Warm Arctic-Cold Eurasia pattern helps predict spring wildfire burned area in West Siberia

Abstract

Extreme wildfires have devastating impacts on multiple fronts, and associated carbon greatly heats the earth’s climate. Whether and how to predict wildfires becomes a critical question. In this study, we find that the preceding-winter “warm Arctic-cold Eurasia” (WACE) pattern significantly enlarges the spring burned area in West Siberia. The winter WACE and accompanying snow reduction result in dryness and vegetation exposure in West Siberia in spring, increasing fire risks. A multiple linear regression model is constructed that successfully predicts the spring burned area in West Siberia one season in advance (R-squared coefficient=0.64). The same predictors also well predict the corresponding fire carbon emissions. Independent predictions for spring burned area in 2019 and 2020 are very close to observations, with a mean absolute percentage error of only 3.0%. The findings of this study provide a possibility for guarding humans against extreme wildfires and predicting sharp rises in carbon emissions.

The authors show that the winter “warm Arctic-cold Eurasia” pattern significantly enhances spring fire activity in West Siberia, contributing to the prediction of wildfire burned area, and resulting CO2 emissions, one season in advance.

Introduction

Extreme wildfires are expected to become more intense and frequent under global warming^1–3. Widespread burning has dramatically impacted the global carbon cycle by emitting carbon dioxide and altering multiple carbon sources and sinks^4,5. It is estimated that approximately 2 billion metric tons of carbon is emitted annually by wildfires into the atmosphere⁶. In recent years, frequent wildfires invaded the boreal forests near the Arctic Circle. The permafrost in this region have the potential to directly melt^7,8, turning it into a vast net carbon source^9,10. The burning of vegetation also degrade forests and hinder the ecosystem recovery^11,12. As a result, the carbon sink capacity of terrestrial ecosystems is weakened, further contributing to global warming¹³. Thus, the interaction between climate warming and wildfires forms a positive feedback cycle^4,14. In addition, the harm of wildfires is not limited to direct effects such as human casualties, economic losses, and biodiversity reduction^15,16. The air pollution from fire smoke can spread hundreds of kilometers and affect long-term human health^17,18. During 2010 to 2019, as many as 2.18 billion people worldwide were exposed to severe fire air pollution for at least one day per year¹⁹. In the spring and summer of 2023, the record-breaking wildfires in Canada caused severe environmental issues around the world²⁰, and the staggering carbon emissions made Canada the fourth largest emitter of greenhouse gases globally in 2023²¹. At the same time, the Kurgan region of West Siberia also experienced intense and large-scale wildfires in the spring, burning an area of 11,000 hectares²². However, the research on wildfires in West Siberia has received much less attention compared to that in other regions, such as the pan-Arctic, North America and Australia^8,20,23.

The frequency and intensity of wildfires are significantly impacted by weather and climate conditions, including hot temperatures, dry conditions, and high wind speeds^4,24. Linkages between wildfires and these factors exist on multiple time scales from synoptic to interannual. The wildfires in the West Coast of the United States and Southeastern Asia experience evident synoptic-scale variations, but the difference is that the former primarily modulated by wind speed and humidity and the latter by rainfall²⁵. On the seasonal to interannual time scale, the large-scale climate modes, such as El Niño–Southern Oscillation, Atlantic sea surface temperatures anomalies, Arctic sea ice melting and Arctic Oscillation can teleconnectively modulate fire activities through the atmospheric processes and cumulative effects of snow^26–29. The “warm Arctic-cold Eurasia” (WACE) is a typical winter climate pattern in the mid-high latitudes^30,31. The opposing temperature anomalies between the Arctic and Eurasia can extend from the surface to the upper troposphere³². The related atmospheric circulations also adjust simultaneously, manifested by the enhanced Siberia high and Ural blocking, as well as the weakened westerly jet³³. These phenomena, from the surface to the troposphere, involving components ranging from temperature to atmospheric circulations, constitute an integrated WACE pattern³⁴. This pattern has significant impacts on the extreme cold and snowfall in East Asia and Europe^35,36 which may impact the spring climate. However, the connections between the winter WACE pattern and the extreme wildfires in the following spring are still unclear.

The preceding climate variability can provide important predictable information for extreme disaster events, contributing to the construction of corresponding prediction models³⁷. The drift or split of the stratospheric polar vortex can significantly contribute to the subseasonal prediction of cold extreme in the mid-low latitudes of the northern hemisphere³⁸. The sea ice loss in the Barents-Kara Seas is also a predictable source of extreme cold events in East Asia³⁹. The oceanic forcing can improve the predictability of snow cover in the northern hemisphere by affecting surface temperature, snowfall, and shortwave radiation⁴⁰. Changes of Arctic sea ice in different regions and the ocean memory effects are also the effective predictors for extreme air pollution such as winter haze aerosols and spring dust^41,42. The developments of these prediction models offer scientific support for anthropogenic responses to extreme climate. However, the wildfire research is currently focused on the satellite retrieval¹⁴ and weather-fire feedback^3,25, with few attempts to predict extreme wildfires 1–3 months in advance using the preceding climate signal.

Given the significant climate impacts of the WACE pattern, we propose investigating the potential relationships between the winter WACE pattern and the spring wildfires in West Siberia, as the results may help build models to predict wildfires. Thus, we first examine the local and simultaneous meteorology influencing spring wildfires in West Siberia, and then explore the pivotal mechanisms how preceding winter WACE pattern modulated subsequent spring wildfires. Based on the physical linkages revealed in this study, an effective prediction model is constructed for the burned area of spring wildfires in West Siberia, as well as for the fire CO₂ emissions.

Results

Meteorological conditions enhancing wildfires

Wildfires are expressed by the burned area from the FireCCI51 Grid product in this study. The fire activities in the mid-high latitudes of western Eurasia climatologically and intensively occur in April and May during the period of 2001–2020 (Supplementary Fig. 1). The spring wildfires in western Eurasia are more climatologically intense and extensive in West Siberia (35–85°E, 52–58°N, the black box in Fig. 1a), and the April–May burned area in West Siberia is defined as BAWS_AM. The BAWS_AM together amounts to about 65% of the total burned area of the year in this region during 2001–2020 (Fig. 1a, see Methods). Across the northern hemisphere, the spring wildfires in West Siberia are also prominent⁴³, whose burned area in April–May accounts for approximately 28% of the burned area in the northern hemisphere during the same period (see Methods). The West Siberia is relatively sparsely populated⁴⁴, especially west of Ural Mountain where severe wildfires burning occur, suggesting that the occurrence of wildfires in this region is mostly regulated by meteorology and climate variability⁴³.

Fig. 1. Wildfire activity over West Siberia and its linkage with meteorological conditions.

a Climate mean of the April–May burned area (unit: 10²ha) from 2001 to 2020. Hatched areas indicate that the burned area in April–May accounts for more than 65% of the annual burned area. The black box indicates the area of West Eurasia. b Correlation coefficients between the April–May burned area in West Siberia (BAWS_AM) and the metrological conditions, including the local synchronous conditions and the cumulative conditions, which were calculated from two datasets of burned area (brown triangles represent the FireCCI51 from 2001 to 2020; blue squares represent the FireCCILT11 from 2001 to 2018). The local synchronous factors include April–May soil moisture (LSM_AM), potential evapotranspiration (LPET_AM), standardized precipitation-evapotranspiration index (LSPEI_AM), precipitation (LPre_AM), and surface air temperature (LSAT_AM) in West Siberia. The cumulative conditions include April local snow cover (LSnowc_A), January snowfall in the mid-high latitudes of western Eurasia (SF_J), February snow depth in the northern west Siberia (SD_F) and February–March local normalized difference vegetation index (LNDVI_FM). Detailed calculation methods of these meteorological factors are provided in Methods. The dashed lines represent the 95% confidence level. The linear trend is removed. Source data are provided as a Source Data file.

The impacts of meteorological conditions on spring wildfires burned area in West Siberia can be divided into the synchronous effects of local April–May meteorological conditions and the cumulative effects of the preceding climate variability. The associations between BAWS_AM and the local April–May meteorological conditions in West Siberia are first identified (Fig. 1b). The linear trends of BAWS_AM and meteorology are removed to amplify the interannual-decadal component (see Methods).

The dry conditions are revealed to play an important role for the wildfires burning ^27,28. The BAWS_AM is remarkably and negatively correlated with the April–May local soil moisture and positively related to local potential evapotranspiration (Supplementary Fig. 2a, c), with correlations coefficients of −0.68 (p < 0.01) and 0.45 (p < 0.05), respectively (Fig. 1b). Low soil moisture and strong potential evapotranspiration signify the dry conditions, which can be reflected in the response of standardized precipitation-evapotranspiration index (SPEI, see Methods). The negative SPEI anomalies in West Siberia indicate drought conditions (Supplementary Fig. 2b), favoring the occurrence of wildfires. Moreover, the April–May precipitation in West Siberia is decreased when the BAWS_AM is larger than normal (Supplementary Fig. 2d), further exacerbating the dry conditions. Many studies also point out that the local hot surface air temperature (SAT) have significant impacts on fires burning^22,24. However, we find the relationship between the BAWS_AM and the April–May SAT in West Siberia (LSAT_AM) is insignificant. A possible reason may be the low base temperature and weak variability in West Siberia during the early 21st century (Supplementary Fig. 3). While under the influences of global warming, the local base temperature and the LSAT_AM variability increased in the past decade, and the correlation between the two begin to rise. For example, in the 2020 year, the local heating had significant impacts on BAWS_AM (Supplementary Fig. 4). The above relationships are also verified by another set of burned area data (FireCCILT11) using an overlapping time period of 2001–2018 (Fig. 1b). Different burned area data both indicate that the April–May dry conditions in West Siberia are the dominant meteorological driving factors for BAWS_AM, especially the local soil moisture.

In addition to the significant response of the local April–May meteorology, the cumulative variations in snow are also remarkably associated with BAWS_AM. When the BAWS_AM is enhanced in spring, the April local snow cover in West Siberia is abnormal decreased (Fig. 1b, Supplementary Fig. 5c), which may be due to the persistent and accumulated effects of the snow variation in the preceding winter. The less snowfall in the mid-high latitudes of western Eurasia in the preceding January (defined as SF_J, see Methods) lead to a continue decrease in February snow depth in the northern West Siberia (defined as SD_F, see Methods; Supplementary Fig. 6a), and the above snow factors are both significantly correlated with BAWS_AM (Fig. 1b, Supplementary Fig. 5a, b). As a result, the cumulative effect of snow reduction in the preceding January–February is stored into the spring, causing an abnormal decrease in snow cover in April (Supplementary Fig. 6). It can be seen that the accumulative variation of snow from preceding winter to spring are all closely related to BAWS_AM. Moreover, the raw material of the fire is also an important condition. Accompanied with the enhanced BAWS_AM, the normalized difference vegetation index (NDVI, see Methods), which quantifies the growth and density of vegetation⁴⁵, shows a favorable vegetation state from February to March in West Siberia (Fig. 1b, Supplementary Fig. 5d), providing sufficient fuel for wildfires. Therefore, whether the spring BAWS_AM, as well as the accumulation of snow and vegetation, are linked to the climate variability in the preceding winter need to be further explored.

Lagged linkage between winter WACE and spring wildfires

The winter climate variability can impact the meteorological conditions in the subsequent spring through the memory effects of less significantly varying climate variables, thereby affecting the occurrence of extreme events⁴⁶. The January–February SAT anomalies in the preceding winter associated with BAWS_AM present a distinct WACE pattern (Fig. 2a). The January–February SAT difference between the Barents–Kara seas (70°–85°N, 40°–100°E) and Eurasia (45°–65°N, 50°–110°E) after detrending is defined as WACE_JF index (the former minus the latter) to represent the winter WACE intensity. The positive WACE_JF represents a WACE phase in winter, while the negative value is expressed as the cold Arctic-warm Eurasia phase (CAWE). The correlation coefficient between WACE_JF and BAWS_AM is 0.50 (p < 0.05), indicating that the significant January–February WACE pattern can greatly enhance the burned area of spring wildfires (Fig. 2d). While in the years when CAWE is remarkable, the BAWS_AM is much weaker than the climate mean. The significant winter WACE or CAWE phase in 12 years match well with the positive or negative BAWS_AM responses except for 3 years (Fig. 2c), and the significant and intensive anomalies of burned area mainly happen in areas west of Ural mountain (Supplementary Fig. 7).

Fig. 2. The relationship between winter Warm Arctic-cold Eurasia pattern and spring wildfire activity.

a Composite January–February surface air temperature (SAT) according to the April–May burned area in West Siberia (BAWS_AM). The black boxes represent the locations of the Barents–Kara seas and Eurasia. The white and gray dots indicate that the results are significant above the 95% and 90% confidence level, respectively. b Correlation coefficients between the BAWS_AM and the integrated winter warm Arctic-cold Eurasia pattern (WACE), including the January–February WACE index (WACE_JF), WACE pattern at 500 hPa, westerly jet, Ural blocking and Siberia high (see Methods), which are calculated from two datasets of burned area (brown triangles represent the FireCCI51 from 2001 to 2020; blue squares represent the FireCCILT11 from 2001 to 2018). The dashed lines represent the 95% confidence level. c Temporal evolution of WACE_JF index (line; unit: °C) and BAWS_AM (bars; unit: Mha) after detrending from 2001 to 2020. d Monthly BAWS_AM (unit: Mha) for 2001–2020 in each year (thin lines, red and blue for each year in the composition), climate mean (black), composite for WACE_JF index > 0.7 standard deviation (significant warm Arctic-cold Eurasia phase, red), and WACE_JF index <−0.7 standard deviation (significant cold Arctic-warm Eurasia phase, blue). The linear trend in (a, b, c) is removed. Source data are provided as a Source Data file.

The connection between the winter WACE and spring BAWS_AM is not limited to surface temperature (Fig. 2b). The January–February temperature anomalies associated with BAWS_AM can extend to the middle troposphere, showing deep warming in the Arctic center and deep cooling in the Eurasia center (Supplementary Fig. 8a). The enhancement of Siberia high in surface and Ural blocking in middle troposphere in January–February, as well as the weakening of westerly jet in upper troposphere, are all significantly correlated with BAWS_AM (Supplementary Fig. 8b, c), exhibiting a positive pressure structure. The atmospheric circulations in winter before the active wildfire period are stronger than that during the active wildfire period (Supplementary Fig. 8d–f). The significant correlation coefficients between BAWS_AM and each key circulation factors in January–February are verified using the two burned area data (Fig. 2b). It can be seen that the winter integrated WACE pattern (WACE_Whole, see Methods), including the temperatures and atmospheric circulations from surface to troposphere, comprehensively modulates the spring BAWS_AM (R = 0.50, p < 0.05). The winter climate varies ahead of the spring wildfires, implying that the winter WACE_Whole is a potential predictor for BAWS_AM.

The integrated winter WACE pattern is not only reflected in the atmospheric signals, but also closely related to the variation of snowfall and snow cover (Supplementary Fig. 9a, b). When the WACE occurs in January–February, the significant Ural high is accompanied by the abnormal descending movements, and the easterly winds in the south of the Ural high disperse the water vapor (Supplementary Fig. 9c). At this time, the January SF_J that closely linked to BAWS_AM is significantly decreased (Supplementary Fig. 9a), with the correlation coefficient between SF_J and WACE_JF at –0.56 (p < 0.01). The SD_F in the subsequent February also continue to decrease, which is also significantly correlated with WACE_JF (R = –0.47, p < 0.05). The SF_J and SD_F show significant linkages with both winter WACE pattern and spring BAWS_AM (Fig. 1b, Supplementary Fig. 9a, b), which implies that the snow variations may be a key bridge linking the winter climate variability and spring fire activity.

The effect of the reduced snowfall in January and thinning snow depth in February can be stored and continued into the spring, causing the less snow cover and snowmelt (Supplementary Fig. 6). The negative snowmelt anomalies reduce the spring soil moisture and increase the wildfire risk²⁷ (Supplementary Fig. 10a). The thinning snow cover makes the snow line withdraw northward earlier and the surface exposed more quickly, which facilitates to potential evapotranspiration from surface to atmosphere and further aggravates the surface dryness²⁸ (Supplementary Fig. 10b). It can be seen that the snow variations mainly regulate the April–May soil moisture and potential evapotranspiration intensity in West Siberia, thus impacting the dry conditions (Supplementary Fig. 9d), which are the major meteorological driving factors of BAWS_AM (Fig. 1b). In addition, the snow covers also impact the vegetation, whose effect on phenology plays a decisive role in West Siberia compared to that of soil moisture⁴⁷. The reduction of snow covers and snowmelt is conducive to the vegetation exposed and growing season advanced, which provides more fuels for wildfires (Supplementary Fig. 10d). The effect of snow on vegetation reflect the process of accumulative change on a seasonal scale. The January–February WACE pattern and the accompanying snow reduction cause a dry and vegetation exposed West Siberia in spring, contributing to the occurrence of wildfires. The WACE-related snow variations (SNOW_WACE, see Methods), characterized as reduced SF_J and SD_F, smoothly explain the lagged relationship between the spring wildfires and the winter WACE pattern, and may also be an important signal for predicting BAWS_AM (R = 0.66, p < 0.01).

Predicting wildfires in advance

Considering the significant physical linkages of spring BAWS_AM with winter WACE_whole and SNOW_WACE (Supplementary Fig. 11), we are motivated to construct a multiple linear regression model to predict the BAWS_AM using these two predictors, and evaluating by the “leave-one-out” cross-validation (see Methods). The R-squared coefficient between the observed and predicted BAWS_AM is 0.50 (Fig. 3a), indicating that this model can effectively predict the BAWS_AM as long as a season in advance. However, the cause of wildfires is complicated, whose related meteorological conditions are not only modulated by the winter WACE pattern. Taking the 2020 year for example, the synchronous April–May high pressure located over the eastern west Siberia also significantly impacts the BAWS_AM by heating local temperatures (Supplementary Fig. 4, 12). Therefore, the predicted April–May high pressure by NCEP Coupled Forecast System Model (defined as Z500_CFS; see Methods) is brought to optimize the prediction performance (Supplementary Fig. 12b). The Z500_CFS can well predict the variation of this atmospheric circulation factor, with a correlation coefficient at 0.59 (p < 0.01).

Fig. 3. Prediction for spring wildfire activity.

Temporal variations of original observed (black) and predicted (blue and orange; evaluated by the “leave-one-out” cross-validation) (a) April–May burned area in West Siberia (BAWS_AM; unit: Mha) and (b) corresponding fire CO₂ emissions (unit: Tg) from 2001 to 2020. The blue lines represent the predicted BAWS_AM and fire CO₂ emissions based on the model using the integrated warm Arctic-cold Eurasia (WACE) pattern (WACE_Whole) and the WACE-related snow variations (SNOW_WACE), and the orange lines represent the predicted BAWS_AM and corresponding fire CO₂ emissions based on the optimized model using WACE_Whole, SNOW_WACE and the predicted April–May high pressure (Z500_CFS). The orange asterisks represent the independent prediction results of 2019 and 2020 based on the optimized prediction model. c, d The spatial distribution of original burned area (unit: 10²ha) observed and predicted based on the optimized prediction model in April–May of 2019 and 2020. The black box indicates the area of West Eurasia. Source data are provided as a Source Data file.

The final prediction model for BAWS_AM shows improvement (Fig. 3a), with the R-squared coefficient between the predicted and observed BAWS_AM increasing to 0.64, and the percentage of the same sign for predicted BAWS_AM anomalies reaches 80% (see Methods). The mean absolute error (MAE) of the predicted BAWS_AM is 1.5 Mha, which is only accounts for 30% of its climate mean. The independent predictions of BAWS_AM in 2019 and 2020 are very closed to the observations (Fig. 3a), with the mean absolute percentage error (MAPE) for these two years at only 3.0%. This effective ability also reflects in the spatial distribution prediction of the burned area (Fig. 3c, d, see Methods), which accurately reproduces the spatial location of the strong wildfire centers. Moreover, the extremes intense BAWS_AM is predicted well. Of the five years with extreme BAWS_AM anomalies (BAWS_AM > one standard deviation), the extremes can be successfully predicted in four of them (Fig. 3a), which manifests as the predicted BAWS_AM also exceeding one standard deviation. The MAE of these four extreme years that can be correctly predicted is 0.6 Mha, and the MAPE is 6.2%, showing a strong ability to predict the extreme BAWS_AM.

Based on the closed linkages between the wildfires and carbon emission, we use the same three predictors above to predict fire CO₂ emissions of BAWS_AM (see Methods). The prediction exhibits a high degree of consistency with observation (Fig. 3b), with a R-squared coefficient of 0.45. The percentage of the same sign for predicted CO₂ emissions anomalies reaches 75%, and approximately 61% of the interannual variability of the fire carbon emissions can be explained by the combined effects of these three predictors. The MAE of the predicted CO₂ emissions is 4.5 Tg. For the four years of extreme high CO₂ emissions resulting from BAWS_AM, three years can be correctly predicted (Fig. 3b), with the MAPE for these three years at 22.9%, indicating a certain capability to capture the extreme CO₂ emission. It can be seen that exploring the linkages between climate variability and wildfires is not only able to predict the intensity of wildfires area in advance, but it is also further helpful in foreknowing sharp rises in the resulting carbon emissions.

Discussion

In this study, we mainly reveal the role of winter WACE pattern in enhancing the spring wildfires in West Siberia (Fig. 4). The reduction of snow associated with WACE in the preceding January–February establishes a bridge for winter WACE to modulate the spring BAWS_AM through the cumulative effect, which provides dry conditions in April–May in West Siberia and increases the fire risks. Most importantly, the winter WACE pattern helps to predict spring burned area of fire activities one season in advance. In addition to the significant effect of the climate variability in mid-high latitudes on wildfires, the El Niño located in tropic is also closely related to large-scale forest fires by impacting precipitation²⁶. However, the effect of tropic-subtropics sea surface temperature anomalies on BAWS_AM is not taken into account in this study. This may explain the spring local precipitation changes that WACE cannot well explain, thus further improving the predicted accuracy of BAWS_AM. For the prediction model of fire carbon emissions, there are many additional factors that are not included, such as the type of vegetation and soil¹³, which may be the reason why our prediction model is not perfect. This study provides a preliminary attempt to predict wildfire area and the resulting carbon emissions, but more comprehensive and scientific predicted methods are needed to deal with the increased extreme wildfires and carbon bombs under climate warming.

Fig. 4. Physical mechanisms of the winter warm Arctic-cold Eurasia (WACE) pattern enhancing the spring wildfires in West Siberia.

The winter overall WACE pattern, including the WACE temperature anomaly extending from the surface to the middle troposphere, the weakened westerly jet, and the strengthened Ural high and Siberia high, has a significantly strong impact on the spring wildfires in West Siberia through the accumulative effect of snow variation. The reduction of snow cover and snowmelt cause a dry and vegetation exposed West Siberia in April–May, increasing the extensive fire risks.

The strong covariation of winter WACE pattern and spring wildfires also show in the changes of trend and variability (Fig. 5). During 2001 to 2012, the January–February SAT in Arctic-Eurasia region showed a significant “Arctic warming-Eurasia cooling” trend, causing an increasing trend and strong interannual variability of BAWS_AM. While after 2012, the trend and variability of WACE_JF both significantly weakened, which modulated the amplitude of BAWS_AM to decrease sharply and the trend to become insignificant. The decreasing of “Arctic warming-Eurasia cooling” trend in the past decade is possibly due to the enhanced subseasonal variability of SAT in Arctic-Eurasia³³. However, the rapid “Arctic warming-Eurasia cooling” may reemerge before 2050s⁴⁸. This means that the intensity of spring wildfires in West Siberia may face an increasing trend and the emergence of extreme strong fire activities in the future.

Fig. 5. Co-variation of the winter warm Arctic-cold Eurasia pattern and the spring wildfires in West Siberia.

Temporal evolution of January–February warm Arctic-cold Eurasia index (WACE_JF; black; unit: °C) and April–May burned area in West Siberia (BAWS_AM; green; unit: Mha) from 2001 to 2020. The thin lines represent the trends in two time periods of 2001–2012 and 2013–2020. The trend values and variability in these two time periods marked in the corresponding periods. Source data are provided as a Source Data file.

In the context of global warming, the extreme wildfires have planted an untimed carbon bomb on the global carbon cycle, not only reflecting in the direct carbon emissions, but also in destroying the sensitive and carbon-rich terrestrial ecosystems. As major carbon sinks, the terrestrial ecosystems have strong interannual variability, and are the key factors in determining the variation of atmospheric carbon dioxide concentration⁴⁹. The findings of this study are also of great significance for predicting the terrestrial carbon sink and the global carbon cycle. As the co-occurrence of droughts and heat waves is increasing globally, the severity and frequency of global wildfires will face an enormous increasing risk in the future^1–3. The explorations on the interaction between wildfires and climate change help to broaden understanding of extreme wildfires. Predictions of future changes in wildfires and fire CO₂ emissions provide the ability to guard against the extreme wildfires and contribute to better responses and adaptations to climate change.

Methods

Burned area and fire emissions

The European Space Agency Fire Disturbance Climate Change Initiative (CCI) project has produced maps of global burned area derived from satellite observations. The FireCCI51 Grid product is the result of summing up burned area pixels and their attributes, as extracted from their original sinusoidal projection⁵⁰, within each cell of 0.25 degrees in a regular grid covering the whole Earth in monthly composites. Wildfires are expressed by the burned area from the FireCCI51 Grid product in this study. The intensity of the April–May wildfires in West Siberia is represented as the April–May burned area in West Siberia (BAWS_AM).

The fire CO₂ emissions dataset is estimated by using Measurements of Pollution in the Troposphere (MOPITT) satellite retrievals of CO and a global atmospheric inversion system⁵¹, with a horizontal resolution of 3.75° × 1.9°. Based on this dataset, the CO₂ emissions from the April–May wildfires in West Siberia to the atmosphere can be obtained (defined as the April–May fire CO₂ emission over 35–85°E, 52–58°N), which provides the basic data to further construct the prediction model for the fire CO₂ emissions related BAWS_AM.

Definition and calculation of multiple indices

To identify the meteorology conditions accompanied the April–May burned area in West Siberia, the fifth generation ECMWF atmospheric reanalysis data (ERA5)⁵² and the Climatic Research Unit data (CRU)⁵³, which assimilates vast amounts of historical observations are collected. The local weather parameters in West Siberia including the area-averaged April–May soil moisture (LSM_AM), potential evapotranspiration (LPET_AM), Standardized Precipitation Evapotranspiration Index (LSPEI_AM), surface air temperature (LSAT_AM) and precipitation (LPre_AM) are analyzed for the linkages between the local synchronous metrological conditions and the extreme wildfires in West Siberia. The “L” in the abbreviation represents “local”, making it more clear that it refers to the local conditions in West Siberia. The Standardized Precipitation Evapotranspiration Index (SPEI) is a multiscalar drought index⁵⁴, which is designed to take into account both precipitation and potential evapotranspiration in determining drought. The lower the SPEI value, the drier it is.

The winter snow variations in the mid-high latitude of western Eurasia related to WACE pattern can persist into spring and have an important role for the BAWS_AM through its memorability and cumulative effects. The area-averaged April local snow cover in West Siberia is defined as LSnowc_A index. The area-average January snowfall in the mid-high latitudes of western Eurasia (57°–70°N, 35°–90°E) is defined as SF_J index, and the area-average February snow depth in the northern west Siberia (53°–65°N, 55°–85°E) is defined as SD_F index. The calculation of the SNOW_WACE index for prediction is through the following equation:

$${\mathrm{SNOW}}_{\mathrm{WACE}}={\mathrm{CC}}_{\mathrm{SF}}\times \mathrm{normalized}{\mathrm{SF}}_{\mathrm{J}}+{\mathrm{CC}}_{\mathrm{SD}}\times \mathrm{normalized}{\mathrm{SD}}_{\mathrm{F}}.$$ 1

where the CC_SF, and CC_SD represent the correlation coefficients of BAWS_AM with SF_J and SD_D, respectively.

The winter integrated WACE pattern includes not only surface temperature anomalies in the Arctic-Eurasia region, but also the temperature anomalies in middle troposphere and key atmospheric circulations. The January–February air temperature difference at 500 hPa between the Barents–Kara seas (70°–85°N, 40°–100°E) and Eurasia (45°–65°N, 50°–110°E) represents the deep temperature feature of WACE pattern, defined as T_JF. The intensities of Siberia high, Ural blocking and westerly jet of the integrated WACE pattern were defined as follows. The January–February area-averaged sea level pressure over Siberia (55°–70°N, 30°–85°E) is defined as the variation in Siberia high (SH_JF). The January–February area-averaged geopotential height at 500 hPa over the Ural Mountains (60°–77°N, 20°–80°E) is defined as the variation in Ural blocking (UB_JF). The January–February area-averaged zonal wind at 200 hPa over the mid-high latitude of Eurasia (55°–70°N, 40°–130°E) is defined as the variation in westerly jet (WJ_JF).

The calculation of the index WACE_Whole was through the following equation:

$${\mathrm{WACE}}_{\mathrm{Whole}}={\mathrm{CC}}_{\mathrm{WACE}}\times \mathrm{normalized}{\mathrm{WACE}}_{\mathrm{JF}}+{\mathrm{CC}}_{\mathrm{SH}}\times \mathrm{normalized}{\mathrm{SH}}_{\mathrm{JF}}+{\mathrm{CC}}_{UB}\times \mathrm{normalized}{\mathrm{UB}}_{\mathrm{JF}}+{\mathrm{CC}}_{\mathrm{WJ}}\times \mathrm{normalized}{\mathrm{WJ}}_{JF}$$ 2

where CC_WACE, CC_SH, CC_UB, and CC_WJ represent the correlation coefficients of BAWS_AM with WACE_JF, SH_JF, UB_JF, and WJ_JF, respectively.

Vegetation cover: NDVI

Normalized Difference Vegetation Index (NDVI) is an indicator that quantifies the growth and density of vegetation by measuring the difference between near-infrared (which vegetation strongly reflects) and red light (which vegetation absorbs)⁴⁴. As shown below, NDVI uses the near infrared (NIR) and red (Red) channels in its formula:

$$NDVI=\frac{\left(NIR-Red\right)}{\left(NIR+Red\right)}$$ 3

NDVI ranges from −1 to +1. A score of −1 indicates the absence of vegetation and 1 the abundance of healthy vegetation. The denser and healthier the vegetation, the closer the NDVI is to 1. The area-averaged February–March local NDVI in West Siberia is defined as LNDVI_FM index.

Data treatment and statistical methods

In exploring the linkage between climate variability and BAWS_AM, the linear trend and its long-term (2001–2020) mean of all the monthly data for each calendar month are removed. The method of detrending is to first calculate the slope and intercept of the least squares linear trend line of the original variables during 2001–2020, and obtain the linear trend term of year(i) as slope×year(i)+intercept, and then remove the linear trend term from the original variables to obtain the detrended results.

In this study, the statistical methods such as correlation coefficient, composite analysis, trends analysis, and significance tests are used. The type of correlation coefficient used in this study is Pearson correlation⁵⁵, which measures the linear relationship between two random variables. The linear trends during the different sub-periods are calculated by the least squares methods. A two-sided student’s t-test is used to test the statistical significance of the composite analysis⁵⁵. The significance of the correlation and the slope rates of the linear trends during the different sub-periods can be tested by using the student’s t-test. The 99% confidence level is denoted by p  <  0.01.

Calculation of two percentages about burned area

Based on the calculation of the FireCCI51 product during 2001–2020, the total burned area of April and May in West Siberia accounts for about 65% of the total burned area of the whole year in this region. The total burned area of April and May in West Siberia is 100.4 million hectares during 2001–2020, and the total burned area of the whole year in West Siberia is 154.4 million hectares during 2001–2020. The result of “65%” can be obtained by dividing the two.

Similarly, the result that April–May burned area in West Siberia accounts for about 28% of the burned area in the Northern Hemisphere in April–May is also calculated based on the FireCCI51 product. The April–May burned area in the West Siberia and in the entire Northern Hemisphere during 2001–2020 are 100.4 million hectares and 352.2 million hectares, respectively. By dividing the two values, the result of “28%” can be obtained.

Identification of the month with maximum burned area

The identification method is to calculate the climate mean of the monthly burned area during 2001–2020 at each grid point, and then the month with the maximum burned area can be obtained through the comparison between each month. As shown in Supplementary Fig. 1, the months for the strongest burned area in West Siberia occur in April or May (the green shading).

Definition of the Z500_CFS

When building the prediction models for BAWS_AM and the resulting CO₂ emissions, we bring in the predicted April–May high-pressure factor (Z500_CFS) as a predictor (Supplementary Fig. 12b). The Z500_CFS is defined as the area-averaged April–May geopotential height at 500 hPa over the eastern west Siberia (45°–70°N, 75°–115°E) predicted by NCEP Coupled Forecast System Model⁵⁶, which is released in March.

“Leave-one-out” cross-validation for prediction

A multiple linear regression model is developed to predict the BAWS_AM and the resulting fire CO₂ emissions, incorporating the January–February integrated WACE pattern (WACE_Whole), the WACE-related snow variations (SNOW_WACE) and the predicted April–May high-pressure factor (Z500_CFS), and using the following equation:

$$\mathrm{Predicted}{\mathrm{BAWS}}_{\mathrm{AM}}=a_1{\mathrm{WACE}}_{\mathrm{Whole}}+a_2{\mathrm{SNOW}}_{\mathrm{WACE}}+a_3{\mathrm{Z}500}_{\mathrm{CFS}}+a_0$$ 4

$$\mathrm{Predicted}{\mathrm{CO}}_2=b_1{\mathrm{WACE}}_{\mathrm{Whole}}+b_2{\mathrm{SNOW}}_{\mathrm{WACE}}+b_3{\mathrm{Z}500}_{\mathrm{CFS}}+b_0$$ 5

where a₀, a₁, a₂, a₃ and b₀, b₁, b₂, b₃ denote the coefficients determined through the multivariable regression procedure for BAWS_AM and the resulting fire CO₂ emissions, respectively. WACE_Whole, SNOW_WACE and Z500_CFS are indices of the three predictors.

The prediction model is employed a more thorough “leave-one-out” cross-validation⁵⁷. For example, to predict the BAWS_AM in 2005, the regression coefficients are calculated using data in all years except 2005: 2001–2004, 2006–2020. Then to predict the year 2006, the model is fitted using 2001–2005, 2007–2020. Repeating this for each year gives the predicted BAWS_AM from 2001–2020, and the a₀, a₁, a₂, a₃ determined through the multivariable regression procedure are different from year to year. The regression models using “leave-one-out” cross-validation provide a more convincing prediction, and better illustrate the effective operational capability of the prediction models for BAWS_AM and the resulting fire CO₂ emissions.

The spatial prediction method for burned area in 2019 and 2020 is to use the same three predictors above to construct a prediction model for the burned area of each grid point separately, thus forming a distribution of predicted burned area.

Evaluation indicators for the prediction model

We use the mean absolute error (MAE), mean absolute percentage error (MAPE) and percentage of same sign (PSS) to evaluate the prediction model. The PSS is obtained by calculating whether the anomalies of observation and prediction are of the same sign, which can assess the predicted accuracy of the abnormal large or small wildfires. MAE and MAPE are indicators used to assess the error between observations and predictions. The formulas for calculating MAE and MAPE are as follows:

$$MAE=\frac{1}{n}{\sum }_{i=1}^n∣\widehat{y_i}-y_i∣$$ 6

$$MAPE=\frac{100\%}{n}{\sum }_{i=1}^n∣\frac{\widehat{y_i}-y_i}{y_i}∣$$ 7

where $y_i$ represent the observation, $\widehat{y_i}$ represent the prediction, and n is the number of years from 2001 to 2020, i.e. 20.

Author contributions

Z.Y. and Y.Z. performed this study, plotted all figures in the main text and supplementary information, and wrote the preliminary manuscript. S.H. and H.W. joined this project and gave discussions on the improvement of this manuscript. All the authors contributed to the writing and reviewing of the manuscript.

Peer review

Peer review information

Nature Communications thanks the anonymous reviewers for their contribution to the peer review of this work. A peer review file is available.

Data availability

The burned area from FireCCI51⁵⁰ are available at 10.5285/58f00d8814064b79a0c49662ad3af537 and from FireCCILT11⁵⁸ are available at https://catalogue.ceda.ac.uk/uuid/62866635ab074e07b93f17fbf87a2c1a. The inversion-based global gridded fire CO₂ emissions data⁵¹ are available at https://figshare.com/ndownloader/files/38634248. Monthly surface meteorological data from ERA5⁵² including the surface air temperature, precipitation, soil moisture, surface wind speed, sea-level pressure, snowfall, snow depth and snowmelt are available at https://cds.climate.copernicus.eu/cdsapp#!/dataset/reanalysis-era5-single-levels-monthly-means?tab=form. Monthly meteorological data on pressure levels from ERA5 including geopotential height at 500 hPa, zonal and meridional winds at 850 hPa, temperature and vertical velocity from 1000 to 200 hPa are available at https://cds.climate.copernicus.eu/cdsapp#!/dataset/reanalysis-era5-pressure-levels-monthly-means?tab=form. Monthly potential evapotranspiration and SPEIdata from CRU⁵³ are available at https://crudata.uea.ac.uk/cru/data/hrg/cru_ts_4.05/ and https://spei.csic.es/index.html and, respectively. Monthly snow cover data from Rutgers University Climate Laboratory⁵⁹ are available at https://climate.rutgers.edu/snowcover/docs.php?target=datareq. Gridded NDVI data from NOAA⁶² are available at https://www.ncei.noaa.gov/metadata/geoportal/rest/metadata/item/gov.noaa.ncdc:C01558/html. The forecast geopotential height at 500 hPa in April and May from NCEP Climate Forecast System version 2⁶⁰ are released in March and available at http://iridl.ldeo.columbia.edu/SOURCES/.NOAA/.NCEP/.EMC/.CFSv2/. Statistical methods are noted in the text and figure captions. The data generated for all figures of this study are provided in the Source Data file. Source data are provided with this paper.

Code availability

The computer codes for analyzing data and drawing plots are developed in NCAR Command Language (available at https://www.ncl.ucar.edu/). The computer codes used in this study are available from the corresponding author on request.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-024-53470-4.