Four decades of drought impacts on Swiss protective forests: a vanishing altitudinal gradient

Swiss protective forests are distributed along a wide altitudinal gradient, where many are already facing the deleterious impacts of repeated drought episodes. How intensifying drought will influence forest structure and, ultimately, their protective effect is still an open question. In this study, we investigated the temporal evolution of forest demographic traits and canopy moisture resilience over four decades, using National Forest Inventories (NFI) and satellite-derived vegetation index (NDMI), examining both growing and declining forest stands across five altitude belts. Our results reveal a general decline in forest stability and productivity in growing stands, evidenced by decreasing stand basal area increment driven by accelerating stem density losses rather than a reduction in average individual tree growth. Declining stands were consistently exposed to higher drought stress, particularly during the two most recent NFI periods. A critical finding is the vanishing influence of the altitudinal gradient on forest canopy resilience components, indicating a convergence of drought acclimation levels across the gradient. In declining stands, fundamental resilience thresholds appear to be exceeded independently of altitude, while the specific resilience component signaling decline shifted from resistance to recovery over time, suggesting a progressive compositional adaptation that nonetheless proved insufficient to prevent continued decline. Drought stress reshaped individual tree growth drivers before inducing stand-scale reorganization. The vulnerability of large, dominant individuals to drought were observed broadly across most periods, although a structural shift in stand growth partitioning toward smaller trees was detectable only in declining stands at the highest altitude and only in the most recent periods, alongside with higher average tree growth rates in that belt. These findings emphasize the urgent need for adaptive management strategies that promote structural heterogeneity, sufficient stem density, and the transition toward drought-tolerant tree cohorts to sustain the forest protective function under intensifying drought.

In Switzerland, protective forests represent the most important biological measure against natural hazards, such as avalanches, rockfall, landslides, flood and sediment/debris transport processes in channels, due to the large area it covers (44% of the national forest cover in the territory; 540'000 ha) (Strauss and Fischer, 2025). Protective forest is defined as a forest that can effectively and continuously protect a recognized damage potential from an existing natural hazard, hereafter referred to as the “protective effect” (Teich et al., 2022). Such forests play a fundamental role in reducing risks for people and infrastructure, and maintain livelihoods in mountain regions. The protective effect depends on forest structural traits, such as the stand basal area, the stem density and/or the canopy cover (e.g., Teich et al., 2012; Dupire et al., 2016; Moos et al., 2016). Its stability depends not only on the current forest structure (Schmid et al., 2024) but also on the trees' capacity to cope with their environment. In this frame, forest management is oriented to the maintenance of forest cover with minimal canopy gaps and a multi-aged forest structure with a variety of tree size (O'Hara, 2006).

Forest management, however, has become increasingly complex due to unprecedented global warming over the last decades with more frequent and extreme drought episodes (Spinoni et al., 2018; Suarez-Gutierrez et al., 2023), amplifying tree mortality as well as long-lasting decline in forest growth (Allen et al., 2010, 2015; Senf et al., 2020; Taccoen et al., 2019), weakening physiologically survival trees (Zlobin, 2024), altering the growth partitioning between trees (Pretzsch et al., 2022), and limiting maximum stand density (Bravo-Oviedo et al., 2018; Condés et al., 2017). The extent to which increasing drought episodes will affect the forest's protective function is still unclear. In Switzerland, protective forests are distributed over altitudinal belts ranging from colline to subalpine superior. A recent review highlighted the influence of altitude on the climate-induced changes of the protective effect (Moos et al., 2023). The majority of the analyzed studies tended toward a decrease of the protective effect in the forest stands located at lower altitude (i.e., less than 1,000 m a.s.l.). However, these findings are primarily based on forest simulations and carry uncertainties due to the limited availability of empirical data of those forests, especially on their resilience to drought disturbances and potential changes in forest structure.

Moreover, the severity of drought impacts on forest and tree growth varies along an altitude gradient. Forest stands at low altitudes generally experience higher water deficits, resulting in reduced tree stem diameter growth rate (Pretzsch et al., 2020; Krejza et al., 2021) and higher tree mortality (DeSoto et al., 2020), potentially inducing a lack of large trees in those altitudes (Bhutia et al., 2019). At intermediate to high altitudes, drought impacts on trees tend to exacerbate in recent years, but at the same time, tree growth rates still increase due to warmer temperatures (Pretzsch et al., 2020; Pompa-García et al., 2021a; Obojes et al., 2024) and increased recovery and resilience of forest canopy density to drought episodes compared to forests at lower altitudes (Li et al., 2023). However, as extreme drought episodes intensify, it remains uncertain whether this historical buffering effect of altitude is currently weakening, potentially disrupting the traditional altitudinal gradient of canopy resilience to drought over time.

In general, drought stress and risk of tree mortality increase with site aridity (Li et al., 2023; DeSoto et al., 2020; Bouwman et al., 2025; D'Orangeville et al., 2018) and depend on tree social status and/or tree size (Vospernik and Nothdurft, 2018; Ma et al., 2023), resulting in an asymmetric growth partitioning within forest stands (Pretzsch et al., 2022; Bose et al., 2022; Nixon et al., 2025; Forrester, 2019; Keyser and Brown, 2016). While drought-related shifts of the size hierarchy pattern and altered stand growth were mostly studied in forests for wood production, empirical evidence on drought impacts on mountain forests in steep terrain remains scarce. Consequently, how drought stress drives stand structure, either by promoting an asymmetric growth partitioning favoring large acclimated trees, or through a more generalized decline across all tree sizes, remains to be quantified.

Climate-induced forest growth changes emphasize the trends at regional and landscape scales. Nonetheless, due to their transitory characteristic, drought episodes explain only a small portion of the observed variability of stand basal area increment (Gazol et al., 2022), particularly when a minor portion of trees within the forest stand was impacted. By focusing solely on the stand basal area increment, the late detection of the deleterious effects on forest growth limits our ability to identify the underlying patterns promoting forest decline induced by drought.

However, such analyses require exhaustive datasets at tree scale over large areas and for long observation periods. Recent studies used satellite imagery to compute vegetation indices, like the NDVI (Normalized Difference Vegetation Index), to detect upper-layer forest canopy greenness in response to drought episodes (Gazol et al., 2018). Jevšenak et al. (2024) reported, though, that NDMI (Normalized Difference Moisture Index) is the most important predictor of individual tree growth, indicating its relevance for capturing forest stands resilience to drought while representing overall individual tree growth. In this vein, by using the indices of resilience components (Lloret et al., 2011), Sturm et al. (2022) were able to map the Swiss forests resilience to the 2018 drought episode and the main climatic and forest parameters contributing to canopy moisture resilience during and after a drought episode. The authors also highlighted a key limitation of satellite-based approaches, which primarily integrate signals from the dominant canopy trees. In multi-layered forest structures typical of protective forests, understory strata may partially compensate for dominant tree decline, potentially introducing a bias in canopy-level resilience metrics that obscures actual stand structural dynamics. Altogether, while the effects of drought on forest canopy condition and stand structure are increasingly documented across environmental gradients, how these responses jointly translate into long-term structural changes in protective forests, where structural continuity is critical, remains poorly understood.

This study establishes an empirical and descriptive baseline aimed at quantifying the spatiotemporal evolution of the influence of altitude on drought stress intensity, forest canopy resilience to drought and subsequent stand structural dynamics over the past forty years in Swiss protective forests. To align the observed responses at both the stand and canopy levels with individual tree dynamics, our approach explicitly couples stand-scale trajectories with tree-level growth responses. We focus on three main objectives: (i) examining if the altitudinal gradient of drought stress and canopy resilience has shifted over time; (ii) determining whether canopy resilience contributes to stand-level growth trajectories; and (iii) identifying the structural and individual mechanisms driving these trajectories.

Using long-term datasets from the Swiss National Forest Inventory (NFI), we first evaluate the spatiotemporal dynamics of forest decline, drought stress intensity (DSI), and canopy moisture resilience across altitude belts and inventory periods. We then analyze how these factors dictate the structural and demographic traits of growing versus declining stands, tracking metrics such as stand basal area increment (Δ BAs), stem density increment (Δ N), and the growth dominance coefficient (GDC). Finally, we aim to disentangle how altitude gradients, canopy resilience to drought, and stand structural traits interact with drought stress intensity to drive individual tree-level growth (Δ BAt) over time.

We hypothesize that: (i) the drought stress is increasing significantly in the mid- and high-altitude belts, leading to a gradual disappearance of the canopy resilience gradient over time; (ii) low canopy resilience to extreme drought episodes is associated with stand decline (Δ BAs < 0). However, the threshold at which low canopy resilience triggers net stand decline varies along the altitude gradient due to baseline environmental constraints. Finally, we also expect that (iii) Δ BAs trajectories (i.e. growing or declining) and their underlying mechanisms differ along the altitudinal gradient. At low altitudes, stand decline is driven by both reduced individual tree growth (Δ BAt) and severe net stem density losses (Δ N), resulting in symmetric growth partitioning (GDC). At high altitudes, stand decline is instead characterized by a growth reduction among the larger trees, albeit from a higher baseline of overall individual tree growth than at low altitudes, reflecting the a buffering effect of lower drought stress intensity.

National forest inventory data

Our dataset is based on Swiss National Forest Inventory data collected between 1983 and 2022. Forest stands are defined as circular plots distributed on a 1.4 × 1.4 km systematic permanent sample grid covering the entire Swiss forests. All trees with a stem diameter at breast height (DBH) >12 cm were measured within a circle of 2 ares, while only trees with a DBH >36 cm were surveyed in a circle of 5 ares. We selected stands only in protective forest areas and excluded stands with an intervention within the 10 years before the survey and with fire traces. In average, we retained 1,380 forest stands per NFI period (from 891 up to 1,543 forest stands) from an altitude of 281.9 m up to 2,218.6 m above sea level. For each stand and each NFI period, we focused on the forest demographic traits defined as the stand and tree basal area, and net stem density increments, the growth dominance coefficient and the tree growth. The relative annual stand basal area and stem density increment between two surveys were computed as (Equations 1, 2): with BAs i and N i, the initial stand basal area (m 2) and net stem density (stem. ha −1) measured in the survey i, respectively. BAs j and N j, the final stand basal area and net stem density measured in the survey j, respectively. Δ BAs ij and Δ N ij the relative annual stand basal area increment (%. yr −1) and net stem density increment (%. yr −1) for the NFI period ij, and nVS the number of vegetation seasons between surveys.

Annual tree growth was quantified as the annual basal area increment (Δ BAt, cm 2. yr −1) of each living trees measured in both NFI surveys for a dedicated period by dividing the difference between the basal area of the tree at NFI i and j by the number of the vegetation season of the NFI period.

Then, for each forest stand and NFI period, the growth dominance coefficient (GDC) was computed following Binkley (2004) (Equation 3): where n is the number of trees in the stand, x k is the cumulative proportion of the initial tree basal area for the k -th tree (with x 0 = 0), and y k is the cumulative proportion of the tree basal area increment for the k -th tree (with y 0 = 0), with trees ordered in ascending order by their initial basal area.

Characterization of drought conditions

Monthly temperature and precipitation were derived from the DaymetCH dataset. This data set contains gridded monthly mean temperature and precipitation values interpolated from daily measurements from MeteoSwiss weather station based on the Daymet method (Thornton et al., 1997) at a spatial resolution of 100 m. We characterized the drought conditions at a specific site based on a drought stress index (DSI). It was calculated using a simple water balance model following (Bugmann and Cramer, 1998), which is a refinement of the Thornthwaite water balance model (Thornthwaite, 1955). Although various more complex methods exist to compute the water balance (e.g., Granier et al., 1999; Speich et al., 2020), Thornthwaite-type methods only require time series of monthly temperature and precipitation (Lutz et al., 2010) and, thus, are usable with limited data and over lager scales. As with most Thornthwaite-type methods, precipitation is treated as immediately available to the soil moisture pool, without an explicit snow accumulation and melt delay module.

In the used water-balance model, water can be intercepted by vegetation, evaporated from bare soil, stored in the soil as soil moisture, taken up and transpired by vegetation, or a surplus that runs off of the surface. The calculation requires two time series, mean monthly precipitation and temperature, and two state variables that can vary in space, latitude and soil water-holding capacity. Monthly potential evapotranspiration (PET m) was estimated using a heat index derived from the monthly mean temperature and corrected for sun angle and day length depending on latitude and the local topography [following Bugmann and Cramer, 1998; adapted from Thornthwaite and Mather (1957) and Pastor and Post (1985)]. These corrections are particularly relevant in mountain environments where topographic shading and seasonal variation in day length substantially influence evaporative demand.

Monthly mean precipitation is partitioned in infiltration and interception (I m). The latter is determined as a constant fraction of monthly precipitation, estimated from mean values for broadleaf and conifer species (Miralles et al., 2010).

By subtracting the interception from PET m, the monthly evaporative demand (D m) is calculated as Equation 4: The evaporative supply rate S m is calculated as a linear proportion of the soil moisture of the previous month (SM m) and the soil water storage capacity (SWC) according to Federer (1982) (Equation 5): Whereas c w is a supply rate constant (i.e., maximum rate of evaporation).

Evapotranspiration from the soil E m is assumed to be the minimum of the demand and the supply (Equation 6): The yearly drought stress index DSI (dimensionless) is finally defined as 1 minus the ratio of the summed monthly evapotranspiration and the summed demand (Equation 7): The site aridity was defined as the cumulative DSI (DSI.c) for a specific NFI period and each forest stand. To compare the drought stress between forest stands that were surveyed at different dates (i.e., with variable NFI period duration), two another metrics related to the drought conditions were analyzed: the median and the maximum DSI values over a specific NFI period. Extreme drought episodes were identified using a threshold of 1.5 on the z score from the DSI time series for each forest stand. Extreme drought episodes were mostly detected in 2003, 2006, 2015, 2018, and 2022 with more than 30% of the entire studied forest stands affected, including the forest stands in subalpine superior areas. Despite the observed increase of DSI in 1990 and 1991, only 16.1% and 15.6% of the total studied forest sites were concerned, respectively (Figure 1).

Satellite imagery

For each forest stand, satellite imagery from 1984 to 2022 were collected using Landsat 5, 7, and 8 collections from SentinelHub platform (Sentinel Hub, Sinergise Solutions d.o.o., a Planet Labs company). We focused on August month as following Sturm et al. (2022). Normalized Difference Moisture Index (NDMI) was used as the indicator of forest canopy moisture content and was computed following (Gao, 1996) (Equation 8): where NIR is the Near Infrared band and SWIR1 is Shortwave Infrared band.

To mitigate bias from differing Landsat satellite sensors, NDMI values were normalized (Sulla-Menashe et al., 2016) by extracting PIFs (pseudo-invariant features) pixels outside vegetation areas, using CORINE layer (https://land.copernicus.eu/en/products/corine-land-cover/clc2018) to identify pixels in continuous urban and industrial zones, under the assumption that these land-use classes remained stable in extent over the 1984–2022 study period, and pairing scenes assessed in August within each pair of overlapping years. Linear regressions using the total least squares method (Markovsky and Van Huffel, 2007) were then performed to estimate the parameters (slope and intercept) to transform the data. As Landsat 7 is the intermediate sensor for our time series, it was used as the reference. Because the Scan Line Corrector (SLC) on the Landsat 7 ETM+ sensor failed on May 2003, resulting in a permanent loss of approximately 22% of scene coverage per acquisition, two linear regressions were fitted between Landsat 5 and 7 (before and after the SLC failure). A third regression was similarly fitted between Landsat 7 and Landsat 8 over their overlap period. Cross-sensor regression fits varied in strength (Supplementary Table S5), reflecting the reduced density and spatial distribution of valid PIF pixels following the 2003 Landsat 7 SLC failure and the more limited temporal overlap between Landsat 7 and 8.

Forest stands responses to extreme drought episodes were quantified using the resilience components proposed by Lloret et al. (2011), and so-called hereafter Rt NDMI, Rc NDMI and Rs NDMI for the resistance, recovery and resilience in NDMI, respectively. While single-season NDMI does not capture multi-year legacy effects of drought on forest growth, this study focuses on within-drought canopy stress response and post-drought recovery at the episode scale, consistent with the Lloret resilience framework. The length of the reference periods (pre- and post- drought episode) were set to a period of four years.

The SentinelHub platform delivers pre-processed satellite images with cloud masking and atmospheric correction procedures without topographic (i.e., terrain-illumination) correction. However, all analyses rely exclusively on the ratio-based resilience components of Lloret et al. (2011) rather than absolute NDMI values. Static terrain-illumination bias for a given pixel is expected to be consistent across years within the same fixed acquisition month (August), and therefore cancels within these intra-pixel ratios, independent of its magnitude or its variation across topography and altitude belts.

Data analysis

Stand basal area increment (Δ BAs) integrates distinct processes: tree growth, mortality and ingrowth, and serves as an essential metric for assessing the protective effect of forests. However, because NFI data are acquired approximately every 10 years, stand-level growth metrics inherently integrate multiple years of climatic fluctuations, structural shifts, and lag effects (Anderegg et al., 2015). Consequently, the relationship between decennial stand-level growth and short-term drought severity is highly non-linear and governed by complex ecosystem thresholds (Vicente-Serrano et al., 2013). To capture these non-linear thresholds, forest stands were split into two distinct growth categories based on their relative Δ BAs during each NFI period. Forest stands with positive Δ BAs were labeled as “growing” (Gr), whereas those with negative Δ BAs were labeled as “declining” (De).

To ensure this binary classification separated stands into distinct demographic regimes rather than artificial groups slightly fluctuating around structural equilibrium (stability), we systematically validated these groups by applying Welch's t -test to compare drought-related metrics (DSI and NDMI resilience components) and forest demographic traits (relative net stem density increment, growth dominance coefficient and average tree growth) between the two categories for each NFI period. This approach ensures that while Δ BAs serves as the overarching structural proxy for the forest's protective function, its constituent demographic drivers are explicitly validated and untangled.

Then, the influence of the altitude belt gradient on the drought metrics, as well as the forest demographic traits was evaluated using Kruskal-Wallis tests. Where significant effect of altitude belt were detected, post-hoc pairwise comparisons were performed using Welch's t -tests with Benjamini-Hochberg p -value adjustments. Given the unequal sample sizes across NFI periods and altitude belts, all statistical tests were performed independently for each period, precluding direct cross-period statistical comparisons.

Moreover, as NDMI reflects canopy moisture status rather than tree growth or biomass directly, the NDMI-derived Lloret resilience components describe changes in canopy condition in response to drought episodes. To empirically evaluate their relevance as proxies of stand structural dynamics, their relationship to stand basal area increment trajectories was explicitly assessed across growth categories and altitude belts.

To quantify the contribution of drought stress conditions, canopy NDMI resilience to extreme drought episodes and initial stem density to individual tree growth rates, linear-mixed effect models were fitted to individual tree basal area increment (Δ BAt) for each NFI periods. The fixed effects included initial tree diameter (DBH.i), canopy NDMI resilience (Rs NDMI), cumulative drought stress index (DSI.c), initial stem density (N.i), slope, aspect, and altitude belt categories (AB). To account for the hierarchical structure of the data, forest stand was included as a random effect.

Two-way interactions between DSI.c and DBH.i, N.i, and Rs NDMI were explicitly included to test how initial structure and canopy resilience modulate individual tree growth (Equation 9).

As protective forests are predominantly mixed-species stands and this study aims to characterize general trends rather than species-specific responses, tree species was not included as a predictor.

All continuous predictors were standardized (scaled and centered) prior to model parametrization. Model performance was evaluated using adjusted R 2, Pearson correlation coefficients between observed and predicted values, and Root Mean Square Error (RMSE). Potential multicollinearity among predictors was rigorously checked by computing the Variance Inflation Factor (VIF), ensuring all values met acceptable thresholds. Spatial autocorrelation was assessed on model residuals using Moran's I index. Finally, model parameter estimates and confidence intervals were derived via bootstrapping, and slope comparisons across altitude belts were performed using estimated marginal means (EMMs), thereby formally testing the interaction effects between the altitude gradient and the drought predictors. where u j ~ N (0, σ u 2) represents the random intercept for each forest site number, and ε i j ~ N (0, σ ε 2) represents the residual error.

All data analyses were performed using R (R Core Team, 2025) with terra (Hijmans, 2025), nlme (Pinheiro and Bates, 2000; Pinheiro et al., 2025), emmeans (Lenth, 2025), rcompanion (Mangiafico, 2025), and parameters (Lüecke et al., 2020) libraries.

Temporal evolution of forest dieback over the altitude gradient

Overall, the proportion of declining stands increased over time across all altitude belts, with the most pronounced and consistent trend observed at low altitudes, while higher belts showed more variable temporal dynamics (Figure 2). At low altitudes, the colline (Cn) belt showed a consistent and progressive increase in declining stands, from 11.1% in NFI 1-2 to 30.4% in NFI 4-5, representing the strongest temporal trend observed across all belts. In the mountain inferior (MI) belt, the proportion remained relatively stable across the first three periods (averaging 17.4%) before slightly increasing in NFI 4-5 (21.4%). The mountain superior (MS) belt showed no clear directional trend, with a transient reduction in declining stands during NFI 2-3 (10.8%) but otherwise similar proportions across the remaining periods. In contrast, the subalpine inferior (SI) belt, despite also reaching its lowest proportion in NFI 2-3 (15.9%), showed a gradual increase in declining stands thereafter. The subalpine superior (SS) belt displayed a distinct pattern, with a decrease in declining stands over the first three periods (from 14.8% in NFI 1-2 to 10.3% in NFI 3-4), followed by an abrupt increase in NFI 4-5 (18.8%).

Impact of drought conditions and extreme drought episodes on stand basal area increment over time

Declining stands were primarily characterized by higher DSI than growing stands in the most recent NFI periods, with DSI values decreasing with altitude across all periods and growth categories, and the altitudinal differentiation becoming less pronounced over time (Figure 3, Table 1 and Supplementary Table S1). At low altitudes, the Cn and MI belts showed similar median DSI values in most NFI periods, with the exception of growing stands in NFI 1-2 and 4-5. However, maximum DSI was consistently lower in the MI belt than in the Cn belt for growing stands across all periods. In declining stands, this difference in maximum DSI between the two lowest belts was significant only during NFI 2-3 (Figure 3, Supplementary Table S1).

Declining stands also exhibited lower canopy Rs NDMI to extreme drought episodes than growing stands, with this trend strengthening over time (Figure 4). However, in declining stands specifically, no altitudinal differentiation was observed in any of the studied periods (Supplementary Table S2). In particular, declining forest stands exhibited significantly lower Rt NDMI values during NFI 2-3 and 3-4, while Rc NDMI only differed in NFI 4-5, resulting in a lower Rs NDMI from NFI 2-3 onward (Table 1). In growing stands, altitude influence was observed during NFI 1-2, with higher Rt NDMI and Rs NDMI in the Cn, MI and MS belts decreasing gradually with altitude, while Rc NDMI showed the opposite trend, despite no significant differences between altitude belts (Supplementary Table S2). This altitudinal pattern reversed in NFI 2-3, with lower Rt NDMI in the Cn and MI belts relative to higher altitude belts, and significant highest Rc NDMI in the Cn belt. During NFI 3-4, the SI belt showed the highest Rc NDMI, followed by the MI and MS belts (Figure 4, Supplementary Table S2).

Evolution of the forest demographic traits in growing and declining forest stands along the altitude gradient

In growing stands, structural traits showed contrasting temporal and altitudinal patterns. Relative stand basal area increment (Δ BAs) decreased over time, with the overall altitudinal effect disappearing in NFI 4-5 as the decrease became widespread across all altitude belts. Relative net stem density increment also declined over time, from an average of 2.84% in NFI 1-2 to 1.92% in NFI 4-5, with no altitudinal effect observed in any period. In contrast, GDC values were significantly more negative with increasing altitude across all NFI periods, indicating a consistent shift from proportional growth contribution at low altitude toward growth dominance by smaller trees at higher-altitude growing stands (Figure 5, Supplementary Table S3). Individual tree growth (Δ BAt) remained stable over time within each altitude belt but differed consistently between belts across all periods, with the lowest rates in the Cn, MI and SS belts (12.65 cm 2.yr -1 on average) and the highest rates in the MS and SI belts (14.96 cm 2.yr -1 on average).

In declining stands, forest structure changes were strongest at low altitude and intensified over time. Δ BAs was less negative with increasing altitude during NFI 4-5, and the Cn belt showed the strongest temporal decrease, from –1.89% in NFI 1-2 to –2.29% in NFI 4-5 (Figure 5, Supplementary Table S3), whereas the SS belt showed the least negative Δ BAs values overall (–1.23% on average). Across all periods, declining stands exhibited lower net stem density increment and individual tree growth, together with higher GDC, than growing stands (Table 1). GDC differed significantly across belts only during NFI 3-4 and 4-5, with values remaining close to zero or positive at low-to-mid altitudes, and shifting toward negative values, indicative of small-tree growth dominance, specifically in the SS belt (Supplementary Table S3). Net stem density increment differed significantly with altitude, except during NFI 3-4 (Figure 5). Finally, Δ BAt differed between altitude belts only during the two most recent periods, with higher rates in the SS belt (Supplementary Table S3).

Drivers of individual tree growth of Swiss protective forests over time

The major parameter explaining positively individual Δ BAt was the initial tree diameter (DBH.i, Figure 6). The drought growth conditions (DSI.c) showed a pervasive negative effect on Δ BAt but its negative interaction with DBH.i was not significant in the NFI 3-4 and 4-5 (Figure 6 and Supplementary Figure S2). Median stand NDMI resilience (Rs NDMI) to extreme drought episodes impacted negatively Δ BAt only during the last NFI period. The interaction between Rs NDMI and DSI.c was positive during the NFI 1-2 but negative in the NFI 4-5 (Figure 6 and Supplementary Figure S1). Aspect did not significantly influence Δ BAt, while slope had a negative impact. The initial stem density (N.i) negatively influenced Δ BAt for all NFI period. The interaction between N.i and DSI.c was significant and positive only in NFI 4-5. The altitude belt had a negative effect on the Δ BAt, with stronger difference in SS belt. Slopes showed significant differences for the studied periods. However, the slopes of intermediate belts (i.e., MI and MS) were similar, while the slope of SI belt differed from lower altitude belt until NFI 4-5.

Temporal shifts in forest demographic traits and the vanishing altitude gradient in drought resilience

Our results highlight the evolution of growth and decline in Swiss protective forests over the past 40 years, emphasizing the influence of drought conditions and resilience components on changes of forest demographic traits. In Switzerland, drought-induced forest decline accounts for only a small proportion of forests dying from “natural” causes [1.6% according to Scherrer et al. (2022)]. In stands where tree growth is sensitive to drought, Gazol et al. (2022) highlighted the critical but partial role of drought episodes as a stress factor relative to other co-occurring drivers. Within this broader context, our results showed a decreasing proportion of growing forest stands over time, particularly during the last NFI period, points to a general decline in forest stability and productivity. This negative trajectory in stand basal area increment Δ BAs is fundamentally driven by lower net stem density increments rather than a drop in average individual tree growth (Figure 5 and Supplementary Table S3), suggesting that stressful conditions trigger stem density losses at the stand scale.

At the tree scale, we observed a temporal shift in the relationship between Rs NDMI and Δ BAt. While no correlation between Rs NDMI to extreme drought episodes and Δ BAt was found in the first NFI periods, a significant positive correlation emerged in the NFI 4-5 (Figure 6). This is likely linked to the higher frequency of extreme drought episodes during the shortest NFI 4-5 period (Figure 1), which increased the likelihood of capturing the deleterious effects of lower stand resilience on subsequent annual tree growth, unlike earlier periods where pre- or post-drought conditions may have buffered these impacts.

In line with our first hypothesis, although drought stress metrics consistently decreased with altitude for all NFI periods, our results indicated a progressive loss of the altitude influence on forest canopy resilience to extreme drought episodes and subsequent forest demographic traits in growing forest stands (Figures 5, 6, Supplementary Table S2). The significant effect of altitude on resilience components gradually vanished over the successive NFI periods: first for Rs NDMI (NFI 2-3), followed by Rt NDMI (NFI 3-4), and finally Rc NDMI (NFI 4-5) (Figure 4). This pattern suggests a convergence of acclimation levels and forest responses to drought episodes across the altitude gradient.

Given the high variability of forest responses to drought, which are typically mediated by tree species composition, topography, and local historical climatic conditions (Müller and Bahn, 2022; Sturm et al., 2022), this homogenization is likely linked to successive spatio-temporal changes in drought stress intensity coupled with shifting forest composition and tree acclimation levels. For instance, forest stands at intermediate and high altitude belts may temporary benefit of the warmer growth conditions (Pompa-García et al., 2021b; Pretzsch et al., 2020), which could explain the higher Rc NDMI and average Δ BAt observed in growing forest stands within these belts during NFI 2-3 and 3-4 (Supplementary Tables S2, S3). However, the legacy effects of more frequent extreme drought episodes may increasingly exceed the recovery capacity of trees across all belts (Müller and Bahn, 2022; Kannenberg et al., 2020).

In contrast, in declining stands, no influence of altitude on the resilience components was observed across all studied periods (Figure 4). This suggests the existence of fundamental drought resilience thresholds that, once surpassed, lead to stand decline largely independently of altitude (Adams et al., 2017; Anderegg et al., 2015). Overall, these results showed that the historical buffering effect of altitude against drought-induced damage is vanishing, suggesting that, should drought stress continue to intensify, stand decline may become increasingly independent of altitude in the future.

Resilience thresholds and the onset of stand decline

Low canopy resilience to extreme drought episodes was consistently associated with stand decline, although the specific resilience components involved shifted over time, confirming our second hypothesis. The drought stress index (DSI) did not increase linearly over the study period. However, declining forest stands were associated with drier growth conditions than growing forest stands during NFI 3-4 and 4-5 (Figure 1). The resilience components linked to stand decline also shifted over time: declining stands initially exhibited lower Rt NDMI (NFI 2-3 and 3-4) which was subsequently replaced by lower Rc NDMI in NFI 4-5 (Figure 4 and Table 1). As canopy moisture status is predominantly determined by dominant trees, the resilience components primarily reflect the level of drought acclimation of the largest individuals in the stand. The observed shift from lower resistance to lower recovery may indicate a transition from stands dominated by trees inherently vulnerable to drought, such as large or coniferous individuals (Forrester, 2019; Bose et al., 2022; Stangler et al., 2022; Krejza et al., 2021; Gazol et al., 2022), toward stands dominated by trees that resist initial drought stress but may become progressively fatigued under repeated drought episodes, as observed during the most recent NFI period (Choat et al., 2018; DeSoto et al., 2020). Together, these results suggest that declining stands underwent a degree of compositional adaptation over time, with the most vulnerable individuals to initial drought stress being progressively filtered out, shifting the limiting resilience component from resistance to recovery. This adaptation, however, appears insufficient to prevent continued decline, suggesting that the fundamental resilience threshold identified in previous section was exceeded even as stand composition shifted toward more drought-resistant individuals. Because tree species was not included in this study, we cannot distinguish whether this compositional adaptation reflects within-species selection (e.g., preferential loss of the most vulnerable individuals) or turnover toward more drought-tolerant species. Since both of these mechanisms are plausible, a more in-depth analysis would be needed to clearly distinguish between them.

Drought-induced shifts in tree-scale growth drivers and their structural consequences along the altitude gradient

Our results reveal that drought stress reshapes the functional relationships between tree size, canopy resilience, and individual growth, ultimately triggering localized structural reorganization exclusively within high-altitude declining stands, validating our third hypothesis.

At tree scale, the individual tree growth models explained a moderate but meaningful share of variance in Δ BAt (adjusted R 2 = 0.52–0.57; Supplementary Table S4), which is consistent with the inherently multi-causal nature of tree growth at large spatial and temporal scales. A substantial portion of the remaining variance likely reflects localized predictors not captured at this scale, such as soil properties, micro-topography, or species-specific responses, whose inclusion would require targeted plot-level investigations beyond the scope of a national inventory framework. The relationship between canopy resilience Rs NDMI and individual tree growth Δ BAt shifted between periods. Rs NDMI only became a significant predictor of Δ BAt in NFI 4-5 (Figure 6). Concurrently, the interaction between Rs NDMI and cumulative drought stress (DSI.c) reversed from positive in NFI 1-2 to negative in NFI 4-5. Since Rs NDMI is primarily driven by the signal of dominant trees, this reversal aligns with the negative interaction observed between initial diameter (DBH.i) and DSI.c across most periods. Large trees exhibit higher vulnerable due to their higher water demand and maintenance costs (Forrester, 2019; Bose et al., 2022; Keyser and Brown, 2016). The negative interaction between Rs NDMI and DSI.c during NFI 4-5 indicates that constrained dominant trees may limit the growth of subordinate trees to maintain baseline functional requirements, shifting competition dynamics in multi-strata protective forests.

Crucially, our results showed that stand-scale growth dynamics (Δ BAs) and structural baselines differ between growth categories (Table 1), modulating how forests adjust to growth conditions. Growing stands maintained stable, inherently zero-to-negative GDC values across all inventory periods (ranging from −0.004 to −0.119, Supplementary Table S3), indicating a trend to allocate growth toward smaller trees. Notably, the MS and SI belts maintained high average tree growth rates in NFI 4-5. This growth partitioning stability likely mediated by net stem density increment under dry conditions (Figure 6). Rather than a static buffering effect, the observed declines in net stem density increments over time in growing stands (Figure 5) suggest that these stands may undergo a shifting stem density-dependent equilibrium to optimize resource allocation among remaining trees under drought stress (Bravo-Oviedo et al., 2018; Bello et al., 2019; Sachsenmaier et al., 2025).

Conversely, declining stands initially exhibited positive baseline GDC values (0.041 to 0.233 during NFI 1-2 and 2-3, Supplementary Table S3), indicating that growth was predominantly allocated to large individuals. This structural configuration may have contributed to their vulnerability to drought stress (Pompa-García et al., 2021b; McQuillan et al., 2022). A localized structural reorganization was also observed in the SS belt during NFI 3-4 and 4-5, suggesting that their stand basal area growth may become increasingly reliant on smaller, potentially more resilient individuals. Indeed, SS belt GDC values were the most negative among declining stands (−0.206 and −0.158, respectively), alongside the highest average Δ BAt (12.91 and 11.65 cm 2. yr −1, respectively). Moreover, with growing and declining stands distinguished by Rc NDMI and Rs NDMI (Table 1), those smaller trees are potentially more drought-tolerant, better positioned to maintain their growth under repeated drought stress. The SS belt therefore emerges as a singular case among declining stands, where growth partitioning toward smaller, potentially more resilient individuals may represent an early signal of compositional turnover under intensifying drought, a pattern not detectable elsewhere along the altitudinal gradient.

The maintenance of stand basal area under intensifying drought stress is therefore a sequential, multi-scale process where stands undergo structural and density adjustments to reach new ecological equilibria. Our results indicate that before a major structural shift in growth partitioning becomes detectable at the stand scale, modifications occur predominantly at the tree scale through drought-induced constraints on dominant trees, coupled with a progressive reduction of net stem density over time. Growing stands are characterized by a structural strategy in which growth is not exclusively allocated to the largest trees, whereas declining stands favor the growth of dominant trees, making them vulnerable during drought episodes. A major shift from this pattern occurs in the SS belt, where the GDC of declining stands dropped to strongly negative values, forcing a structural reorganization in favor of smaller individuals. Therefore, these results indicate that stand-scale responses are nonlinear, determined by differences in baseline growth dominance among growth categories and by the ability of stands to structurally adjust their stem density limits. Although this study isolated general, large-scale forest demographic trajectories, future investigations explicitly quantifying how the interaction between local vertical stratification and stand density modulate these sequential processes under dry growth conditions will be crucial for comprehensively understanding the mechanisms underpinning growth maintenance at the stand scale.

Implications for the Swiss protective forests

The increasing number of declining forest stands at lower altitudes and the general decrease of stand basal area increment (–29.57% and –17.58% between NFI 1-2 and 4-5 in Cn and SS belts, respectively) may result in a significant decrease in their protective effects in the long-term. Stand basal area can be considered as good proxy for the protective effect of a forest against various natural hazard processes (Moos et al., 2017; Cislaghi et al., 2021). Thus, its decrease is directly linked to a potentially diminishing protective effect of forests with ongoing climate change. Our study showed not only a decreasing trend of relative basal area increment, but also net stem density increment. The low rate of tree regeneration could critically jeopardize the stability and resilience of the protective forests in the longer term (Strauss and Fischer, 2025). This is in line with recent simulation-based studies, predicting critical declines in stand basal area and thus, the protective effect, of forests at lower elevations (e.g., Moos et al., 2021; Hillebrand et al., 2023). However, Hillebrand et al. (2023) found that changing temperature and drought conditions predominantly affect protective forests at lower elevations, whereas high-altitude forests exhibited weaker responses. Similarly, Huber et al. (2021), using the dynamic forest model ForClim, predicted an increasingly negative influence of drought on lower-elevation forest stands across Switzerland. At high elevation, forest response highly depended on soil conditions. The present study corroborates these findings, indicating that changes in drought conditions increasingly impact also forests at higher elevation.

Ensuring the long-term protective effect of these forests will require a proactive, targeted management. Regular silvicultural interventions can promote regeneration and accelerate the transition to more drought-tolerant individuals (e.g., Elkin et al., 2015; Huber et al., 2021). This is also supported by the positive influence of higher initial stem density and lower individual tree stem diameter on tree growth under dryer conditions found in this study. Nonetheless, the extent to which management can reduce the drought vulnerability of protective forests remains subject for further investigations.

Limitations

Our study summarized the long-term trends in the multiscale responses to drought in Swiss protective forests and their resulting stand basal area increment. Improving the characterization of water stress along an elevation gradient by incorporating an explicit snow module (for example, snowmelt models based on degree-days) is a key improvement for better representing mountain hydrology. In fact, the calculated DSI index, like most water balance indices based on the Thornthwaite model (including the widely used Standardized Precipitation-Evapotranspiration Index, or SPEI), does not explicitly account for snowpack accumulation or the delayed release of water from snowmelt. In high-altitude zones, this could lead to an underestimation of water availability early in the season if winter precipitation is retained as snow cover rather than infiltrating immediately, thereby delaying the actual replenishment of soil moisture relative to the time of snowmelt. In this case, water stress would therefore be overestimated for subalpine zones. We note that this limitation is not specific to the DSI method applied: the standard SPEI formulations share the same assumption of immediately available precipitation; it is therefore a constraint common to Thornthwaite-type approaches at monthly resolution, rather than a shortcoming specific to the index applied here.

The varying quality of calibration between sensors, depending on sensor pairs and time periods (Supplementary Table S5), introduces uncertainty into the NDMI-derived resilience components for drought episodes occurring near or across sensor transitions, most notably the 2003 extreme drought. The Landsat 5–7 regression was indeed notably weaker after the SLC failure in 2003 than before it, and the Landsat 7–8 regression was overall the weakest of the three regressions performed. This factor, as well as the absence of explicit topographic correction mentioned in the “Methods” section, should be taken into account when interpreting the magnitude of the resilience components, especially for the 2003 drought episode. Nonetheless, the patterns and relative responses observed along the altitudinal gradient for a given drought episode remain representative of the variability in canopy response.

Furthermore, the net stem density increment was interpreted as a net demographic signal rather than decomposed into mortality and ingrowth components, as the NFI data available for this study record only the resulting net change, without distinguishing between individual mortality and recruitment events. A stable or moderately negative net stem density increment could therefore mask mutually offsetting dynamics of mortality and ingrowth that cannot be disentangled from the current dataset.

Finally, this study characterizes overall structural and demographic trends across all protective forests and does not take into account tree species composition, which may itself vary systematically along the elevation gradient and be associated with different levels of resilience or vulnerability to drought. Since the protective effect of these forests relies primarily on their structure rather than their species composition, analyzing species-specific responses was beyond the scope of this study. However, the composition and regeneration of tree species as a function of elevation could contribute to the observed altitudinal patterns and represent an important area of future investigations.

Despite these limitations, the consistency of the observed patterns across four decades of NFI data, multiple independent drought episodes and forest canopy moisture resilience metrics, and the full altitude gradient supports the robustness of the general trends reported here.

Over the last 40 years, Swiss protective forests have shown a decline in forest growth performance, particularly at lower altitudes, partially driven by increasing drought stress, reduced resilience, and declining net stem density.

Our findings underscore that this vulnerability is undergoing a critical transition with a progressive loss of the influence of altitude on forest resilience to drought. The historical advantage of higher altitudes against drought stress is diminishing, suggesting a convergence of acclimation levels where drought stress thresholds are increasingly exceeded across all belts, regardless of altitude. This convergence, however, does not yet translate into uniform structural responses. Declining stands exceeded fundamental resilience thresholds independently of altitude, while the specific resilience component signaling this threshold shifted from resistance to recovery over time, suggesting a progressive compositional adaptation that nonetheless proved insufficient to prevent continued decline.

At the structural level, our results indicate that drought stress reshapes individual tree growth drivers before inducing stand-scale reorganization. Tree-level penalties on large, dominant individuals are detectable broadly and across most periods, while a corresponding stand-level structural shift (toward growth allocated to smaller trees) remains, to date, detectable only in stands already exhibiting basal area loss at the highest altitude, and only in the most recent periods. Whether this localized response represents a viable long-term structural recovery pathway or a delayed manifestation of the decline trajectory observed at lower altitudes remains an open question.

Ensuring the long-term protective function of these forests requires a proactive and spatially differentiated management approach. Our results support prioritizing the maintenance of sufficient stem density and the promotion of structurally diverse, multi-layered stands to enhance local microclimate buffering under intensifying drought. Accelerating the transition toward drought-tolerant tree cohorts through targeted silvicultural intervention will be critical, particularly at low and intermediate altitudes where decline trends are most pronounced and the protective effect most immediately at risk.