1 INTRODUCTION

Climate change and land-use change are the main drivers of past and current plant diversity. These drivers and their interaction are leading to shifts in species distribution, the extinction of the most vulnerable species, and a reshuffling of existing communities (Franklin et al., 2016; Kuhn & Gégout, 2019; Pecl et al., 2017; Thomas et al., 2004). The flora of the forest understorey makes up 80% of forest vascular plant diversity and plays a key role in many ecosystem functions (Landuyt et al., 2019). Likewise, understorey flora is being increasingly considered in forest management decisions in the face of climate change (Blondeel et al., 2021; Gilliam, 2007; Landuyt et al., 2019), and in global forest conservation and restoration efforts (Stanturf et al., 2014).

The influence of free-air temperature (the macroclimate) on understorey communities is buffered by the cover of local overstorey trees (De Lombaerde et al., 2021; Godefroid et al., 2006; Maclean et al., 2015; Zellweger et al., 2020). Tree cover in temperate forests creates an understorey microclimate characterized by cooler maximum temperatures and warmer minimum temperatures. Microclimate depends on local stand conditions, but is also driven by broader scale factors. Topography for example is an important factor that can influence temperature through elevation, aspect, and cold air pooling, as is macroclimate, which is determined by drivers such as latitude, solar radiation and distance to the coast. The effects of local tree cover on the understorey microclimate and species communities are increasingly under study; however, less is known about the effect of landscape-scale forest cover on understorey composition.

The forest habitats in central European landscapes have typically undergone intensive logging in the past, which has resulted in the current mosaic of forest patches of different sizes. These patches, where forest understorey species may persist under existing tree cover, are embedded in an extensive agricultural matrix (Institute for Environment and Sustainability et al., 2013). The influence of landscape-scale forest cover on the regional climate is still under debate and is currently the subject of many studies, since forest cover at the landscape scale involves two processes pulling in opposite directions. On the one hand, temperatures may increase with increasing forest cover because a forest's albedo is lower than other land cover types like grasslands and croplands. This warming effect is most apparent in cooler seasons, when forests do not retain snow cover and release a latent heat flux at night. On the other hand, forests have a cooling influence during warmer seasons, stemming from their higher evapotranspiration, which directly cools the air and promotes cloud formation (Bonan, 2008; Hesslerová et al., 2013; Pokorny et al., 2010). The growing season is a critical period for both annual and perennial cold-adapted species. One could therefore expect that landscape-scale forest cover could benefit plant species by cooling the hotter and drier (mean and extreme) conditions in spring and summer that could induce the dieback of vulnerable species. In these landscapes, the community could also be comprised of cold-adapted species because they can outcompete the warm-adapted species located at the cold edge of their distributions (Sanczuk et al., 2022). Highly forested landscapes could also influence plant community composition and favour cold-adapted species through other means. For example, large forest patches in central Europe historically grow on low-nutrient soils unsuitable for agriculture, and are less influenced by fertilization from nearby croplands than small forest patches (Bergès et al., 2016). This is relevant as the cold-adapted species in Europe are also adapted to poorer soil conditions (Ewald, 2003). Taken together, these two characteristics increase the potential for highly forested landscapes to conserve cold-adapted species.

In this study, we investigated the influence of forest cover in the surrounding landscape on plot-scale (i.e., local) plant community composition in agriculture–forest mosaics in the temperate biome. We carried out pairwise comparisons of forest plots with contrasting landscape-scale forest cover (1-km area), and used the community temperature index (CTI) as a proxy for community adaptation to climate. CTI is calculated as the average thermal optimum of the recorded species of a plot. The thermal optimum of a species is estimated from its maximum probability of occurrence along the temperature gradient and therefore reflects the climate that the species experiences in its biogeographic area. CTI may be used to compare a species’ or a community's tolerance to the warming climate; a large difference between the current climate and a species’ optimum can be an early warning sign of local or regional extinctions (Kuhn & Gégout, 2019).

We hypothesized that (a) plant communities surrounded by highly forested areas would have a lower CTI than those located in landscapes with little forest cover; that (b) there would be a significant landscape-forest cover effect, which, (c) together with soil factors, would explain the differences in CTI.

2 MATERIALS AND METHODS

2.4 Study area, plot selection, and geographic pairing of the plots

Our study site was located in the French temperate broadleaf and mixed-forest biome where 31% of the land cover is forested (IGN, 2019). The climate is oceanic to continental (with larger temperature amplitudes) and the MAT ranges from 8 to 11 °C (Météo France weather stations; Piedallu et al., 2019). The study area included mountain ranges, but the strict selection procedure described hereafter mainly focused on lowland areas characterized by alternating expanses of croplands, grazing land and deciduous forest patches of different sizes.

We selected NFI plots exclusive to this biome, and refined our selection with two criteria: (a) the plots had to be forested – as opposed to open land or recent clear-cuts, and (b) to ensure the quality of the CTI value, the plots had to host more than five species for which thermal optimum information was available.

The plots with more than 80% of forested area within a 1-km radius (251 ha) were classified as ‘forested’ (F) and those with less than 30% (94 ha) were classified as ‘non-forested’ (NF). We selected pairs of ‘forested’ versus ‘non-forested’ plots based on two constraints: (a) the distance between the plots was < 5 km, and (b) the elevation difference was < 50 m. These criteria allowed us to minimize the macroclimatic differences between plots. In order to give the same weight to each plot, a given plot was only included in one pair. We used an algorithm to maximize the number of possible pairs respecting the two constraints: we first paired isolated plots, then paired the plots with many neighbours (see the Data Availability section for codes). This process resulted in a larger sample size of paired plots, although the mean distance between the plots of a pair was not minimized.

The final dataset contained 4,024 plots arranged into 2,012 pairs (Figure 1a). The mean distance between the plots in a pair was 3.6 km (see Figure 1b for an example). The maximum and minimum distances between plots were 5 and 1 km (NFI grid resolution), respectively. F and NF plots were similar in terms of macroclimatic MAT, with a mean average difference of only 0.06 °C across all plots (Table 1). The plot-scale basal area and canopy cover differed by 2.1 m²/ha and 0.6%, respectively (Table 1). Average pH and mean Ellenberg N values differed between the two landscape classes: they were higher in NF than in F plots by 0.8 and 0.9, respectively (Table 1).

TABLE 1. Environmental characteristics of forest (F) versus non-forest (NF) plots.

Landscape of the plot	Number of plots	Mean annual temperature (°C)	Elevation (m a.s.l.)	Landscape forest cover (ha)	Landscape forest cover (%)	Mean Ellenberg N	Bioindicated pH	Basal area (m2/ha)	Canopy cover (%)
Non-forested (NF)	2,012	11.00 (9.68–12.91)	252 (61–633)	62 (19–92)	19.9 (6.2–29.2)	5.3 (3.4–6.7)	6.2 (4.6–7.2)	24.7 (4.6–49.7)	78.3 (30–100)
Forested (F)	2,012	10.94 (9.69–12.81)	262 (71–639)	281 (254–312)	89.4 (80.8–99.4)	4.4 (2.7–5.9)	5.4 (4–6.8)	22.8 (5.2–42.6)	78.9 (30–100)

• Note: Mean of environmental, landscape and stand variables per plot classification (5th and 95th quantiles in parentheses). Landscape forest cover is estimated within a 1-km radius around the plot.

3 RESULTS

The plots in forested landscapes (F) had an average CTI 0.26 °C lower than the plots in non-forested landscapes (NF) (p < .001). The difference in CTI between F and NF plots was highly variable (0.81 °C SD). In 63% of the plot pairs, F plots had a lower CTI (the difference was negative), and for 17% of the pairs this difference was more than −1 °C (Figure 2). Conversely, for 37% of the plot pairs, the F plots had a higher CTI, and the difference was more than +1 °C for 6% of the pairs. Differences in CTI ranged from −3.5 to +3.0 °C (Figure 2).

We found only minor effects, that is significant but negligible effects (effect size < 0.03 °C), of ΔElevation, ΔMAT, ΔEllenberg L and ΔYear on pairwise differences in CTI (Table 2). However, the effects of ΔBioindicated pH and ΔEllenberg N on ΔCTI were greater: −0.24 and 0.11 °C, respectively (Table 2). This greater effect size was caused by the large difference between the F and NF plots for the two soil parameters (Table 1). In our subset of species, thermal optimum was positively correlated, with low statistical significance, with the pH indicator value (coefficient: .16, p < .10, Supporting Information Figure S1 and Table S3). Thermal optima and N Ellenberg values were not correlated (p = .40, Supporting Information Table S3).

We found a significant effect of local canopy cover on ΔCTI. This effect however contributed marginally (effect size = 0.001 °C) to lower CTI in F plots because F and NF plots had on average the same local canopy cover (Table 1 and Supporting Information Table S1). We found no significant effect of distance to forest edge on ΔCTI (p = .42, Supporting Information Table S2). It should be noted that these two results do not imply that canopy cover or distance to forest edge are unrelated to CTI. Rather, they show that the balance of these variables created by the pairwise selection successfully reduced their effect size to be negligible.

We fitted our linear model with only pairwise differences as a predictor. As a result, setting the differences to 0 meant that the model compared the F and NF categories with all other factors considered to be equal. Thus, the significant intercept we found indicated a decrease in CTI in F plots not explained by any other factors than the difference in landscape classification (NF versus F, Table 2). The lower CTI in the F plots was robust to different F–NF classification thresholds (e.g., highly forested landscapes, Figure 3) but lower CTIs were especially apparent in landscapes with > 70% total forest cover. Effect sizes of the other factors playing a significant role were also robust to changes in NF–F classification thresholds. The decrease in ΔCTI and the contribution of the other factors to it were linear; we did not detect a saturation effect of increasing landscape-scale cover effect.

4 DISCUSSION

Our results indicate that the CTI of vascular plants in forest understories is 0.26 °C lower in forested landscapes than in unforested landscapes. This result can be compared with the recent community thermophilization rates (increasing CTI over time) of c. 0.1 °C per decade found in temperate forest plant communities (Dietz et al., 2020; Martin et al., 2019; Richard et al., 2021), and with recent air temperature warming rates of c. 0.26 °C per decade (2002 to 2018) in our study area (Dietz et al., 2020). We found a significant spatial pattern for the CTI of forest understorey plants equivalent to three decades of thermophilization, or one decade of macroclimatic warming. By controlling for canopy cover and distance to forest edge, we show that the lower CTI in highly forested landscapes is explained by differences in soil conditions (pH and nutrient content) that favour cold-adapted species. Finally, the structure of our sampling allowed us to observe a landscape-scale forest cover cooling on CTI all other factors being equal.

In our study, bioindicated soil characteristics (pH and nutrient content) drove 50% of the difference in CTI between F and NF plots (0.13 °C). In landscapes with less than 30% forest cover, the understorey plant communities in our plots were characterized by species requiring nutrient-rich soils and high pH, and by nitrophilous species. These species had higher thermal optima, thus creating a significant (albeit weak) correlation between soil and climate preference (Supporting Information Figure S1 and Table S3). On average, soil pH and mean N Ellenberg values were respectively 0.8 and 0.9 lower in F plots (Table 1). These differences may have affected the CTI, as an increase of 1 in a pH indicator value increases the thermal optimum of a species by 0.16 °C (Supporting Information Table S3). Correlations of thermal optima and soil preferences have been documented before (Ewald, 2003); they may result from adaptation of to poor soils found in cold forests (Osman, 2013).

Landscape forest cover was therefore linked to local soil conditions we observed in the F and NF classes (Table 1). In addition, pH and nitrogen are both very sensitive to past agricultural practices. Small forests are more likely to be younger forests growing on past agricultural lands in our study region, as shown by historical maps made in circa 1750 (Vallauri et al., 2012). The smaller (and generally younger) forests display high nutrient loads because they are located on former agricultural lands, and conversely, the larger forests historically grow on less fertile soils (Bergès et al., 2016). Furthermore, smaller forest patches and forest edges are more affected by horizontal fertilization from nearby fields, and by atmospheric N deposition (Bergès et al., 2016). Agricultural mosaics make up most European forests' immediate surroundings (Institute for Environment and Sustainability et al., 2013). Our findings may therefore be applicable to most temperate European forests, where the species pool and agriculture mosaics are similar.

The remaining effect of 0.13 °C at the intercept of the model represents an effect that none of the other included variables explained, and which can be attributed to the difference in landscape forest cover. The lower CTI in the forested landscapes implies that the species within the community originate from cooler biogeographic region. Such species can be present due to legacy effects; cold-adapted species are associated with old-growth forests (Bodin et al., 2013; Dupouey et al., 2002; Ewald, 2003). The remnants of old-growth forests in France tend to be large forests that fall into our definition of forested landscapes (Vallauri et al., 2012). Cold-adapted forest species also have limited dispersal capacity (Dupouey et al., 2002). The larger forests in the landscapes with high forest cover could favour species with low dispersal capacity since connectivity is increased in large, closely knit forest habitats (Saura et al., 2014). We cannot exclude the possibility that our findings may in part be a result of the temporal dynamic of CTI, that is, thermophilization, as our study is set in a fixed time span and could be a snapshot that masks different thermophilization rates (Richard et al., 2021).

In our analyses, we controlled for microclimatic effects. We carried out balanced plot pairing and added two well-documented determinants of forest microclimate – immediate canopy cover and distance to forest edge – to our sub-models (Chen et al., 1999; De Frenne et al., 2019; Meeussen et al., 2021). The effect of the landscape-scale forest cover remained significant with these controls. This indicates that landscape-scale forest cover may have a cooling effect on regional temperatures, with a subsequent effect on the understorey. Forests have higher evapotranspiration than other land cover types such as cropland and grasslands, thereby lowering the air temperature and promoting cloud formation. The forest cover can cool the regional climate during the growing season (Bonan, 2008; Hesslerová et al., 2013; Pokorny et al., 2010), and this service is critical for the survival of understorey plants, especially cold-adapted species, during the hot summer months. Indeed, the conditions in the understorey depend not only on local stand structure, but also on the regional climate (De Frenne et al., 2021).

Our results show a large variability in CTI differences between the F and NF plots (Figure 1). Indeed, 37% of the F plots had a higher CTI than the NF plots. This is most likely due to sampling conditions. CTI is sensitive to changes in community composition when the number of occurrences is low. Furthermore, though we selected geographically close plots with contrasting landscape-scale forest cover to balance the environmental and stand variables of the two large classes F and NF for the whole dataset (Table 1), some individual pairs no doubt differed in stand conditions (e.g., canopy cover). Such conditions influence the microclimate (De Frenne et al., 2013) and contributed to the large variability in differences we found in the full sample. Finally, we do not exclude the possible impact of other landscape elements such as water bodies, urbanized areas, complex edge structures or topographic variations that could significantly influence climate and community dynamics (Meeussen et al., 2021).

Our study relied on plant species indicator values but their use is controversial; indicators can be poorly correlated with actual measurements as they are sometimes derived from expert knowledge (Marrec et al., 2022; Szymura et al., 2014). We strengthened our analysis by including the pH indicator value from Gégout et al. (2005), which was calibrated with soil measurements and a floristic survey database. We maintain that it was useful to include these indicator variables in our study as a proxy for excess fertilization from nearby previous and current agricultural activities. Similarly, we combined Ellenberg L (light) values, when available, and canopy cover data (in one of the sub-models) to better account for canopy density and light conditions.

We used the mean of the recorded understorey thermal optima as our response variable. The thermal optima in ClimPlant were computed from distribution maps and grids of macroclimatic temperature (Vangansbeke et al., 2021). These indices are valuable to infer the biogeographic origin of a species. We interpreted cooler CTI within a highly forested landscapes as a regional cooling favouring cold-adapted species. However, the potential of CTI to infer direct microclimatic temperature is limited (Marrec et al., 2022); future research including microclimatic measurements or analysis of readily available weather station data will be critical to further elucidating the climatic versus soil effects contributing to the persistence of cold-adapted species observed in our study.

Current climate warming is likely to have particularly harmful effects on cold-adapted species, which may escape regional warming by retracting their range to climate refugia with locally cooler and more suitable microclimates (Corlett & Westcott, 2013; Hylander et al., 2022; Kuhn & Gégout, 2019). For forest species, these climatic refugia can be topographic features (e.g., cold air pooling inside valleys; Stark & Fridley, 2022), dense forests (Frey et al., 2016) or even hedgerows in open landscapes (Vanneste et al., 2020). Our study highlights that highly forested landscapes also promote the presence of cold-adapted species, a type of refugium that is expected to last as the buffering capacity of forest will stay constant or increase with climate change (De Lombaerde et al., 2021). Our results however should not be used to undermine the importance of small forest patches in agriculture mosaics (Valdés et al., 2020). Forests in landscapes with limited forest cover harbour on average more warm-adapted species, which are more suited to the warming climate and are more resilient in the face of disturbances that could rapidly remove the forest buffering capacity (Christiansen et al., 2021; Hylander et al., 2022). In addition, our results (Figure 3) show a linear decrease, without saturation, in CTI with increasing landscape-scale forest cover; as a result, any amount of forest cover in the landscape can have an effect on CTI. This implies that landscape-scale forest cover diversity also matters for understorey plant diversity by providing a diverse set of soil and thermal conditions for a variety of species. Acknowledging such heterogeneity and the potential of large forests to act as refugia is one of the keys to successful forest biodiversity conservation at the landscape scale (Hylander et al., 2022). We acknowledge that part of the difference in CTI could be driven by the colonization of warm-adapted generalist species in edges and low forested landscapes. We chose to emphasize the presence of cold-adapted species in ‘forested’ plots as they are the most threatened by climate change, and are most discussed in recent conservation literature (Hylander et al., 2022). In addition, our regional cooling interpretation complements the current literature on the potential protection forest microclimate offers in the warm edge of the distribution of cold-adapted species (De Frenne et al., 2021; Sanczuk et al., 2022).

Current land-use changes will likely drive changes in forest cover and forest distribution (Doelman et al., 2018; Ellis, 2021). We demonstrate that the mean thermal optimum of an understorey plant community is sensitive to the amount of forest around it. Large forest masses harbour on average more cold-adapted species, and landscapes with forest patches of contrasting sizes may provide a suite of opportunities for different species. As the climate continues to warm, guaranteeing the availability of both forested and diverse landscapes will be key to ensuring biodiversity protection and ecosystem adaptation and resilience in the near and distant future.

CONFLICT OF INTEREST STATEMENT

The authors declare that they have no competing financial interests or personal relationships that could influence the work reported in this paper.

BIOSKETCHES

Jeremy Borderieux investigates the response of forest plant communities to macroclimate, microclimate and fragmentation through space and time. He uses data science techniques to synthesize forest inventory data, compilations of historical and recent plant databases, together with meteorological data, land cover maps, and in-situ measurements of microclimate (Twitter: @JeremBorderieux).

Josep M. Serra-Diaz investigates the response of forest diversity and ecosystem dynamics to global change (Twitter: @JMSerraDiaz).

Jean-Claude Gégout compiles historical and recent floristic databases to study long-term vegetation changes and test several methods of soil properties bio indication via floristic surveys.