Strong winds drive grassland fires in China

Accounting for 41.7% of China’s total land area, grasslands are linked to the livelihoods of over 20 million people. Although grassland fires cause severe damage in China every year, their spatiotemporal patterns and climate drivers are not well understood. In this study, we used grassland fire record forms provided by the National Forestry and Grassland Administration and grassland fire location data from the Wildfire Atlas of China to examine the spatiotemporal patterns and and seasonality of fires in China for the period from 2008 to 2020. We found that most grassland fires occurred in Inner Mongolia in northern China, specifically in the Hulun Buir and Xilingol grasslands. We found distinct differences in fire seasonality in northern China, which has a major fire season in April, versus southwestern China, where the major fire season occurs in February, March and April. April grassland fires in northern China are the result of strong winds, typically from the west, and spring drought. A secondary fire season in northern China occurs in October and is also driven by strong winds. The fire season in southwestern China seems to be less shaped by climatic factors such as wind speed, precipitation, and drought. This study provides support for decision-making by fire prevention and fire management authorities in China.


Introduction
The risks and suppression costs of wildfires are increasing worldwide, as is the damage they cause to ecosystems, watersheds, and human infrastructure (Burke et al 2021). Yet, in many terrestrial ecosystems around the world, fire is an important ecological process, and many plants and animals depend on fire for regeneration, growth, and suitable habitat (Bowman et al 2020). Grassland fires, an important type of wildfire, pose a serious threat to people who reside in proximity to rural and forested areas and can result in injuries, fatalities, and property loss (Moinuddin et al 2018). Grasslands account for 41.7% of China's 3.7 million square miles and directly affect the livelihoods of over 20 million people (Zhou et al 2020).
Although grassland fire disasters threaten both human lives and the functioning of ecosystems, the scientific exploration of fire in China has mainly focused on forest fires (Zong et al 2021). As a result, very little is known about the interactions between grassland fires and climate change. There is a serious lack of quantitative fire data for grassland fires in China, making it difficult to track spatiotemporal grassland fire patterns. Quantitative fire data, including fire records and confirmed fire hotspots, cannot be replaced by the global fire data detected by satellites (Giglio et al 2006). The lack of grassland fire records limits our ability to examine the key characteristics and drivers of grassland fires in China.
Accounting for over one-third of Earth's vegetation, grasslands are the most fire-prone ecosystem, and 80% of all fires worldwide are grassland fires (Leys et al 2018), among which savannas alone account for 60% of total global fire emissions (Russell-Smith et al 2017). Although temperate and alpine grasslands comprise most of China's land cover, most fire studies have focussed on fires in forests and savannas (Russell-Smith et al 2013, Laris 2021. In contrast to savannas distributed in tropical and subtropical climate zones, which have distinct dry and wet seasons, grasslands in China are mainly distributed in the temperate continental semi-arid climate zone and in the Qinghai-Tibet Plateau alpine climate zone. Moreover, alpine grasslands are found primarily in China , Zhou et al 2020. Improving our understanding of wildfires in these distinct grassland ecosystems could improve the global perspective on the characteristics of grassland fires (Russell-Smith et al 2017).
One of the important features of fire regimes is their seasonality (Johnstone 2016). The concept of fire seasonality has evolved with the increase in fire data obtained via remote sensing approaches, as has our understanding of the drivers of fire, including climate factors and human activities (Platt et al 2015). Satellite-based measurements can now provide data such as the total number of fires and the total area burned, which can be used to describe the seasonality of fires and the fire season duration (Archibald et al 2013). In addition to ignition sources such as lightning and human activity (Slocum et al 2007), fire weather indices are also used to describe the fire season (Wotton et al 1993, Lawson andArmitage 2008). Understanding fire seasonality is important for fire suppression and prevention operations, especially in highly populated countries like China (Fang et al 2021).
The Intergovernmental Panel on Climate Change recently concluded that severe fire weather is very likely to increase under the projected global warming (Arias et al 2021). The temporal and spatial patterns of the fire attributes of any regional vegetation mainly depend on weather conditions (Flannigan and Harrington 1988, Trouet et al 2009, Abatzoglou and Kolden 2013. Temperature, relative humidity, precipitation, and wind speed independently and/or synchronously contribute to an increase in both the frequency and severity of fire globally (Flannigan and Harrington 1988, Jolly et al 2015, Holz et al 2017. Yet, due to the lack of grassland fire data in China, the major drivers of grassland fires in China are still not fully understood. In this paper, we present an overview of grassland fires in China. The objectives of our study were: (a) to examine the grassland fire records of China, including their seasonality and spatiotemporal patterns and (b) to determine the major climate drivers of grassland fires in China.

Grassland fire data and grassland geographic data
We obtained grassland fire data from two sources: the grassland fire forms reported to the National Forestry and Grassland Administration (NFGA; 2008-2020) and grassland fire location data from the Wildfire Atlas of China (WFAC;2005. The NFGA forms contain information about the number of grassland fires and the total area burned for each month in each of the 14 provinces (Gansu, Heilongjiang, Hebei, Inner Mongolia, Jilin, Liaoning, Ningxia, Qinghai, Xinjiang, Shaanxi, Shandong, Shanxi, Sichuan, and Tibet) in which most grasslands are found. We also extracted fire location data for 1029 grassland fires from the WFAC data set (Fang et al 2021) as a separate source of fire numbers, fire locations, and the observation and extinguishment times of fires. Remotely sensed fire locations recorded in the WFAC are confirmed by local emergency departments and classified as agricultural fires, prescribed burning, slash-andburn activity, forest fires, industrial fires, grass fires, fires in a neighboring country, shrub fires, wasteland fires, etc. Between 2005 and 2018, there were a total of 128 548 fires in the nine categories listed above, of which 0.8% were classified as grassland fires. The top three fire classes were agricultural fires (32.9%), slashand-burn activity (29.0%), and forest fires (17.3%).
We obtained grassland distribution data from the Chinese Land Use data set (CLUDs) for the period 1980-2015 (Liu et al 2014). The CLUDs can be downloaded from the Resource and Environment Science and Data Center cloud platform (www.resdc.cn/). This data set categorizes grasslands as areas of high (>50% coverage), medium (20%-50%), or low (5%-20%) coverage.

Climate data
The main climate variables we used were the monthly temperature, precipitation, zonal (U) wind, meridional (V) wind, the wind speed without direction at a height of 10 m, and the standardized precipitation evapotranspiration index (SPEI). These variables are generally considered to be important climate drivers of grassland fires (Daubenmire 1968). Temperature, precipitation, and wind data for the same period as that of the NFGA fire data were obtained from the fifth generation ECMWF atmospheric reanalysis of the global climate (ERA5) (Hersbach et al 2020;2008-2020; the SPEI was obtained from Climatic Research Unit gridded Time Series (CRU TS) 4.04 (Harris et al 2020;2008-2020. The wind speed is the root of the sum of the u 2 and v 2 components. The ERA5 data set contains reanalyzed data at a spatial resolution of 0.25 • × 0.25 • , whereas the CRU TS 4.04 data set is generated by interpolating weather station data that have a spatial resolution of 0.5 • × 0.5 • . In addition to the above climate variables, commonly used fire weather indices (the fire weather index, the build up index, the danger risk, the drought code, the Duff moisture code, the initial fire spread index, the fine fuel moisture code, the fire daily severity rating, the fire danger index, the Keetch-Byram drought index, the spread component, the energy release component, the burning index, and the ignition component) produced by three different models developed in Canada, the United States, and Australia (https://doi.org/10.24381/cds.0e89c522) were also used (Vitolo et al 2020). The fire weather indices were calculated using weather forecasts from historical simulations provided by the European Centre for Medium-Range Weather Forecasts (ECMWF) ERA5 reanalysis (Hersbach et al 2020). We further used the monthly averages of each variable to obtain the spatial correlation between the burned areas and the fire weather and wind speed indices during April and October. Details of the fire weather indices are available in the supporting materials (see table S1).

Seasonality clusters
We used the grassland fire report data set to characterize fire seasonality in each province. In this study, we defined the fire season as the months in which the sum of the monthly percentage of the burned area was above 70%. We calculated the normalized area burned and number of fires as a time series, then clustered the time series based on the dissimilarity between the fire time series of each province (Montero and Vilar 2015). For this purpose, we performed a hierarchical cluster analysis using the time series distance and the cluster function of the R package 'TSclust' (Montero and Vilar 2015).

Climate-fire relationship
To investigate the climatic factors driving the variability of the burned area, we conducted a spatial correlation analysis of the monthly burned area in Inner Mongolia, the northern region, the southwestern region (Result 3.1) and the gridded climate variables of the corresponding month. The climate variables included the monthly zonal (U) wind, meridional (V) wind, wind speed, SPEI, temperature, and precipitation. More specifically, we calculated Pearson correlations between the monthly burned area time series and the monthly gridded climate data from January to December for the period from 2008 to 2020 using the Koninklijk Nederlands Meteorologisch Instituut (KNMI) Climate Explorer online software (https://climexp.knmi.nl; Trouet and Van Oldenborgh 2013) and R Core Team (2021). Similar analysis procedures were employed to explore the links between the monthly burned areas and the fire weather indices (Vitolo et al 2020).
To further investigate the climatic drivers for the monthly burned areas in Inner Mongolia, which accrued most of the grassland fires and experienced the largest grassland burns in China, we generated probability density functions (PDFs) of the climate variables for the three years with the highest (2013,2014,2020) and lowest (2009,2016,2019) annually burned areas, as well as for years of near-average annually burned areas (2008,2010,2011,2012,2015,2017,2018) for the region from 43 • N to 52 • N and from 110 • E to 121 • E. We selected this region based on the spatial distribution of the grass fire occurrences as well as the significance (p < 0.1) of the spatial correlation between the fires and the climate data. We extracted the climate data for this region using the R package 'raster' (Hijmans et al., 2015) and then used the 'normal' function to estimate the PDFs. We conducted all data analysis and plots using R Core Team (2021). We also calculated a two-sample t-test to explore the climatic differences between large (small) burned area years and near-average burned area years (R Core Team 2021).

Spatiotemporal patterns of grassland fires
The NFGA data reveal that 1291 grassland fires were reported in China between 2009 and 2020, accounting for 497 258.13 ha of burned area (table S2). Inner Mongolia accounts for most of the burned area in China, with 26.6% of the total number of fires and 89.2% of the burned grassland area (table S2). Other areas with high numbers of fires, but more limited burned areas, include Qinghai, Jilin, Sichuan, and Gansu (table S2,

Seasonality of grassland fires
In seven regions in northern China, including Inner Mongolia, the number of fires and the burned  October corresponds to 10.8% of the burned area. In southwestern China, there is a single major fire season in the spring that lasts longer, from February to April, accounting for 70% of the burned area.

Climate-fire relationship
We found wind speed to be the primary driver of interannual variability in the burned areas in April and October in northern China (figure S2), and specifically in Inner Mongolia (figure 2), whereas we detected no distinct climate driver in southwestern China (figure S3). In Inner Mongolia, the region representing almost 90% of the grassland burned area in China, i.e. the area burned in April (the core fire season) is significantly (p < 0.1) correlated with regional zonal winds, while the area burned in October (the secondary fire season) is significantly (p < 0.1) correlated with regional meridional winds (figure 4). We found no significant correlations for the other months outside the two main fire seasons in Inner Mongolia, or for other climate variables (temperature, precipitation, and SPEI) (figures S4-S8).
Given that more than 93% of all grassland burned in northern China occurs in Inner Mongolia (table S2), it is not surprising that we found similar fire-climate relationships for the two regions (figures S2, S9-S13). In southwestern China, on the other hand, we notably found no significant relationship between the burned area and any of the climate variables (figures S3, S14-S18).
The mean zonal and meridional winds in the high fire activity area (43 • N-52 • N, 110 • E-121 • E) of Inner Mongolia in April increase as the per-year burned areas increase from small to normal to large (figure 5). In October, the mean zonal and meridional winds are smaller in the small grassland fire years than in the years of normal and large grassland fires. The PDFs indicate a significant increase in the mean values of the zonal wind, the meridional wind, and the wind speed in April and October for the the years of small, normal, and large burned areas. Consistent trends in both April and October can be identified in the plots of U, V, and wind speed. Additionally, a two-sample t-test revealed that the values of U, V, and wind speed in April and October during years of small and large burned areas were lower and higher, respectively, than those of normal burned area years (p < 0.001) (figure 5).
The burned areas in Inner Mongolia during April and October showed stronger links with the wind speed than with the other widely used fire weather indices (figures S20-S33). In addition, there seemed to be no significant correlations between the wind speed and the other fire weather indices in Inner Mongolia (Hulun Buir and Xilingol grasslands, p > 0.1). Similar results held true for the grassland burned areas of northern China (not shown).

Discussion
In China, the largest grassland acreage (82 051 942 ha) and the highest percentage of grassland area (68.1%) occur in Tibet , yet we find that grassland fires occur infrequently in this region (figure 2). The majority (89% of area burned) of grassland fires in China occur in Inner Mongolia, the province with the second-largest grassland acreage (78 804 483 ha, or 68.81% of its total area). Grasslands in Inner Mongolia are distributed as temperate meadow steppe, temperate typical steppe, and temperate desert steppe from the northeast to the southwest . The majority of grassland fires occur in northeastern Inner Mongolia in Hulun Buir and Xilingol grassland, where typical temperate steppe dominates.
Grassland fire in northern China, which is dominated by the fires in Inner Mongolia, shows a twoseason pattern, in which the main fire season occurs in April and a secondary season occurs in October. Fire seasonality is primarily driven by a seasonal cycle of climate conditions, flammability, fuel accumulation, and human activities (Magi et al 2012). In early spring, low precipitation, drought, and high wind speeds create the conditions for high fire risk in northern China (Qu et al 2010, Zhou et al 2016. Most of the precipitation in this region occurs in the summer, leading to a new fuel load in the autumn (Qu et al 2010, Zhou et al 2016. The fire risk decrease in late autumn is attributed to a reduction in the fuel load due to mowing in early autumn, which is mainly performed for grassland restoration (Wang et al 2018). Unlike the three-season model of the savannas of southern Florida, which is influenced by an annual combination of drought, intense solar radiation, low humidity, and warm air temperatures (Platt et al 2015), grassland fire seasonality in northern China is driven primarily by high westerly wind speeds and a local drought in April and by high northerly wind speeds in October. In southwestern China, fire seasonality is more likely to be related to human activities (Fang et al 2021) than climate drivers. Moreover, the human causes of the grassland fires in the two regions are different. The burning of waste land (48.6%) and worship (15.3%) are two major causes of grassland fire in northern China, while domestic cooking (21.2%), juveniles playing with fire (18.5%), and worship (14.5%) are three major causes of grassland fires in southwestern China (figure S19). The grass type and topographies are different between the two regions. The grassland in northern China consists of typical temperate steppe in Hulun Buir and Xilingol in the Mongolian Plateau at an elevation of roughly 1000-1500 meters, while the grassland in southwestern China (Sichuan province) is mainly alpine meadows in the Hengduan mountains, which consist of many mountain ranges, most of which run roughly north-south and range from 1300 to 6000 meters .
Wind is generally considered to be a major driver of grassland fires, because wind speeds influence the quantity of biomass consumed and decrease the fuel moisture by increasing evapotranspiration (Beer 1991). Large, uncontrolled fires are usually driven by a combination of high wind speeds and drought conditions (Keeley et al 2011). Unlike strong Foehn winds or topography-driven downslope winds, the westerly winds cross the Hulun Buir grassland at elevations ranging from 650 m to 700 m before they reach the Daxing'anling mountains. Only then are the winds slowed or stopped by a wall of mountains. The large annual mean wind speed distribution in northern China increases from northeastern Inner Mongolia to Xinjiang province and the Tibetan Plateau and ranges from 2.0 to 6.3 m s −1 ; the weakest annual mean wind speed is in the Sichuan Basin (Guo et al 2011). The interannual seasonality of the wind speed means that the largest wind speeds are observed in the spring, while consistently low wind speeds are observed in the autumn, winter, and summer (Guo et al 2011, Zhang andWang 2020). The wind directions in the four seasons are roughly similar in China, with the exception of the Tibetan plateau in the summer (Zhang et al 2019). The high wind speed season in northern China overlaps with the season of high grassland fire activity.
The fire weather indices showed weaker links with burned areas than the wind speed. A possible reason for this could be that the fire weather indices include  (2010,2018,2020), seven normal-fire years (2008,2009,2011,2013,2014,2016,2017) and three large-fire years (2012,2015,2019) were selected for Inner Mongolia. The region identified as having a high incidence of grassland fires (43 • N-52 • N, 110 • E-121 • E) corresponds to a high density of grassland fire locations. many climate variables that obscure the importance of the main factor responsible for grassland fires, i.e. wind speed, in this study in China. In particular, drought metrics such as the drought code and the Keetch-Byram drought index, which are included in the fire weather indices, had only a low correlation with the burned areas. The main climate force that drives fires is the wind speed in the Inner Mongolian grasslands, while the dominant climate driver is drought in nearby vegetation in northeastern China (Yao et al 2017), Mongolia (Hessl et al 2016), and southeastern Siberia (Wang et al 2021). Another explanation for the low correlation between the fire weather indices and the burned area could be that the widely used fire weather indices may perform better in forests than in grasslands. The global pattern of correlation between the monthly fire weather index (FWI) and the burned area shows that the correlation in grassland regions was lower than in forest regions (Jones et al 2022). Moreover, FWIs have been widely used to characterize forest fires in the Daxing'anling mountains near the Hulun Buir and Xilingol grasslands (Tian et al 2011).
In this study, we focused on the top-down controls (weather and climate) responsible for grassland fires rather than the bottom-up controls (fuel and topography), despite the importance of fuel, topography, and socio-ecological systems for fires (Liu et al 2013). Population density and human activities such as anthropogenic fires, fire detection, and firefighting responses affect not only the incidence of fires but also their spread (Knorr et al 2014, Lasslop and Kloster 2017, Abatzoglou et al 2018, Wu et al 2022. The population density is much lower in Inner Mongolia (0.19 people per ha) than in Sichuan, southwest China (1.73 people per ha). However, the burned area in Inner Mongolia is much greater than that in southwest China, which could be a result of the longer time taken to notice and combat fires. Our study emphasizes the climate forces responsible for the grass fires in Inner Mongolia, which represent 89% of the grass fires in China, while the drivers of the grass fires in southeast China still need further investigation.

Conclusions
In this study, we used a new and robust grassland fire record data set obtained from the NFGA for the period from 2008 to 2020 and a grassland fire location data set obtained from the NFAC to examine the spatiotemporal patterns and fire seasonality in China for the period from 2005 to 2018. The spatiotemporal pattern of grassland fires in China highlights that Inner Mongolia in northern China, specifically, the Hulun Buir and Xilingol grasslands, is a region of severe fires. The distinct differences in fire seasonality and grassland fire drivers between northern China and southwestern China provide a guide for fire prevention in China. Our study fills a gap in the scientific knowledge about grassland fires in China and provides an insight into the forces responsible for grassland fires as well as support for decision-making in fire prevention and fire management. There are 3264 professional forest firefighting teams nationwide in China. The fire season identified in this study provides a more precise description of fire seasonality and fire danger regions, leading to better performances in fire prevention, suppression, and prescribed burning.

Data availability statement
All data that support the findings of this study are included within the article (and any supplementary files).