Main

Climate change has strong impacts on global ecosystems1,2. These impacts are likely to result in high economic costs for society3. Recent studies estimated economic losses related to climate impacts on global ecosystems to several trillion dollars4, with income reductions of 19%5, several hundred billions of dollars needed to compensate for loss and damage6, and reductions in the gross domestic product of 1.2% per 1 °C increase in global mean temperature7. In particular, the already observed8 and projected future9,10,11 increases in frequency and intensity of extreme events, such as droughts, wildfires and floods, have severe consequences for the global economy4,12.

Forest ecosystems are particularly prone to climatic extremes because trees are sessile and long lived13,14. As forest products are central to a bio-based economy15,16,17, changing extreme events pose a major challenge for a wider use of bio-based materials. A major concern in this regard are forest disturbances, that is, large-scale pulses of tree mortality from wildfires, pest outbreaks and storms18,19. Forest disturbances have increased in frequency and severity in many parts of the globe in recent decades20,21 and are expected to further increase under continued climate change22. A hotspot of changing forest disturbances is Europe, where disturbance rates doubled in less than 20 years for major disturbance agents20, and a massive recent pulse of tree mortality was unprecedented in at least 170 years23. Societies in Europe are strongly dependent on forests for providing jobs, supporting rural livelihoods, and contributing to environmental and economic well-being24. Yet, the continental-scale economic impacts of climate-mediated disturbances remain unclear so far.

Understanding the economic impacts of disturbances is complex, as interactions with other climate-induced changes have to be considered. For instance, climate change affects forest productivity, with decreases projected for water-limited regions25 but broad-scale increases expected due to CO2 fertilization25,26,27 and an extension of the growing season particularly in boreal and mountain ecosystems28,29. Changing forest productivity interacts with disturbance costs: increasing productivity increases the growing stock, which in turn enhances the timber value of forests. Simultaneously, productivity-related increases in harvest levels lead to higher returns that might partly or completely offset the economic costs of increasing disturbances30,31. These interactions require a joint evaluation of the economic consequences of changing disturbance and productivity30,31.

Here we present an estimate of the current and future economic costs of forest disturbances at continental scale, accounting for climate-mediated changes in both forest productivity and disturbance in Europe. Specifically, we investigated the historical costs of disturbance on the timber-based forest value (under the climate conditions of the reference period 1981–2005) and how they are likely to evolve under scenarios of future climate change (representing the climate conditions expected for the period 2076–2100). We assessed hotspots of economic disturbance impact in Europe and whether increasing forest productivity can offset economic losses from disturbance under climate change.

To address our questions, we coupled three crucial elements, (1) spatially explicit (16 ×16 km) forest growth simulations at the level of individual tree species, (2) >150,000 Monte Carlo simulations of forest disturbances informed by latest remote sensing data and (3) economic models to quantify the costs of changing productivity and disturbance across continental Europe focusing on the commercially most relevant species (a total of 91 million ha, or two-thirds of Europe’s forest area). We studied a total of 1,536 scenarios under three different representative concentration pathway (RCP) scenario families and quantified the cost of disturbance by comparing scenario simulations to the counterfactual of simulations without disturbance under the same climate conditions. Economic effects were quantified by converting simulated time series of planned (that is, business-as-usual management) and unplanned (that is, disturbance-related) timber harvests to economic cashflow, which was subsequently discounted and summed to obtain forest value (Extended Data Fig. 1). We translated the present value of total disturbance costs into annual costs by multiplying with a discount rate of 1.5%. Disturbances are discrete events in space and time, and averaging over extended spatiotemporal scales masks their immediate local effects12. Hence, we report costs for both the average (mean across all stochastic simulations) and the extreme case (defined as the average of the worst 5% of simulations, conditional value at risk), with the latter being particularly informative for planning under the precautionary principle32.

Climate change doubles the costs of natural disturbances

Under historical conditions (1981–2005), the economic costs of natural disturbances in Europe (loss of forest values) were €115 ± 3 billion (mean ± s.d., Fig. 1), with an average annual cost of €1,729 ± 48 million yr−1 and an average cost per unit area of €1,265 ± 35 ha−1. These costs were the result of on average 74.5 million m3 of timber disturbed per year (Table 1). Disturbances reduced the total forest value of Europe by 28.6 ± 0.7% compared with the counterfactual of no disturbance. Under climate change, total future timber harvest increased by 17.1%, from 259.6 to 304.1 million m3 yr−1 (for the period 2076–2100) under scenario RCP4.5 (increases of 12.9% and 28.0% under RCP2.6 and RCP8.5, respectively), resulting from both increasing productivity (increasing planned harvest) and increasing disturbance (increasing unplanned harvest; Table 1). Distinct increases in unplanned, that is, disturbance-induced, harvests increased costs of disturbance to €186 ± 8 billion under RCP4.5, and up to €247 ± 15 billion under RCP8.5 (RCP2.6: €146 ± 3 billion). This reduced the potential forest value relative to the counterfactual of no disturbance by up to 42.1 ± 2.5%, and translated to average annual costs of €2,783 ± 116 million yr−1 under RCP4.5 (RCP2.6: €2,191 ± 42 million yr−1; RCP8.5: €3,711 ± 218 million yr−1), and average costs per unit area of €2,037 ± 85 ha−1 (RCP2.6: €1,603 ± 31 ha−1; RCP8.5: €2,715 ± 160 ha−1). Simulating a moderate shortening of the rotation period by 10 years as a measure to adapt to changing climate and disturbance regimes reduced the cost of disturbance by up to €10 billion under scenario RCP8.5 (Extended Data Fig. 2). All economic results were sensitive to varying discount rates, with higher costs of disturbance at lower discount rates (Extended Data Fig. 3). Different climate model projections within the same RCP family resulted in similar trajectories (Extended Data Fig. 3).

Fig. 1: The current and future cost of natural disturbances in Europe’s forests.
figure 1

Bars show the timber-based forest value losses from disturbances relative to the counterfactual of undisturbed forest development under the same climate scenario. Data denote mean ± s.d. across all simulations (N = 300). ‘Historical’ assumes climate conditions from 1981–2005, while RCP scenarios are for projected future climate conditions for the period 2076–2100.

Full size image
Table 1 Planned and unplanned timber harvests under climate change
Full size table

Central Europe as hotspot of future disturbance costs

The economic costs of natural disturbances varied widely across Europe (Fig. 2 and Extended Data Table 1). Hotspots of future disturbance costs were mainly located in Central Europe (especially in parts of Germany, Austria, Switzerland and the Czech Republic; Fig. 2c), where disturbances lowered the economic value of forests on average by €3,233 ha−1 under moderate climate change (RCP4.5), and €2,460 ha−1 and €4,375 ha−1 under mild (RCP2.6) and severe (RCP8.5) climate change, respectively (Fig. 2a). Extreme costs in this region (that is, the average of the economically worst 5% of simulations, conditional value at risk) were €17,067 ha−1 under RCP4.5 (€13,932 ha−1 and €19,885 ha−1 under RCP2.6 and RCP8.5, respectively). In contrast to the high disturbance costs in Central Europe, forests in Northern Europe had 60–65% lower economic costs of disturbance, with average losses of €1,164 ha−1 under RCP4.5 (€966 ha−1 and €1,523 ha−1 under RCP2.6 and RCP8.5, respectively). Nonetheless, local extremes were also high in Northern Europe, exceeding €10,000 ha−1 (RCP2.6: €9,747 ha−1, RCP4.5: €12,370 ha−1, RCP8.5: €14,664 ha−1, Extended Data Table 1). Generally, the continental-scale differences in disturbance costs decreased when considering extreme values (Fig. 2). Disturbance costs in Southern Europe (Fig. 2c) were between those in Northern and Central Europe (Extended Data Table 1).

Fig. 2: Hotspots of future disturbance costs in Europe’s forests.
figure 2

The colour of each point represents the cost per hectare of forest within a 16 × 16 km cell, and the size of each point corresponds to the forested area within that cell. a, Average disturbance costs across all simulations. b, Extreme costs, expressed as the mean over the 5% scenarios with the highest costs. Histograms illustrate the distribution of disturbance costs across all scenarios, with the values for RCP4.5 corresponding to the data shown in the maps. Vertical lines within each histogram indicate average disturbance costs. Only cells with a forest cover of at least 5% are displayed. c, Map showing the European regions considered in this study: Northern Europe (N), Central–Western Europe (CW), Central–Eastern Europe (CE), South–Western Europe (SW) and South–Eastern Europe (SE). All maps use the ETRS89-LAEA Europe projection (EPSG:3035). See Extended Data Fig. 4 for maps assuming other climate scenarios. Credit: shape file by Andy South.

Full size image

Productivity gains offset disturbance losses

The increase in forest productivity under climate change (Table 1) overcompensated the economic losses from disturbances in Europe overall, but regional variation was high. The productivity-related increase in forest value can be attributed to two effects: higher initial growing stocks and increased sustainable harvest levels from elevated tree growth (Extended Data Fig. 5). These effects were strongest under scenario RCP8.5, yet in this scenario, disturbance-induced losses also increased most strongly (Fig. 3a). The economic effects of productivity and disturbance increased at similar rates across climate scenarios, resulting in little variation in overall forest value with climate change (RCP2.6: €337.9 ± 12.6 billion, RCP4.5: €328.6 ± 20.8 billion, RCP8.5: €340.8 ± 34.9 billion). However, scenario uncertainty was lowest under RCP2.6 and increased considerably with increasing severity of climate change (Fig. 3b). Productivity-related offsets of disturbance losses varied distinctly across Europe. In Northern Europe, economic gains from increasing productivity clearly outweighed disturbance-mediated losses under climate change (Fig. 3c,d). In contrast, productivity stagnated or declined in Southern Europe while disturbances increased, resulting in decreasing net forest values under climate change. In Central Europe, productivity was projected to increase, albeit at a lower rate than in Northern Europe. Here, economic losses from disturbances are compensated by gains in productivity, yet the offset capacity decreases with increasing severity of climate change (Fig. 3c,d). Only under the mild climate scenario RCP2.6 was the forest value of all European regions (and 92% of all European countries, Extended Data Fig. 6) projected to increase relative to historical levels, when considering the net effects of both changing productivity and disturbance.

Fig. 3: Effects of changing forest productivity and disturbance on forest value in Europe.
figure 3

a,c, Changes in forest value at the continental (a) and regional (c) scale under future climate compared to historical values. Historical assumes climate conditions from 1981–2005, while RCP scenarios are simulated under the climate conditions projected for the period 2076–2100. Blue bars represent losses due to increasing disturbances, green bars show gains from increasing productivity. Dots indicate the net change in forest value compared to historical values. b,d, Continental (b) and regional (d) forest values under different climate scenarios. In all panels, data show the mean ± s.d. across all simulations in the respective stratum. Regions were defined as in Fig. 2c. Icons in a adapted from OpenMoji (https://openmoji.org/) under a CC-BY-SA 4.0 licence.

Full size image

Discussion and conclusions

Here we provide a continental-scale estimate of forest disturbance costs for Europe. Our findings suggest that disturbance costs could more than double under climate change. Under severe climate change, the annual losses estimated here correspond to up to 15% of the current gross value added of the forestry sector in Europe24. We furthermore highlight considerable differences in the economic costs of forest disturbances throughout Europe, identifying particular hotspots in Central Europe. Local-scale studies from this region show that disturbance costs could be even higher than estimated here, when only considering disturbance impacts on the economically most valuable tree species32. Nonetheless, our results of disturbance-based losses on the timber-based forest value of up to €19,885 ha−1 suggest that disturbances could posit major economic challenges for timber-based forestry in the future. Our results also indicate that simultaneous increases in forest productivity under climate change could offset economic losses from increasing disturbances. The finding that Northern Europe is the main beneficiary of climate change, with increasing forest values in Europe’s boreal zone, is in agreement with previous studies on the economic effects of climate change33. In contrast, the timber-based forest value in Southern Europe is already considerably lower than in other parts of the continent24 and will decrease further under climate change. The negative impacts of climate change on Southern Europe identified here correspond well with previous assessments9,25,33,34.

Important limitations need to be considered when interpreting our results. First, our estimates of the cost of disturbances are likely conservative, because we only focused on timber-related forest values. Although timber remains the main marketable good from forests throughout Europe, disturbances have broad impacts beyond timber35, and the societal costs of disturbance impacts on non-marketable ecosystem services were not considered here. Similarly, we focused on the economically most important tree species and did not study the effect of tree species change in response to changing climatic conditions. More broadly, the consideration of alternative silvicultural strategies was beyond the scope of our analysis. Rather, we assumed even-aged management as the most widely used silvicultural strategy across the continent24,36,37 and considered local differences in productivity and their effect on rotation periods. While this approach is not able to capture the full variability of forest management regimes applied in Europe24,36, it approximates the dominant management of the simulated tree species well36,38, and provides a robust and consistent baseline for quantifying continental-scale disturbance costs. We also omitted income from thinning, likely underestimating Europe’s forest value. However, previous studies showed that considering thinnings does not substantially reduce disturbance costs32. The increasing volatility resulting from disturbances may also drive risk-averse forest owners to exit the market39, which can have considerable economic effects beyond the ones considered here. Moreover, we only incorporated the three most important forest disturbance agents in Europe, namely, wildfires, windstorms and bark beetle outbreaks. The advent of novel disturbances such as invasive alien pests and pathogens could considerably alter future disturbance regimes at the continental scale40 and further increase economic costs of disturbance. Another important assumption in our analysis is that some of the disturbed timber is salvage harvested and thus enters the timber market. This is the current default management response to disturbances in Europe41,42, yet the practice is increasingly criticized for its ecological impacts43,44. Lower salvage rates would likely further increase the economic costs of forest disturbances32,43.

While our quantification of disturbance costs is likely conservative, the estimate of the compensatory effect of increasing productivity might be optimistic. Productivity gains offset disturbance losses because they simultaneously increase initial forest values and periodic returns in our analysis (Extended Data Fig. 5), yet the assumptions with regard to initial values are uncertain (that is, similar forest age structure at higher productivity levels). Furthermore, effects of reduced water use efficiency45 or acclimation effects in response to elevated CO246 were not considered here. In fact, recent studies already indicate declining forest growth even in some areas of Northern Europe47. While we here used best available modelling approaches at continental scale, these uncertainties call for further research on economic effects of both changing productivity and disturbance34,48.

The high economic costs of disturbance identified here have important implications for forest policy and management. First, losses from disturbances need to be considered more explicitly in forest planning and the economic valuation of different forest management strategies. While even-aged coniferous forests were propagated throughout Europe on the basis of economic grounds in the past, these considerations ignored the substantial disturbance risk of these silvicultural systems13,49, and hence also the associated costs. As disturbances are likely to increase further under climate change, future considerations of forest planning and management need to explicitly account for their impacts50. Second, we show that considerable adaptation efforts are needed to reduce disturbance impacts in Europe’s forests. While these efforts (including measures such as planting less disturbance-prone tree species and managing for structured and mixed forests49,51) require resources, they will also reduce disturbance risk49,51 and thus disturbance-related costs. This highlights that there is not only an ecological but also an economical imperative to climate change adaptation in forestry. We show that moderately reducing rotation lengths might reduce disturbance costs, illustrating the potential of climate change adaptation measures in silviculture. However, such measures could also have unintended consequences on other ecosystem services such as forest carbon storage and habitat value36,52, and should thus only be applied after careful consideration of local trade-offs. Efforts should particularly focus on hotspots of disturbance impacts in Central and Southern Europe, where the expected rise in disturbance-related costs could become increasingly prohibitive for regular forest management in the absence of adaptive measures. Lastly, our results underline the importance of climate change mitigation. Only when climate change was limited to mild levels (scenario RCP2.6) did net positive economic effects of changing productivity and disturbance emerge throughout the European continent. In contrast, largely unmitigated climate change (scenario RCP8.5) could lead to severe economic losses from disturbances, particularly in Southern and Central Europe. We thus conclude that mitigating climate change can avoid substantial disturbance-related costs in the forestry sector.

Methods

We modelled forest development explicitly for 16 × 16 km grid cells across Europe at 10-year time steps using a matrix model (Extended Data Fig. 1). In each grid cell, we considered the forested area of the four economically most important tree species in Europe38, namely, European beech (Fagus sylvatica L.), Norway spruce (Picea abies (L.) H. Karst), Scots pine (Pinus sylvestris L.) and deciduous oaks (Quercus robur L. and Quercus petraea Matt. Liebl.). Together, these species account for more than 95% of the European timber market53 and two-thirds of Europe’s forest area (91 million hectares). Stand age distributions were extracted from an analysis based on remote sensing54, and tree species shares for each 10-year age-class bin were considered proportional to the species composition at grid-cell level. For each 16 × 16 km grid cell, the forest area was thus distributed to 104 classes (4 tree species × 26 age classes).

For the sake of parsimony and to consistently compare the effects of disturbance and productivity change across Europe, we assumed a single silvicultural system for all simulated forests. As the clear-cut system is still the dominant silvicultural system in Europe for the tree species considered here24,36,37, we simulated even-aged stand development followed by clear cutting and planting. We accounted for local variation in management intensity across Europe by deriving rotation lengths at the cell level. Specifically, we calculated economically optimal rotation lengths (that is, the age of final cutting that maximizes forest value) contingent on tree species and local site productivity for each cell (see Supplementary Table 1 and section ‘Calculation of the economically optimal rotation age’ in Supplementary Methods for more details). For the calculation of optimal rotation lengths, we assumed timber production to start from bare ground and we disregarded disturbances.

Forest managers are actively adapting their management to the emerging changes in environmental conditions. While a comprehensive assessment of the economic effects of climate change adaptation measures is beyond the scope of our analysis, we considered two elements of climate change adaptation in our simulations: First, we simulated an adaptation of optimal rotation periods to the emerging changes in productivity under climate change. This was implemented by deriving optimal rotation periods separately for each climate scenario and grid cell (see section ‘Climate data’ in Supplementary Methods for more details). Thus, rotation periods are effectively shortened in locations where productivity increases, and rotation periods are extended in areas where productivity decreases, simulating dynamic adaptation of managers to changing environmental conditions. Second, we tested the effect of adapting the rotation length from its local economic optimum value in a sensitivity analysis, varying rotation period length by up to ±20 years from the economic optimum (Extended Data Fig. 2). In Europe, reducing rotation period length is a measure that is frequently discussed in the context of dampening the impacts of disturbances52,55, while extending rotation periods can increase forest carbon storage and habitat value36. Thinnings were disregarded in our simulations32, and we simulated no tree species change, that is, species were replanted in their current proportions after final harvesting or disturbance (but see refs. 53,56).

We assessed the effect of climate change by considering two time slices, one representing historical climate (1981–2005) and one future climate (2076–2100). For each time slice, we simulated 500 years of forest development starting from current forest conditions, with climate conditions averaged for each time slice. We chose this approach over transient scenario simulations as it better quantifies the long-term economic consequences of changing climate and disturbance regimes, given that discounting in economic analyses strongly reduces the weight of future changes in transient analyses. We chose a simulation period of 500 years, as it contains multiple rotations, and aggregated cash flows beyond this period have little influence on the forest value because of discounting. Additional to the time slice representing historical climate, we considered 3 climate scenarios (representing different radiative forcing levels, that is, RCP2.6, RCP4.5 and RCP8.5 from CMIP5) each derived from 3 different climate models, downscaled to our 16 × 16 km grid cells (for details see section ‘Climate data’ in Supplementary Methods). This resulted in 12 different climate scenarios considered in the analyses. Climate affected both productivity and disturbance in the simulations, as described in the following paragraphs.

Forest productivity and its response to climate change was quantified by dynamically simulating the potential net primary productivity (NPP) per grid cell and species. Potential NPP was defined as the NPP of a fully stocked pure stand of a species under a given climate, during the stand development stage in which tree growth culminates. Potential NPP for each grid cell and scenario was estimated using a deep neural network trained on data generated by the process-based simulation model iLand57,58. To derive training data for the deep neural network, we used iLand to simulate NPP values of the four tree species across the full climate and soil gradients of Europe and under the 12 different climate scenarios considered. In iLand, NPP increases with temperature (as long as temperatures are below optimal temperatures for photosynthesis) and atmospheric CO2 concentration, while water availability (in the atmosphere and soil) and plant-available nitrogen limit the carbon uptake of trees (a detailed description of iLand and the mechanisms used to calculate NPP can be found in refs. 57,58 and on the model website https://iland-model.org). We ran simulations on the basis of daily climate data for each climate scenario (see section ‘Climate data’ in Supplementary Methods) and extracted the resulting annual NPP values. We developed a feedforward convolutional deep neural network with 13 layers and 189,000 trainable parameters. We subsequently trained the deep neural network on 14 million datapoints derived from iLand to predict annual potential NPP values for the four tree species under study, contingent on the soil and climate conditions prevailing at a grid cell (see section ‘Forest productivity under climate change’ in Supplementary Methods). The deep neural network was well able to learn the responses of the underlying process-based model and generalized well between climate change scenarios. Predicted potential NPP values for current climate conditions were evaluated against independent observations from satellite data (see section ‘Forest productivity under climate change’ in Supplementary Methods). To derive merchantable timber volume in our matrix model simulations, we used species-specific yield table estimates (see section ‘Yield tables’ in Supplementary Methods for more details), dynamically calculating the respective yield class from potential NPP values per cell. Specifically, we used a quadratic link function to capture the nonlinear relationship between NPP and yield class (see section ‘Mapping of NPP values to yield tables’ in Supplementary Methods), assuming correspondence between the range of potential NPP values predicted in simulations under historical climate and the range of yield class values covered in yield tables. In this way, we combined the robust estimates of merchantable timber volume from yield tables with dynamically simulated productivity changes under climate change, resulting in realistic projections of future timber production at cell level.

To comprehensively assess the impact of natural disturbances, we considered two types of disturbances in the simulation: first, climate-sensitive ‘background’ disturbances were derived on the basis of statistical survival probabilities estimated on the basis of continental-scale forest inventory data59 (see section ‘Background disturbances’ in Supplementary Methods). This disturbance type represents small- to medium-scale events that happen regularly (for example, small-scale mortality from drought or insect infestations, small-scale windthrow), but do not lead to larger-scale economic implications such as market crashes. They were sensitive to changing mean annual temperature, maximum temperature of the warmest month, minimum temperature of the coldest month, annual precipitation sum and precipitation sum of the warmest quarter in our simulations59. For each grid cell and tree species, we simulated these disturbances within each time step by resetting a proportion of the cell’s area to age class zero and replanting the area, with the proportion affected by disturbance derived from a climate-sensitive hazard probability function59. Second, to account for economic risks from rare but large disturbance events, we simulated stochastic, landscape-scale extreme events, such as severe fires, windstorms and bark beetle outbreaks, using biome-specific remote sensing data60,61,62. Extreme events were defined for each scenario as those affecting more than twice the area disturbed in an average year of the scenario. To estimate the occurrence of such extreme disturbance events, we first calculated historical biome-specific average disturbance rates (1986–2005). We subsequently developed scenarios of potential future changes in average disturbance rates based on expected increases in drought intensity and frequency in Europe2, as well as on other best available estimates for future disturbance change20,22 and the already observed responses to recent climate change (Supplementary Table 2). On the basis of these analyses, we assumed mean disturbance increases by a factor of 2, 4 and 6, respectively, for RCP2.6, RCP4.5 and RCP8.5, and investigated the sensitivity of our results to a wide range of different potential future disturbance changes (Supplementary Figs. 7 and 8). Subsequently, the specific frequencies and magnitudes of extreme events were derived using Taylor’s power law equations fit to remote sensing data63 (Supplementary Fig. 1), estimating the temporal variance of disturbance rates per agent and biome under historical and increased future mean disturbance rates. The occurrence of extreme events in a simulated time step was determined at the level of biomes by drawing from Poisson distributions. The affected area in each biome was calculated by multiplying the number of extreme events in each time step by their magnitude. Affected cells were then drawn randomly until the estimated disturbed area was reached. In disturbed cells, depending on the disturbance agent (see section ‘Extreme disturbances’ in Supplementary Methods), all or most of the forest area was reset to age class zero.

Revenues from timber-based forest management were calculated from tree diameter- and species-dependent timber prices and establishment costs53 (Extended Data Table 2). Revenues were discounted and summed over the whole simulation period to quantify forest value32,53 (default discount rate of 1.5%, but see Extended Data Fig. 3 for a sensitivity analysis of different discount rates). Simulated increases in productivity under climate change influenced forest value via increased timber harvests, which positively affected forest value. Simulated disturbances had a range of nuanced economic impacts in the simulation (Extended Data Table 3). For climate-sensitive background disturbances, we assumed that the impacted timber is sold with a 50% decrease in net revenue due to a loss of timber quality and increased harvesting costs (see section ‘Prices’ in Supplementary Methods). For extreme disturbance events, revenues from disturbed timber were set to zero for the affected grid cell, assuming a collapse of the timber market in the region as a result of large quantities of disturbed timber flooding the market32. For extreme events, planting costs for the next cohort of trees were assumed to increase by a factor of two, to reflect higher post-disturbance expenses from planting large areas and to account for typical shortages in nurseries after large disturbance events. To assess the sensitivity of our results to additional market effects of disturbances (that is, price drops from extreme disturbance events radiating out to larger areas), we conducted an auxiliary analysis in which we reduced the timber prices of regular harvests in cells adjacent to those affected by extreme disturbance events (see section ‘Prices’ in Supplementary Methods and Supplementary Fig. 7).

To account for the stochastic nature of disturbances, we used Monte Carlo simulations32 to consider a wide range of potential future disturbance impacts. This method involved running multiple simulations for each studied scenario. Within each simulation, different random disturbance events were generated to capture a wide range of disturbance sequences and their corresponding economic outcomes. This allowed us to quantify the variability of the economic impacts of disturbances. Monte Carlo approaches have been widely used in economics and ecology64,65,66,67, and are particularly suited to capture the impacts of highly variable events32. Specifically, we ran 100 Monte Carlo simulations for each scenario and cell with different random numbers used for (1) drawing extreme disturbance events in each time step and (2) choosing which cells are affected by these events. To assess the economic impact of disturbances, we calculated losses as the average difference in economic value between simulations with and without disturbances across all simulations. In other words, simulations without disturbances in the respective climate scenarios served as the counterfactual to quantify the economic impacts of disturbance. Since disturbances, particularly extreme events, disproportionately impact the fat tail of skewed economic damage distributions (Supplementary Fig. 2), and mean comparisons cannot fully capture the effects of increasing frequencies of rare events12, we also analysed extreme values from our Monte Carlo simulations. Specifically, for each scenario and simulated cell, we calculated the average of the worst 5% of economic losses across all Monte Carlo runs (conditional value at risk). To transform losses in net present value into annual costs, we calculated an annuity by multiplying the losses by the assumed discount rate of the simulation (1.5%). Regional analyses were conducted for European regions as suggested in the Forest Europe report24 (Fig. 3c). Within each region, hotspot areas were defined as cells above the 99th percentile of all values in that region (Extended Data Table 1). To estimate economic costs per unit forest area, the total costs were divided by the forested area for each cell.

We successfully evaluated our simulation approach against independent data for several key metrics. The simulated standing timber volume and annual timber extraction rates under historical climate matched observed values well (Supplementary Figs. 5 and 6). The model estimated that 28.7% of the annual timber harvest was due to natural disturbances under historical climate, which amounted to 74.5 million m3 timber yr−1 (Table 1). These values are well within the range of empirical data from recent decades, reporting 42.6–78.5 million m3 of timber disturbed annually and 12%–32% of unplanned canopy openings from disturbances relative to the total harvested timber volume for the period 1986–200520,68. These results highlight the robustness of our simulations; full details of the evaluations conducted are provided in Supplementary Methods (see sections ‘Standing timber volume’ and ‘Amount of extracted timber’). We used TensorFlow69 in combination with Keras API70 in Python for implementing the deep neural network. All other simulations and analyses were done using the R programming environment71 in R Studio72 v.2023.12.1.402.