Main

According to the Global Wildfire Information System1, on average, four million km2 of land burn every year (2002–2023), an area almost equal to the European Union. Wildfires are an integral part of many ecosystems around the globe, playing key roles in ecosystem dynamics and retention of species that have evolved in response to fire2. However, this global phenomenon is often responsible for environmental, social and economic losses, which combined with land abandonment, droughts, absence of appropriate land management and urban development are expected to aggravate land degradation3,4. In addition, wildfires will become a persistent threat because fire risk is expected to increase in a context of a warmer and drier climate5.

Led by elevated temperatures in topsoil, fire-induced changes in soil physical and chemical properties reduce water infiltration and ground-cover protection and alter soil aggregate stability6,7. When combined, these factors result in significant increases in soil erodibility, varying between geographical regions and burn severity8,9. High-severity burns increase on-site soil erosion but also lead to offsite impacts downstream of the burned area in the form of destructive floods and debris flows6,10 and transport of ash and sediment into downstream water bodies11,12. Moreover, the importance of post-fire soil erosion events is likely to increase in response to ongoing and anticipated increases in fire activity and rainfall intensity10.

Wildfires have been regarded as a hydrological and geomorphological agent7; until now post-fire soil erosion mostly has been assessed from plot to catchment scale by means of field measurements and from slope to regional scales using modelling approaches13. Global-scale studies, however, have the potential to compare impacts of fires in different biomes10,14,15 and to put into perspective land-management actions16. Current model-based soil erosion assessments at global scales17,18 already gave steps forward, providing this knowledge for informed land-management decisions, however, these have not yet considered the effects of wildfires. Whereas initial large-scale post-fire soil erosion estimates8 provide an idea of the impact of wildfires in European soils, this approach remains limited to a single disturbance (one fire year) and a short recovery period (five years), showing that a global assessment accounting for cumulative fire impacts and a long-term perspective is still missing8,10.

Hence the aim of this study is to provide the estimation of post-fire soil erosion at a global scale and to analyse geomorphological changes driven by fire over the past two decades under a land-management perspective. To accomplish this, we estimated the global post-fire soil erosion by adapting the Revised Universal Soil Loss Equation (RUSLE) model to account for cover changes induced by global wildfires. We used the largest global burn severity database available (MOSEV19) for an 18-year period and compared it against unburned conditions. This allowed to estimate the cumulative impact of several wildfire occurrences in the global soil erosion budget and identify the main trends for post-fire soil erosion at continental scale. Additionally, this allowed us to analyse the recovery of fire-affected ecosystems under systematic pressure, but also project trends into the future in combination with the anticipated changes in rainfall.

Results

Burned area, severity and erosion

Over the study area and period, annual burned areas averaged 2.9 million km2. Throughout 18 years, the trend in burned areas has been declining significantly globally (Mann–Kendall trend test p < 0.0005), with the exception of an increase in North America (Extended Data Fig. 1). Africa is the most-affected continent, with 66% of the annual burned area identified mostly among low (67%) and moderate (33%) severities. High-severity burns are more pronounced in North America (10%), Asia (5%) and Europe (4%) but are rare in other continents (Extended Data Fig. 1).

Burned area and severity were used as inputs for soil erosion estimations with RUSLE. However, to account for the long-term contribution of fires to the global erosion budget, our estimations consider the cumulative effect of various wildfires. Therefore, we have estimated the post-fire soil erosion from each individual fire, in every fire-affected pixel throughout the 18-year timeframe, summing up the erosion amounts at annual scale (Fig. 1a–c).

Fig. 1: Concept of cumulative post-fire impacts and mean annual estimates of background and post-fire soil erosion for the period 2006–2019.
Fig. 1: Concept of cumulative post-fire impacts and mean annual estimates of background and post-fire soil erosion for the period 2006–2019.
Full size image

a, The window of disturbance model for a single fire event7, where post-fire soil erosion increases after the fire and decreases with recovery. b, The window of disturbance for multiple wildfires with similar impacts if considered individually. c, The cumulative impact, achieving a plateau if impacts are similar (black line) and two alternative patterns indicating potential degradation (red line) whereas this plateau is achieved later or a positive recovery (blue line) indicating a decrease in impacts with time. d, Mean annual post-fire soil erosion estimates by continent (logarithmic scale). e, Mean annual post-fire soil erosion estimates globally (Table 1). Error bars represent standard deviation (n = 13). Post-fire erosion (orange bars) refers to the estimated soil loss following fire, accounting for fire-induced changes and based on ground-cover observations. These values serve as important references for emergency response and mitigation measures in fire-affected areas. Background erosion (green bars) represents the estimated soil erosion under undisturbed vegetation conditions before fire. These estimates consider all areas affected by fires but use pre-fire conditions for parameterization. This provides reference values for soil erosion under natural, undisturbed conditions. The difference between post-fire and background erosion estimates (shown as grey bars) represents the increase in erosion attributable to fire events because of ground-cover removal and fire-induced changes in soil properties. This approach enables assessment of post-fire erosion trends independent of local intrinsic factors such as topography, soil properties and climate.

To evaluate the mean and trends of post-fire soil erosion, we identified six years as the minimum period needed to capture the cumulative effects. This was estimated through the time to reach a plateau for background erosion (unburned conditions) and within the range of observed temporal effects provided by literature6,7,8. Using this metric, the fire-affected areas targeted in this study are estimated to result in 8.1 ± 0.72 Pg of soil losses annually over 2006–2019 (Table 1 and Fig. 1d). This value represents an additional 5.1 ± 0.56 Pg of soil losses annually when compared to background soil erosion (Fig. 2). The global erosion budget revealed that only 31% (2.50 Pg yr−1) of the annual soil erosion can be attributed to the most recent fire year, whereas the remaining results from previous fire events (Table 1).

Fig. 2: Global pre-fire and post-fire soil erosion for the year 2019.
Fig. 2: Global pre-fire and post-fire soil erosion for the year 2019.
Full size image

a, Global pre-fire soil erosion estimates. b, Global post-fire soil erosion estimates. The model provides erosion estimates for burned areas and cumulative impacts since 2001, covering land surfaces across 177 countries and approximately 12% of Earth’s terrestrial area. Soil erosion rates are categorized into seven classes based on the European Soil Bureau classification and the Global Soil Erosion Modelling Framework (GloSEM)18. Basemaps from Natural Earth (naturalearthdata.com) (projected coordinate system: sphere Robinson). Data from ref. 50.

Table 1 Mean annual soil erosion in Pg yr−1 for post-fire, background and additional erosion per continent and at global scale
Full size table

Global post-fire erosion is mostly driven by African burned areas (Table 1 and Figs. 1 and 2), which corresponds to 62% of the soil losses, followed by Asia (12%), South America (11%) and Oceania (10%), with minor contributions from North America (4%) and Europe (1%) (Fig. 3), resulting in a global soil erosion rate of 9.53 Mg ha−1 yr−1 for the first post-fire year (Supplementary Section 1). Post-fire soil erosion estimates are mostly originated from tropical, savannah (60%) climatic zones, which are the largest contributor to erosion across most continents (Table 1). In Europe, the dominant climate types contributing to soil erosion are cold, no dry season, warm summer; and temperate, dry summer, hot summer, with similar (21–23%) share (Table 1).

Fig. 3: Mean recovered areas by time since fire per continent.
Fig. 3: Mean recovered areas by time since fire per continent.
Full size image

a–f, Dots concern average share of recovered areas (%) for given time since fire (years) between 2001 and 2019 and per continent: Africa (a), Asia (b), Europe (c), North America (d), Oceania (e) and South America (f). Line expresses tendency of data by locally weighted smoothing; shaded area represents standard error with 0.95 confidence interval (ggplot2, RStudio). Recovered areas are defined as areas that achieve pre-fire vegetation cover conditions. RCOVER is determined for each continent following a single fire event and throughout the study period. Recurrent fires in the same pixel were removed from long-term assessment and reclassified as another event.

Post-fire recovery and long-term soil erosion trends

Within our study it was possible to assess recovery of the affected land through the RCOVER8, which represents the progress of vegetation recovery when compared to pre-fire conditions. As expected, the overall recovery dynamics at continental scale reveal complex pattern whereas factors such as burn severity, ecosystem resilience and post-fire climate play major roles10.

Results show that following a single fire, continents such as Oceania and South and North America present the highest recovering performance, whereas Europe and Africa reveal modest recovery rates (Fig. 3). Additional disturbances beyond wildfires might interfere with recovery, thus explaining why the maximum recovered areas after a single fire event was as much as 83% for Oceania, 75–76% for North and South America, 67% for Asia, 56% for Europe and 46% for Africa. These results should also include the effect of precise management actions, such as logging or post-fire mitigation measures20 through visible changes in vegetation indices.

When investigating recovery considering cumulative impact over several fire years (Extended Data Fig. 2), the complexity between affected land cover, climate and biome emerge. At a global level, recovered areas represent 39% of the 2019 baseline, whereas areas under high level of pressure (RCOVER [0.0–0.3[) stand in the 13% (Extended Data Fig. 2d). However, these values tend to be more favourable in forest-dominated landscapes (Recovered = 44%, [0–0.3[ = 5%) or the contrary as is the case in mixed land cover (Recovered = 36%) or grassland areas ([0–0.3[ = 17%). The most-affected climate (Aw, 41%) or biome (TSGSS, 54%) seems to drive global RCOVER distribution (Recovered = 33–35%, [0–0.3[ = 11%).

These results suggest that burned areas not only are vulnerable immediately after the fire, but also a substantial portion of them never managed to recover to pre-fire conditions, as shown by the 39% recovered areas when considering long and recent fire impacts (Extended Data Fig. 2d) or of 48% recovery for areas burned 6 to 18 years (Fig. 3). As a result, the estimated additional soil erosion shows an increasing trend (Fig. 4), contradicting with the overall burned-area reduction trend. At a continental level, the increasing trend in annual post-fire soil erosion is only statistically significant for North America (Mann–Kendall trend test p < 0.05), whereas the increasing trend for Africa, Europe and Oceania, and the declining trend for Asia and South America, are statistically non-significant.

Fig. 4: Additional post-fire soil erosion and trend per continent for consecutive fire years 2006–2019.
Fig. 4: Additional post-fire soil erosion and trend per continent for consecutive fire years 2006–2019.
Full size image

a–f, Bars concern annual estimates of additional soil erosion between 2006 and 2019 and per continent: Africa (a), Asia (b), Europe (c), North America (d), Oceania (e) and South America (f). Additional post-fire soil erosion corresponds to the difference between post-fire soil erosion and background soil erosion. Black line indicates soil erosion trend by linear smoothed conditional means, and shaded area represents standard error with 0.95 confidence interval (ggplot2, RStudio). Note statistical significance for increasing trend for North America (Mann–Kendall trend test, two-sided, p < 0.008595), and non-significant trends for remaining continents.

Unexpectedly, it was not possible to determine a standard recovery time within the areas affected by fire under this methodology, nor under the time range (2–7 years) provided by the scientific community6,7,10. This analysis was explored under vegetation indices but also in terms of soil erosion estimations; however, FCOVER observations from before and after the fire revealed differences even in the longest time series available (2001 to 2019) after a single fire (38–69% recovery), illustrating the variability of time required to achieve recovery for fire-affected ecosystems, which can also be a result from additional disturbances besides the fire, such as land management8, pest outbreaks21 or droughts10,22, which have not been considered in this study.

Future projections

Using 2019 as baseline year, we have assessed the impact of the expected changes in rainfall erosivity23, combining changes in rainfall amounts and intensities (Extended Data Figs. 3 and 4). The computation of all Representative Concentration Pathway (RCP) scenarios results in an increase in soil erosion globally from 11% (RCP2.6) to 23% (RCP8.5) for 2050 and from 23% (RCP2.6) to 28% (RCP8.5) for 2070. The greatest contribution comes from Africa where the burned areas have a larger surface, resulting between 54% (RCP8.5) and 60% (RCP2.6) of the total post-fire soil erosion globally in 2050 and between 54% and 55% for 2070. The increase in rainfall erosivity in the burned areas in Asia, however, results in the highest relative increases between 45% and 52% for 2050 and 49% and 67% for 2070 in relation to the baseline.

Discussion

The results from this study indicate that soil erosion following wildfires worldwide represent 19% of the latest global soil erosion estimation and 35% of the soil erosion generated in agricultural land17. This global estimation can illustrate the dimension of the problem and trigger the development of solutions to prevent further impacts10,24.

Despite being based in field data8, the comparison and validation of post-fire soil erosion estimations with field measurements is still challenging. Most of the post-fire soil erosion-monitoring field studies are focused on short-term assessments8,20, whereas over long(er) terms, they have only been done as non-continuous assessments25,26, often targeting processes that are scale dependent10. Fewer studies consider the impacts of various fires in soil erosion rates27 nor consider the history of prior disturbances in the experimental design28. The predicted global soil erosion rates for the first post-fire year (9.53 Mg ha−1 yr−1) are one to two orders of magnitude above the tolerable soil losses (0.3–1.4 Mg ha−1 yr−1)29, which is a frequently used metric in the assessment of soil health in European soils30.

To assess the robustness of our estimations, we have compared those against post-fire soil erosion field observations from plot to catchment scale20,30. Measured data shows a notable variability from no response to several orders of magnitude in soil erosion rates measured under different burn severities, monitoring periods and within individual countries (Fig. 5). Our estimations at country level for first post-fire year are within the range of most field observations, with some possible overestimations (Australia, Portugal and North Korea). It should be highlighted this is not an exhaustive dataset (Supplementary Section 2) and that the global post-fire soil erosion-measurements dataset is still missing13.

Fig. 5: Field-measured annual post-fire soil erosion rates in Mg ha−1 under different burn severities, time since fire and from individual countries.
Fig. 5: Field-measured annual post-fire soil erosion rates in Mg ha−1 under different burn severities, time since fire and from individual countries.
Full size image

a, Soil erosion observations (n = 525) per burn-severity classification (low, moderate, high) for burned areas without severity classification (Burned_no class) and for unburned conditions. b, Post-fire soil erosion observations (n = 516) per time since fire for first, second, three to four and more than five years after fire. c, Post-fire soil erosion rates by country, including post-fire soil erosion-rate estimates for first post-fire year for comparison (n = 516). Error bars depict soil erosion standard deviation as reported in original study, if available. Data from ref. 51.

The use of RUSLE in the assessment of post-fire soil erosion globally might represent only a portion of the post-fire impacts in geomorphic processes occurring after a fire13. Whereas RUSLE can grasp the most frequent hillslope processes occurring after a fire7,10 (sheet, inter-rill and rill erosion), estimating the potential of sediment transport to channels, it lacks connectivity and temporal scale to address processes at larger spatial scales such as mass movements10 (for example, debris flows, landslides) or sediment transport and deposition. Additionally, model adaptation to burn conditions still require further refinement on the impact of fire-induced changes to soil erodibility13 and how these factors change with time. Simultaneously, this approach allows the identification of areas with higher risk for erosion and therefore allows a prioritization of hotspots for the application of land-management actions13,31, while its connection to other models at smaller timescales and with higher spatial resolution would allow prediction of high-risk events with consequences offsite from the burned area11,32,33.

Regarding the use of MOSEV as a model input, this methodology can induce some underestimation by not identifying burned areas with sizes inferior to 25 ha (Supplementary Section 3). Despite acknowledging that large fires drive most of the impacts34, the size of the fire is relevant for global fire regimes and a key attribute in the definition of pyromes35. We highlight the importance of having available high-resolution data on fire occurrence, burned area and burn severity at global scale for a more accurate impact assessment, similarly to what is currently being done at EU scale1. The limited vegetation recovery observed in this study, especially for Europe and Africa, is in line with previous research22 on European forests, evidencing increasing trends of forest disturbances, such as drought or abiotic factors. This is also highlighted at global scale36, indicating a reduction of forest resilience under managed and intact forests. Under this work36, it was hypothesized that current (2000–2020) observed changes in forests globally could already be a result of climate-induced changes. Nonetheless, within this work it was not possible to find a comprehensive study integrating multiple local forest disturbances and to assess how such pressures interact altogether globally21 to fully justify the limited recovery found. In contrast, our observations for limited recovery are not aligned with others37, using Enhanced Vegetation Index (EVI) and Kernel Normalized Differential Vegetation Index (kNDVI) indices to assess vegetation recovery, which observed a regain of vegetation productivity to pre-fire levels in 87% of the fire-affected areas within 2 years. Such inconsistencies might be related to differences in the definition of recovery (‘Adaptation for post-fire conditions’ section in Methods), differences in the indicators since FCOVER targets ground cover, but also to the coarser scale used (10 km) in comparison with this study (500 m).

The concept of window of disturbance model38 already has been extensively investigated by the post-fire community6,7,39 and still drives research questions on the post-fire hydrological and erosive response40 and post-fire recovery for soil properties41. However, and similar to what was found in this study, no clear standard recovery time has been found yet. In fact, the results of this study are in line with the recently suggested geomorphic resilience framework10, whereas after wildfires, the ecosystem response can persist, change and recover or lead to an alternate state. On the basis of our results, either 18 years of observations do not allow us to fully capture the change and recovery of fire-affected areas or the share of areas under alternate state might be greater than anticipated ( > 50%). In addition, the variability in climate, vegetation type, burn severity and history of past disturbances are known to affect post-fire recovery and thus explain the observed recovery variability.

The results of this study also show a non-significant statistical trend of increasing post-fire soil erosion globally despite the decreasing burned area observed from 2001 to 2019. This may be driven by a global (non-significant) increase in burn severity42 but also by the limited recovery of some areas as a potential consequence of land degradation4. Moreover, post-fire soil erosion is estimated to aggravate in the future, as 2050–2070 projections indicate an increase of rainfall erosivity23. These considerations, however, do not include the anticipated changes in fire activity43, which when combined with the anticipated rainfall erosivity changes23, were projected to increase by 68% for post-fire debris flow events, in locations where those events already took place in the past10.

The results of this study combined with the limited number of countries (USA, Spain) undertaking operational post-fire management actions globally20 highlight the need for further knowledge on forest disturbances to plan and implement adequate land management in the mitigation and restoration of burned areas21. Field studies have shown that pre-fire fuel treatments44, post-fire soil mitigation treatments20 or post-fire restoration45 have shown positive results in the recovery of vegetation but also for soil condition46 after wildfire.

Despite the uncertainties (Supplementary Section 4), we provide an estimation of the post-fire soil erosion globally, contributing with two new blocks (forest and fire) to the existing high-resolution global soil erosion estimates and projections (GLoSEM)17. However, further research on how additional disturbances impact post-fire soil erosion and recovery is still required, namely, for how long wildfire affects soil properties41, what is the feedback of those fire-induced changes in forest recovery47 and how well vegetation indices represent those dynamics. Ultimately those knowledge gaps can be implemented in soil erosion models, thus achieving more robust predictions at multiple scales and a more informed decision-making. Moreover, further investigations on offsite impacts due to the post-fire soil erosion, including the carbon sink ability of soil organic carbon redistribution by erosion after fire48 or the impact of the transport of sediments and ash to the downstream water bodies11 are necessary.

Conclusion

This study estimates soil erosion under post-fire conditions at the global scale. The main findings of this modelling exercise with respect to the 2001–2019 burned area are as follows:

  • Our study estimates global post-fire soil erosion to be 8.1 Pg yr−1, considering cumulative effects of 18 years of fire disturbances.

  • Post-fire soil erosion estimations correspond to 19% of the global estimation for all land and 35% of the erosion for agricultural land17.

  • Africa is the continent that is impacted the most in terms of post-fire soil erosion (62%), given its substantially larger burned area (67% total).

  • The standard recovery time could not be determined under this study, given the substantial proportion of areas ( > 50%) presenting less vegetation cover in comparison to pre-fire conditions.

  • The 2006–2019 period reveals an increasing post-fire soil erosion trend, and when using projections for 2050–2070 rainfall erosivity, post-fire soil erosion is predicted to further increase, with 11–23% and 23–28% increase, for 2050 and 2070, respectively.

  • Post-fire recovery showed limited recovery after fire, especially in regions such as Africa and Europe. Countries involved should prioritize post-fire actions to prevent land degradation and adapt to future climate demands.

These results reinforce that limited attention has been given to the prevention and mitigation of wildfire impacts to forest soils. We suggest using these results to improve climate change assessment reports, and to inform about wildfires contributions to land-degradation reporting. Moreover, improving global forest resilience is urgent to halt land degradation and enhance climate change adaptation.

Methods

Methodological approach

To accomplish a global post-fire soil erosion assessment, this study used as main data sources land and climate data products from previous soil erosion assessments at European8,52 and global scales18,53 and remote sensing data for the determination of wildfiresʼ impact on ecosystems and the correspondent vegetation dynamics from pre-fire conditions to post-fire recovery (Fig. 5).

The data are used to feed the Revised Universal Soil Loss Equation (RUSLE)54,55 for forest, shrubland and grasslands56, following up the latest EU post-fire soil erosion assessment8 methodology. However, this study goes a step further by upscaling the previous methodology to the global spatial scale and by increasing the assessment temporal scale from 5 to 18 years. In addition, it also looks into the cumulative impacts of wildfires on soil erosion for several fire years (Fig. 1a–c), instead of following the recovery of a single year of fires, which was highlighted as a major underestimations aspect (Fig. 1d,e).

With the increase in the spatial and temporal scale, several adaptations were required as a result from data availability restrictions over two main data inputs, the biophysical parameter fraction of vegetation cover (FCOVER) and the normalized burn ratio (NBR). The methodological approach presents a balance between keeping a reasonable input resolution and the increase in data processing for a global study (Extended Data Figs. 5 and 6).

The FCOVER determination was required for each individual fire throughout the 18 years because the FCOVER products from Copernicus were not available before 2014. As a result, this indicator had to be estimated for all burned areas in the entire globe57 (30-m resolution), which also gave us the opportunity to compensate the resolution loss from now using MOSEV (500 m) database19. The decision to use MOSEV database was driven by being the only open-access product compiling burned-area delineation at a global scale and by the fact that such a dataset already contained NBR classification. Moreover, a recent study has also revealed the use of FCOVER as having a good performance assessing burn severity58. Because such changes could lead to new uncertainties, and because it was possible to replicate the original8 assessment under this new methodology, the soil erosion results under both methodologies were compared to assess quality (Supplementary Section 3).

RUSLE model and pre-fire conditions

This study used the RUSLE54,55 to model the average soil erosion potential in the absence of wildfire by including ground-cover changes described by ref. 59. The RUSLE model uses the following factors to estimate soil loss:

$$A=R\times K\times \mathrm{LS}\times C\times P$$
(1)

where A is the soil loss (Mg ha−1 yr−1), R is the rainfall and runoff factor (MJ mm h−1 ha−1 yr−1), K is the soil erodibility factor (Mg h MJ−1 mm−1), LS is the slope length and the slope steepness factor (dimensionless) and C × P is the cover management practice factor (dimensionless).

The K and R factors were expressed spatially using the latest maps created by the Joint Research Centre of the European Commission53. The C—cover and management—factor was originally developed for agricultural croplands. It has to be adjusted to forest, shrub and grasslands ground-cover characteristics to be able to spatially differentiate the various cover species, canopy cover densities and ground-floor vegetation52. Considering the size of the study area, the C factor used to predict the soil loss potential8 was quantified by manipulating the biophysical parameter Fraction of Vegetation Cover (FCOVER60). This provided an estimation of the fraction of the vegetation that is visible vertically, allowing to differentiate a C factor for undisturbed conditions (CU) between bare or protected soil for forest, shrub and grasslands (Extended Data Fig. 5). Management practices (P) were not considered under this assessment.

Adaptation for post-fire conditions

With regard to the wildfire disturbance, the state of the art identified that post-fire soil erosion rates and vegetation recovery are closely related to burnt severity and bare soil cover9,20. Notwithstanding, as shown by previous research61,62, forests, shrubland and grasslands are distinctively impacted by wildfires, whereas contrary to the undisturbed condition, burned forests present much higher post-fire soil erosion rates in comparison to shrubland and grasslands due to higher fire residence times and burn severities. Therefore, the C factor for burned conditions (Fig. 1) was approximated as follows:

$${\mathrm{Cmax}}_{{\mathrm{B}}\_{\mathrm{Forest}}}={\mathrm{Cmax}}_{{\mathrm{U}}\_{\mathrm{Forest}}}\times {B}_{\mathrm{severity}}\times 10\,\mathrm{for}{\ \mathrm{FCOVER}}=0.2,\,\max =0.5$$
(2)
$${\mathrm{Cmax}}_{\mathrm{B\_Shrub}}={\mathrm{Cmax}}_{\mathrm{U\_Shrub}}\times {B}_{\mathrm{severity}}\,\mathrm{for}\,\mathrm{FCOVER}=0.2,\,\max =0.3$$
(3)
$${\mathrm{Cmax}}_{\mathrm{B\_Grass}}={\mathrm{Cmax}}_{\mathrm{U\_Grass}}\times {B}_{\mathrm{severity}\,}\mathrm{for}\,\mathrm{FCOVER}=0.2,\,\max =0.3$$
(4)

where, for equations (2), (3) and (4), the maximum C factor for burned conditions (CmaxB) was estimated by multiplying the maximum C factor for undisturbed conditions (CmaxU) and the burn-severity anomaly factor (Bseverity) as previously derived through effect size determination for erosive response between burned and unburned conditions9. Such a maximum value was determined for ground-cover (FCOVER) values of 0.2, similarly to previous approach63. Regarding the individual cover types, for forest (equation (2)) the C factor was increased one order of magnitude for all the severities and limited to a maximum of 0.5, while shrubland (equation (3)) and grasslands (equation (4)) were represented by low and moderate severities and limited to a maximum of 0.3. In this way, it was possible to accommodate the impacts determined for distinctive severities9, and also the conceptual model62 for hydrological response and observations on soil properties61 for distinct cover types (Extended Data Fig. 6).

The recovery of the burned areas was addressed by using vegetation cover indicators as proxies, resulting our definition for recovery8, which considers that an area is recovered once the ground cover reaches the pre-fire conditions. In this way, it was possible to determine the erosion reduction with the increase of the protective cover but also to consider the initial characteristics of the local vegetation before the wildfires.

For the first point, a minimum C factor was estimated:

$${\mathrm{Cmin}}_{{\rm{B}}}={\mathrm{Cmin}}_{{\rm{U}}}\,\mathrm{for}\,\mathrm{FCOVER}=1.0$$
(5)

Being the minimum C factor for burned conditions, CminB approximated to the CminU (ref. 52) for each land cover (forest = 0.0001, shrubland = 0.003 and grassland = 0.01) whenever the ground cover (FCOVER) reached its maximum value of 1.

This formulation allowed for the inclusion of the impact of the burn severity in the soil erosion estimations for when protective cover presented the minimum value according to the burn-severity impact and the approximation of the recovery of the burned areas under an exponential curve in the function of FCOVER over the monitoring period (Extended Data Figs. 5 and 6 and Supplementary Section 5). The choice of an exponential base for such estimations goes in line with the conceptual model after observing a nonlinear relationship between sediment yields and bare soil cover62 and a sharp increase of the sediment yields on the 60–70% bare soil cover9.

Therefore a recovery rate (RCOVER) was calculated, and several classes (Extended Data Table 1) were defined as follows:

$${{\rm{RCOVER}}}_{t}=\frac{{{\rm{FCOVER}}}_{t}}{{{\rm{FCOVER}}}_{{\rm{pre}}{\textstyle \text{-}}{\rm{fire}}}}$$
(6)

The consideration of a RCOVER class allowed us to adopt a succession of C factor equations according to the status of the vegetation-recovery progress (Supplementary Section 5). For the first two classes ‘No Recovery’ and ‘Ongoing Recovery’, the C factor would be calculated as described for equations (2)–(5), according to the initial burn-severity classification and land-cover attributes. In the case of ‘Near Recovery’ class, this would lead to the adoption of the equation with one burn-severity level lower than the original one; likewise the ‘Recovered’ class would lead to adoption of the equation corresponding to unburned conditions. This formulation allowed us to use FCOVER as a constant input for C estimations, while reducing the impact of severity proportionally to the pre-fire conditions.

Spatial data processing

The burned-area delineation, wildfire date and burn severity (NBR) were retrieved from the MOSEV database19 with a resolution of 500 m and global coverage, for the 2000–2020 period for all fires with an area of at least 25 ha. Given the incompleteness of the dataset for the years of 2000 and 2020, these were removed from our analysis, whereas burn severity was then converted to low, moderate and high burn-severity classes64.

To target the land-cover types affected by wildfires, the reference dataset for land-cover factions, the dynamic land-cover map at 100-m resolution (CGLS-LC10056) was used, which allowed estimation of the individual contributions of soil erosion for forest, shrub and grassland with a reference year of 2015. Despite a more recent version of this dataset existing (2019), the choice to keep 2015 was made to keep the comparability with the previous assessment8, but also because of the inexistence of comparable datasets before this date, which would allow accommodation of land-cover changes throughout the entire temporal scale of this study. Nonetheless, to ensure consistency, both land-cover datasets for 2015 and 2019 were compared for the total area affected by fire (17 million km2), resulting in land-cover changes (difference in cover share (%) in each pixel) under the ± 10% threshold. Moreover, cropland was also excluded from this analysis due to the lack of consolidated knowledge on the cumulative effect from intensive soil management operations and wildfire8, whereas at the same time, soil erosion in these areas have already been assessed extensively17.

The ground-cover changes were assessed with FCOVER60, which provides a measure of the fraction of ground covered by green vegetation. The change in vegetation cover at 30-m resolution was determined by Normalized Differential Vegetation Index (NDVI57), using as a reference the date the first Landsat imagery was available following each fire record. The repeated measurement of NDVI was performed annually (2001–2019) to follow up the recovery of each burned area. Whenever a repeated wildfire occurred in the same location, the procedure would start again considering the latest fire as a reference and assuming the recovery was sufficient to allow the most recent fire to occur.