(1) Remote sensing

To estimate the spatial boundary of a wildfire, NextSat-2 SAR imaging was used to distinguish the burned area. To measure wildfire damage severity, the Relative differenced Normalized Burn Ratio (RdNBR) was calculated from Sentinel-2 multispectral imagery using bands B8 and B12. Using Google Earth Engine, images from 14 March 2025 and 29 March 2025 were acquired, and pixels classified as clouds were excluded from the analysis. Cloud masked pixels were gap-filled using corresponding pixels from the nearest cloud-free acquisition date. RdNBR was computed as,

and was adopted to enable relative comparisons among different vegetation structures. Severity thresholds were then applied to delineate three classes of burn severity. Although previous studies commonly interpret low, moderate, and high burn severity in terms of fire behavior, typically ranging from surface fires through crown fires to bole charring, we focused on a definition that is more directly usable for rapidly estimating carbon storage loss.

Specifically, we adopted the minimum and maximum RdNBR values for the lowest and highest burn-severity classes from earlier studies and then selected the intermediate threshold so that the three burn-severity classes contained an approximately equal numbers of pixels[8, 9]. Accordingly, regardless of forest type, grid cells with RdNBR values greater than or equal to − 2 and less than 0.069 were classified as low-severity burns, those with values greater than or equal to 0.315 and less than 0.9 as moderate-severity burns, and those with values greater than or equal to 0.9 and less than 2 as high-severity burns (Fig. 2). All remaining analyses were conducted in R 4.5.1 using the terra package[10].

(2) InVEST Carbon storage

In this study, the InVEST Carbon Storage and Sequestration model was used to quantitatively evaluate changes in carbon storage before and after the wildfire. The InVEST carbon model is a spatially explicit tool that estimates the total carbon stored in a landscape at a given time by combining a land-use/land-cover (LULC) map with carbon density values for four carbon pools: aboveground biomass (AGB), belowground biomass, soil, and dead organic matter[11]. Here, AGB refers to all living biomass above the soil surface, including stems, branches, bark, and foliage, while belowground biomass includes live root biomass. Soil organic carbon represents the stock of organic matter stored in soils and is known to constitute the largest carbon pool in terrestrial ecosystems. The dead organic matter pool encompasses not only litter but also both downed and standing dead wood. To estimate post-fire carbon storage, a post-fire LULC map was first constructed that incorporated the mapped burn-severity classes. Carbon storage was then recalculated by subdividing the original forest types (broadleaved, coniferous, and mixed forests) into three fire-severity levels (low, moderate, and high) according to the burn-severity classification and assigning corresponding carbon densities to each class. In this study, the level-2 land-cover map produced in 2022, by the Ministry of Climate, Energy and Environment was used as the LULC input.

Because there is uncertainty in post-fire aboveground carbon residuals, we defined two scenarios, A - higher-residual aboveground (HRA) and B - lower-residual aboveground (LRA), that reflect alternative assumptions about the residual fraction. The residual fractions of each carbon pool by forest type (Table 1) were determined by synthesizing analyses presented in the Monitoring of Ecological Damage and Risk in the Jirisan Hadong Wildfire Area report[12] and in previous studies[13–15].

For the aboveground pool (scenario A), severity-specific residual rates were derived using Live Burning Efficiency (LBE) values reported in previous studies[14, 15]. Here, LBE is defined as the proportion of live biomass actually consumed by a wildfire relative to the amount of available fuel per unit area. Studies focusing on coniferous and mixed forests have reported that approximately 25–65% of live aboveground biomass is lost under low-, moderate-, and high-severity fires. Using the Spanish National Forest Inventory combined with dNBR, Balde et al.[15] estimated LBE for conifer forests to be about 0.44, 0.55, 0.60, and 0.81 for low, moderate–low, moderate–high, and high severity, respectively.

In this study, these values were integrated to derive representative loss rates for each severity class, resulting in residual fractions (1 – LBE) of approximately 0.69, 0.49, and 0.32 for low-, moderate-, and high-severity fire, respectively. We then rounded these values and assumed that 70%, 50%, and 30% of aboveground carbon storage remains in areas classified as low, moderate, and high burn severity. These assumptions correspond to the aboveground scenario A in Table 1 and represent combustion-based residual fractions derived from LBE studies.

In contrast, the aboveground scenario B is a field-based residual fraction derived from the wildfire damage survey conducted as part of the ecosystem damage and risk monitoring of the Jirisan Hadong fire by the Korea National Park Research Institute. Using the proportion of surviving trees and the degree of crown damage by forest type and burn severity, the actual proportion of stem and crown biomass remaining immediately after the fire was estimated. Based on these estimates, the aboveground residual fractions for low, moderate, and high severity areas were set to 58%, 15%, and 0%, respectively.

Relative to the higher-residual aboveground scenario A, the lower-residual scenario B represents a more conservative combustion assumption, intended to avoid overestimating the amount of remaining aboveground biomass. The residual fractions for belowground, soil, and dead wood–litter pools were determined with reference to the analysis by Sweeney et al.[13]. Belowground carbon was assumed to remain at 100% across all burn-severity classes, highlighting that it is unlikely to be completely combusted over a short time. Soil organic carbon was assumed to retain 99% of its pre-fire stock in low- and moderate-severity areas and 95% in high-severity areas. Because dead wood and the litter layer are more directly exposed and sensitive to combustion and heating, their residual fractions were set to 79%, 76%, and 65% for low-, moderate-, and high-severity fire, respectively.

Accordingly, the aboveground scenario A represents a higher-residual case, synthesized from previous studies, whereas the aboveground scenario B represents a lower-residual case based on post-fire field observations reported by the Korea National Park Research Institute. The remaining pools such as belowground, soil, and dead wood–litter with incorporate pool-specific residual fractions were derived from Sweeney et al.[13]. These carbon pool residual fractions were applied to the InVEST carbon storage model and compared to the total carbon storage before and immediately after the wildfire, thereby quantifying carbon losses by burn severity and forest type.

Environmental values can be expressed in monetary terms using four broad classes of valuation methods: direct market price methods, indirect market price methods, non-market valuation methods, and value transfer approaches[16, 17]. Among these, we adopted a direct market price method, applying the allowance prices observed in the K-ETS. This choice allows rapid estimation in the aftermath of a disaster without additional surveys or model assumptions, and directly reflects institutional carbon prices, thereby providing a loss estimate that is closely linked to actual policy and market conditions.

When converting the loss of carbon storage into an economic loss, we determined the market-price approach based on K-ETS prices to be the most appropriate. We expressed the wildfire-induced loss of aboveground carbon (tC) in CO₂ units (tCO₂) using the molecular mass ratio 1 tC = 3.667 tCO[18]. We then applied the average K-ETS allowance price over approximately one month starting on 22 March 2025: the date of the fire. On this basis, the economic value used in this study was 8,793 KRW per metric ton of CO₂.

Of the total burned area, approximately 62% was forest, which was analyzed by forest type and burn severity. Based on the RdNBR thresholds, burn severity was classified into three levels (low, moderate, and high) and combined with forest type to yield nine categories in total (Table 2). The total forest area affected by the wildfire was 105,623.91 ha. Of this, coniferous forests accounted for 56,255.91 ha (53.3% of the burned forest area), while broadleaved and mixed forests covered 34,575.86 ha (32.7%) and 14,792.14 ha (14.0%), respectively. By burn severity, low, moderate, and high classes occupied 27,492.59 ha (26.0%), 44,977.01 ha (42.6%), and 33,154.31 ha (31.4%), respectively, with the three classes being relatively evenly distributed, although the moderate class was the most prevalent, as shown in Table 2.

The distribution of burn severity and area differed among forest types. Coniferous forests not only occupied the largest area but also showed pronounced moderate and high severity. The low-severity area in coniferous forest was 15,299.34 ha, corresponding to 27.2% of the burned coniferous forest, while moderate severity covered 20,732.51 ha (36.9%) and high severity 20,224.06 ha (36.0%), indicating similar shares for the moderate and high classes. In broadleaved forests, the low-severity area was only 7,718.70 ha (22.3%), whereas moderate severity covered 17,105.94 ha (49.5%) and high severity 9,751.22 ha (28.2%), suggesting that roughly half of the burned broadleaved forest experienced moderate-severity fire. Mixed forests occupied the smallest share of the burned forest overall. However, their burn-severity distribution was similar to, or slightly more moderate than, that of broadleaved forests. In mixed forests, the low-severity area was 4,474.55 ha (30.2%), moderate severity 7,138.56 ha (48.3%), and high severity 3,179.03 ha (21.5%), with about half of the area in the moderate class and a relatively lower proportion of high severity compared with coniferous forests.

Taken together, the distribution of burned areas by forest type and severity indicates that the wildfire produced particularly concentrated moderate- and high-severity damage in coniferous forests, whereas broadleaved and mixed forests were characterized by a predominance of moderate-severity burns, as shown in Fig. 3.

Before the wildfire, the total aboveground carbon storage in the study area was estimated at 18,591,817.72 tC, corresponding to an average of approximately 176 tC ha⁻¹ over the burned area (105,623 ha. Figure 4). After the fire, it decreased to about 14.7–12.9 MtC, depending on the assumed aboveground residual scenario, and the mean density declined to roughly 140–123 tC ha⁻¹. Thus, aboveground carbon losses attributable to the wildfire were 3,846,681.10–5,655,446.45 tC (about 3.85–5.66 MtC), equivalent to approximately 21–30% of the pre-fire aboveground stock, as shown in Fig. 5. On a per-area basis, this corresponds to an average loss of 38.7–56.8 tC ha⁻¹, or about 142–209 tCO₂ ha⁻¹ when expressed in CO₂-equivalent terms.

Spatial patterns of carbon storage indicate that, prior to the wildfire, high aboveground carbon densities of 150–240 tC ha⁻¹ were broadly distributed across forested mountain areas, whereas lower values of 30–90 tC ha⁻¹ were observed in riparian zones, croplands, and urban areas. Overall, the Uiseong wildfire reduced the average carbon storage capacity of the burned area by roughly one-fifth to nearly one-third, with particularly pronounced declines in zones experiencing high-severity fire.

Based on the aboveground carbon storage loss (3,846,681.10–5,655,446.45 tC, approximately 3.85–5.66 MtC) estimated from pre- and post-fire stocks, we converted this amount to CO₂ equivalents and applied the average emissions trading price of Korea to estimate the associated economic cost. The carbon mass (tC) was converted to tCO₂ by multiplying the molecular mass ratio of carbon dioxide to carbon (3.667) as reported by Pearson et al.[18], and a unit price of 8,793 KRW per tCO₂ was applied.

As a result, the pre-fire economic value of aboveground carbon storage in the study area was estimated at approximately 599.5 billion KRW (₩599,473,287,728), while the post-fire value decreased to about 475.4–417.1 billion KRW (₩475,441,167,761–₩417,119,462,620), depending on the assumed residual scenario (Fig. 6). The difference between these two estimates represents the loss of carbon storage capacity and corresponds to roughly 14.1–20.7 MtCO₂. This loss is equivalent to an economic cost of approximately 124.0–182.4 billion Korean won (₩124,032,119,967–₩182,353,825,108) at the applied ETS price. The burned forest area (105,623.91 ha) lost roughly 1.2–1.8 million KRW of carbon value per hectare. Aboveground carbon stocks declined by 21–30% relative to pre-fire storage, showing that a single megafire can eliminate a large portion of forest carbon within a short period.

We evaluated the accuracy of the InVEST carbon storage model by comparing its estimates with the aboveground biomass from the European Space Agency’s Climate Change Initiative (ESA CCI Biomass). The ESA CCI product provides annual, 100m AGB maps for global forests, along with per-pixel standard-deviation layers for selected years[19, 20]. For comparison, we resampled the ESA CCI AGB layers to the resolution of the InVEST outputs and computed pixel-wise differences, as shown in Fig. 7. Differences were defined as ESA CCI carbon storage minus InVEST carbon storage. Positive values therefore indicate higher ESA estimates, and negative values indicate higher InVEST estimates. Most differences range from − 25 tC (broadleaved forests, red; coniferous forests, green) to − 50 tC (mixed forests, blue), suggesting that InVEST generally produces larger carbon estimates. Part of this discrepancy reflects the inclusion of belowground carbon in InVEST, although additional sources of error also contribute.

The 2025 wildfire that started in Uiseong, Gyeongsangbuk-do was the largest recorded wildfire in the Republic of Korea, and our analysis suggests that carbon storage in the burned forests was reduced by up to about 100 tC ha⁻¹ in the most severely affected areas. Given that many of the burned stands stored roughly 100–200 tC ha⁻¹ of aboveground carbon prior to the fire, this implies that a substantial portion of the local carbon storage capacity was lost. Total aboveground carbon stocks in the study area decreased from about 18.59 MtC before the fire to approximately 14.75–12.94 MtC after the fire, depending on the residual fraction scenario, corresponding to a loss of 3.85–5.66 MtC. With carbon stock expressed in CO₂-equivalent units, this loss amounts to roughly 14.1–20.7 MtCO₂. Applying the 2025 average price in the Korean ETS (8,793 KRW tCO₂⁻¹) yields an estimated economic loss of about 124.0–182.4 billion KRW. This is comparable to the annual emissions of approximately 3.2–4.7 million passenger vehicles, highlighting the large impact that a single wildfire can have on regional carbon budgets and the national carbon economy.

Globally, forests store an estimated 662 GtC as of 2020, with roughly half of this stock contained in aboveground biomass and soil carbon pools[21]. Over recent decades, forest biomass carbon in northern ecosystems has generally increased. However, since around 2016 it has shifted from a positive to negative trend, largely associated with wide spread drought, wildfires and other disturbances[22]. In this context, our finding that a single wildfire in Uiseong removed a sizable portion of the carbon storage accumulated over several decades suggests that the stability of forest carbon storage is far more uncertain than previously assumed. With the disturbance of soil seed banks, losses of soil organic carbon, and delayed emissions from dead wood, wildfires represent more than a short-term carbon release. They also consume future carbon sequestration that would have accumulated over the coming decades.

The unexpected depletion of carbon storage also has important implications for the national greenhouse gas (GHG) inventory and carbon neutrality strategy. As a Party to the UNFCCC and the Paris Agreement, the Republic of Korea is required to submit an annual national GHG inventory, and its net emissions, including the land use and forest sector, were reported to be approximately 686.5 MtCO₂ eq in 2022[23].

In this context, the loss of carbon storage capacity associated with the Uiseong wildfire was estimated at about 14.1–20.7 MtCO₂ which corresponds to roughly 37–55% of the annual net removals in 2022, and reported for the Land Use, Land-Use Change and Forestry (LULUCF) sector. This suggests that a single megafire can partially offset or distort the annual carbon sequestration of forest sinks recorded in the national inventory. Under climate conditions where the frequency and magnitude of wildfires are increasing, such events may become a major driver of year-to-year variability in the inventory.

The IPCC Guidelines in 2006 and the LULUCF Good Practice Guidance adopt the managed land proxy whereby emissions and removals in the land sector are reported as total fluxes from managed lands[5]. This approach represents a pragmatic compromise, recognizing the practical difficulty of disentangling anthropogenic and natural fluxes in the land sector. However, some countries, including Canada and Brazil, have introduced supplementary approaches that explicitly distinguish natural disturbances such as wildfires in their reporting. Korea is likewise discussing institutional improvements to refine its LULUCF statistics, and there is a need for further debate on how to incorporate emissions from disturbance events like wildfires, as well as their long-term recovery trajectories, into the national inventory. Our results offer a foundation for separate disturbance accounting and for interpreting national inventory outcomes in years affected by major wildfires.

4.3. Linkages with the Paris Agreement, the Framework Act on Carbon Neutrality and Green Growth, and the K-ETS

The structure of the national GHG inventory and LULUCF statistics is, at a higher level, shaped by the Paris Agreement and Korea’s domestic carbon neutrality legislation and institutions. The Paris Agreement aims to reduce global net emissions to near zero around 2050 and, in Article 5, explicitly calls for the conservation and enhancement of sinks and reservoirs, including forests. In line with this, Korea has submitted a Nationally Determined Contribution (NDC) targeting a 58–61% reduction in emissions by 2035 compared with 2018 levels and has enacted the Framework Act on Carbon Neutrality and Green Growth for Coping with Climate Crisis, which codifies its 2050 carbon neutrality vision. This Act mandates the establishment of national and local carbon neutrality master plans. It also requires climate change impact assessments, and the creation of a carbon neutrality fund. In addition, it identifies the conservation and expansion of sinks, such as forests and agricultural lands, as key policy instruments.

Korea has also operated the Korea Emissions Trading Scheme (K-ETS) since 2015, which has become a core mitigation instrument covering major emitters in the power, industry, building, and transport sectors, namely around 800 large point sources that together account for more than 70% of national emissions. As of 2024, the average auction price in the K-ETS is reported to be about 10,355 KRW tCO₂⁻¹, and the average secondary-market price to be about 9,238 KRW tCO₂⁻¹, with cumulative auction revenues reaching roughly 1.4 trillion KRW. In this study, we applied a comparable market price of 8,793 KRW tCO₂⁻¹ to estimate the value of the carbon storage loss caused by the Uiseong wildfire at approximately 124.0–182.4 billion KRW. This implies that a single wildfire effectively destroys a volume of carbon assets whose value is equivalent to a substantial portion of the fiscal resources accumulated over several years through the ETS, thereby eroding the practical room for maneuver afforded by carbon-neutral policy instruments.

Therefore, the Uiseong wildfire suggests that the assumed stability of forest sinks, which underpins the implementation of Korea’s NDC under the Paris Agreement, the Framework Act on Carbon Neutrality and Green Growth, and domestic mitigation policies centered on the K-ETS, should be reconsidered. As disturbances such as wildfires, drought and pest outbreaks become more frequent, it may be unrealistic to assume that forest sinks will remain stable over time. More conservative sink scenarios that explicitly account for wildfire risk are therefore needed.

As wildfire risk increases and uncertainty surrounding forest carbon sinks grows, the need for assessment tools that can more precisely quantify wildfire-induced carbon losses is becoming urgent. Nevertheless, in practical national inventory work, it remains difficult to incorporate wildfire damage at high spatial resolution due to limitations in data availability, cost, and time. In Korea as well, forest carbon stocks are estimated, using sources such as the National Forest Inventory and basic forest statistics. However, a systematic method for capturing carbon losses by burn severity in the immediate aftermath of large wildfires has yet to be established.

This study evaluated carbon loss by delineating the extent and severity of fire damage using the satellite-based RdNBR index and then applying severity-specific residual fractions to the pre-fire carbon storage map produced by the InVEST carbon model. This approach enables the estimation of post-fire reductions in carbon storage within a relatively short period after a wildfire, and the resulting estimates can serve as a useful baseline for national inventory agencies or local governments when applying temporary correction factors after large fires or when establishing separate accounts for natural disturbances.

Given that this study was designed to propose a rapid, simplified assessment method, it entails several important limitations. First, the accuracy of the carbon storage estimates derived from the InVEST model and remote sensing are limited. Although the analysis included a broad set of carbon pools, the residual fractions for these pools were taken from representative values reported in previous studies and applied uniformly across forest types. In contrast, for aboveground biomass, we distinguished between (A) a baseline residual assumption derived from previous LBE studies and (B) a more conservative residual assumption that assumes greater combustion in high-severity classes based on field surveys by the Korea National Park Research Institute. We then applied these differentially by burn severity. Consequently, the residual fractions for belowground biomass, soil carbon, and dead organic matter reflect broad averages rather than conditions specific to Korean forests or differences across severity classes. Developing more detailed and locally calibrated biophysical tables will therefore be essential for improving the accuracy of these pool estimates.

Furthermore, transitions between carbon pools were not explicitly simulated. Wildfire-induced tree mortality results in both immediate emissions and transfers of carbon from live biomass into dead organic matter pools such as standing dead wood and litter. A simple stock-difference approach, however, cannot distinguish between these processes and may implicitly treat all reductions in biomass as atmospheric emissions. In this study, severity-specific residual fractions were applied to estimate post-fire changes in dead organic matter and soil pools. However, we did not separate carbon transferred into these pools from carbon lost through combustion or decomposition. Moreover, the distinction between residual scenarios (A and B) was applied only to aboveground biomass. Belowground, soil, and dead organic matter pools shared the same residual values in both scenarios. As a result, the comparison between scenarios does not fully capture uncertainties in other carbon pools or the transition processes among them.

Second, the severity-specific residual fractions do not adequately capture the structural heterogeneity of Korean forests, including differences in species composition, stand age, diameter class, and stand density. For developing a rapid, nationally applicable assessment method, we assumed a uniform stand structure and varied the residual fractions only by burn severity. As a result, variation in combustion patterns among forest types within the same severity class, for example, between young plantations, mature conifer stands, or old broadleaved stands, was not represented, causing potential for over- or underestimating carbon losses, particularly in old conifer forests or very dense stands.

Future research should therefore (1) refine residual fractions by forest type and stand age through integration with existing forest survey data, such as the National Forest Inventory and surveys in national parks, (2) develop nation-wide carbon models for soil and dead biomass pools that incorporate domestic observational and experimental data and (3) conduct long-term carbon budget analyses that account for wildfire occurrence probabilities under climate scenarios and repeated-burn scenarios. As such work accumulates, evaluating how large wildfires affect Korea’s carbon-neutral pathway and national GHG inventory will be more precise, and policies can be designed under the Framework Act on Carbon Neutrality and Green Growth and the K-ETS that systematically incorporate wildfire risk.