Skip to main content
NCBI home page
Scientific Reports logoLink to Scientific Reports
. 2026 Feb 3;16:7034. doi: 10.1038/s41598-026-36589-w

Quantitative risk assessment for the compound drought-wildfire disaster

Kyunghun Kim 1,2,✉, Hung Soo Kim 3,✉
PMCID: PMC12920799  PMID: 41634073

Abstract

Compound drought-wildfire (CDW) disasters, driven by the interaction of drought and wildfire, are increasing worldwide. Despite this trend, there is still a lack of methodologies for quantitatively assessing CDW disaster risk, which makes it difficult to develop response plans. Therefore, this study proposes a quantifiable methodology for CDW risk assessment by applying a drought scenario to simulate drought induced weather conditions and the Fire Weather Index (FWI) to quantify potential wildfire spread and severity. The drought scenarios represent a 12-month drought with a 30-year return period, whereas the normal scenarios are based on typical regional climate conditions. Based on each scenario, DFWI and NFWI were calculated to quantify the risks of CDW disasters and single wildfire disasters. The estimated DFWIs showed the quantitative risk of CDW disasters in each region and monthly risk maps were developed to highlight high risk areas. By comparing DFWI and NFWI, the impact of drought on wildfire risk was analyzed and presented quantitatively. These results provide essential information for disaster response strategies for CDW disasters and emphasize the need for integrated drought and wildfire response management by clearly demonstrating the impact of drought on wildfire risk.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-36589-w.

Keywords: Compound disaster, CDW, FWI, Drought scenario, DFWI, NFWI

Subject terms: Climate sciences, Environmental sciences, Hydrology, Natural hazards

Introduction

According to data from the Emergency Events Database (EM-DAT, https://www.emdat.be/) maintained by the Centre for Research on the Epidemiology of Disasters in UCLouvain1, there has been a significant increase in the average number of victims and damage costs per wildfire event. Specifically, between the 2000–2012 and 2013–2024 periods, the average number of victims experienced a 6.5-fold increase (from 13,359 to 86,837), while the average damage costs showed a 2.5-fold increase, rising from $2.67 billion to $6.82 billion. A variety of factors contribute to the increasing severity of wildfires, but many researchers suggested that the interaction of drought and wildfire is the primary cause2–4. Consequently, wildfires have been redefined as a compound drought-wildfire disaster (CDW disaster), rather than a single disaster4,5.

Drought exacerbates wildfire risk by drying out vegetation and reducing humidity, both of which are critical factors that enhance fuel availability and facilitate fire spread6–8. These drought-induced impacts on wildfire behavior demonstrate the interaction between drought and wildfire, which leads to their classification as a compound disaster (For details, see the supplementary material). A notable example of the CDW disasters is the 2022 California wildfire crisis, where prolonged drought conditions directly fueled the ignition and explosive spread of massive fires, resulting in catastrophic loss. Other significant examples include the 2019–2020 Australian wildfires and the 2023 Canadian wildfires.

CDW disasters share several common consequences. These include widespread destruction of ecosystems, significant loss of human life, severe degradation of air quality, and profound economic impacts. Furthermore, CDW disasters tend to accelerate climate change by releasing large amounts of greenhouse gases, thus creating a vicious feedback loop. Consequently, CDW disasters exert immense ripple effects far beyond the location where the disaster occurs. These CDW disasters are no longer confined to specific regions bur are occurring globally, including in Australia, Korea, Spain, Brazil and so on9–12. Moreover, as the frequency of global drought is projected to increase due to climate change13, the frequency and magnitude of CDW disasters are also expected to worsen, demanding urgent action to address them. However, CDW disasters remain insufficiently studied. Existing research predominantly focuses on the individual relationships between drought and wildfire, the development of wildfire models using drought indices or the estimation of wildfire risk indices14–23. Until now, no study has explicitly incorporated the interaction between drought and wildfire, nor has any proposed a methodology for quantifying the risk of CDW disasters. This critical gap remains as a substantial deficit in compound disaster response and poses a significant obstacle to both risk assessment and the formulation of effective response strategies.

To address this gap, this study develops a methodology for quantitative risk assessment of CDW disaster in South Korea. To assess the risk of the CDW disasters, it is essential to simulate the drought induced dry weather conditions and quantify the risk of wildfires based on these conditions. First, we define a drought scenario and use Fire Weather Index (FWI) for the assessment of CDW disaster risk. The drought scenario simulates precipitation time series by employing a bivariate frequency analysis based on the duration and severity of drought events. The FWI is a wildfire risk index developed by the Canadian Forest Fire Danger Rating system24. It is based on meteorological factors that affect wildfires and is used to quantitatively assess the potential for wildfire spread and severity. Consequently, when the FWI is calculated based on the dry weather factors simulated in the drought scenario, it represents the risk of wildfire under the drought, which corresponds to the risk of the CDW disaster (we call this Drought FWI or DFWI). Next, to analyze the impact of drought on wildfire risk, this study defines a normal scenario using the average precipitation as the weather condition. The calculation of the normal scenario uses the same process as that of the drought scenario. Consequently, when the FWI is calculated from the simulated meteorological factors in the normal scenario, it represents the risk of wildfire under normal conditions, which means the risk of a single wildfire disaster (Normal FWI or NFWI). A comparative analysis of the FWIs calculated from each scenario enables the quantification of the impact of drought on the potential spread and severity of wildfires. Ultimately, this study establishes a novel framework for quantifying CDW risks by explicitly incorporating the interaction between drought and wildfire. This approach not only provides an innovative solution to the critical research gap in compound disaster quantification but also delivers essential, data driven indicators for the establishment of effective and evidence based CDW disaster response strategies.

Results

Calculation of FWI based on observation data

The suitability of the FWI in representing wildfire risk within the study area was evaluated by comparing observation based FWIs with actual wildfire statistics, including the frequency of events, burned area, and economic losses. Figure 1 shows the average FWI for each watershed, with darker red indicating higher wildfire risk. The highest and lowest average FWIs were observed in watershed 4016 (3.88) and 4002 (1.88), respectively. Watersheds with higher risk are concentrated in the eastern region, characterized by lower precipitation and relative humidity alongside higher temperatures and wind speeds relative to other areas (Supplement Fig. 1). In particular, the FWI distribution shows a spatial trend similar to relative humidity (Supplement Fig. 1b), which has been identified as the most critical meteorological factor for wildfire occurrence in South Korea25.

Fig. 1.

Fig. 1

Open in a new tab

Comparison of Average FWI of watersheds (January 1, 1981 to December 31,2023). The average of daily FWI in each watershed, (a) is calculated using the four meteorological data (precipitation, relative humidity, temperature, and wind speed) for the period of January 1, 1981, to December 31, 2023. The shaded watersheds correspond to the areas where the 2022 Gangneung-Donghae wildfire event, which caused the greatest wildfire damage, occurred. The highest average FWI is 3.88 for watershed 4016, and the lowest average is 1.88 for watershed 4002.

Next, the calculated FWIs were compared with the wildfire statistics from the Korea Forest Service’s annual report26. Notably, 2022 exhibited disproportionately high damage area and economic loss (Supplement Fig. 2b, c) due to the Gangneung-Donghae wildfire, which affected five watersheds in the eastern region of the peninsula (shaded areas in Fig. 1). Covering 20,676 hectares over 213 h, this event stands as the longest and largest wildfire disaster in South Korea’s history before 2025.

To statistically confirm the FWI’s reliability in the study area, Spearman’s rank correlation analysis was conducted (Supplement Table 1). All p-values were below 0.05, confirming the statistical significance of the correlations at a 95% confidence level. The correlation coefficients ranged between 0.5 and 0.7, indicating a moderate correlation according to Rovai et al27.. These results validate the FWI as a reliable indicator of wildfire risk within the study area and areconsistent with previous regional findings28,29.

Calculation of DFWI and NFWI

To calculate the DFWI and the NFWI, both drought and normal scenarios were calculated for each watershed. Three drought statistics (number of drought events, average drought duration, and average drought severity) of each watershed were calculated in the drought scenario (Supplement Fig. 3). The average number of drought events in each watershed was 52, with a duration and severity of 5.55 months and 132.35 mm, respectively. The number of drought events was found to be higher in the northern watersheds compared to the southern watersheds (Supplement Fig. 3a). In the duration of drought events (Supplement Fig. 3b), watersheds in the coastlines had a longer duration than others. However, there was no clear spatial pattern in either the number or duration of drought events. In contrast, a distinct spatial pattern was observed in terms of average severity (Supplement Fig. 3c), with watersheds in the southern region experiencing greater average severity compared to others.

The average annual precipitation under the drought scenario was 685.64 mm, representing approximately 52% of the 1,319 mm/year observed under normal conditions. This confirms that the simulated drought conditions result in a significant precipitation deficit of nearly 50%. The spatial distribution of precipitation in the drought scenario reveals that the highest totals occurred in the southern coastal watersheds (Fig. 2). The largest precipitation deficit was observed in watershed 2502, where total precipitation decreased by 1,259 mm/year. Conversely, watershed 2101 showed the smallest reduction, with a decrease of only 335.2 mm/year. Under drought conditions, the average daily precipitation was 1.89 mm/day with 246.5 non-precipitation days, compared to 3.62 mm/day and 232.4 days under the normal scenario. Notably, while the average daily precipitation decreased by approximately 49%, the number of non-precipitation days increased by only 6% (Supplement Fig. 4). The spatial patterns of these variables closely follow the distribution of total precipitation; watersheds with higher total precipitation exhibited greater daily averages and fewer non-precipitation days.

Fig. 2.

Fig. 2

Open in a new tab

Total precipitation of both drought and normal scenarios for each watershed. (a-b), Total precipitation of drought scenario (mm/year) (a), and Total precipitation of normal scenario (b). The total precipitation of drought scenarios, (a) is calculated by a bivariate frequency analysis of the duration and severity of drought events in each watershed. The total precipitation of normal scenarios, (b) is average annual total precipitation of each watershed based on observation data (1981–2023). The watershed 2502 has the most significant decline in total precipitation between drought and normal scenarios, in contrast, the watershed 2101 has the smallest reduction.

Based on the scenarios, the mean and standard deviation for DFWI were 5.19 and 1.87, respectively. For NFWI, these values were 1.71 and 0.33, indicating that NFWI exhibits significantly less spatial variability than DFWI. Regarding the spatial distribution of the DFWI (Fig. 3a), watersheds in the southeastern region with high NFWI values (Fig. 3b) also exhibited high DFWI values. NFWI values from June to September were notably lower than the overall average (Supplement Fig. 5). This trend is attributed to the influence of the North Pacific air mass, a maritime tropical system that brought hot, humid weather and approximately 70% of the annual precipitation during this period. Consequently, these weather conditions were highly unfavorable for wildfires, resulting in suppressed NFWI values. In contrast, NFWI values for the remaining months were higher than the overall average, with the peak values observed in April and May. These months constitute the primary wildfire season in South Korea, accounting for approximately 32% of all wildfire events26. April and May served as a transition period from the cool to the warm season, during which relative humidity and the moisture content of soil and vegetation declined. Furthermore, these months were characterized by frequent high wind speeds and significant accumulations of fallen leaves from the preceding autumn. Given that precipitation was also comparatively low during this time, the elevated NFWI values were expected, as these months represented the period most vulnerable to wildfires. Monthly DFWI exhibited a pattern consistent with the monthly average NFWI. From June to September, when NFWI values were at their minimum, DFWI also remained below the period average. In comparison, the remaining months (excluding January) displayed elevated DFWI levels that exceeded the overall average.

Fig. 3.

Fig. 3

Open in a new tab

Results of DFWI and NFWI and the Difference and Ratios between the Two Risk Indexes. (a-d), Average DFWI according to watershed over the entire period (a), Average NFWI according to watershed over the entire period (b), Difference of average DFWI and NFWI about each watershed (c), and Ratio of average DFWI and NFWI about each watershed (d). Average two risk indexes according to watershed over the entire period, ( a-b) show high-risk watersheds in red and low-risk watersheds in blue. The watersheds located in the southeastern part of the study area show high risk levels of both risk indexes. Monthly average two risk indexes for the entire study area. Difference between DFWI and NFWI, (c) is derived by subtracting NFWI from DFWI for each watershed. Ratio between DFWI and NFWI, (d) is calculated by dividing DFWI by NFWI for each watershed.

The impact of drought on wildfire risk was analyzed by comparing DFWI and NFWI. Across the entire study area, drought intensified wildfire risk by approximately 3.03 times compared to the overall average. Monthly analysis (Supplement Table 2) indicates that DFWI was, on average, 3.49 higher than NFWI. This difference was most pronounced in April and May, whereas it was minimal in January, July, and August. Furthermore, the monthly DFWI/NFWI ratio reveals that DFWI was approximately 1.49 times greater than NFWI on average. This ratio was notably higher during the winter months (December to February) and relatively lower from July to September. The average DFWI and NFWI values were compared by watershed group (Supplement Table 3), with the difference and ratio calculated. For these groups, the average difference and ratio were 3.38 and 3.04, respectively. At the individual watershed level, the average difference was 3.51; watersheds with the largest differences closely aligned with those exhibiting high DFWI values in the southeastern region (Fig. 3c). The average ratio was 3.16, indicating that DFWI is more than three times higher than NFWI across all watersheds. Notably, the highest ratio values were concentrated in the southern and southwestern regions (Fig. 3d). A consistent finding across all comparisons is that DFWI is approximately three times greater than NFWI. The primary distinction arises from the influence of drought; specifically, drought induced dry weather and the resulting reduction in fuel moisture increases wildfire risk by approximately threefold. This demonstrates that drought significantly amplifies the potential for large scale wildfire disasters.

To validate the scenario-based impact of drought on wildfire risk against actual occurrences, the study analyzed CDW disasters within the area. Wildfire records from 2017, specifically location and duration, were cross referenced with the drought events identified during scenario development. Through this process, four representative CDW cases were identified, notably including the catastrophic 2022 Gangneung-Donghae wildfire (Supplement Table 4). The FWIs were calculated using observational data from the affected watersheds during these disasters. These values were then compared to the average FWIs for the corresponding periods in years without CDW events. The average FWI calculated for the same periods in non CDW years represents the baseline condition for each watershed. During the 2022 Gangneung-Donghae wildfire, the average FWI across all affected watersheds exceeded 6, with some areas reaching high-risk values above 10. Conversely, the average FWI for the same periods in non CDW years remained below 3.5 (Supplement Table 4 d). Across all analyzed CDW cases, the differences and ratios ranged from 1.02 to 11.55 (mean: 5.29) and 1.23 to 3.85 (mean: 2.59), respectively. This robust quantitative consistency between the scenario-based risk increase (averaging over 3-fold) and empirical findings from actual CDW events (averaging 2.59-fold) validated the proposed methodology and supported its generalizability for assessing CDW risk.

Compound drought-wildfire disaster risk map

To identify high risk areas for CDW disasters, a risk map was developed. Monthly DFWI values for all watersheds were categorized into ten risk groups using consistent criteria (Fig. 4). A monthly comparison revealed that January and the period from June to September exhibited a higher proportion of blue shaded watersheds, indicating lower risk. Conversely, most watersheds were shaded red in April and May, coinciding with the peak wildfire season. The remaining months were predominantly characterized by yellow shading. This spatial pattern aligned with the seasonal characteristics of the study area as described in Supplement Fig. 5. The monthly risk maps demonstrate that watersheds in the southeastern region consistently face a high risk of CDW disasters. These findings highlight that the southeastern part of the study area is particularly vulnerable to such compound events; therefore, priority should be given to these watersheds when developing disaster response and management strategies.

Fig. 4.

Fig. 4

Fig. 4

Open in a new tab

Monthly risk maps of CDW. a-l, January risk map (a), February risk map (b), March risk map (c), April risk map (d), May risk map (e), June risk map (f), July risk map (g), August risk map (h), September risk map (i), October risk map (j), November risk map (k), and December risk map (i). The monthly DFWIs for all watersheds were categorized into 10 risk groups based on the same criteria. High risk watersheds are shown in red, whereas low risk watersheds are shown in blue. a shows generally low risk levels, except for some watersheds in the east and south-east. (b) shows an overall increase in risk levels compared to (a), with watersheds in the southeast showing a higher risk than others. Most watersheds have a higher risk in (c) than (b), including watersheds in the northern region. In (d), most watersheds are at high risk, except for a few. In (e), the risk continues to increase, especially watersheds in the north and southeast having more risk than others. (f) shows a sharp decline in risk across most of the watersheds, whereas some watersheds in the southeast remain at high risk. (g-h) show low risk than other months. In (i), the risk increases for most watersheds, but not for those in the northwest. By (j), the risk is higher for most of the watersheds, especially those in the southeast. In (k), the risk level of watersheds in the southeast either increases or stays the same, while other watersheds go down overall. In (i), the watersheds in the southeast have a higher risk than the rest, which remain at relatively lower risk.

Compared to other regions, high risk watersheds receive less precipitation, which exacerbates aridity during drought events. This characteristic is driven by both geographic and climatic factors. These watersheds are situated in basin topography surrounded by the Taebaek and Sobaek Mountains. These high-altitude ranges act as barriers that prevent humid atmospheric air from reaching the basin. Specifically, their elevation inhibits the inland movement of moisture, leaving the interior areas less influenced by humid air masses. Climatically, these watersheds are located far from the coast and are dominated by a continental climate. This climate exhibits greater diurnal and annual temperature fluctuations and lower relative humidity than oceanic climates, resulting in lower overall and less frequent precipitation than other areas. Consequently, the combined effects of mountain shielded topography and a dry continental climate create a structural environment of moisture deficit. Together, these factors collectively explain why the southeastern watersheds exhibit such heightened vulnerability to CDW disasters compared to the rest of the study area.

Discussion

This study proposed a risk assessment methodology of the CDW disaster by simulating drought weather conditions with the drought scenario and calculating the potential spread and severity of wildfire with the FWI. To the best of our knowledge, this is the first study to provide a quantitative assessment of CDW disaster risks. The key findings of this study across the Korean Peninsula are as follows:

  1. During the defined drought period (duration: 12 months, frequency: once every 30 years), only 52% of the average annual total precipitation occurred in the study area.

  2. The monthly DFWI calculations indicate that the risk of CDW disasters is highest in April and May, and lowest from June to September.

  3. Spatial analysis reveals that watersheds located in the southeastern part of the peninsula are identified as high-risk areas.

  4. A comparison between the DFWI and NFWI indicates that wildfire risk increases due to drought induced dry conditions.

Quantitative risk assessment of CDW disasters is an essential component of evidence-based disaster planning. The risks derived from this study enable governments to evaluate regional risk levels, establish precise response plans, and prioritize mitigation investments. This approach addresses the current challenge of developing mitigation strategies due to the lack of quantitative risk for CDW disasters. Furthermore, this research moves beyond theoretical discussions by providing numerical evidence that quantifies the impact of drought on wildfire risk. Such quantification is vital for refining future compound disaster models and serves as a foundation for integrated response strategies. Moving away from separate management approaches toward an integrated mitigation plan will enhance management efficiency and facilitate proactive, rather than reactive, responses to CDW events.

Limitations and future research directions

While this methodology provides significant quantitative evidence for CDW risk, its comprehensive interpretation and global applicability require a transparent discussion of the study’s inherent methodological and data limitations, which must guide the direction of future research. First, the structural and input constraints of the FWI must be acknowledged. Originally developed for Canadian forest conditions, the FWI may not fully capture the specific ecosystem characteristics of the study area. Although preliminary correlation analyses with local wildfire statistics suggest broad suitability, future efforts should focus on reparametrizing the index to better reflect regional environmental conditions. Crucially, as the FWI relies solely on meteorological inputs, it overlooks non meteorological drivers essential for fire occurrence. This includes anthropogenic ignition factors, which account for over 71% of wildfire occurrences in the study area26. The exclusion of ignition triggers means that even under environmental conditions highly favorable for wildfire spread (indicated by high FWI values), a CDW disaster cannot materialize without an initial spark. Furthermore, the risk method overlooks solar radiation, which directly influences the surface energy balance, temperature rise, fuel drying rates, and evapotranspiration. Solar radiation plays a pivotal role in amplifying wildfire intensity and spreads, particularly when fuels are already under drought induced stress30,31. The absence of these two factors suggests that the wildfire risk represented by the FWI may be incomplete or potentially distorted. Therefore, future research must aim to modify the FWI by integrating socio-economic variables related to anthropogenic ignition (such as population density and forest accessibility) and explicitly account for the radiation. Lastly, caution must be exercised in interpreting absolute FWI values. The FWI is an index designed to quantify the potential for wildfire spread and intensity, rather than to serve as a direct proxy for actual fire damage. While high FWI values indicate extreme fuel dryness and rapid flame propagation factors that dramatically increase suppression difficulty and the potential for large scale fires32, the realized damage is governed by diverse socio-economic and managerial factors, including population density, asset value, and initial response efficacy. Consequently, the index must be interpreted primarily as an indicator of potential severity rather than a deterministic predictor of loss.

Second, the spatial uniformity of the drought scenario applied across all watersheds presents a methodological limitation. This study utilized a standardized drought event, officially recognized by the Korean government, as a consistent baseline for all watersheds. However, as illustrated in Supplement Fig. 1, each watershed exhibits distinct meteorological characteristics, suggesting that the frequency and intensity of actual drought conditions vary significantly by watershed. Due to these regional variations, the official drought scenario applied here may not perfectly align with the actual conditions observed during historical CDW events. For future work, it is necessary to cluster watersheds based on similar climatic profiles and perform separate drought analyses for each cluster. Identifying the most frequently occurring drought patterns in this manner will facilitate the establishment of regionalized drought scenarios that reflect local meteorological realities.

Third, the study’s validation process was constrained by the limited number of confirmed historical CDW events. This scarcity is due both to the inherent rarity of CDW disasters and the lack of a standardized, officially recognized methodology for identifying these events. While this small sample size necessitates caution when generalizing the results, it is noteworthy that the FWI increase rates observed in the four verified cases demonstrated strong quantitative consistency with the scenario analysis results. This alignment provides significant proof of concept for the proposed framework. Nevertheless, future validation efforts must aim to incorporate a broader range of CDW events by leveraging long term, multi decadal meteorological and wildfire records to strengthen the statistical robustness of the findings.

Finally, the proposed quantitative methodology for CDW risk necessitates adaptation for global application. The current study relies on area averaged and point source data formats specific to the South Korean administrative context. Transitioning to a global scale would require the integration of grid-based climate reanalysis or satellite derived observation datasets. A critical subsequent step involves customizing both the drought scenarios and the FWI for diverse climatic zones. Furthermore, validating the effectiveness of the FWI within these specific regions is a crucial prerequisite. Since absolute FWI values possess relative meaning based on local fuel types and baseline climates, the interpretation of the final DFWI should be converted into a standardized, percentile-based index.

As demonstrated by Kudláčková et al33., utilizing regional FWI distributions rather than absolute thresholds is essential for accurately classifying risk across different global environments. Implementing these processes would facilitate the creation of a global CDW risk map, providing the objective evidence needed to identify international high-risk areas and prioritize the efficient allocation of global resources for wildfire suppression and drought mitigation. This necessity is exemplified by Kudláčková et al33., who utilized distinct thresholds for fire risk classification across varying climate zones, confirming that the FWI serves as a relative indicator of potential risk within a specific region rather than a global absolute standard. Successfully implementing these processes would enable the global quantification of CDW risk, providing the objective evidence needed to develop a global disaster risk map. Ultimately, such a tool would facilitate the identification of international high-risk zones and prioritize the efficient allocation of global resources for wildfire suppression and governmental drought mitigation efforts.”

Methods

Study area and data

The study area encompasses 107 watersheds across the Korean Peninsula (Supplement Fig. 6), selected to address the urgent need for a quantitative assessment of CDW disaster risks in South Korea. Among the total medium sized watersheds, four on Jeju Island and six in North Korean or border regions were excluded due to insufficient meteorological data, leaving 107 watersheds for analysis. Currently, South Korea is experiencing a rapidly escalating wildfire threat exacerbated by climate change. Over the past decade (2015–2024), the annual average frequency of wildfires (546 cases) and the resulting burned area (4,003 ha) have significantly surpassed the previous 20 year averages (1995–2014; 452 cases and 2,480 ha, respectively)26,34. Large scale wildfires (defined as events exceeding 100 ha or lasting over 24 h) have become an annual occurrence since 2017. A notable example is the Andong wildfire on March 22, 2025, which decimated 100,678.08 ha in a single day, marking it as one of the most severe events on record. Evidence suggests that antecedent drought conditions have significantly intensified the severity of these large scale fires35. With climate change scenarios predicting increased drought frequency and intensity36, the wildfire risk is expected to expand, raising the potential for catastrophic CDW events. Given South Korea’s extensive forest cover and dense Wildland-Urban Interface, CDW disasters pose a severe threat to life and property, necessitating a quantitative risk indicator for proactive national level disaster preparedness.

To calculate the DFWI and NFWI, a comprehensive daily meteorological dataset was compiled for the period from January 1, 1981, to December 31, 2023. The variables required for the FWI system include daily precipitation, mean temperature, mean wind speed, and mean relative humidity. Area averaged daily precipitation data for each watershed was obtained from the Water Management Information System (WAMIS), providing spatially aggregated precipitation values. For the remaining variables, data was sourced from 54 Automated Synoptic Observing System (ASOS) stations operated by the Korea Meteorological Administration (KMA). The spatial distribution of these ASOS stations is illustrated in Supplement Fig. 7. Detailed specifications regarding the dataset are provided in the Data Availability Statement.

Drought and normal scenario

The drought scenarios were generated using a six-step procedure, as detailed in Supplement Fig. 8a. The initial stage involved the construction of consistent daily and monthly precipitation series, followed by the conversion of point source data into Mean Areal Precipitation (MAP) using the Thiessen polygon method to ensure spatial representativeness across the 107 watersheds. Subsequently, drought events were identified by calculating monthly precipitation anomalies, defined as the difference between the observed monthly precipitation and the 6 month cumulative monthly average. In this study, a drought event was defined by a sustained period during which the anomaly remained at or below zero. To account for the stochastic and multidimensional nature of the drought events, a bivariate drought frequency analysis was then conducted using copula functions. This approach allowed for the simultaneous consideration of both drought duration and severity. After identifying the best fit marginal probability distribution for each variable, the optimal copula functions were integrated into the bivariate frequency analysis framework proposed by Shiau and Shen37 to determine the joint return periods and establish representative drought scenarios.

graphic file with name d33e630.gif 1

Where, FD(d) and FS(s) are the probabilities of drought duration (D) and severity (S), and C(∙) is the copula function. TDS represents the combined return period of a drought above a certain duration (d) and severity (s), while E(L) is the average interval between drought events.

By applying the specific duration and return period of a selected drought into the bivariate framework, the total precipitation deficit for a given scenario can be estimated. This study utilized a drought event characterized by a year duration and a 30 year return period, consistent with the standard criteria employed in South Korea’s national drought vulnerability assessments38. Subsequently, the total annual precipitation calculated from the bivariate frequency analysis was disaggregated into monthly values. To this end, monthly weights were derived by calculating the ratio of the long term average precipitation for each month to the total average annual precipitation. Finally, these monthly totals were converted into daily precipitation sequences using a Hidden Markov Model. To account for the stochastic variability in daily precipitation distributions inherent in the model, 1,000 distinct drought scenarios were simulated for each watershed. This ensemble approach ensures that the resulting DFWI and NFWI values reflect a robust range of potential daily weather patterns under the defined drought conditions.

To quantify the incremental increase in wildfire risk attributable to drought, a normal scenario was established as a baseline. This scenario represents the expected precipitation levels for each region under typical climatic conditions. The construction of the normal scenario follows a procedural framework like that of the drought scenario (Supplement Fig. 8b), but with key methodological distinctions. Instead of employing bivariate frequency analysis, the normal scenario utilizes long term average monthly precipitation derived from historical observation data. Finally, the daily precipitation time series for the normal scenario was generated using the Hidden Markov Model.

DFWI and NFWI

To quantify the risk of CDW disasters, this study integrated drought induced meteorological simulations with wildfire risk modeling. The Drought Fire Weather Index (DFWI) was derived by inputting the previously described stochastic drought scenarios into the FWI system. This process ensures that the resulting risk metrics explicitly reflect the intensified fuel dryness and atmospheric conditions characteristic of the predefined drought events. The FWI utilizes four meteorological variables: precipitation, mean temperature, wind speed, and relative humidity. While the direct impacts of drought on temperature and wind speed variance remain complex and less definitively established in this context, relative humidity is highly sensitive to drought induced fluctuations in precipitation and soil moisture. Such changes directly modulate the atmospheric water balance and fuel moisture content39. Therefore, this study reflected the impacts of drought by focusing specifically on the modification of precipitation and relative humidity. For the remaining variables, observed historical values were maintained across all scenarios.

To estimate relative humidity under drought scenarios, this study used an Artificial Neural Network (ANN) model. The ANN was selected for its superior ability to capture complex, nonlinear interactions among meteorological variables, which conventional linear models often fail to represent40,41. Furthermore, previous research has demonstrated the effectiveness of ANN based humidity estimation within the same study area42. The model used precipitation, temperature, and wind speed as input variables, with relative humidity as the target output. To account for distinct seasonal variability, the dataset was partitioned by month, and independent monthly relative humidity models were developed. The structure of the ANN, specifically the number of layers, was optimized using an objective function based trial and error method. The objective function was defined by the Root Mean Squared Error (RMSE) to minimize the discrepancy between model estimates and observed data. Taking the January relative humidity model for watershed 1001 as a representative example, the January specific data (n = 1,364) were extracted and split into training (70%), validation (15%), and testing (15%) subsets. During the training phase, the maximum number of iterations was set to 1,000, and the configuration yielding the lowest RMSE during validation was selected. The final performance was evaluated using the remaining 15% of the data (test set). As shown in Supplement Table 5, the evaluation of architectures ranging from 1 to 10 layers indicated that a 4-layer configuration produced the minimum RMSE. Consequently, a 4-layer ANN structure was finalized for estimating January relative humidity in watershed 1001.

The calculation of the NFWI followed a procedural framework identical to that of the DFWI, with the critical distinction being the input of the normal scenario rather than the drought scenario. To maintain methodological consistency, the same ANN based estimation model is utilized to derive relative humidity for the normal weather conditions. The NFWI serves as a baseline, representing the inherent risk of an isolated wildfire event without the influence of prolonged moisture deficits. In contrast, the DFWI explicitly integrates the cascading impacts of drought, characterizing the risk within a CDW disaster context. Consequently, a comparative analysis between the NFWI and DFWI provides a robust, quantitative basis for isolating and determining the specific impact of drought on potential wildfire spread and severity across the study area.

Supplementary Information

Below is the link to the electronic supplementary material.

Supplementary Material 1 (8MB, docx)

Acknowledgements

This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korea government (MSIT) (No. 2022R1A2C2091773) and the INHA University Research Grant.

Author contributions

H.S.K conceived and designed the study. K.K. processed the methodologies and collected meteorological data. K.K and H.S.K contributed to the data analysis and interpretation. K.K. wrote the manuscript. All authors edited the manuscript.

Funding

This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korea government (MSIT) (No. 2022R1A2C2091773) and the INHA University Research Grant.

Data availability

Weather stations located in the study area were checked to collect data sets on four meteorological factors (precipitation, temperature, wind speed, and relative humidity) to calculate the two scenarios, DFWI and NFWI. Among the checked stations, those with observation data for the meteorological factors for the period January 1, 1981, to December 31, 2023, resulting in a total of 54 ASOS stations (Supplement Fig. 7). The data for the 54 stations are available from the KMA Weather Data Service-Open Met Data Portal ([https://data.kma.go.kr/resources/html/en/aowdp.html](https:/data.kma.go.kr/resources/html/en/aowdp.html)) operated by the KMA. Area-averaged daily precipitation data for each individual watershed were collected from the Water Management Information System (WAMIS). The WAMIS (https://www.wamis.go.kr/) managed by Han River Flood Control Office, is the Korean government’s integrated platform providing essential hydrological and meteorological data, including precipitation, discharge, watershed characteristics for various watersheds across the country. The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.

Declarations

Competing interests

The authors declare no competing interests.

Footnotes

Publisher’s note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Contributor Information

Kyunghun Kim, Email: kkh0904@inha.ac.kr.

Hung Soo Kim, Email: sookim@inha.ac.kr.

References

  • 1.Delforge, D. et al. EM-DAT: the emergency events database. Int. J. Disaster Risk Reduct.124, 105509 (2025). [Google Scholar]
  • 2.Littell, J. S., Peterson, D. L., Riley, K. L., Liu, Y. & Luce, C.H. A review of the relationships between drought and forest fire in the United States. Glob. Change Biol. 22, 2353–2369 (2016). [DOI] [PubMed] [Google Scholar]
  • 3.Turco, M. et al. On the key role of droughts in the dynamics of summer fires in Mediterranean Europe. Sci. Rep.81(7), 1–10 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 4.Richardson, D. et al. Global increase in wildfire potential from compound fire weather and drought. npj Climate Atmospheric Sci.23(5), 3–12 (2022). [Google Scholar]
  • 5.Batllori, E. et al. Compound fire-drought regimes promote ecosystem transitions in Mediterranean ecosystems, Journal of Ecology, 107, (2019).
  • 6.Flannigan, M. D., Krawchuk, M. A., de Groot, W. J., Wotton, B. M. & Gowman, L. M. Implications of changing climate for global wildland fire. Int. J. Wildland Fire18, 483–507 (2009). [Google Scholar]
  • 7.Pereira, P., Úbeda, X. & Martin, D. A. Fire severity effects on ash chemical composition and water-extractable elements. Geoderma191, 105–114 (2012). [Google Scholar]
  • 8.Gouveia, C. M., Bastos, A., Trigo, R. M. & DaCamara, C. C. Drought impacts on vegetation in the pre- and post-fire events over Iberian Peninsula. Nat. Hazard.12(10), 3123–3137 (2012). [Google Scholar]
  • 9.Walden, L., Joseph, B. F., Katinka, X. R., George, M. & Richard, J. H. Drought then wildfire reveals a compound disturbance in a resprouting forest. Ecol. Appl.33(2), e2775 (2023). [DOI] [PubMed] [Google Scholar]
  • 10.Kim, K., Ahn, H., Kim, J., Lee, T. & Kim, H. S. Study on forest fire risk analysis based on drought scenarios. J. Korean Soc. Hazard Mitigation24(1), 7–18 (2024). [Google Scholar]
  • 11.Chaparro, D., Piles, M., Vallllossera, M. & Camps, A. Surface moisture and temperatures trends anticipate drought conditions linked to wildfire activity in the Iberian Peninsula. European J. Remote Sensing49(1), 955–971 (2016). [Google Scholar]
  • 12.Pontes-Lopes, et al. Drought-driven wildfire impacts on structure and dynamics in a wet Central Amazonian forest. Proc. R. Soc. B288, 1951 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 13.Vicente-Serrano et al. A. Global drought trends and future projections, Philos. Trans. R. Soc. A: Math., Phys. Eng. Sci., 380, 2238 (2025) [DOI] [PMC free article] [PubMed]
  • 14.Littel, J. S., Peterson, D. L., Riley, K. L., Liu, Y. & Luce, C. H. A review of the relationships between drought and forest fire in the United States. Glob. Change Biol.22, 2353–2369 (2016). [DOI] [PubMed] [Google Scholar]
  • 15.Turner, M. G., Hargrove, W. W., Gardner, R. H. & Romme, W. H. Effects of fire on landscape heterogeneity in Yellowstone National Park Wyoming. J. Vegetation Sci.5(5), 731–742 (1994). [Google Scholar]
  • 16.Westerling, A. L. & Swetnam, T. W. Interannual to decadal drought and wildfire in the western United States. EOS Trans. Am. Geophys. Union84(49), 545–555 (2003). [Google Scholar]
  • 17.Reinhard, M., Rebetez, M. & Schlaepfer, R. Recent climate change: Rethinking drought in the context of Forest Fire Research in Ticino. South of Switzerland, Theoretical Appl. Climatol.82(1), 17–25 (2005). [Google Scholar]
  • 18.Xiao, J. & Zhuang, Q. Drought effects on large fire activity in Canadian and Alaskan forests. Environ. Res. Lett.2(4), 044003 (2007). [Google Scholar]
  • 19.Dimitrakopoulos, A. P., Bemmerzouk, A. M. & Mitsopoulos, I. D. Evaluation of the Canadian fire weather index system in an eastern Mediterranean environment. Meteorol. Appl.18(1), 83–93 (2011). [Google Scholar]
  • 20.Taufik, M. et al. Amplification of wildfire area burnt by hydrological drought in the humid tropics. Nat. Clim. Chang.7(6), 428–431 (2017). [Google Scholar]
  • 21.Ruffault, J., Curt, T., Martin-StPaul, N. K., Moron, V. & Trigo, R. M. Extreme wildfire events are linked to global-change-type droughts in the northern Mediterranean. Nat. Hazard.18(3), 847–856 (2018). [Google Scholar]
  • 22.Ellis, T. M., Bowman, D. M. J. S., Jain, P., Flannigan, M. D. & Williamson, G. J. Global increase in wildfire risk due to climate-driven declines in fuel moisture. Glob. Change Biol.28(4), 1544–1559 (2022). [DOI] [PubMed] [Google Scholar]
  • 23.Pokharel, B. et al. Amplified drought trends in Nepal increase the potential for Himalayan wildfires. Clim. Change17(2), 17 (2023). [Google Scholar]
  • 24.de Groot, W. J. et al. Developing a global early warning system for wildland fire. For. Ecol. Manage.234(1), S10 (2006). [Google Scholar]
  • 25.Jeon, B. & Chae, H. A study of analysis on relationship between Korea forest fire occurrence and weather factor. J. Korean Soc. Hazard Mitigation17(5), 197–206 (2017). [Google Scholar]
  • 26.Korea Forest Service. 2023 wildfire statistic yearbook https://eng-nifos.forest.go.kr/kfsweb/cop/bbs/selectBoardArticle.do;jsessionid=UW1qecX11avigttK1IUt9lIWvo7HRMsKq51J3PuuPNHJITfevJWKWNGG1eWQ8nR4.frswas01_servlet_engine5?nttId=3205390&bbsId=BBSMSTR_1008&pageUnit=10&pageIndex=1&searchtitle=title&searchcont=&searchkey=&searchwriter=&searchWrd=&ctgryLrcls=&ctgryMdcls=&ctgrySmcls=&ntcStartDt=&ntcEndDt=&mn=NKFS_02_15_01 (2024).
  • 27.Rovai et al. of referencing in Social science research design and statistics: A practitioner’s guide to research methods and IBM SPSS Analysis (ed. Rovai, A.P.) 101–120 (Watertree Press LLC, 2013).
  • 28.Park, H. S., Lee, S. Y., Chae, H. M. & Lee, W. K. A study on the development of forest fire occurrence probability model using Canadian forest fire weather index – occurrence of forest fire in Kangwon province. J. Korean Soc. Hazard Mitigation3(6), 95–100 (2009). [Google Scholar]
  • 29.Lim, C. J. & Chae, H. M. Application of the Canadian fire weather index for forest fire danger assessment in South Korea. Forests16, 1059 (2025). [Google Scholar]
  • 30.Gomes, J. F. P. & Radovanovic, M. Solar activity as a possible cause of large forest fires – A case study: Analysis of the Portgues forest fires. Sci. Total Environ.394(1), 197–205 (2008). [DOI] [PubMed] [Google Scholar]
  • 31.Ma, B., Liu, X., Tong, Z., Zhang, J. & Wang, X. Coupled effect of high temperature and droughts on forest fires in Northeast China. Remote sensing16, 3784 (2024). [Google Scholar]
  • 32.Van Wagner, C. E. (1987). Development and structure of the Canadian Forest Fire Weather Index System (Forestry Technical Report No. 35). Canadian Forestry Service, Petawawa National Forestry Institute, Chalk River, ON. Available at FRAMES resource catalog.
  • 33.Kudláčková, L. et al. Assessing fire danger classes and extreme thresholds of the Canadian Fire Weather Index across global environmental zones: a review. Environ. Res. Lett.20, 013001 (2025). [Google Scholar]
  • 34.Korea Forest Service. 2014 wildfire statistic yearbook https://nfsv.forest.go.kr/kfsweb/cop/bbs/selectBoardArticle.do;jsessionid=32xM8Xg6HJDAJJrNTaYyvyawlISrKVP46chSstvtl541yc7Qbw2hOIqiPbDPvPwg.frswas01_servlet_engine5?nttId=3051495&bbsId=BBSMSTR_1008&pageUnit=10&pageIndex=29&searchtitle=title&searchcont=&searchkey=&searchwriter=&searchWrd=&ctgryLrcls=&ctgryMdcls=&ctgrySmcls=&ntcStartDt=&ntcEndDt=&mn=UKFV_05_05 (2015).
  • 35.Kim. K. (2025). Analysis of compound drought-wildfire disaster risk and concurrent drought disaster using complex network [Doctoral dissertation]. Inha University.
  • 36.Lee, S. J. et al. Construction of GIS database for wildfire in the Korean Peninsula using MODIS data. J. Korean Cartographic Assoc.16(30), 129–137 (2016). [Google Scholar]
  • 37.Shiau, J. T. & Shen, H. W. Recurrence analysis of hydrologic droughts of differing severity. J. Korea Water Resour. Planning Managem.127(1), 30–40 (2001). [Google Scholar]
  • 38.Ministry of Environment. Guidelines for the Drought Vulnerability Map (Unpublished international report. Ministry of Environment, 2022). [Google Scholar]
  • 39.Farahmand, A., AghaKouchak, A. & Teixeira, J. A vatange from space can detect earlier drought onset: an approach using relative humidity. Sci. Rep.5, 8553 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 40.Yasar, A., Simsek, E., Bilgili, M., Yucel, A. & Ilhan, I. Estimation of relative humidity based on artificial neural network approach in the Aegean Region of Turkey. Meteorol. Atmos. Phys.115(1), 81–87 (2012). [Google Scholar]
  • 41.Shad, M., Sharma, Y. D. & Singh, A. Forecasting of monthly relative humidity in Delhi India, using SARIMA and ANN models. Modeling Earth Syst. Environ.8(4), 4843–4851 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 42.Choi, H., Moon, T., Jung, D. H. & Son, J. E. Prediction of air temperature and relative humidity in greenhouse via a multilayer perceptron using environmental factors. J. Bio-Environ. Control28(2), 95–103 (2019). [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplementary Material 1 (8MB, docx)

Data Availability Statement

Weather stations located in the study area were checked to collect data sets on four meteorological factors (precipitation, temperature, wind speed, and relative humidity) to calculate the two scenarios, DFWI and NFWI. Among the checked stations, those with observation data for the meteorological factors for the period January 1, 1981, to December 31, 2023, resulting in a total of 54 ASOS stations (Supplement Fig. 7). The data for the 54 stations are available from the KMA Weather Data Service-Open Met Data Portal ([https://data.kma.go.kr/resources/html/en/aowdp.html](https:/data.kma.go.kr/resources/html/en/aowdp.html)) operated by the KMA. Area-averaged daily precipitation data for each individual watershed were collected from the Water Management Information System (WAMIS). The WAMIS (https://www.wamis.go.kr/) managed by Han River Flood Control Office, is the Korean government’s integrated platform providing essential hydrological and meteorological data, including precipitation, discharge, watershed characteristics for various watersheds across the country. The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.

Articles from Scientific Reports are provided here courtesy of Nature Publishing Group

RESOURCES