Construction and assessment of a fire risk index system for typical grasslands in Xinjiang, China

Abstract

Background

Fire hazards have a substantial impact on grassland ecosystems, and they are becoming more frequent and widespread because of global changes and human activities. However, there is still a lack of a widely accepted or practical method to evaluate grassland fire risk. In our study of typical grasslands in northern Xinjiang, we selected 18 evaluation indicators for grassland fires from three aspects of hazard, exposure, and vulnerability. Employing the analytic hierarchy process, weighted comprehensive evaluation method, and standard deviation classification, we determined the fire risk level thresholds, aiming to develop efficient and precise methods for assessing grassland fire risks, and ultimately created a grid-based map of grassland fire risk levels.

Results

The risk level of grassland fires is determined by the combined spatial heterogeneity of fire-causing factors’ hazard and fire hazard-bearing bodies’ vulnerability and exposure. The hazard of grassland fire and fire hazard-bearing bodies’ vulnerability and exposure are dominated by medium level and medium–low level. Most areas of grassland fire risk levels are medium–low, medium, or medium–high risk, with few areas being high risk or low risk. The grassland fire risk exhibits a spatial distribution characterized by higher risks in the western and lower in the eastern; high and medium–high risk areas are primarily distributed in the western and some northeastern regions of the study area. The simulate result effectively represents the spatial distribution of grassland fire in the research area.

Conclusion

We established a grassland fire risk index system and model, creating a spatial distribution map of grassland fire risk levels based on grid. Few grassland areas have fire risks and show a patchy distribution. The results generally reflect the spatial distribution pattern of grassland fire risks in the study area. This research provides technical support for scientifically formulating local grassland fire disaster prevention and relief strategies.

Resumen

Antecedentes

Los riesgos de incendios tienen un impacto substancial en ecosistemas de pastizales, y están siendo más frecuentes y extensos debido al cambio Climático Global y a las actividades humanas. Sin embargo, todavía faltan métodos prácticos que sean ampliamente aceptados para evaluar el riesgo de incendios en pastizales. En nuestro estudio de pastizales típicos en el norte de Xinjiang, China, seleccionamos 18 indicadores de evaluación de incendios de pastizales desde tres aspectos: de riesgo, de exposición, y de vulnerabilidad. Empleando el proceso de Análisis Jerárquico, el Método de Evaluación Comprensiva Ponderada (Weighted Comprehensive Evaluation Method), y la clasificación de la desviación estándar, determinamos los límites del nivel de riesgo, tendiente a desarrollar métodos eficientes y precisos para determinar el riesgo de incendios en pastizales, y finalmente, crear un mapa basado en grillas donde se muestren estos niveles de riesgo.

Resultados

El nivel de riesgo de incendios en pastizales está determinado por la combinación de la heterogeneidad espacial de los factores causantes del riesgo y la vulnerabilidad y exposición de los portadores-conductores de esos riesgos. El riesgo de incendios de pastizales y la vulnerabilidad y exposición de sus conductores son dominados por niveles medios y medios-bajos de esos riesgos. En la mayoría de las áreas, los niveles de riesgo de incendios son medio-bajos, medios y medio-altos, con muy pocas áreas de riesgo alto, o bajo. El riesgo de incendios de pastizales exhibe una distribución espacial caracterizada por alto riesgo en el Oeste y bajo riesgo en el Este, y áreas de riesgo alto y medio-alto están primariamente distribuidas en el Oeste y en algunas regiones del Noreste del área de estudio. El estudio de simulación representa efectivamente la distribución espacial de fuegos de pastizales en el área estudiada.

Conclusiones

Establecimos un Sistema y Modelo de Índice de Riesgo para pastizales, creando un mapa de distribución espacial de niveles de riesgo basado en una grilla. Pocas áreas presentan riesgo de incendios y muestran una distribución en parches. Los resultados reflejan de manera general el patrón de distribución de los riesgos de incendio de pastizales en el área de estudio. Esta investigación provee de soporte técnico para formular estrategias de prevención y alivio en el caso de desastres por incendios.

Introduction

Grassland ecosystems are important part of terrestrial ecosystems, playing an important role in maintaining biodiversity, regulating the global carbon cycle and climate. They also serve as essential resources for husbandry and economic development in relatively remote minority regions (Liang et al. 2012; Zhou et al. 2024a, b). Although the importance of grasslands is recognized worldwide, and various protective measures have been taken, managing and conserving grasslands remains a significant challenge (Hu and Nacun 2018; Leys et al. 2018). Wildfires are among the most frequent natural disasters in grasslands, with destructive power and substantial impact on grassland ecosystems (Dixon et al. 2022), causing significant economic losses to nations. The release of substantial amounts of greenhouse gasses and aerosols from grassland fires can alter the chemical composition of the Earth’s atmosphere, leading to air pollution and even global changes. Additionally, sustained fires directly impact the carbon cycle and carbon balance (Veldman et al. 2015; Zheng et al. 2023). In particular, extreme fire behaviors like smoldering combustion, once they occur, can cause severe ecological damage and even lead to explosive combustion (Watts and Kobziar 2013). Admittedly though, grassland fires, although destructive in some regions, can also be beneficial in others, for instance, improving vegetation productivity, releasing nutrients from litter, and sustaining the ecosystem’s stability and diversity (Dickinson and Ryan 2010).

According to the statistics, annually 80% of the world’s fires occur in grasslands (Leys et al. 2018). China is significantly affected by grassland fires, with one-third of its 400 million hm² of grasslands being prone to fires, one-sixth of the area frequently experiences fires, especially during the dry and windy conditions in the spring and autumn seasons (Wei et al. 2020). In recent years, China has implemented a series of grassland conservation projects, leading to effective vegetation recovery and a notable increase in fuel load. Concurrently, as the global climate warms (Brown et al. 2023; Jones et al. 2022), the risk level of grassland fires is escalating, showing a new trend of extending from primarily occurring in spring and autumn to all year round (Zhang et al. 2015). Along with the accelerated economic and social development in pastoral regions, the increase in facilities and equipment on grasslands also leads to escalating direct economic losses caused by grassland fires (Jiang et al. 2018). With the longest land border of any province in China, Xinjiang borders many countries, further increasing the fire risk in this province. Grassland fires in Xinjiang could destroy the productivity of these areas, severely impacting the development of grassland animal husbandry and the sustainable income growth of herders (Fang et al. 2021).

Grassland fires are influenced by multiple factors linked to the hazard of fire-causing factors and the exposure and vulnerability of fire hazard-bearing bodies (Appendix A). The relationships between these factors are complex. Increased temperatures, for example, can enhance evaporation of moisture from dead plants in grassland surface, heightening fuel’s temperature and flammability, thus raising fire occurrence probabilities (Chuvieco et al. 2004). At the same time, in arid and semi-arid regions, rising temperatures may decrease combustibles, reducing fire risks. Therefore, the formation mechanism of grassland fire is unclear, and quantitatively assessing these risks has been a long-standing challenge (Palaiologou et al. 2020; Soubry et al. 2021). In practical scenarios, acquiring the numerous factors for evaluation is challenging, and in many remote areas, the absence of data makes it even more difficult to assess grassland fire risks. As a result, the timely assessment and updating of grassland fire risks have emerged as key issues in grassland fire research (Wang et al. 2023).

Presently, the common methods for grassland fire assessment mainly contain the single-factor method and the composite index method (Liu et al. 2019). The single-factor method utilizes one crucial factor that impacts grassland fires, for example, the moisture content of fuels or drought index, to assess the likelihood of grassland fire occurrence (Dennison et al. 2003; Schunk et al. 2017). This method does not account for the influence of various risk factors. Subsequently, the composite index method was proposed, providing a good approach for fire risk assessment. The elements of grassland fire risk assessment include environmental factors (ignition sources, fuels, meteorological conditions, and terrain), characteristics of combustibles (moisture content of fuels, fuel load capacity, and fuel continuity), and human factors (population density, distance from roads, distance from settlements). Notable progress has been made in fire risk assessment based on multiple factors (Michael et al. 2021; Mishra et al. 2024; Sari 2021). Michael et al. (2021) used the AHP for determining the weights of indices and integrated a fuzzy weighted sum model for prioritizing fire risks in fuzzy sets, ultimately producing a fire risk map that combines decision analysis methods with fuzzy functions.

Presently, the Keetch-Byram Drought Index (KBDI) and Grassland Fire Danger Index (GFDI) are utilized to build a grassland fire risk assessment system. Numerous studies have been conducted to assess grassland fire risks in different areas (Khastagir et al. 2018; Laneve et al. 2020), but the selections of indices are relatively limited, omitting some factors that may cause grassland fires, such as fuel types, the amount of flammable houses and fire-fighting equipment, population, and GDP. Therefore, the existing selections of grassland fire risk assessment indices are still incomplete. Research on quantitative assessment of grassland fires is insufficient, and the precision of these risk assessments requires further improvement. The grassland fire index system is incomplete, currently emphasizing only on hazard assessments while overlooking attributes of the fire hazard-bearing bodies themselves, as well as social and human influences. Additionally, in practical scenarios, the multiple factors utilized in the evaluation are difficult to acquire. In many remote areas, because of the lack of grassland fire indicator data, it is often more challenging to assess fire risks. Therefore, the quickly assessment and updating grassland fire risks have become a popular topic current research (Liu et al. 2019; Wang et al. 2023). The advancement of satellite monitoring technologies and their application in grassland fire surveillance in recent years have significantly improved the capabilities for monitoring, early warning, and assessment of grassland fires, driving the evolution of grassland fire management into a wider scope (Pragya et al. 2023; Sharma et al. 2022). However, grassland fire risk assessment and early warning research involves theories and methods from interdisciplinary field including grassland science, geographic information science, disaster science, ecology, and climatology, necessitating a comprehensive approach that integrates multiple elements for assessing grassland fire risks (Oliveira et al. 2018).

In summary, there remains scope for further research in grassland fire risk assessment. Currently, frequent extreme weather events and intensified human activities have a significant influence on the management of local grassland husbandry production. In this study, we chose a typical grassland in northern Xinjiang as the research area. Utilizing factors categorized into hazard, exposure, and vulnerability, and based on interdisciplinary theories and methods from remote sensing technology, ecology, grassland science, husbandry, and disaster science, we developed a grassland fire risk assessment system and conducted grid-based grassland fire risk assessments, aiming to offer a scientific and technical support for management decisions and sustainable grassland development.

Materials and methods

Study area

The study area is located in the northern part of the Xinjiang Uygur Autonomous Region, China, between 82.989438°–86.424817° E and 45.942742° ~ 48.066148° N (Fig. 1). The terrain is higher in the west and lower in the east, with considerable topographic undulations. The study area has a temperate continental climate with abundant sunshine (annual sunshine duration of 2500–3500 h) and a dry climate. The average temperature ranges from − 4 to 9 °C annually, with January (or February in mountainous areas) being the coldest month and July the hottest. The average annual precipitation does not exceed 200 mm. The study area has 67,644.06 hm² of grassland, with a grassland coverage rate of 77.89%. There are various types of vegetation in the study area. According to the classification standard and system of grassland types in China, the grasslands in the area can be divided into temperate desert, temperate steppe desert, lowland meadow, mountain meadow, alpine meadow, temperate steppe, temperate meadow steppe, and temperate desert steppe (Zhang et al. 2019).

Fig. 1

Overview of the study area showing county boundaries

Data sources

Grassland field survey data

The field survey was conducted during the vegetation senescence season from September to October 2021, and 326 sample points were collected (Table 1). In sample points, the quadrats were laid out using the nine-point method (Fig. 2). Three sampling transects, each with a length of 100 m, were set at 120° angles and were centered at the center point of each sample plot. The herb and dwarf shrub quadrats (9 quadrats) were placed at 30 m, 60 m, and 90 m along each transect from the center point. The shrub and tall herb quadrats (3 quadrats) were set up along each sampling line at a distance of 50 m from the center point, and three sub-quadrats were set up within each quadrat.

Table 1 Distribution of the vegetation types and the actual sampling sites in the study area

Fig. 2

Schematic diagram of sampling points

• (1) Above-ground fuel dry weight collection.

All the above-ground parts of herbs in each quadrat were harvested by flush mowing (Zhang et al. 2023), and their fresh weight was measured to an accuracy of 10 g. After the harvested plants were thoroughly mixed, all or a certain proportion of them were packed into sample bags, brought back to the labs, and dried in an oven at 65 °C for 48 h until constant weight to obtain the dry weight of grassland above-ground fuel. Then, the average dry weight of all quadrats was used to represent the dry weight of above-ground fuel for the entire sample point.

• (2) Surface litter dry weight collection.

All litter within the sample plots was collected using a rake and its fresh weight was measured. For each plot, after thoroughly mixing the litter, all or a portion of it was placed in sample bags, brought back to the labs, and dried in an oven at 65 °C for 48 h until constant weight to obtain the dry weight of surface litter fuel load.

Above-ground fuel load data

Based on MOD09GA_GQ data (GS image fusion of MOD09GA and MOD09GQ data), along with topography, meteorology and soil, and 48 other factors, the Boruta variable screening method was used to screen six factors, GNDVI, Longitude, Prep_8-9 (cumulative precipitation for the first 2 months of August–September), Clay_1 (clay content from 0 to 30 cm depth), DEM, and Temp_9 (average temperature in September), which were ranked high in importance, as independent variables, and above-ground fuel dry weight of 326 sample points surveyed on the ground as the dependent variable; the random forest (RF) algorithm in machine learning method was utilized to construct an inversion model for grassland above-ground fuel load (Zhou, Zhang, et al. 2024) and to produce a spatial distribution map of grassland above-ground fuel load.

Surface litter fuel load data

The average weight of surface litter in all plots of the same grassland type was calculated as the surface litter fuel load of this type of grassland.

Meteorological data acquisition and processing

Meteorological data were obtained from the China’s National Earth System Science Data Center (http://www.geodata.cn/data/). DEM data with a spatial resolution of 500 m were used as the co-data, and a gridded meteorological dataset was generated for the study area through spatial interpolation by Anusplina software. The data mainly included average month wind speed, average monthly rainfall, average monthly temperature, maximum monthly temperature, and minimum monthly relative humidity.

Field ignition sources investigation

The information on sources of wildfire occurring within the study area in recent years was obtained using diverse methods such as data sharing, data collection, consulting with fire control experts from local grassland fire prevention and disaster management departments, inquiry survey, field investigation, and information reporting. All data were sorted and summarized to form a regional wildfire source distribution map.

Data on fire hazard-bearing bodies

The distribution data of grassland resources, population, economic data, buildings, and fire prevention facilities in the study area were obtained from the standard gridded data shared by the Xinjiang Bureau of Statistics, with a grid cell of 1000 m × 1000 m. The data collected mainly included exposure data (above-ground biomass, number of buildings and fire prevention facilities, population, and GDP) and vulnerability data (proportion of combustible grassland area, proportion of combustible buildings, proportion of elderly and young human populations, and vulnerability of economic activities). Using inversion and utilizing data on exposure, vulnerability, and other factors, we developed a geographic information database detailing the quantity and spatial distribution of fire hazard-bearing bodies.

Historical fire location data

Investigated data on grassland fires that have occurred in recent years include grassland fire archive data and grassland fire statistical data. The data is sourced from the China Forest and Grassland Fire Prevention and Control website (https://slcyfh.mem.gov.cn/), the Global Fire Atlas Data website (https://daac.ornl.gov/CMS), and the grassland fire prevention offices and archives of various counties and cities in the study area. The collected data was compiled and summarized, resulting in a total of 172 grassland fire incident records. This dataset covers the total number of grassland fires that occurred annually in the study area from 2003 to 2022, the number of grassland fires of different levels, and detailed information such as relevant latitude and longitude coordinates.

Fire product data

MCD64A1 burned area data were obtained from the National Aeronautics and Space Administration (NASA) Earth Observing System Data and Information System (EOSDIS, https://ladsweb.modaps.eosdis.nasa.gov/search/, accessed on 18 December 2023), with a spatial resolution of 500 m, the time series from 2000 to 2021. The MCD64A1 burned area dataset records information such as the location of the fire, start date, end date, burned area, and confidence level, with the fire date represented in Julian days. The data preprocessing includes the following: (1) Using the MODIS Reprojection Tool (MRT) software, projection transformation (using WGS84 ellipsoid) and mosaicking preprocessing were performed. (2) On the basis of MCD64A1, we extracted burned area data at all confidence levels.

Research methods

Hazard model

Grassland fire hazard refers to the hazard posed by factors that cause grassland fires and includes their different intensities and occurrence probabilities. These factors include but not limited to fuels, meteorological conditions, and wildfire ignition sources. In this study, considering the regional environment, ten hazard indices affecting grassland fires, including grassland above-ground fuel load, were selected to build a grassland fire hazard assessment model, which assesses and estimates the likelihood and intensity of potential grassland fires (Table 2).

Table 2 Hazard indices and their abbreviations

Exposure model

Exposure of fire hazard-bearing bodies indicates the susceptibility of various entities like people, buildings, and properties to be affected by grassland fires. The greater the number of people and properties exposed to various hazards in a region, the higher the value density of assets at risk, increasing the potential for loss and elevating disaster risk (Lecina-Diaz et al. 2021). In this study, three social exposure and one grassland resource exposure indices were selected to construct an exposure assessment model for fire hazard-bearing bodies (Table 3).

Table 3 Exposure indices and weights of fire hazard-bearing body

Vulnerability model

Grassland fire vulnerability of fire hazard-bearing bodies refers to their sensitivity to grassland fire damage under certain natural or socio-economic conditions. It reflects the recovery ability of the burned areas and how well the local residents cope with and adapt to the threat of fires (Miller and Ager 2013). The vulnerability indices include many variables discussed in other studies, such as flammable grassland area, housing construction, and GDP. Based on the conditions of life and social development in the northwestern part of Xinjiang, this study selected four vulnerability indices to construct a vulnerability assessment model for fire hazard-bearing bodies (Table 4).

Table 4 Vulnerability indices and weights of fire hazard-bearing body

Grassland fire risk assessment method

Grassland fire risk assessment is performed to evaluate and estimate the likelihood of grassland fires and the resulting losses using a risk ranking method. The fire risk assessment is based on regional disaster system theory. According to this theory, disaster risk is the outcome of the combined impact of hazard (H), exposure of fire hazard-bearing bodies (E), and vulnerability of hazard-bearing bodies (V), with each factor being equally important in risk modeling (Shi 2019). Following the aforementioned analysis, the formula for calculating grassland fire risk is defined as:

$$R=\sqrt[3]{H\times E\times V}$$

(1)

where R is the grassland fire risk index, H is the grassland fire hazard index, E is the exposure index of the grassland fire hazard-bearing bodies, and V is the vulnerability index of the fire hazard-bearing bodies.

AHP

The analytic hierarchy process (AHP) is a qualitative and quantitative analysis method commonly used to solve evaluation problems (Sivrikaya and Küçük 2022). Based on the selected of hazard, exposure, and vulnerability indicators, in this study, the AHP was utilized to calculate the weight values of each indicator. The process generally includes four steps:

• (1) Building a recursive hierarchical model

• (2) Construction of a comparison judgment matrix

• (3) Calculation of the relative and combined weights of each comparison factor

• (4) Testing for consistency

The pairwise comparison matrix was constructed by the nine-scale method, and the weight of each index was calculated as follows:

$${M}_{i}={\left(\prod_{i=1}^{n}{a}_{ij}\right)}^{1/n}$$

(2)

$${w}_{i}=\frac{{M}_{i}}{\sum_{i=1}^{n}{M}_{i}}$$

(3)

where ${a}_{ij}$ is the element in the ith row and the jth column in the decision matrix, n is the order of the decision matrix, ${M}_{i}$ is the geometric mean of the elements in the ith row, and ${w}_{i}$ is the weight of each influencing factor.

To test the consistency of the comparison matrix, the consistency index and ratio were calculated as follows:

$${\text{CI}}=\frac{{\lambda }_{\text{max}}-n}{n-1}$$

(4)

where ${\lambda }_{\text{max}}$ represents the largest eigenvalue in the matrix and CI is the consistency index. Additionally,

$$\text{CR}=\frac{\text{CI}}{\textrm{RI}}$$

(5)

where RI is the random consistency index and CR is the consistency ratio. When CR < 0.1, the comparison matrix satisfies the consistency requirement.

Normalization processing method

For comparisons, the data were normalized, with the processed values ranging between [0,1].

Weighted comprehensive evaluation method

Because different indices have different normalized values, there are differences in the extent to which different indices affect the evaluated object. Calculate by multiplication of the normalized values of different indices by their corresponding weights. The index values in this study are calculated as follows:

$$C=\sum_{i=1}^{n}{W}_{i}\times {X}_{i}$$

(6)

where C is the index value; ${X}_{i}$ is the normalized value of the ith index; ${W}_{i}$ is the weight of the ith index; and n is the number of indices.

Standard deviation classification

The method is based on the mean (μ) and standard deviation (σ) of the sample set, using the mean ± various multiples of standard deviation (σ) as boundary values for classifying the sample set (Chan 2014). This study employs μ − σ, μ, μ + σ, and μ + 2σ as boundary values, dividing hazard, exposure, vulnerability, and grassland fire risk into five levels (Table 5).

Table 5 Intervals determined by standard deviation classification

Model validation

In this study, three validation methods were utilized to verify the accuracy of the grassland fire risk map. (1) Overlay the MCD64A1 burned area data with the grassland fire risk map to analyze the overlap between burned areas and different fire risk level areas. (2) Overlay the historical fire location data with the grassland fire risk map to analyze the overlap between historical fire locations and different fire risk level areas. (3) Using ArcGIS software to randomly produce 172 random points, matching the number of historical fire location in the study area. Grassland fire locations are represented by the number 1, and locations without recorded ignition or fire by 0. The fire risk values corresponding to these points were extracted using the sample tool in ArcGIS software. Using SPSS software, an ROC curve was plotted, and the model’s evaluation accuracy was analyzed based on the ROC curve (AUC) (Ju et al. 2023).

Result

Determination of index weights

Figure 3 shows the calculation result of the weights of each index of the hazard, exposure, and vulnerability models. The results of hazard model show that the weight of FL is the highest (0.20). This is followed by LF (0.18), FR (0.12), and DS (0.12). Conversely, AS has a relatively small impact on igniting grassland fires, with the lowest weight (0.05). For exposure model, the highest weight is given to AB (0.44), and the lowest to GDP (0.15). In terms of vulnerability model, the highest weight is for PA (0.46), and the lowest is for VA (0.14).

Fig. 3

Grassland fire assessment index system and the weights of each index

The consistency test results show that the CR for hazard (CR = 0.0080), exposure (CR = 0.0172), and vulnerability (CR = 0.0172) are all less than 0.05, indicating that the judgment matrices meet the consistency requirements.

Grassland fire risk assessment

Hazard, exposure, and vulnerability analysis

In the study area, the grassland fire hazard levels vary greatly and exhibit a patchy distribution (Fig. 4a). Medium–low hazard levels are the most widely distributed in the study area, accounting for 46.38% of the total area, mainly in the central and northeastern regions of the study area. This is followed by a medium hazard level (21.18%) and medium–high hazard level (15.42%). High hazard areas are least distributed, covering only 3.21% of the study area, and mainly distributed in the northeastern and some western regions of the study area. The exposure level of fire hazard-bearing bodies overall decreases from the northwestern and central-eastern parts outwards (Fig. 4b). Grids with high exposure are primarily located in the northwestern and central-eastern regions, with scattered distribution in the northeastern part, comprising 3.57% of the study area. Grids with medium high exposure are mainly found in the western and northeastern parts, accounting for 7.43% of the study area. Medium exposure grids are predominantly situated in the northeastern and western regions, having the largest area share of 34.60%. Medium low exposure grids are mainly located in the central and southwestern areas, with some in the northeastern region, making up 32.28% of the study area. Grids with low exposure are primarily found in the central and southeastern regions, with scattered distribution in the western and northeastern areas, comprising 22.12% of the study area. High vulnerability grids for fire hazard-bearing bodies are located in the central area between the northeastern and western parts (Fig. 4c), making up 5.99% of the study area. Medium vulnerability grids are distributed throughout the study area and have the largest area share of 52.57%. Medium low vulnerability grids are mainly located in the southern and southeastern parts, with a few in the northeastern part, covering 31.83% of the study area.

Fig. 4

Grassland fire a hazard, b exposure, c vulnerability, and d grassland fire risk

Grassland fire risk analysis

Based on a comprehensive analysis of the three elements of hazard (H), exposure of fire hazard-bearing bodies (E), and vulnerability of fire hazard-bearing bodies (V), the spatial pattern of grassland fire risk was determined (Fig. 4d). The area with a medium–low level of grassland fire risk is the most extensive in the study area, accounting for 39.25% of the total area, primarily distributed in the central and southeastern regions of the study area. The next largest category is the medium risk area, covering 35.83% of the study area, mainly located in the western and northeastern parts. Next are the medium–high risk areas, accounting for 10.67%. The high risk (3.74%) areas have minimal distribution in the study area. High and medium–high risk areas are primarily block-like pattern in the western and central-eastern parts and in dot-like pattern in the northeastern region.

Validation of the analytical accuracy

The accuracy of the simulated grassland fire risk results was estimated using three validation methods, including calculating the ROC curve and overlay analysis of burned area data, historical fire data with grassland fire risk maps. Using burned area data to verify the simulation accuracy showed that the rate of overlap of high and medium–high fire risk areas with burned areas was 75.63%, while the rate of overlap of low and medium low fire risk areas with burned areas was 14.94% (Fig. 5a). Using historical fire location data to verify the simulation accuracy showed that the rate of overlap of high and medium–high fire risk areas with historical fire location was 54.65%, while the rate of overlap of low and medium low fire risk areas with historical fire location was 21.51% (Fig. 5b). When the ROC curve (AUC value) was used to validate the simulation accuracy (Fig. 6), AUC = 0.712.

Fig. 5

The overlap between the a burned area data, b historical fire location data, and the grassland fire risk map

Fig. 6

ROC curve

Considering all three results, the simulated grassland fire risk results can adequately represent the grassland fire situations in the study area.

Discussion

Selection of the grassland fire risk index system

Grassland fire is a complex system. The accuracy of its assessment and prediction relies on the model’s reliability, the completeness of indices, and the support from remote sensing and geographic information systems (Morgan et al. 2014). Constructing a scientifically sound fire risk index system is crucial (Gong et al. 2023). Too few key indices can lead to inaccuracies in fire risk inversion models due to incomplete considerations, while too many can cause multicollinearity of input variables and model overfitting, along with difficulties in efficiently collecting and organizing data (Sari 2021). Previous fire risk assessments often focused only on a narrow range of factors (Snyder et al. 2006). Researchers typically used weather indices or historic ignition data to assess the hazard of grassland fires (DaCamara et al. 2014; Liu et al. 2010), overlooking the attributes of the fire hazard-bearing bodies and socio-economic status (Wang et al. 2013). This resulted in inadequately developed index systems and overly simplistic selection of factors (Naderpour, Rizeei, and Ramezani 2021). Differing from past studies, this research took into account a broader array of factors that could cause grassland fires when selecting indices for the hazard of fire-causing factors. In addition to fuels (above-ground fuel load, surface litter fuel load) and meteorological conditions (average monthly wind speed, average monthly temperature, maximum monthly temperature, average monthly rainfall, and minimum monthly relative humidity), it also detailed data on wildfire ignition sources (fire frequency, density of important ignition sources, density of population with no civil capacity and limited civil capacity). Furthermore, this study took a comprehensive approach by considering the sensitivity and exposure of fire hazard-bearing bodies to grassland fire, choosing multiple indices like above-ground live biomass, the number of buildings and fire prevention facilities, the proportion of combustible grassland area, and the proportion of combustible buildings. This makes the index system more complete and easier to assess and understand, offering a very broad application prospect. Given the variability in different regional environments, future research should consider additional more potential factors into the grassland fire risk assessment index system, especially those related to disaster prevention and mitigation capabilities, such as professional fire-fighting teams, fire material reserves, fire barrier systems, and fire department access roads, to make the grassland fire assessment index system more comprehensive and practical.

Factors influencing grassland fire risk and countermeasures

The hazard distribution map indicates that the northeastern and some western regions of the study area have a higher potential for grassland fires. Compared to other areas, high hazard regions have higher average monthly temperature and maximum monthly temperature, and the rainfall is low, which increases the temperature and flammability of the fuel, thereby creating conditions favorable for grassland fire ignition and spread (Chuvieco et al. 2004; Heyerdahl et al. 2008). Additionally, fires that occur during extreme weather situations, for example, high temperatures and persistent heat, are harder to control. Once grassland fires occur in these regions, they will spread rapidly to the surrounding areas, leading to large-scale fire events. This finding is similar to previous research on the climate-driven factors of fire spread (Hantson et al. 2022). The exposure distribution map shows that the exposure of fire hazard-bearing bodies to grassland fires in the study area is primarily impacted by the population density. Grids with high exposure are concentrated in densely populated areas of the northwest and central-eastern regions. Compared to other areas, regions with higher population densities experience greater exposure due to more frequent human activities. Notably, the GDP in densely populated areas is also significantly higher than in other regions, further increasing the level of exposure (Chuvieco and Justice 2010). The vulnerability distribution map shows that the vulnerability to grassland fires in the study area is mainly impacted by the types of vegetation and buildings, with grids of high vulnerability mostly situated on meadow grasslands. In contrast to desert regions, meadow grassland areas boast higher grassland productivity and can sustain more livestock. Fires in such areas have a greater impact on grassland husbandry (Boisramé et al. 2017). Additionally, in areas with high grassland productivity, where grazing is dense, herders’ temporary shelters are often built with flammable materials. Should a grassland fire occur, the potential damage in these areas could be substantially greater than in other regions.

The high-risk areas of grassland fires are mainly plate-shaped and distributed in the western and central-eastern regions of the study area, indicating a higher likelihood of future grassland fires in these areas. These areas overlap somewhat with portions of high hazard and exposure areas, where precipitation is relatively abundant and grassland vegetation growth conditions are favorable, providing ample fuel for grassland fires (Chuvieco et al. 2004). Additionally, the density and frequent human activities (such as burning paper for tomb sweeping and smoking) in the densely populated areas of the western and central-eastern regions provide more opportunities for potential arson incidents (Chang et al. 2023); the synergy of these factors increases the potential likelihood of grassland fires. In these areas, it is crucial to improve the control of human-caused fires, increase awareness of fire safety, and promptly remove combustible materials from the grasslands. The western and central-eastern areas, being at the intersection of agriculture and pastoralism, have complex and high exposure of fire hazard-bearing bodies, resulting in high grassland fire risk. Special attention to grassland fires is needed in these areas, along with effective measures to reduce the likelihood of such fires, including establishing early fire detection systems, formulating policies for population and vegetation management, and enhancing fireproofing technologies in buildings. In the central region, where high hazard areas are distributed in a block-like pattern, but the overall grassland fire risk is not considered high, suggesting that the exposure and vulnerability of fire hazard-bearing bodies in these areas are relatively low. This further indicates that grassland fires are a complex and dynamic process, influenced by factors such as hazard, exposure, and vulnerability during their development into disasters (Podschwit et al. 2022).

This research employed dataset of MCD64A1 data, grassland fire locations data, and ROC curve for accuracy validation. While the validation results of grassland fire risk levels demonstrate considerable feasibility, there is still significant room for improvement. A part of the grassland fire locations is distributed in medium risk areas, as the recorded ignition or fire only indicates the location and time of ignition on grasslands, without considering the extent of post-fire losses. For instance, certain grassland areas with lower vulnerability and exposure experience lesser damage from fires, thus falling into the medium risk category. It is necessary for future research to include more comprehensive historical data, including economic damages and casualties caused by fires, to validate grassland fire risk assessments more effectively. Additionally, the study area is located in northern Xinjiang, where several counties and cities border other countries (Hao and Liu 2012). Fires originating outside these borders and spreading into China could lead to large-scale grassland fires, affecting fire risk assessments. This study did not consider the impact of external fire sources. Thus, one of the critical challenges for future research is to identify, describe, and quantify the sources of uncertainty in fire risk research.

Conclusion

The result showed that the risk level of grassland fires is determined by the combined spatial heterogeneity of fire-causing factors’ hazard and fire hazard-bearing bodies’ vulnerability and exposure. Few grassland areas have higher grassland fire risk levels, mainly concentrated in the western part of the study area, with most areas having a medium–low level of grassland fire risk. Specifically, the risk of grassland fires is higher in the western and certain northeastern regions of the study area, whereas the southeastern and central-western regions have a lower potential for grassland fires. The most significant contributions to grassland fire risk come from factors like grassland above-ground fuel load, surface litter fuel load, above-ground live biomass, and the proportion of combustible grassland area. These indices relate to the growth status and flammability of grasslands. Measures like timely mowing and medium grazing in the northwestern and northeastern areas with better grassland growth and higher flammability can help reduce the fuel load and thus decrease the incidence of grassland fires. The population and the number of buildings and fire prevention facilities also significantly contribute to the risk level of grassland fires. In densely populated western and central-eastern areas, management and awareness of human-caused fires should be strengthened, such as establishing early fire monitoring systems and formulating population management and fire prevention policies. The grassland fire risk grid levels developed in this study effectively represent the local grassland fire scenarios, aiding in a detailed understanding of grassland fires in the study area and providing guidance for the rational distribution of resources required for grassland fire management. In future grassland fire risk studies, machine learning algorithms, multi-criteria decision analysis, and other methods can be considered to establish grassland fire risk assessment models to obtain more accurate assessment results. At the same time, based on grassland fire risk assessment studies, grassland fire early warning research should be conducted to provide more scientific guidance for grassland fire prevention. In regions prone to frequent grassland fires, conduct grassland fire risk assessment research with higher spatial resolution (e.g., grid cells of 100 m × 100 m) to precisely tackle various levels of grassland fire risks.