Introduction

Forests represent about 30% of the worldwide land area (FAO 2020) providing key ecosystem services for human well-being (Gamfeldt et al. 2013; Díaz et al. 2018). However, global change is altering forest structure and demography worldwide (Anderegg et al. 2022). Particularly, climate and stand characteristics are impacting forest ecosystems, especially at the global scale and more intensively in Mediterranean regions (Carnicer et al. 2011; Sánchez-Salguero and Camarero 2020). Climate change is characterised by increasing temperatures as well as a higher occurrence and intensity of droughts and heatwaves (Allen et al. 2015). This interacts with the abandonment of forest management leading to increased competition and alterations to forest structure, as well as an increase in vulnerability and exposure to biotic and abiotic disturbances (e.g., Jump et al. 2017; Seidl et al. 2017). These alterations can cause a range of forest responses, such as changes in tree phenology (Yuan et al. 2020), increases in canopy damage (Moreno-Fernández et al. 2021) and altered demographic rates (Allen et al. 2015; Gazol et al. 2017). Those responses are already inducing important changes in forest composition and ecosystem functions (McDowell et al. 2020; Brodribb et al. 2020).

The “decline spiral model” suggests that tree mortality depends on multiple interactive biotic and abiotic factors, acting together as drivers leading an individual tree to die (Franklin et al. 1987; Manion 1991). These factors can be classified as predisposing factors (i.e., those reducing the tree resistance), inciting factors (i.e., those initiating the tree mortality process) and contributing factors (i.e., those leading to tree mortality; Manion 1991). Forest structure is a key predisposing factor. For example, greater stand biomass increases the degree of competition from neighbours (Gómez-Aparicio et al. 2011), reduces growth (Ruiz-Benito et al. 2017a) and increases mortality likelihood (Jump et al. 2017; Changenet et al. 2021). Another predisposing factor can be tree size, which is related with tree mortality so that small trees are particularly vulnerable to mortality (Ruiz-Benito et al. 2013; Andivia et al. 2020). Drought is a key inciting factor due to its interaction with forest structure which strongly reduces tree vigour (Choat et al. 2012; McDowell and Allen 2015).

Tree mortality associated to forest decline is a complex and hardly predictable process due to its stochastic nature and the multiple interactive drivers controlling the process (Hartmann et al. 2018; Trugman et al. 2021). Background tree mortality can be defined as the occurrence of tree mortality without extreme disturbances, linked to forest succession and species ontogeny (Franklin et al. 1987; Bréda and Badeau 2008). On the other hand, die-off mortality is related to events where a large proportion of the trees die as a consequence of extreme abiotic and biotic disturbances (e. g., fires, droughts or pests) (Hammond et al. 2022). Forest die-off events are being increasingly observed worldwide due to global change (Jump et al. 2017; Hammond et al. 2022), pointing to the need of investigating broad-scale mortality events to understand forest dynamics under changing conditions (Allen et al. 2015; Jump et al. 2017).

Tree mortality events occur at different spatial scales (Allen et al. 2010). Tree mortality affects specific trees at local scales, yet die-off mortality can affect a larger proportion of trees at landscape scales (Jump et al. 2017; Changenet et al. 2021). When regional die-off events occur, the spatial autocorrelation of tree damage and mortality decreases with distance, informing about the spatial extent of these events (Gazol et al. 2022). Furthermore, declining trees tend to occur close to each other and, therefore, the spatial autocorrelation of tree damage can be used as an early warning signals of die-off events (Camarero et al. 2015). These early-warning signals include extraordinary growth reductions, crown defoliation and discoloration (Dobbertin 2005) or partial canopy dieback (Jump et al. 2017). However, the spatial patterns of tree damage and mortality at large spatial extents and their variation with forest structure and climate effects are largely unexplored.

Tree vulnerability in terms of tree damage and mortality is strongly linked to the functional characteristics of tree species (Greenwood et al. 2017). In this sense, gymnosperms can be highly sensitive to increased temperature and drought (Carnicer et al. 2013; Anderegg et al. 2015b), but angiosperms are not exempt for suffering drought-induced decline events (Camarero et al. 2021). Angiosperm species are more competitive and tend to dominate under mesic conditions and late successional stages. While gymnosperms are more sensitive to competition in juvenile stages and dominate in more xeric environments and during early successional stages (Zavala et al. 2000). Despite these contrasting responses, whether tree damage and mortality spatial patterns differ between these different functional groups are barely known.

Mediterranean forests are highly exposed to climate change because they are vulnerable to the concomitant occurrence of summer heat and drought (Gazol and Camarero 2022), and the predicted increase on aridity (IPCC 2022). On the one hand, the abandonment of forest management and forest expansion in old agricultural fields are leading to biomass accumulation and densification, which increases forest vulnerability to pests, pathogens, and fires (Bradford and Bell 2017; Cruz-Alonso et al. 2019). On the other hand, afforestation practices have led to dense and monospecific forests (Ruiz-Benito et al. 2012; Vadell et al. 2016), which are also very vulnerable to disturbances (Ruiz-Benito et al. 2013; Sánchez-Cuesta et al. 2022).

Here, we quantified the spatial patterns and the underlying drivers of tree damage and mortality in Mediterranean forests and analysed how these spatial patterns and the effects of the drivers are changing over time. For this, we used three consecutive censuses from the Spanish National Forest Inventory since the 1980s decade. Our specific objectives were to: (i) analyse the spatio-temporal patterns of tree damage and mortality by forest type (angiosperm, gymnosperm, and mixed forests) in Mediterranean forests of the Iberian Peninsula; (ii) assess to what extent these patterns are spatially related to structural and climatic drivers; and (iii) investigate the temporal dynamics in the spatial dependence of tree damage and mortality on structural and climatic drivers. We hypothesised that there is spatial aggregation in tree damage and mortality events, being stronger in tree damage than tree mortality and in gymnosperm forests than in angiosperm and mixed forests. In addition, we expected a spatial dependence between tree mortality/damage and their drivers, aggregation increasing with basal area and drought and decreasing with tree size. Finally, we hypothesised that the spatial aggregation of tree damage and mortality increased over time as well as the effect of stand structure and drought. The assessment of spatio-temporal trends in tree damage and mortality patterns is key for understanding the effects of global change and the identification of vulnerable areas required to design cost-effective adaptation measures.

Materials and methods

Study area

We focused on forests under Mediterranean climate in Spain. In the East of the Iberian Peninsula, a temperate and humid climate with dry and hot summers predominates. The mean temperature characterizing the selected distribution area is 17–18 °C and approximate precipitation is 400–450 mm. A more continental Mediterranean climate with dry and warm summers and frequent frosts in winter is the most common in the inner Iberian Peninsula, with mean temperatures around 15 °C and annual precipitations of 420 mm. In the southern and south-eastern area, the climate is semi-arid, with a mean temperature of 18 °C and annual precipitation of 300 mm (AEMET and IMP 2011). More than a half of the selected forest area is represented by angiosperm forests, dominated by Quercus ilex L., Quercus suber L., and Quercus pyrenaica Wild. Gymnosperm forests occupy 38% of the selected forest area with Pinus halepensis Mill. and Pinus pinaster Ait. as the most common species (MITECO 2020).

National Forest Inventory data

To assess the temporal and spatial patterns of tree damage and mortality in Mediterranean forests, we used data from the Spanish National Forest Inventory (hereafter SFI). This dataset comprises plots systematically distributed in areas considered as forests (tree coverage greater than 5%) on a 1-km2 cell grid (Villaescusa and Díaz 1998). Four censuses of SFI are currently available: 1SFI (1965–1974), 2SFI (1986–1996), 3SFI (1997–2007) and 4SFI (2008-present). The data between SFI censuses is only comparable in permanent plots established since the 2SFI, being available for the entire country in the 2SFI and 3SFI. In addition, despite the information of the 4SFI is not yet available for the whole country, we used the most updated data (see inset map of Fig. 1c) covering a large environmental gradient. Therefore, to assess the spatio-temporal trends in tree damage and tree mortality we used two different datasets: (1) the permanent plots between the second and third SFI to assess the spatial patterns of tree damage and mortality (covering the entire continental Spain), (2) the permanent plots between the second, third and fourth SFI to assess the temporal patterns of change in tree damage and tree mortality (in this case not covering the entire continental Spain with a wide spatial gradient from the Atlantic northwester Spain to the wet and dry Mediterranean areas of northeaster and southeaster Spain, respectively). From the SFI plots, we removed the Atlantic, Alpine and Macaronesian forests (from 76,665 to 48,305 plots in 23SFI and from 12,981 to 8,789 comparable plots in 234SFI), which are in the temperate broadleaved and mixed forest region following the ecoregion classification in Olson et al. (2001).

Each SFI plot follows a nested design of four concentric circular subplots of 5, 10, 15 and 25 m radius. In these subplots, trees are sampled according to their diameter at breast height (tree size): ≥ 7.5 cm, ≥ 12.5 cm, ≥ 22.5 cm and ≥ 42.5 cm, respectively (Villaescusa and Díaz 1998). For each sampled tree, height, size, species identity, the origin and magnitude of tree damage and the status (alive, dead) are recorded. The origin of the damage is classified as unknown (c. 11% of all damaged trees in the 3SFI), biotic (e.g., fungi, insects, dominance; c. 64% of all damaged trees in the 3SFI), abiotic (e.g., snows, winds, droughts or fires, c. 20% of all damaged trees in the 3SFI) or anthropogenic (e.g., machinery damage; c. 5% of all damaged trees in the 3SFI). For this study, we discarded the latter group. Therefore, we retained biotic (pests and pathogens) and abiotic (i.e. climatic, fires) damage. The magnitude of the damage is classified as low, medium or high, but here we only considered medium or high tree damage to follow a conservative approach (c. 66% of the damaged trees were classified as low). To reduce the bias in the damage and mortality estimation introduced by silvicultural activities, plots with any signal of silviculture between censuses (e.g., thinning or cutting) were removed from the analysis (from 48,305 to 37,128 plots).

We quantified absolute and relative tree damage and mortality through basal area of sampled trees (DBH > 7.5 cm), calculated as the sum of the cross-sectional area of the trunk from the diameter at breast height of each living sampled individual (m2 ha−1). Absolute tree mortality was calculated as the sum of the basal area of dead trees divided by the number of years between consecutive inventories (23SFI and 34SFI; m2 ha−1 year−1). Absolute tree damage was calculated as the sum of basal area of all trees that experienced medium or high tree damage in the 3SFI (m2 ha−1) and the 4SFI (m2 ha−1). To calculate relative tree damage (%) and mortality (% year−1), we divided absolute values by the total basal area of the plot (the second census in the case of mortality). Initially, we explored absolute and relative tree damage and mortality because they could have different patterns between close and open forests (the later particularly common in SW Spain, see Moreno-Fernández et al. 2019). However, we retained for subsequent analysis absolute tree damage and mortality because the correlation between absolute and relative values was high (r ~ 0.6 for both tree damage and mortality; see Online Appendix A, Fig. A.1).

After the plot filtering, we used selected permanent plots between the second and third inventories (23SFI) to assess tree damage and mortality. To investigate the role of forest type on tree damage and mortality, we classified each plot into angiosperm, gymnosperm, or mixed forests. Angiosperm and gymnosperm forest types correspond to those plots with dominance equal or greater than 80% in total basal area for angiosperms or gymnosperm species, respectively, while the rest of the plots were classified as mixed forests. We analysed 37,128 permanent plots from the 23SFI (12,799 plots of angiosperm, 15,847 plots of gymnosperm and 8,482 plots of mixed forests; Online Appendix B, Fig. B.1). To study the temporal changes in the spatial patterns of tree damage and mortality, we used permanent plots across the three consecutive inventories (8789 permanent plots in the 2-3-4SFI; 4106 correspond to angiosperm, 3771 to gymnosperm and 912 to mixed forests).

Underlying drivers of tree damage and mortality

We considered stand characteristics and spatio-temporal variations in climate as key drivers of tree damage and mortality. Initially, stand structure was characterised through stand basal area (m2 ha−1), stand tree density (No. trees ha−1), mean tree size (mm), coefficient of variation of tree diameter (ratio of the standard deviation to the mean of the diameters of all trees in the stand) and slenderness (ratio between mean DBH and tree height). We retained as proxies of stand structure stand basal area and mean tree size, due to the high correlation with other variables of stand structure, low correlation with climatic variables and similar spatial dependences with tree damage and mortality when correlation between potential structural variables was high (e.g., Online Appendix C, Fig. C.1 and C.2).

The spatial variations in climate were characterised through water availability, calculated as the difference between annual precipitation and potential evapotranspiration, divided by potential evapotranspiration (%) for the period between 2 and 4SFI. Thus, high values of water availability correspond to wet regions and low values to arid regions. Water availability was calculated using data from Moreno and Hasenauer (2016) and Rammer et al. (2018) databases with the easyclimate R package (Cruz-Alonso et al. 2023). Temporal variations in drought intensity were characterised through changes in the minimum Standardised Precipitation Evapotranspiration Index (SPEI, Vicente-Serrano et al. 2010). SPEI is a dimensionless index with low values related to dry conditions and high values to wet climatic conditions. We calculated the 12-months minimum SPEI per plot for each month for the period between 23 and 34SFI and kept the lowest annual value in the series using the SPEIbase v2.7. database and the SPEI package (Vicente-Serrano et al. 2010; Beguería and Vicente-Serrano 2017). With the lowest annual value, we calculated the minimum annual value between NFIs and the mean of the minimum annual SPEI. When compared across sites these two variables reflect different aspects of drought: the first variable reflecting the occurrence of extreme events between inventories, and the second reflecting pervasive drought conditions between inventories. Despite correlation between both were relatively high (see r > 0.6 in Online Appendix C, Fig. C.3) the autocorrelation of the mean of the minimum was greater and, therefore, we decided to retain this variable as an indicator of the intensity of drought for subsequent analyses.

Statistical analyses to assess the spatio-temporal patterns of tree damage and mortality

To describe the spatial patterns of tree damage and mortality by forest type (angiosperm, gymnosperm, and mixed forests), we used spline correlograms. Spline correlograms are a generalization of spatial correlograms which calculate the relationship between two distance matrices and provide a global measure of the spatial autocorrelation of events in space (Fortin and Dale 2014). Values can indicate, at a given spatial lag, a negative spatial correlation (i.e., − 1 indicates high and low values of the target variable tend to occur close in space, that is, repulsion), or a positive spatial correlation (i.e., 1 means that spatially nearby values of a variable tend to be similar, that is aggregation). Values close to 0 represent the absence of spatial correlation (Burian 2012; Fletcher and Fortin 2018). For tree damage and tree mortality in each forest type, we calculated correlograms as a continuous function of distance being 50 km the maximum distance considered (Bjørnstad and Falck 2009). We used 999 iteration bootstrapping (i.e. resampling technique used for inferring uncertainty) to generate 95% confidence intervals of autocorrelation or cross-correlation values (Fletcher and Fortin 2018). To investigate the temporal dynamics in tree damage and mortality, we calculated the spatial autocorrelation in tree damage and tree mortality in permanent plots available across the three last censuses of the SFI corresponding to a timeframe of approximately 30 years (see Online Appendix B, Fig. B.2).

Spatial dependence between tree damage and mortality with underlying structural and climatic factors, and its variation over time

To test the effect of the underlying drivers on tree damage and mortality as well as the associated spatial patterns between the events and the drivers, we followed a two-step approach. First, we fitted hurdle-gamma models (Mullahy 1986). We modelled tree damage and mortality in each stand considering stand variables (basal area and tree size) and climate (water availability and drought intensity) as fixed factors, X and Y coordinates as covariates to control the spatial autocorrelation of the response, a ziGamma conditional error distribution, and a log link function. Afterwards, the importance of the variables was estimated using the Akaike Information Criterion (AIC; Burnham and Anderson 2002). We compared the full model (i.e., including all structural and climatic variables as fixed effects) with models in which each explanatory variable from the fixed effects was removed. To characterise the goodness of fit, we observed the distribution of the residuals of the most parsimonious model and calculated pseudo-R2 (Nakagawa and Schielzeth 2013; Online Appendix D, Fig. D.1 and D.2). We used glmmTMB (Brooks et al. 2017), DHARMa (Hartig 2022), pscl (Jackman 2020) and bbmle (Bolker 2022) packages in R 4.1.2. (R Core Team 2021).

Second, we performed cross-correlogram analyses to assess the spatial dependence between tree damage or mortality and the structural and climatic drivers (Bjørnstad and Falck 2009). The cross-correlogram uses a modified Moran’s I statistic to evaluate the relationship between two variables at different spatial scales (Fortin and Dale 2014) and, therefore, inform about the spatial dependence between co-existing events (Rossi et al. 1992). This can be interpreted as segregation (negative values) or aggregation (positive values) between variables (Larson and Franklin 2006). Further than 50 km, the spatial patterns did not vary, so we interpreted the results at distances lower than 50 km. All correlograms and cross-correlograms were computed using the package ncf (Bjørnstad 2022).

Finally, to assess if the underlying factors of tree damage and mortality varied over time, we performed four hurdle-gamma models with the permanent plots available in the three consecutives censuses of the Spanish Forest Inventory, i.e., one model for each response variable and period (23SFI and 34SFI for tree mortality, 3SFI and 4SFI for tree damage). We modelled tree damage and mortality as a function of basal area, tree size, water availability and drought intensity. We used Gamma conditional error distribution using a log link function. The most parsimonious model for each response variable and period was chosen based on the ΔAIC. We checked the distribution of the residuals of the most parsimonious model and calculated pseudo-R2 (Nakagawa and Schielzeth 2013; see Online Appendix D, Fig. D.1 and D.2). Finally, we calculated cross-correlograms to study the variation over time in the spatial dependence between tree damage and mortality with the relevant structural and climatic drivers (Bjørnstad and Falck 2009).

Results

Spatio-temporal patterns of tree damage and mortality in Mediterranean forests

Overall, we found a mean increase of 63% in tree damage between 3 and 4SFI; and 23% in tree mortality between 23 and 34SFI, with patches of high damage and mortality distributed across the Iberian Peninsula (Fig. 1a, b). Mixed forests had the highest tree damage and mortality, whereas angiosperm forests had more damage and less mortality than gymnosperm forests in the studied periods (Fig. 1c, d). Tree damage increased between consecutive inventories in all studied forests and mortality increased in gymnosperm forests (Fig. 1c, d). Mixed forests had the highest increase in tree damage (from 1.24 ± 2.20 to 2.32 ± 3.04 m2 ha−1) followed by angiosperm forests (from 1.59 ± 4.57 to 2.53 ± 4.87 m2 ha−1, Fig. 1c). Gymnosperm forests had a significant increase in tree mortality between SFI periods (from 0.12 ± 0.33 to 0.17 ± 0.35, Fig. 1d).

Fig. 1
figure 1

a Tree damage (m2 ha−1) and b tree mortality (m2 ha−1 year−1) distribution across Mediterranean forests in the Iberian Peninsula. Boxplots of c tree damage and d tree mortality in angiosperm (brown), gymnosperm (green) and mixed forests (blue) in the third, and fourth Spanish Forest Inventory are shown. The inset in (c) represents regions in which permanent plots in the three consecutive Spanish Forest Inventories are available. The asterisks (*) represent significant differences (P < 0.05) between SFI periods for the studied forests type; capital letters indicate significant differences (P < 0.05) between the different types of forest within the same SFI period

Full size image

We found a positive spatial autocorrelation in tree damage and mortality, being stronger for tree damage (with values ranging between 0.32 and 0.43 at short distances, i.e., 1–3 km) than tree mortality (0.08–0.21; Fig. 2a, b). Furthermore, while tree damage had spatial aggregation at distances shorter than 20 km, tree mortality had spatial autocorrelation only for gymnosperm forests at distances shorter than 10 km (see that the confidence intervals does not overlap the zero-line until km 10, Fig. 2a, b). Gymnosperm forests showed the highest spatial aggregation in tree mortality at short distances while angiosperms and mixed forest the lowest (Fig. 2b and Online Appendix A, Fig. A.2). At longer distances there were no differences among the three forest types (Fig. 2a, b). We did not find important differences in the autocorrelation of tree damage and tree mortality over time (Fig. 2c, d). The only exception was found for tree damage at c. 10 km, with higher autocorrelation in the second period (Fig. 2c).

Fig. 2
figure 2

Spatial autocorrelation of a, c tree damage (m2 ha−1) and b, d tree mortality (m2 ha−1 year−1) in gymnosperm (green), angiosperm (brown) and mixed forests (blue) and 23SFI or 3SFI (solid line) and 34SFI or 4 SFI (dashed line). Filled area represents 95% confidence intervals

Full size image

Underlying drivers of the spatial patterns in tree damage and mortality

The selected models for tree damage and mortality included all potential explanatory variables, except mean tree size for tree mortality (see ΔAIC in Table 1). Basal area, water availability and drought intensity had positive cross-correlation with tree damage and mortality at short distances. Water availability showed the strongest positive cross-correlation with both tree damage and mortality (i.e., correlation between 0.1 and 0.2, decreasing until distances up to 50 km, Fig. 3a, b) followed by drought intensity for tree damage (Fig. 3a), and basal area for tree mortality (Fig. 3b). Mean tree size had a low and negative spatial cross-correlation with both tree damage and mortality (Fig. 3a, b).

Table 1 Comparisons of alternate models of tree damage and mortality based on Akaike information criterion (AIC) to test the main effects of structure and climate
Full size table
Fig. 3
figure 3

Spatial cross-correlation between a tree damage (m2 ha−1) and b tree mortality (m2 ha−1 year−1) with basal area (brown), mean tree size (green), water availability (red) and drought intensity (blue). Solid lines and filled area represent the mean correlation and 95% confidence intervals

Full size image

Temporal variations in tree damage and mortality and their underlying drivers

The selected model for tree damage included basal area, tree size, water availability and drought intensity for both time periods (Table 2). However, tree size was not included in either of the two models for tree mortality (Table 2). The spatial cross-correlation of tree damage and mortality with the underlying drivers changed over time in some cases (Fig. 4). The spatial cross-correlation of tree damage with basal area, water availability and drought intensity significantly increased from the first to the second period (Fig. 4a, e, g). We observed no significant differences between periods in the cross-correlation of tree mortality with basal area, tree size and water availability (Fig. 4b, d, f). Nonetheless, the spatial dependence between tree mortality and drought intensity increased over time (Fig. 4h).

Table 2 Comparisons of models of tree damage for 3SFI and 4SFI and mortality for 23SFI and 34SFI, i.e., the consecutive periods of the Spanish Forest Inventory data, based on Akaike information criterion (AIC) to test the main effects of structure and climate
Full size table
Fig. 4
figure 4

Spatial cross-correlation between tree damage (m2 ha−1) and basal area (a), tree size (c), water availability (e) and drought (f); and between tree mortality (m2 ha−1 year−1) and the explanatory variables (b, d, f, h) by SFI period. Solid lines represent the mean correlation of 23SFI period or 3SFI, dashed lines correspond to the mean correlation of 34SFI period or 4SFI and filled area represent 95% confidence intervals. Grey lines and filled areas represent variables that do not have a significant effect on tree mortality or tree damage

Full size image

Discussion

We found positive spatial autocorrelation of tree damage and mortality in Spanish Mediterranean forests. Tree damage showed higher autocorrelation and longer distances than tree mortality. As hypothesised, our analyses revealed distance-dependent spatial relationships between tree damage and mortality with stand basal area, water availability and drought intensity. However, the spatial autocorrelation of tree damage and mortality did not vary over time, neither in extension nor in magnitude. Despite this unexpected result, the spatial dependence between tree damage and their underlying drivers increased over time, while the spatial dependence of tree mortality on drivers only increased with drought intensity. Altogether, our results provide key information about the spatio-temporal variation of tree damage and mortality patterns in the Iberian Peninsula and the role of stand characteristics, water availability and drought intensity in these spatial patterns at landscape scale.

Spatio-temporal patterns of tree damage and mortality in Mediterranean forests

Tree damage was greater in magnitude and showed higher spatial autocorrelation than mortality (see Fig. 2). Despite scarce previous information on the spatial aggregation of tree damage and mortality at large spatial extents, these results agree with studies reporting aggregated patterns of tree decline (Prieto-Recio et al. 2015) and mortality at finer scales (Silver et al. 2013; Baguskas et al. 2014). Moreover, tree mortality due to insect and pathogens have shown greater spatial autocorrelation (e.g., 25 km, Calvão et al. 2019) and rapid increase over time (Kautz et al. 2017) partially due to increases in temperature and occurrence of droughts (Weed et al. 2013). In fact, drought-induced tree damage and mortality has been largely documented in the study area (Astigarraga et al. 2020; Gazol and Camarero 2022). However, we did not distinguish the cause of tree damage and mortality on our analyses, because despite the high importance of drought-induced responses we found a low number of trees where drought-induced tree damage was identified and a high number of trees where dominance, pest and pathogens were the main cause (1% versus 24% and 33% of all damaged trees in the 3SFI plots, respectively). The low importance of drought-induced tree damage further supports our theoretical framework and, therefore, considering all potential causes of tree mortality as a spiral of events where several drivers might be affecting in sequence (i.e. it is hard to attribute tree damage and mortality to a unique driver as dominance, pest and pathogens or drought interact driving the mortality process, see Franklin et al. 1987). However, further analysis could consider specific easily identifiable causes as fire, storm, biotic or snow (e.g. Barrere et al. 2023), but here we were interested in the combined effect of multiple drivers (see Seidl et al. 2017). This approach will require the continuous monitoring of target trees over time rather than re-surveying plots every ten years but at the cost of reducing the spatial extent and, therefore, understanding spatial patterns.

Forest decline and mortality events are occurring up to regional extents worldwide (Allen et al. 2015). In agreement, we found a greater magnitude in the spatial autocorrelation for tree damage than for tree mortality. This suggests that tree damage can be considered an early-warning indicator in the spiral of tree mortality, because tree damage generally precede tree mortality events (i.e., pitch defence or recovery; Franklin et al. 1987; Jump et al. 2017), which is relevant information for the use of large-scale information available. Furthermore, the interaction of drought with biotic agents and fires can trigger tree mortality and determine its spatial extent (Moreno-Fernández et al. 2021; Camarero et al. 2015; Romeiro et al. 2022), and, therefore, tree damage can be used as early warning signals for potential die-off events.

Mixed forests had the greatest tree damage and mortality, with the greatest increase over time in tree damage for mixed forests and for tree mortality in gymnosperm forests (Fig. 1c). On the one hand, diverse sites often allow relatively high stand biomass and tree growth for certain site conditions due to complementarity effects (e.g. Ruiz-Benito et al. 2014). However, as stand structure develops and stand basal area and tree size changes, the increased tree density and competition can be leading to increased tree mortality linked to higher functional diversity (Searle et al. 2022). On the other hand, tree mortality due to biotic and abiotic factors (i.e., bark beetles, exotic pathogens or fires) is increasing in European forests since the 60 s (Patacca et al. 2022; Adame et al. 2022). An example is the bark beetle outbreaks that are affecting in recent years as a consequence of the Vaia storm in northern Italy in 2018 (Bozzini et al. 2023). The increases in tree mortality rates over time agree with temporal trends in Spain indicating higher mortality rates in gymnosperm forests (Astigarraga et al. 2020; Gazol et al. 2022), which can be related to their high sensitivity to increasing temperatures and droughts (Ruiz-Benito et al. 2017b).

Contrary to our hypothesis, we did not find support for increases in the spatial autocorrelation of tree damage and mortality over time (see Fig. 2). This result could be due to two plausible explanations. First, widespread die-off and mortality events are relatively rare and constitute small proportions of the landscape where specific networks of measurement in the clusters of change will be required (Trumbore et al. 2015). Second, a mean period of ~ 10 years is relatively short to study temporal changes in forests composed by long-lived species (see e.g., Ruiz-Benito et al. 2020). The combination of the two potential explanations could be causing the spatial trends in tree damage and mortality, because we observed an increase in the magnitude of tree damage and mortality, but we did not observe a variation in their spatial extent because regional die-off is strongly dependent on drought-induced extreme events (Allen et al. 2010; Brodribb et al. 2020) and a period longer than 10 years continuously measured across the space might be required to further assess the temporal trends of these patterns.

Effects of stand structure, water availability and drought intensity on the spatial patterns of tree damage and mortality

Water availability had the greatest effect on the spatial patterns of tree damage and mortality, which can be related to the steep spatial gradient of aridity in the Mediterranean region of the Iberian Peninsula (i.e., from 100 to 700 mm of precipitation and 5 to 17.5 °C of average temperature, Costa et al. 2005). Previous studies have observed that wetter areas are experiencing higher levels of tree damage and mortality than drier areas (Ruiz-Benito et al. 2013; Benito-Garzón et al. 2013), but for the best of our knowledge this work is the first showing evidence of the dependency of damage and mortality spatial aggregation on water availability. Greater mortality in wet sites have been linked to the specific biotic and abiotic conditions of productive sites, such as sporadic events of droughts or high temperatures and higher competition (see e.g., Ruiz-Benito et al. 2013; Changenet et al. 2021). Furthermore, despite tree mortality being high in drier areas, tree species adaptation to Mediterranean conditions can buffer the effects of adverse climatic conditions at large spatial extents (Bréda et al. 2006; IPCC 2022) and local increases can be less representative than those caused by the steep gradient in water availability.

We found support for our hypothesis of increasing droughts and stand basal area leading to more ample aggregation in tree damage and mortality (Jump et al. 2017). On the one hand, trees under more frequent and intense droughts have greater probabilities of die-off and mortality events (Allen et al. 2010; Neumann et al. 2017; Gazol and Camarero 2022). Under dry conditions, water can abruptly change from liquid to gas creating xylem cavitations and, ultimately, embolism (Tyree and Zimmermann 2002; Choat et al. 2018). This process interrupts water flow and diminishes the transport of water to the canopy, causing hydraulic failure and tree mortality (Nardini et al. 2013). On the other hand, trees in dense forests are more vulnerable to experiencing tree mortality, because of increased competition for light and soil resources (Bradford and Bell 2017; Jump et al. 2017; Astigarraga et al. 2020). Therefore, all this suggests that forests under increased climatic stress and competition have a greater probability of experiencing tree damage and mortality at wide spatial extents (e.g., Allen et al. 2015; Jump et al 2017).

Increased spatial dependence over time of tree damage and mortality with stand structure, water availability, and drought intensity

The spatial dependency of tree damage with basal area, water availability and drought intensity increased over time, but no significant increases were found for tree mortality except for drought intensity. Our result agrees with increased negative effects of global change on the spatial patterns of tree damage over time, that could be enhanced by increased sensitivity to climate as forests show particularly high increases for tree damage over tree mortality (i.e., increased effect of stand basal area and drought, McIntyre et al. 2015; Astigarraga et al. 2020). The increase of spatial dependence of tree damage in the last Spanish Forest Inventory might be explained due to a stronger role of stand structure and climate that might be increasing tree susceptibility to water limitation (Jump et al. 2017). Mediterranean forests are suffering warmer and dryer summers, with subsequent effects on tree demography and damage (Ruiz-Benito et al. 2013; Anderegg et al. 2015a; Díaz-Martínez et al. 2023). The increase in the spatial dependence of tree damage and mortality with drought over time, which remains strong as distance increases, highlights a clear drought-induced effect with intensified sensitivity over time (Jump et al. 2017; Anderegg et al. 2019; Astigarraga et al 2020). Therefore, the increased sensitivity to drought over time on the spatio-temporal patterns of tree damage and mortality suggests the importance of further assessing both response variables and their dependence over time and space when evaluating die-off events.

Conclusions

The spatial patterns of tree damage and mortality in Iberian forests are changing over time and differential patterns are emerging at landscape extents. These processes are driven by multiple biotic and abiotic factors, particularly water availability, drought intensity and stand structure. Furthermore, we demonstrate that the role of drought intensity on the spatial patterns of tree damage and mortality is increasing over time. The Spanish Forest Inventory emerges as a key tool to assess large-scale tree damage and mortality patterns. However, to further attribute causal factors or understand the effects of climate change on the spatial patterns of tree damage and mortality, it is necessary the establishment of coordinated monitoring networks replicated over time or a combination of methods combining field and remote sensing data (i.e. Bozzini et al. 2023) to increase the spatial and temporal resolution (Trumbore et al. 2015; Hartmann et al. 2018). Analysing spatio-temporal patterns of tree damage and mortality and the underlying drivers provides key information to identify sensitive areas to disturbances and can be a useful tool to design effective management measures.