected the impacts of climate change on ﬁre weather in the 2030s (2021–2050) and 2080s (2071–2099), which would be helpful for improving wildﬁre management and adapting to future climate change in the region. The study area included ﬁve countries: Kazakhstan, Kyrgyzstan, Tajikistan, Uzbekistan, and Turkmenistan. The study area could be divided into four subregions based on vegetation type: shrub (R1), grassland (R2), mountain forest (R3), and rare vegetation area (R4). We used the modiﬁed Nesterov index (MNI) to indicate the ﬁre weather of the region. The ﬁre season for each vegetation zone was determined with the daily MNI and burned areas. We used the HadGEM2-ES global climate model with four scenarios (RCP2.6, RCP4.5, RCP6.0, and RCP8.5) to project the future weather and ﬁre weather of Central Asia. The results showed that the ﬁre season for shrub areas (R1) was from 1 April to 30 November, for grassland (R2) was from 1 March to 30 November, and for mountain forest (R3) was from 1 April to 30 October. The daily burned areas of R1 and R2 mainly occurred in the period from June–August, while that of R3 mainly occurred in the April–June and August–October periods. Compared with the baseline (1971–2000), the mean daily maximum temperature and precipitation, in the ﬁre seasons of study area, will increase by 14%–23% and 7%–15% in the 2030s, and 21%–37% and 11%–21% in the 2080s, respectively. The mean MNI will increase by 33%–68% in the 2030s and 63%–146% in the 2080s. The potential burned areas of will increase by 2%–8% in the 2030s and 3%–13% in the 2080s. Wildﬁre management needs to improve to adapt to increasing ﬁre danger in the future.


Introduction
Wildfires are a dominant disturbance in most forests and are strongly influenced by climate [1]. Climate warming has recently caused changes in the fire regime in the Northern Hemisphere [2], which has experienced extreme wildfire seasons and fire frequency increases in forests. Notably, high-intensity fires have occurred in summer in some regions. During the summer of 2010, climate warming caused several hundred wildfires and burned areas of approximately 5 million ha in Russia [3]. In the summer of 2017, British Columbia, Canada, experienced the worst wildfire, which caused a burned area of 1.2 million ha [4]. In addition, in the boreal forest of North America, climate warming has led to greater and more severe wildfire activity, increased fire frequency and fire sizes, and longer fire seasons [5]. The large-scale wildfires in the United States in 2019 and Australia from 2019-2020 attracted the attention of global society. Central Asia is located in the arid and semiarid zone, which includes

Study Area
Central Asia is located in central Eurasia and is adjacent to China in the west and the Caspian Sea in the west. Its geographical range is 35 • 08 -55 • 25 N, 46 • 28 -87 • 29 E. The total area is approximately 3970 million ha. It has a temperate continental climate and uneven distribution of rainfall. The annual precipitation ranges from 100-400 mm, and it can be more than 500 mm in high mountain areas and less than 200 mm on plains [7].
The forest coverage rate in the study area was 1.6%, which was mainly distributed in the northeast. Shrubs were mainly distributed in the western and central parts of the study area, with a coverage rate of 22.4%. Grassland and farmland accounted for 23.8% and 20.3%, respectively ( Figure 1).

Study Area
Central Asia is located in central Eurasia and is adjacent to China in the west and the Caspian Sea in the west. Its geographical range is 35°08′-55°25′N, 46°28′-87°29′E. The total area is approximately 3970 million ha. It has a temperate continental climate and uneven distribution of rainfall. The annual precipitation ranges from 100-400 mm, and it can be more than 500 mm in high mountain areas and less than 200 mm on plains [7].
The forest coverage rate in the study area was 1.6%, which was mainly distributed in the northeast. Shrubs were mainly distributed in the western and central parts of the study area, with a coverage rate of 22.4%. Grassland and farmland accounted for 23.8% and 20.3%, respectively ( Figure  1).

Climate Data Processing
For the missing temperature and dew point temperature in historical observation data for the 2001-2015 period, we used the sliding average of the before and after five days to replace the missing data. Due to the lack of daily precipitation data, data from neighboring meteorological stations were used.

Climate Data Processing
For the missing temperature and dew point temperature in historical observation data for the 2001-2015 period, we used the sliding average of the before and after five days to replace the missing data. Due to the lack of daily precipitation data, data from neighboring meteorological stations were used. The precipitation frequency from the historical climate data was used to correct the simulated precipitation frequency of the simulated climate data. We assumed that the daily precipitation frequencies of the simulated data were identical to the historical data (2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015). We obtained the precipitation threshold based on its frequency and set the data less than the threshold to zero. Then, we used the coefficient between the simulated annual precipitation and reserved data to correct the daily precipitation.
The mid-day temperature was replaced with the maximum temperature minus the average difference (2 • C). The dew point temperature was calculated from the daily minimum temperature. The calculation formula was developed with the dew point temperature above 0 • C and the daily minimum temperature in each month of the historical observation data.

Fire Weather Indices Calculation
The modified Nesterov index was calculated with the following equation [29]: where MNI(n − 1) and MNI(n) are the fire weather index on days n − 1 and n, respectively. T is the mid-day temperature, d is the dew point temperature, K(n) is a scale coefficient that controls the index change when precipitation occurs on day n [33] (Table 1).

Vegetation Zone
The study area can be divided into four zones based on the vegetation types, which include shrub (R1), grassland (R2), mountain forest (R3), and rare vegetation zone (R4) (Figure 2). This paper focuses on zones with vegetation, such as R1, R2, and R3. The precipitation frequency from the historical climate data was used to correct the simulated precipitation frequency of the simulated climate data. We assumed that the daily precipitation frequencies of the simulated data were identical to the historical data (2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015). We obtained the precipitation threshold based on its frequency and set the data less than the threshold to zero. Then, we used the coefficient between the simulated annual precipitation and reserved data to correct the daily precipitation.
The mid-day temperature was replaced with the maximum temperature minus the average difference (2 °C). The dew point temperature was calculated from the daily minimum temperature. The calculation formula was developed with the dew point temperature above 0 °C and the daily minimum temperature in each month of the historical observation data.

Fire Weather Indices Calculation
The modified Nesterov index was calculated with the following equation [29]: where MNI(n−1) and MNI(n) are the fire weather index on days n−1 and n, respectively. T is the midday temperature, d is the dew point temperature, K(n) is a scale coefficient that controls the index change when precipitation occurs on day n [33] (Table 1).

Vegetation Zone
The study area can be divided into four zones based on the vegetation types, which include shrub (R1), grassland (R2), mountain forest (R3), and rare vegetation zone (R4) (Figure 2). This paper focuses on zones with vegetation, such as R1, R2, and R3.

Burned Areas
The fire season and land cover data were used to filter the false burned areas of the MODIS-MCD64A1 products. Then, we obtained the daily burned areas of each vegetation zone.

Data Processing
SPSS software was used to analyze the correlation between the fire weather index and burned area. The MNI of each zone was interpolated with the kriging method.

Burned Areas
The fire season and land cover data were used to filter the false burned areas of the MODIS-MCD64A1 products. Then, we obtained the daily burned areas of each vegetation zone.

Data Processing
SPSS software was used to analyze the correlation between the fire weather index and burned area. The MNI of each zone was interpolated with the kriging method.

Fire seasons for Each Vegetation Zone
The daily MNI of each vegetation zone was roughly normally distributed. The MNI increased slowly for 90 days and decreased rapidly from the peak. The MNI of R1 increased from the 95th day (Julian day), reached a maximum on the 280th day, and remained at a very low level after the 325th day ( Figure 3(a1)). In R2, the MNI increased from the 95th day (peak on the 283rd day) and was maintained at a very low value after the 342nd day. In R3, the MNI increased from the 125th day and decreased to a very low value after the 300th day (maximum on the 248th day).
slowly for 90 days and decreased rapidly from the peak. The MNI of R1 increased from the 95th day (Julian day), reached a maximum on the 280th day, and remained at a very low level after the 325th day ( Figure 3a1). In R2, the MNI increased from the 95th day (peak on the 283rd day) and was maintained at a very low value after the 342rd day. In R3, the MNI increased from the 125th day and decreased to a very low value after the 300th day (maximum on the 248th day).
The fire season of each vegetation zone could be defined by the fire weather index and the daily burned areas. The daily burned areas for R1, R2, and R3 showed apparent changes during the periods of days 150-280, 140-300, and 90-305, respectively (Figure 3b1-3). However, there were some grass fires during days 60-140. The dates of the increases in daily burned areas were earlier than that of the increases in MNI. In early spring, the fuels are dry and cured grass, which can burn at low MNI conditions. The date when the daily burned areas decreased to zero was also earlier than the date when the MNI reached a very low value. The MNI only indicates fire weather, which does not reflect the seasonal status of live fuel. Although daily burned areas showed a relationship with the MNI (r > 0.23), their changes were not completely consistent.
The fire season of each zone was defined based on the process of daily MNI and burned areas, which were 1 April-30 November for R1, 1 March-30 November for R2, and 1 April-31 October for R3.      The fire season of each vegetation zone could be defined by the fire weather index and the daily burned areas. The daily burned areas for R1, R2, and R3 showed apparent changes during the periods of days 150-280, 140-300, and 90-305, respectively (Figure 3(b1-b3)). However, there were some grass fires during days 60-140. The dates of the increases in daily burned areas were earlier than that of the increases in MNI. In early spring, the fuels are dry and cured grass, which can burn at low MNI conditions. The date when the daily burned areas decreased to zero was also earlier than the date when the MNI reached a very low value. The MNI only indicates fire weather, which does not reflect the seasonal status of live fuel. Although daily burned areas showed a relationship with the MNI (r > 0.23), their changes were not completely consistent.
The fire season of each zone was defined based on the process of daily MNI and burned areas, which were 1 April-30 November for R1,    (Figure 5b1). The mean annual burned area in R2 was 389,020 ha, which was much higher than that in the other zones. The maximum monthly burned area was 133,452 ha in August, and the minimum was 1211 ha in November. The mean annual burned area in R3 was 150 ha. The months in the fire season with the maximum and minimum burned areas were September and June, respectively (Figure 5b3).
The monthly burned areas and MNI in the fire season for R1 and R3 did not show a strong correlation (r = 0.36), but they showed a strong correlation for R2 (r = 0.6) as follows: where, B(n) and MNI(n) are the burned areas and mean MNI in month n.    Figure 5(a1)). The monthly maximum burned areas occurred in May, with 26,603 ha, and the minimum occurred in November, with 250 ha (Figure 5(b1)). The mean annual burned area in R2 was 389,020 ha, which was much higher than that in the other zones. The maximum monthly burned area was 133,452 ha in August, and the minimum was 1211 ha in November. The mean annual burned area in R3 was 150 ha. The months in the fire season with the maximum and minimum burned areas were September and June, respectively ( Figure 5(b3)).
The monthly burned areas and MNI in the fire season for R1 and R3 did not show a strong correlation (r = 0.36), but they showed a strong correlation for R2 (r = 0.6) as follows: where, B(n) and MNI(n) are the burned areas and mean MNI in month n. Grass fires and shrub fires mainly occurred in the period from June to September, and they were usually distributed in the plains and hilly areas with elevations less than 500 m. There were no wildfires in the areas with elevations greater than 2000 m. The fires in mountain forests mainly occurred in the April-June and September-October periods and were usually distributed the areas with low elevation (<500 m). A few wildfires also occurred in high-altitude areas (>1500 m ASL) during July and August ( Figure 6). Grass fires and shrub fires mainly occurred in the period from June to September, and they were usually distributed in the plains and hilly areas with elevations less than 500 m. There were no wildfires in the areas with elevations greater than 2000 m. The fires in mountain forests mainly occurred in the April-June and September-October periods and were usually distributed the areas with low elevation (<500 m). A few wildfires also occurred in high-altitude areas (>1500 m ASL) during July and August ( Figure 6).

Climate Change in the 2030s and 2080s
The mean daily maximum temperature in the fire season of R1 was 23.2 °C during the baseline period. It will be 26.4 °C band 28.2 °C in the 2030s and 2080s, respectively, which will be an increase of 14% and 21% compared with the baseline (p = 0.00). The precipitation in the fire season of R1 was 180 mm at the baseline and will be 219, 174, 198, and 180 mm under the RCP2.6, RCP4.5, RCP6.0, and RCP8.5 scenarios in the 2030s, respectively. The precipitation will increase by 7% to 193 mm in the 2030s. However, the increase was not significant (F-test, p = 0.12). In the 2080s, the precipitation will be 215, 195, 178, and 214 mm under the RCP2.6, RCP4.5, RCP6.0, and RCP8.5 scenarios, respectively. The mean precipitation will significantly increase by 11% to 200 mm in the 2080s (F-test, p = 0.04) (Figure 7).
The mean daily maximum temperature in fire season of R2 was 15.7 °C in the baseline period. It will significantly increase by 21% and 34% in 2030s and 2080s, respectively (p = 0.00). The precipitation in fire season of R2 was 226 mm at the baseline, and it will be 293, 206, 256, and 284 mm in the 2030s under the RCP2.6, RCP4.5, RCP6.0, and RCP8.5 scenarios, respectively. The mean precipitation will increase 15% in 2030s (p = 0.00). In addition, the precipitation in the fire season will significantly increase by 21% in 2080s (p = 0.00), and the precipitation will be 294, 231, 260, and 303 mm under the RCP2.6, RCP4.5, RCP6.0, and RCP8.5 scenarios, respectively.

Climate Change in the 2030s and 2080s
The mean daily maximum temperature in the fire season of R1 was 23.2 • C during the baseline period. It will be 26.4 • C band 28.2 • C in the 2030s and 2080s, respectively, which will be an increase of 14% and 21% compared with the baseline (p = 0.00). The precipitation in the fire season of R1 was 180 mm at the baseline and will be 219, 174, 198, and 180 mm under the RCP2.6, RCP4.5, RCP6.0, and RCP8.5 scenarios in the 2030s, respectively. The precipitation will increase by 7% to 193 mm in the 2030s. However, the increase was not significant (F-test, p = 0.12). In the 2080s, the precipitation will be 215, 195, 178, and 214 mm under the RCP2.6, RCP4.5, RCP6.0, and RCP8.5 scenarios, respectively. The mean precipitation will significantly increase by 11% to 200 mm in the 2080s (F-test, p = 0.04) (Figure 7).

MNI Changes in the 2030s and 2080s
The mean MNI in the fire season of R1 was 3812 at the baseline, and it will increase by 33% and 63% in the 2030s and 2080s, respectively (p = 0.00) (Figure 8). In R2, the mean MNI of the fire season was 1759 at baseline, and it will increase by 42% in the 2030s and 73% in the 2080s (p = 0.00). The  The mean daily maximum temperature in fire season of R2 was 15.7 • C in the baseline period. It will significantly increase by 21% and 34% in 2030s and 2080s, respectively (p = 0.00). The precipitation in fire season of R2 was 226 mm at the baseline, and it will be 293, 206, 256, and 284 mm in the 2030s under the RCP2.6, RCP4.5, RCP6.0, and RCP8.5 scenarios, respectively. The mean precipitation will increase 15% in 2030s (p = 0.00). In addition, the precipitation in the fire season will significantly increase by 21% in 2080s (p = 0.00), and the precipitation will be 294, 231, 260, and 303 mm under the RCP2.6, RCP4.5, RCP6.0, and RCP8.5 scenarios, respectively.

MNI Changes in the 2030s and 2080s
The mean MNI in the fire season of R1 was 3812 at the baseline, and it will increase by 33% and 63% in the 2030s and 2080s, respectively (p = 0.00) (Figure 8). In R2, the mean MNI of the fire season was 1759 at baseline, and it will increase by 42% in the 2030s and 73% in the 2080s (p = 0.00). The mean MNI of the fire season for R3 was 713 at baseline, and it will increase to 1195 in the 2030s and 1752 in the 2080s (p = 0.00).

MNI Changes in the 2030s and 2080s
The mean MNI in the fire season of R1 was 3812 at the baseline, and it will increase by 33% and 63% in the 2030s and 2080s, respectively (p = 0.00) (Figure 8). In R2, the mean MNI of the fire season was 1759 at baseline, and it will increase by 42% in the 2030s and 73% in the 2080s (p = 0.00). The mean MNI of the fire season for R3 was 713 at baseline, and it will increase to 1195 in the 2030s and 1752 in the 2080s (p = 0.00).  (Figure 9a1). In the 2030s, the MNI values in western and northern R1 showed a slight increase but will increase from 23-30% in central areas and 37%-59% in the south. The mean MNI will increase by 19%, 28%, 20%, and 33% in the 2030s under the RCP2.6, RCP4.5, RCP6.0, and RCP8.5 scenarios, respectively. The MNI will increase obviously under scenarios RCP4.5 and RCP8.5, and the maximum increase will reach 68% and 82%, respectively (Figure 9b3,b5). In the 2080s, MNIs will increase by more than 37% in central and southern R1, while MNIs in western R1 Most areas of R1 showed low MNI values (<4691) at the baseline, and only southern areas had high MNI values (8735-17,156) (Figure 9(a1)). In the 2030s, the MNI values in western and northern R1 showed a slight increase but will increase from 23%-30% in central areas and 37%-59% in the south. The mean MNI will increase by 19%, 28%, 20%, and 33% in the 2030s under the RCP2.6, RCP4.5, RCP6.0, and RCP8.5 scenarios, respectively. The MNI will increase obviously under scenarios RCP4.5 and RCP8.5, and the maximum increase will reach 68% and 82%, respectively (Figure 9(b3,b5)). In the 2080s, MNIs will increase by more than 37% in central and southern R1, while MNIs in western R1 will increase by only 15% (Figure 9(c1)). The mean MNI will increase by 15%, 39%, 46%, and 64% over the baseline under the RCP2.6, RCP4.5, RCP6.0, and RCP8.5 scenarios, respectively. Under the RCP2.6 scenario, the MNI will increase significantly in the south, while the MNI will increase in the central and southern regions significantly under the other scenarios. 599) at the baseline. The mean MNI will increase by 49%, 77%, 49%, and 68% in the 2030s over the baseline under the RCP2.6, RCP4.5, RCP6.0, and RCP8.5 scenarios, respectively. In the 2080s, the mean MNI will increase by 32%, 84%, 127%, and 146% under the RCP2.6, RCP4.5, RCP6.0, and RCP8.5 scenarios, respectively. A greater increase in MNI will be distributed in the northwestern and central areas under the RCP2.6 scenario, while MNI will increase clearly in the northwest of R3 under the other scenarios.

Discussion
The MNI can effectively reflect drought weather and fuel moisture in Central Asia [33]. In the study area, there was less precipitation and higher temperatures from June-September, and the MNI reached its peak in August or September. Burned area is the most important indicator of fire regime. We used the daily burned areas from 2001-2015 from MODIS data to describe their distributions spatially and temporally [34]. There have been some studies on wildfires in Central Asia based on remote sensing data in recent years [21,35,36]. We determined the fire season for each vegetation zone according to the MNI and daily burned areas. We considered the characteristics of fire regime and The MNI values in western R2 were high at baseline. The mean MNI will increase by 23%, 35%, 26%, and 42% in the 2030s under the RCP2.6, RCP4.5, RCP6.0, and RCP8.5 scenarios, respectively. The increase is significant for the RCP4.5 and RCP8.5 scenarios. The MNI in the southern area will increase clearly (+50%), but the values in the western areas will increase by only 10%-38% in the 2030s. The MNI will increase by 20%, 51%, 61%, and 73% in the 2080s under the RCP2.6, RCP4.5, RCP6.0, and RCP8.5 scenarios, respectively. The MNI values in the western and southern R2 will increase significantly under RCP2.6 scenario, and the values will increase significantly in the western, central, and southern areas under the other scenarios.

Discussion
The MNI can effectively reflect drought weather and fuel moisture in Central Asia [33]. In the study area, there was less precipitation and higher temperatures from June-September, and the MNI reached its peak in August or September. Burned area is the most important indicator of fire regime. We used the daily burned areas from 2001-2015 from MODIS data to describe their distributions spatially and temporally [34]. There have been some studies on wildfires in Central Asia based on remote sensing data in recent years [21,35,36]. We determined the fire season for each vegetation zone according to the MNI and daily burned areas. We considered the characteristics of fire regime and fire weather. The defined fire season would be an important indicator of fire dynamics and fire management in the region.
Grassland fire and shrub fire mainly occurred in summer (June-September), and the largest burned areas in grassland usually occurred in August or September [21]. The fires in mountain forests in the low altitude regions mainly occurred in spring (April-June) and autumn (September and October). Nevertheless, the fires mainly occurred in summer (July and August) for the forests at high-altitude areas (>1500 m ALS). Fires in mountain forests are usually distributed in areas near farmland or towns, which indicates that human activities impact these occurrences. The references indicated that most forest fires were ignited by humans in low-altitude areas and fires in high-altitude areas in summer were mainly caused by lightening [4,18].
Although both the temperature and temperature in the fire season of the vegetation zones will increase in the 2080s under future climate scenarios, their MNIs will still increase clearly over the baseline. This indicates that fire danger will increase in the future, which is consistent with the results of Liu et al. [9]. They also believed that the mean daily maximum temperature and fire danger rating of Central Asia (HadCM3 model with A2a scenario) would increase from 2071-2100, but they projected that the annual precipitation would decrease.
Climate change will also affect the vegetation of the study area. In the study we did not simulate the vegetation change resulting from the future climate change. Vegetation distribution was affected by many factors, such as climate, anthropic activities, and natural disturbances. In fact, forest and grassland decreased during the period 1992-2003 and increased slightly during 2003-2015. However, vegetation types and their spatial distribution was no changed clearly in the period [37]. We assumed that the vegetation will not signification change in the coming decades. This point will not have influences on the judgment of fire weather changes under the future climate scenarios.
The fewer meteorological stations (18) available from 2001-2015, and their uneven spatial distribution, may affect the interpolated results of fire weather. However, each meteorological factor and fire weather index showed very similar processes in those years. The influence would not affect the reliability of the results.
Fire weather affects ignitions and fire spread. The MNI showed a positive correlation with the monthly burned areas (r < 0.3). Based on the correlation, we projected that the potential burned areas would increase in the future. The potential burned areas of R1, R2, and R3 in the 2030s will increase by 4%, 8%, and 2% over the baseline and will increase by 6%, 13%, and 3%, respectively, in the 2080s.
The wildfires were mainly distributed in shrub and grassland areas. The annual burned area has generally declined since 2010. This trend reflects the role of fire management activities. The governments in Central Asia established fire agencies and promulgated laws and regulations on wildfire management in this century [38]. The wildfires were still serious in some years (such as 2011). The vegetation in the study area plays an important role in the regional ecological environment and biodiversity protection [39]. It is necessary to strengthen wildfire management in the region to adapt to future climate change.

Conclusions
Fire seasons are different for each vegetation zone. Grassland has the longest fire season, and mountain forests have the shortest fire season. The fire seasons of grassland, shrub, and mountain forest are 1 March-30 November, 1 April-30 November, and 1 April-31 October, respectively. Most wildfires in the study area mainly occurred in shrub and grasslands. Most shrub and grass fires occurred in the period from June-September, and fires in mountain forests occurred mainly in the April-June and September-October periods.
The MNI index is a good indicator of fire danger for Central Asia. In the 2030s, the mean daily maximum temperature in the fire season for vegetation areas will increase significantly over the baseline, and the precipitation will increase by 7%-15%. The MNI will increase by 33%-68% for vegetation areas in the 2030s. The mean daily maximum temperature, precipitation and MNI of vegetation areas will increase significantly in the 2080s. The MNI will increase by 63%-146%, and the potential areas will increase by 3%-13% for each vegetation zone.
Author Contributions: X.Z. and X.T. conceived and designed the experiments; X.Z. wrote the original draft and edited the manuscript. X.T. wrote, reviewed and edited the manuscript; Y.Y. contributed on data collection and reviewed the manuscript. All authors have read and agreed to the published version of the manuscript.

Conflicts of Interest:
The authors declare no conflicts of interest.