Maintaining Forest Cover to Enhance Temperature Buffering under Future Climate Change

Depósito de investigación de la Universidad de Sevilla https://idus.us.es/ “This is an Accepted Manuscript of an article published by Elsevier in SCIENCE OF THE TOTAL ENVIRONMENT on 1 March 2022, available at: https://doi.org/10.1016/j.scitotenv.2021.151338” 1 Maintaining forest cover to enhance temperature buffering under future climate 1 change 2 Abstract 3 Forest canopies buffer macroclimatic temperature fluctuations. However, we do not know if and how the 4 capacity of canopies to buffer understorey temperature will change with accelerating climate change. Here 5 we map the difference (offset) between temperatures inside and outside forests in the recent past and 6 project these into the future in boreal, temperate and tropical forests. Using linear mixed-effect models, we 7 combined a global database of 714 paired time series of temperatures (mean, minimum and maximum) 8 measured inside forests vs. in nearby open habitats with maps of macroclimate, topography and forest cover 9 to hindcast past (1970-2000) and to project future (2060-2080) temperature differences between free-air 10 temperatures and sub-canopy microclimates. For all tested future climate scenarios, we project that the 11 difference between maximum temperatures inside and outside forests across the globe will increase (i.e. 12 result in stronger cooling in forests), on average during 2060-2080, by 0.27 ± 0.16 °C (RCP2.6) and 0.60 ± 0.14 13 °C (RCP8.5) due to macroclimate changes. This suggests that extremely hot temperatures under forest 14 canopies will, on average, warm less than outside forests as macroclimate warms. This knowledge is of 15 utmost importance as it suggests that forest microclimates will warm at a slower rate than non-forested 16 areas, assuming that forest cover is maintained. Species adapted to colder growing conditions may thus find 17 shelter and survive longer than anticipated at a given forest site. This highlights the potential role of forests 18 as a whole as microrefugia for biodiversity under future climate change. 19 Keywords: forest microclimate, temperature offsets, canopy, climate change, future 20 climate projections, paired sensor data 21 22 2 Introduction 23 Warming temperatures and changing precipitation regimes are influencing ecosystems across the globe 24 (IPCC, 2018). To date, ecological research assessing the impact of anthropogenic climate change has 25 predominantly relied on macroclimatic data. These data are typically based on a global network of weather 26 stations established at approximately 1.5 to 2.0 m above the soil surface in open habitats (e.g. above short 27 grass) (World Meteorological Organization, 2018). Forest organisms living below and within tree canopies, 28 however, experience microclimatic conditions distinct from those in open habitats (Chen et al., 1999; De 29 Frenne et al., 2021; Geiger et al., 2009). Below tree canopies, lower radiation, wind and evapotranspiration 30 rates often translate into lower temporal variation in air temperature and humidity compared to open 31 environments (Davis et al., 2019; Geiger et al., 2009; Von Arx et al., 2013). In particular, temperature 32 extremes are often strongly attenuated in forest interiors, with lower maxima and higher minima compared 33 to open environments (De Frenne et al., 2019; Li et al., 2015). Studies have already shown that such 34 microclimatic buffering can mediate the response of forest communities to climate change (De Frenne et al., 35 2013; Dietz et al., 2020; Lenoir et al., 2017; Stevens et al., 2015; Zellweger et al., 2020). Despite the increasing 36 evidence that ecosystem dynamics and processes are more likely to be related to forest microclimates than 37 to macroclimate (Chen et al., 2018; De Frenne et al., 2021; De Smedt et al., 2021; Frey et al., 2016a), 38 microclimates are still seldom incorporated in ecological research (e.g. in species distribution models) 39 (Lembrechts et al., 2019) and ignored by dynamic global vegetation models (DGVMs; e.g. Thrippleton, 40 Bugmann, Kramer-Priewasser, & Snell, 2016) that simulate the effects of future climate change on natural 41 vegetation and its carbon and water cycles. In particular, we do not know how forest microclimates will 42 change in the future as macroclimate changes (Lembrechts and Nijs, 2020). 43 Advances in studies on the effects of climate change on different organisms living below or in forest canopies 44 have often been limited by the availability of suitable microclimatic data (De Frenne et al., 2021). One robust 45 way to study forest microclimates is to use microclimate measurements from paired (inside vs. outside 46 forests) sensor networks to calculate temperature offsets, i.e. the absolute and instantaneous difference 47 between temperature inside (i.e., microclimate) and free-air temperatures outside forests (i.e., 48 3 macroclimate) (sensu De Frenne et al., 2021). Negative offset values thus reflect cooler and positive offsets 49 warmer forest temperatures compared to outside forests. These empirical offset values for temperature can 50 be related to readily available environmental data using statistical modelling approaches, and these models 51 can then be used to interpolate and extrapolate microclimate across entire mapped landscapes (Frey et al., 52 2016b; Greiser et al., 2018). Differences between macro‐ and microclimate (i.e., temperature offsets) result 53 from processes operating at many scales that influence incoming solar radiation, air mixing, soil properties 54 or evapotranspiration (reviewed in De Frenne et al., 2021). Macroclimatic conditions (e.g., mean temperature 55 and rainfall), topographic variation in the landscape (e.g., elevation and aspect) and variation in canopy cover 56 and vegetation height have been reported to be the main drivers of the understorey temperatures in forests 57 (De Frenne et al., 2021, 2019; Greiser et al., 2018; Macek et al., 2019; Zellweger et al., 2019). With the advent 58 of global forest microclimate data (De Frenne et al., 2019; Zellweger et al., 2020), this type of modelling now 59 enables the prediction of forest microclimates across forest types under future climate change. 60 Here we map forest microclimate temperature offsets based on (i) paired sensor measurements below the 61 canopy vs. the open-air temperature at a given site and (ii) landscape- and canopy-scale predictors 62 throughout the year for the Earth’s dominant forested ecosystems across five continents and at a spatial 63 resolution of ~1 km. More specifically, our objectives were to (1) make predictions for mean, minimum and 64 maximum temperatures using past macroclimatic data (1970-2000), and, (2) make projections for 65 temperature offsets for the future (2060-2080) macroclimatic conditions. We hypothesised that the 66 buffering capacity of forest canopies results in slower future warming of forest below-canopy temperatures 67 compared to the warming observed in standard meteorological weather stations (macroclimate). 68 69 4 Material & Methods 70 Paired plot data 71 We used a unique data set with 714 temperature offset data points involving paired plots from 74 studies 72 spread across 5 continents (Supplementary Material Fig. S1; Data available in De Frenne et al., 2019). Focus 73 was on air temperature below tree canopies (~72% of observations) and the temperature of the topsoil 74 (~28%), given their importance for responses of forest organisms and ecosystem functioning to macroclimate 75 warming. A key asset of this database is the paired nature of the data, which always combines below-canopy 76 temperature data at a given forest site with open-air temperature data from a neighbouring reference non- 77 forest site. Temperature measurement were performed by various logger types such as HOBO loggers (~15% 78 of observations), iButton loggers (~10%), full weather stations (~5%) and various other logger types (e.g. 79 cylindrical thermistor, Hanna thermohygrometer, thermocouples, etc.; ~70%). Reference sites were a nearby 80 open site equipped with the same type of (shielded) temperature loggers (~82% of observations), a nearby 81 weather station (~14%) (provided the distance did not conflict with the temperature offset of the canopy, 82 e.g., due to significant topographic differences) or a logger placed above the upper canopy surface (~4%). We 83 specifically refrained from using additional data on forest microclimate conditions that were not strictly 84 paired with free-air conditions from a neighbouring site using the exact same design (same sensor, same 85 logger, same shielding material, same height). 86 The data points were collated from the scientific literature in a systematic and reproducible manner (see De 87 Frenne et al., 2019 for full details). Temperature offsets were calculated as the temperature inside the forest 88 minus the temperature outside the forest, or extracted directly from the original study; negative values 89 reflect cooler temperatures below tree canopies while positive values reflect warmer understorey 90 temperatures. This was done for three temperature response variables, i.e. mean, maximum, and minimum 91 temperature (further referred to as Tmean, Tmin and Tmax, respectively) that were computed during a specific 92 time period that could differ between sites but that was exactly the same between paired sensors installed 93 outside and inside the forest at a given site. Multiple forest sites (at least several kilometres apart), seasons 94 (meteorological seasons, later aggregated to growing versus non-growing season) and temperature metrics 95 5 (maximum, mean, minimum, air or soil temperatures) originating from the same study were entered into 96 different rows of the database but tagged under the same study ID. Temperature values of long time series 97 were always aggregated per season and/or year, which means that several temperature values for Tmean, Tmin 98 or Tmax could be generated for the same study site. Temperature measurements were classified as having 99 taken place during the growing season, the non-growing season or throughout the whole year. This 100 classification was performed on the basis of reported meteorological seasons and/or climate information in 101 the original study. The dry and winter season were classified as the non-growing season in tropical and 102 temperate biomes, respectively. Estimates of uncertainty (standard error, standard deviation, coefficient of 103 variation or confidence intervals) of the temperature measurements were only reported for a small minority 104 (13.6%) of offset values in the database and were thus not included in our analyses. See De Frenne et al. 105 (2019) for more details on the literature search, inclusion criteria and the empirical data used in this study. 106 Predictor variables 107 To predict the offsets for the three temperature variables (Tmean, Tmax, Tmin) across all forests at a global extent, 108 we gathered global maps of predictor variables related to macroclimate, topography and forest cover. These 109 three sets of predictor variables were selected based on their importance for forest microclimate, and on the 110 spatial resolution and extent of the available data. All the predictor maps we used are raster maps with a 111 spatial resolution of 30 arcsec (~1 km) and are available at the global extent (i.e., from 80°N to 56°S in latitude 112 and from 180°E to 180°W in longitude). Values for all predictor variables were extracted using the 113 geographical coordinates for each plot pair. 114 Macroclimate. Global raster maps of mean, minimum and maximum free-air temperature (°C; Tmacro), 115 on a monthly basis, as well as monthly precipitation (mm) raster maps, averaged for the climatology 116 1970-2000, were collected from WorldClim version 2.1 (Fick and Hijmans, 2017). In addition, we 117 gathered future projections (2060-2080) for the exact same set of temperature and precipitation 118 variables described in the previous sentence but based on the contrasting “very stringent” 119 representative concentration pathway (RCP) 2.6 and “worst case” RCP 8.5 from three different 120 general circulation models (GCMs) with minimal interdependency, based on Sanderson et al. (2015), 121 6 i.e. HadGEM2-ES, MPI-ESM-LR and MIROC5 (downscaled CMIP5 data from WorldClim; 30 arcsec 122 resolution). 123 Topographic variables and distance to the coast. We gathered six variables related to topography 124 using raster layers derived from the Global Multi-resolution Terrain Elevation Data 2010 125 (GMTED2010) dataset at 30 arcsec resolution (Amatulli et al., 2018). Maps on northness and 126 eastness, elevation (m a.s.l.), elevational variation (EleVar) and topographic position index (TPI) were 127 collected. Northness and eastness are the sine of the slope, multiplied by the cosine and sine of the 128 aspect, respectively. They provide continuous measures describing the orientation in combination 129 with the slope (i.e., a circular variable is transformed into a continuous one, ranging from -1 to 1). In 130 the Northern Hemisphere, a northness value close to 1 corresponds to a northern exposition on a 131 vertical slope (i.e., a slope exposed to very low amount of solar radiation), while a value close to -1 132 corresponds to a very steep southern slope, exposed to a high amount of solar radiation. Aspect 133 values for the Southern Hemisphere were inverted so that a value of 1 in the Southern Hemisphere 134 also means very low amount of solar radiation. Variables EleVar (1) and TPI (2) capture topographic 135 heterogeneity within a 1 km² grid cell around each pair of measurements (inside and outside forest): 136 (1) the standard deviation of elevational values aggregated per 1 km² grid cell (further referred to as 137 elevational variation) and (2) the median of the topographic position index (TPI) values across each 138 1 km² grid cell. The TPI is the difference between the elevation of a focal cell and the mean elevation 139 of its eight surrounding cells. Positive and negative values correspond to ridges and valleys, 140 respectively, while zero values correspond to flat areas (Amatulli et al., 2018). We also produced a 141 map with the distance from each land pixel to the nearest coastline (Dist2Coast) using the coastline 142 map data from Natural Earth (free vector data from naturalearthdata.com). 143 Forest cover and forest height. We used the tree canopy cover (defined as canopy closure for all 144 vegetation taller than 5 m in height) map for the year 2000 by Hansen et al. (2013). This high- 145 resolution global map layer was re-projected and aggregated from 30 m to 30 arcsec using the 146 average of the aggregated raster cells. This canopy cover map is the only available map spanning a 147 global extent at this high resolution. By using this data product, we make the strong assumption that 148 7 canopy cover at the time of temperature measurements is similar to the cover in the year 2000. We 149 consider this assumption as reasonable as the median year of the temperature measurements for all 150 data points is approximately 1996 (range between 1943 and 2014). Finally, we used estimates of 151 canopy height at 1 km resolution derived from the ICESat satellite mission based on 2005 (Simard et 152 al., 2011). 153 Data analysis 154 All statistical analyses were performed in the open-source statistical software environment of R, version 4.0.2 155 (R Core Team, 2021). The temperature offsets for Tmean, Tmax and Tmin were modelled (274, 184 and 202 plot 156 pairs respectively), after removing missing values for sensor height, i.e. not mentioned in the original study, 157 and data points with canopy cover zero (based on the tree canopy cover map introduced above; Hansen et 158 al., 2013) using linear mixed-effect models with random intercept (LMMs) (lme4 package; Bates et al., 2015). 159 In our main models, we combined the seasonal (growing vs. non-growing and annual) time series and 160 performed additional analyses for the different three different time periods (see further and Supplementary 161 Material Appendix S2). We included ‘study ID’ as a random intercept term to account for non-independence 162 between samples within studies. For each of the three studied response variables, we started our modelling 163 protocol from the full model: 164 Toffset ~ Tmacro + Precipitation + Elevation + Eastness + Northness + EleVar + TPI + Dist2Coast + Canopy cover + 165 Forest height + Sensor height + random effect ‘study ID’ 166 For Tmacro, we used the monthly average for either Tmean, Tmax and Tmin temperature during the period 1970- 167 2000 depending on the studied response variable of T offset (Tmean, Tmax or Tmin). Sensor height was also 168 included in the models (continuous variable, in metres above or below the soil surface), as this significantly 169 impacts the magnitude of the temperature offset (De Frenne et al., 2019; Supplementary Fig. S2; Table S1). 170 Sensor height is positive for aboveground and negative for belowground sensors. Data points with sensor 171 height > 2 m were excluded as our aim was to model forest microclimate near the ground. To avoid 172 collinearity in predictor variables and improve model performance, we excluded variables that showed a 173 correlation r ≥ |0.7| (Pearson's product-moment correlation; Supplementary Fig. S3) and variance inflation 174 8 factor ≥ 4 (Zuur et al., 2010). Forest height was therefore removed from all models due to high correlation 175 with canopy cover; for Tmean offset, EleVar was also dropped from the model due to high correlation with TPI. 176 All predictors were standardized by subtracting the mean and dividing by the standard deviation prior to 177 modelling. For each response variable, the single best model was selected based on the Akaike Information 178 Criterion (AIC) using the automated dredge-function of the package MuMIn (Barton, 2009). Goodness of fit 179 was calculated following Nakagawa and Schielzeth (2013). 180 To test for non-linear relationships, we also used generalized additive mixed-effect models (GAMMs) (cf. the 181 gamm4 package) (Wood and Scheipl, 2014) on the same dataset. We applied smoothers to the same set of 182 fixed-effect terms, included the same random intercept term ‘study ID’ and followed the same model 183 selection procedure as for the LMMs. For each of the three studied response variables (Tmean, Tmax, Tmin) and 184 for each of the two modelling approaches, we performed a leave-one-out cross validation (LOOcv) and 185 compared root mean square errors (RMSE) among models (LMMs vs. GAMMs). We found no difference (t 186 test, p-value > 0.05) in RMSE between LMMs and GAMMs, justifying our choice of LMMs (see also 187 Supplementary Fig. S4). Furthermore, we checked spatial autocorrelation in the model residuals for the 188 LMMs using Moran s I-test from the ape package (Paradis and Schliep, 2019). No spatial autocorrelation was 189 detected (p-value > 0.05) in the model residuals. Additionally, we tested the effect of season of sampling 190 (annual, growing and non-growing season; see above) on each response variable. We included season as a 191 categorical variable to the full models described above and followed the same model selection procedure. 192 However, due to the low number of observations for each category (but growing season being the dominant 193 category), results including season were only included in the Supplementary Material Appendix S2. 194 Using the single best LMMs for each of our three response variables, we made predictions for Tmean, Tmax, and 195 Tmin offsets for forest across the globe using the collected map data for all predictor variables retained in the 196 models, setting sensor height to 1.0 m and not considering variation included in the random intercept. 197 Temperature offsets were predicted for all raster pixels (30 arcsec resolution) with canopy cover >50% as this 198 largely concurs with the global distribution of forest areas in the terrestrial ecoregions map by Olson et al. 199 (2001). To assess model performance, we performed spatially blocked k-fold cross-validation (k = 10; folds 200 assigned randomly, with spatial blocks of size 50 km²; Valavi et al., 2019). Furthermore, we made predictions 201 15 in predictions increased towards tropical and boreal forests which is likely caused by extrapolation outside 299 the environmental range included in our data. These biomes were underrepresented in the data, hence, 300 future research should focus on setting out networks of paired temperature sensors in these regions 301 (Lembrechts et al., 2021b). 302 Our projections for both the “very stringent” RCP2.6 as well as the “worst-case” RCP8.5 scenario indicate 303 that buffering by forest canopies for Tmean and Tmax temperature may increase, but minimum temperature 304 offsets will decrease, especially in temperate and boreal regions as ambient temperatures become less cold. 305 This suggests that under climate change, free-air temperatures are likely to have a larger-magnitude increase 306 than the corresponding forest microclimate temperatures, which would reinforce the idea of divergent 307 warming (decoupling) between macroclimate and microclimate (De Frenne et al., 2019; Lenoir et al., 2017). 308 Offsets may even become lower (resulting in increasing or decreasing buffering for Tmean or Tmin, respectively) 309 despite projected decreases in precipitation in some regions (Supplementary Fig. S8). It is possible that finer- 310 grained microclimatic heterogeneity could buffer the impact of a changing macroclimate even further 311 (Maclean et al., 2017). This inference relies, however, on the strong assumption that forest cover and 312 composition will remain stable in the future. Such stability is however unlikely, as climate change itself as 313 well as forest management and disturbances can either increase or decrease forest canopy cover in the 314 future. For example, climate change is however likely to cause increased tree mortality owing to, for instance, 315 repeated and more severe disturbances such as droughts, fires, pathogens and insect outbreaks (Curtis et 316 al., 2018; Senf et al., 2021; Senf and Seidl, 2020). The resulting reduction in tree canopy cover can lead to a 317 sudden loss (i.e. a tipping point) of canopy buffering and increased microclimate warming (Alkama and 318 Cescatti, 2016; Findell et al., 2017; Lembrechts and Nijs, 2020; Richard et al., 2021; Zellweger et al., 2020). 319 On the other hand, strong efforts are being made worldwide to increase forest cover and implement climate- 320 smart forestry practices (Bastin et al., 2019; Di Sacco et al., 2021). How these forest cover changes will affect 321 future forest temperature buffering should be a topic for future forest microclimate research. 322 We projected temperature buffering capacities of forests across the globe under future climate change 323 scenarios. Assuming no change in forest composition, we predicted that forest buffering of Tmean and Tmax will 324 increase in the future (2060-2080), whereas buffering of Tmin will be reduced due to changes in macroclimate 325 16 conditions. Our results indicate that the refugial capacity of cool and dense forest might last longer than 326 anticipated in a warming climate. This knowledge has important implications for forest biodiversity 327 conservation. Forest managers and policymakers could, for example, aim to optimise forest functioning and 328 biodiversity goals by identifying areas in which reducing or retaining canopy cover may have larger impacts 329 on the prevailing microclimate than anticipated under future climate change (Wolf et al., 2021). The paired 330 nature of the data allowed us to model absolute temperature offsets across a global extent with fair accuracy. 331 Gridded microclimate products such as ours, especially when paired with new, well-designed networks of 332 microclimate measurements (Lembrechts et al., 2020) serve ecological and environmental modelers with a 333 more scale-relevant set of products for making predictions and drawing inference. At the regional and even 334 continental scale, novel high-resolution data on forest structure and composition based on remote sensing 335 imagery (e.g. GEDI LiDAR data) are becoming available (De Frenne et al., 2021; Lembrechts et al., 2019; 336 Randin et al., 2020; Zellweger et al., 2018). Including these microclimate measurements and novel spatial 337 map data (e.g. Haesen et al., 2021; Lembrechts et al., 2020) in future models and mapping efforts will increase 338 accuracy of future predictions (Lembrechts et al., 2021a). Our study illustrates that forest microclimates 339 themselves are subject to climate change, which will have important consequences for forest-dwelling 340 species and must hence not be neglected. 341 Data availability: 342 The dataset analysed in the current study is available in the Figshare repository, with the identifier 343 10.6084/m9.figshare.7604849 (de Frenne et al., 2019). 344 Acknowledgements: 345 EDL, PV, PDF, LD, PS, CM and TV received funding from the European Research Council (ERC) under the 346 European Union’s Horizon 2020 research and innovation programme (ERC Starting Grant FORMICA 757833). 347 FZ received funding from the Swiss National Science Foundation (grant number 193645). SH received funding 348 from a FLOF fellowship of the KU Leuven (project nr. 3E190655). Funding for DHK was provided by the 349 National Science Foundation Graduate Research Fellowship Program (DGE-1842473). JA acknowledges 350 Academy of Finland Flagship funding (grant no. 337552). KH received funding from the Swedish Research 351 Council Formas (grants 2014-530 and 2018-2829) and the Bolin Centre for Climate Research, Stockholm 352 University. SG was supported by the Research Foundation Flanders (FWO) (project G0H1517N). JJL received 353 funding from the Research Foundation Flanders (FWO) (project 12P1819N). JL received funding from the 354 Agence Nationale de la Recherche (ANR) within the framework of the IMPRINT project “IMpacts des 355 17 PRocessus mIcroclimatiques sur la redistributioN de la biodiversiTé forestière en contexte de réchauffement 356 du macroclimat” (grant number: ANR-19-CE32-0005-01). KDP received funding from the Research 357 Foundation Flanders (FWO) (project ASP035-19). FRS was supported by the VI Plan Propio de Investigación 358 of Universidad de Sevilla (VI PPITUS). 359 360 361 18 References 362 Alkama, R., Cescatti, A., 2016. Climate change: Biophysical climate impacts of recent changes in global 363 forest cover. Science (80-. ). 351, 600–604. https://doi.org/10.1126/science.aac8083 364 Amatulli, G., Domisch, S., Tuanmu, M.N., Parmentier, B., Ranipeta, A., Malczyk, J., Jetz, W., 2018. Data 365 Descriptor: A suite of global, cross-scale topographic variables for environmental and biodiversity 366 modeling. Sci. Data 5, 1–15. https://doi.org/10.1038/sdata.2018.40 367 Ashcroft, M.B., Gollan, J.R., 2012. Fine-resolution (25 m) topoclimatic grids of near-surface (5 cm) extreme 368 temperatures and humidities across various habitats in a large (200 × 300 km) and diverse region. Int. 369 J. Climatol. 32, 2134–2148. https://doi.org/10.1002/joc.2428 370 Barton, K., 2009. MuMIn: Multi-Model Inference. 371 Bastin, J., Finegold, Y., Garcia, C., Mollicone, D., Rezende, M., Routh, D., Zohner, C., Crowther, T., 2019. The 372 global tree restoration potential. Science (80-. ). 365, 76–79. https://doi.org/10.1126/science.aay8060 373 Bates, D., Maechler, M., Bolker, B.M., Walker, S., 2015. Fitting Linear Mixed-Effects Models Using lme4. J. 374 Stat. Softw. 67, 1–48. https://doi.org/10.18637/jss.v067.i01 375 Chen, J., Saunders, S.C., Crow, T.R., Naiman, R.J., Brosofske, K.D., Mroz, G.D., Brookshire, B.L., Franklin, J.F., 376 1999. Microclimate in forest ecosystem and landscape ecology: Variations in local climate can be used 377 to monitor and compare the effects of different management regimes. Bioscience 49, 288–297. 378 https://doi.org/10.2307/1313612 379 Chen, Y., Liu, Y., Zhang, J., Yang, W., He, R., Deng, C., 2018. Microclimate exerts greater control over litter 380 decomposition and enzyme activity than litter quality in an alpine forest-tundra ecotone. Sci. Rep. 8, 381 1–13. https://doi.org/10.1038/s41598-018-33186-4 382 Curtis, P.G., Slay, C.M., Harris, N.L., Tyukavina, A., Hansen, M.C., 2018. Classifying drivers of global forest 383 loss. Science (80-. ). 361, 1108–1111. https://doi.org/10.1126/science.aau3445 384 Davis, K.T., Dobrowski, S.Z., Holden, Z.A., Higuera, P.E., Abatzoglou, J.T., 2019. Microclimatic buffering in 385 forests of the future: the role of local water balance. Ecography (Cop.). 42, 1–11. 386 https://doi.org/10.1111/ecog.03836 387 De Frenne, P., Lenoir, J., Luoto, M., Scheffers, B.R., Zellweger, F., Aalto, J., Ashcroft, M.B., Christiansen, 388 D.M., Decocq, G., Pauw, K. De, Govaert, S., Greiser, C., Gril, E., Hampe, A., Jucker, T., Klinges, D.H., 389 Koelemeijer, I.A., Lembrechts, J.J., Marrec, R., Meeussen, C., Ogée, J., Tyystjärvi, V., Vangansbeke, P., 390 Hylander, K., 2021. Forest microclimates and climate change : Importance, drivers and future research 391 agenda. Glob. Chang. Biol. 1–19. https://doi.org/10.1111/gcb.15569 392 de Frenne, P., Lenoir, J., Rodriguez-Sanchez, F., 2019. Global buffering of temperatures under forest 393 canopies data and code. Figshare. https://doi.org/10.6084/m9.figshare.7604849tem 394 De Frenne, P., Rodríguez-Sánchez, F., Coomes, D.A., Baeten, L., Verstraeten, G., Vellend, M., Bernhardt- 395 Römermann, M., Brown, C.D., Brunet, J., Cornelis, J., Decocq, G.M., Dierschke, H., Eriksson, O., Gilliam, 396 F.S., Hédl, R., Heinken, T., Hermy, M., Hommel, P., Jenkins, M. a, Kelly, D.L., Kirby, K.J., Mitchell, F.J.G., 397 Naaf, T., Newman, M., Peterken, G., Petrík, P., Schultz, J., Sonnier, G., Van Calster, H., Waller, D.M., 398 Walther, G.-R., White, P.S., Woods, K.D., Wulf, M., Graae, B.J., Verheyen, K., 2013. Microclimate 399 moderates plant responses to macroclimate warming. Proc. Natl. Acad. Sci. U. S. A. 110, 18561– 400 18565. https://doi.org/10.1073/pnas.1311190110 401 De Frenne, P., Zellweger, F., Rodríguez-Sánchez, F., Scheffers, B.R., Hylander, K., Luoto, M., Vellend, M., 402 Verheyen, K., Lenoir, J., 2019. Global buffering of temperatures under forest canopies. Nat. Ecol. Evol. 403 3, 744–749. https://doi.org/10.1038/s41559-019-0842-1 404 De Smedt, P., Boeraeve, P., Baeten, L., 2021. Intra-annual activity patterns of terrestrial isopods are 405 19 tempered in forest compared to open habitat. Soil Biol. Biochem. 160, 108342. 406 https://doi.org/10.1016/j.soilbio.2021.108342 407 Di Sacco, A., Hardwick, K.A., Blakesley, D., Brancalion, P.H.S., Breman, E., Cecilio Rebola, L., Chomba, S., 408 Dixon, K., Elliott, S., Ruyonga, G., Shaw, K., Smith, P., Smith, R.J., Antonelli, A., 2021. Ten golden rules 409 for reforestation to optimize carbon sequestration, biodiversity recovery and livelihood benefits. Glob. 410 Chang. Biol. 27, 1328–1348. https://doi.org/10.1111/gcb.15498 411 Dietz, L., Collet, C., Dupouey, J.L., Lacombe, E., Laurent, L., Gégout, J.C., 2020. Windstorm-induced canopy 412 openings accelerate temperate forest adaptation to global warming. Glob. Ecol. Biogeogr. 2067–2077. 413 https://doi.org/10.1111/geb.13177 414 Fick, S.E., Hijmans, R.J., 2017. WorldClim 2: new 1-km spatial resolution climate surfaces for global land 415 areas. Int. J. Climatol. 37, 4302–4315. https://doi.org/10.1002/joc.5086 416 Findell, K.L., Berg, A., Gentine, P., Krasting, J.P., Lintner, B.R., Malyshev, S., Santanello, J.A., Shevliakova, E., 417 2017. The impact of anthropogenic land use and land cover change on regional climate extremes. Nat. 418 Commun. 8, 1–9. https://doi.org/10.1038/s41467-017-01038-w 419 Frey, S.J.K., Hadley, A.S., Betts, M.G., 2016a. Microclimate predicts within-season distribution dynamics of 420 montane forest birds. Divers. Distrib. 22, 944–959. https://doi.org/10.1111/ddi.12456 421 Frey, S.J.K., Hadley, A.S., Johnson, S.L., Schulze, M., Jones, J.A., Betts, M.G., 2016b. Spatial models reveal the 422 microclimatic buffering capacity of old-growth forests. Sci. Adv. 2. 423 https://doi.org/10.1126/sciadv.1501392 424 Geiger, R., Aron, R., Todhunter, P., 2009. The climate near the ground. Rowman & Littlefield. 425 Greiser, C., Meineri, E., Luoto, M., Ehrlén, J., Hylander, K., 2018. Monthly microclimate models in a 426 managed boreal forest landscape. Agric. For. Meteorol. 250–251, 147–158. 427 https://doi.org/10.1016/j.agrformet.2017.12.252 428 Haesen, S., Lembrechts, J.J., De Frenne, P., Lenoir, J., Aalto, J., Ashcroft, M.B., Kopecký, M., Luoto, M., 429 Maclean, I., Nijs, I., Niittynen, P., Hoogen, J., Arriga, N., Brůna, J., Buchmann, N., Čiliak, M., Collalti, A., 430 De Lombaerde, E., Descombes, P., Gharun, M., Goded, I., Govaert, S., Greiser, C., Grelle, A., Gruening, 431 C., Hederová, L., Hylander, K., Kreyling, J., Kruijt, B., Macek, M., Máliš, F., Man, M., Manca, G., Matula, 432 R., Meeussen, C., Merinero, S., Minerbi, S., Montagnani, L., Muffler, L., Ogaya, R., Penuelas, J., Plichta, 433 R., Portillo‐Estrada, M., Schmeddes, J., Shekhar, A., Spicher, F., Ujházyová, M., Vangansbeke, P., 434 Weigel, R., Wild, J., Zellweger, F., Van Meerbeek, K., 2021. ForestTemp – Sub‐canopy microclimate 435 temperatures of European forests. Glob. Chang. Biol. 1–13. https://doi.org/10.1111/gcb.15892 436 Hansen, M.C., Potapov, P. V, Moore, R., Hancher, M., Turubanova, S.A., Tyukavina, A., Thau, D., Stehman, S. 437 V, Goetz, S.J., Loveland, T.R., Kommareddy, A., Egorov, A., Chini, L., Justice, C.O., Townshend, J.R.G., 438 2013. High-Resolution Global Maps of 21st-Century Forest Cover Change. Science (80-. ). 342, 850– 439 854. https://doi.org/10.1126/science.1244693 440 Hijmans, R.J., van Etten, J., 2012. raster: Geographic analysis and modeling with raster data. 441 IPCC, 2018. Global Warming of 1.5 °C - SR15 - Summary for Policy Makers, IPCC Climate Change Synthesis 442 Report. 443 Lembrechts, J.J., Aalto, J., Ashcroft, M.B., De Frenne, P., Kopecký, M., Lenoir, J., Luoto, M., Maclean, I.M.D., 444 Roupsard, O., Fuentes-Lillo, E., García, R.A., Pellissier, L., Pitteloud, C., Alatalo, J.M., Smith, S.W., Björk, 445 R.G., Muffler, L., Ratier Backes, A., Cesarz, S., Gottschall, F., Okello, J., Urban, J., Plichta, R., Svátek, M., 446 Phartyal, S.S., Wipf, S., Eisenhauer, N., Pușcaș, M., Turtureanu, P.D., Varlagin, A., Dimarco, R.D., Jump, 447 A.S., Randall, K., Dorrepaal, E., Larson, K., Walz, J., Vitale, L., Svoboda, M., Finger Higgens, R., 448 Halbritter, A.H., Curasi, S.R., Klupar, I., Koontz, A., Pearse, W.D., Simpson, E., Stemkovski, M., Jessen 449 Graae, B., Vedel Sørensen, M., Høye, T.T., Fernández Calzado, M.R., Lorite, J., Carbognani, M., 450 Tomaselli, M., Forte, T.G.W., Petraglia, A., Haesen, S., Somers, B., Van Meerbeek, K., Björkman, M.P., 451 20 Hylander, K., Merinero, S., Gharun, M., Buchmann, N., Dolezal, J., Matula, R., Thomas, A.D., Bailey, J.J., 452 Ghosn, D., Kazakis, G., de Pablo, M.A., Kemppinen, J., Niittynen, P., Rew, L., Seipel, T., Larson, C., 453 Speed, J.D.M., Ardö, J., Cannone, N., Guglielmin, M., Malfasi, F., Bader, M.Y., Canessa, R., Stanisci, A., 454 Kreyling, J., Schmeddes, J., Teuber, L., Aschero, V., Čiliak, M., Máliš, F., De Smedt, P., Govaert, S., 455 Meeussen, C., Vangansbeke, P., Gigauri, K., Lamprecht, A., Pauli, H., Steinbauer, K., Winkler, M., 456 Ueyama, M., Nuñez, M.A., Ursu, T.M., Haider, S., Wedegärtner, R.E.M., Smiljanic, M., Trouillier, M., 457 Wilmking, M., Altman, J., Brůna, J., Hederová, L., Macek, M., Man, M., Wild, J., Vittoz, P., Pärtel, M., 458 Barančok, P., Kanka, R., Kollár, J., Palaj, A., Barros, A., Mazzolari, A.C., Bauters, M., Boeckx, P., Benito 459 Alonso, J.L., Zong, S., Di Cecco, V., Sitková, Z., Tielbörger, K., van den Brink, L., Weigel, R., Homeier, J., 460 Dahlberg, C.J., Medinets, S., Medinets, V., De Boeck, H.J., Portillo-Estrada, M., Verryckt, L.T., Milbau, 461 A., Daskalova, G.N., Thomas, H.J.D., Myers-Smith, I.H., Blonder, B., Stephan, J.G., Descombes, P., 462 Zellweger, F., Frei, E.R., Heinesch, B., Andrews, C., Dick, J., Siebicke, L., Rocha, A., Senior, R.A., Rixen, 463 C., Jimenez, J.J., Boike, J., Pauchard, A., Scholten, T., Scheffers, B., Klinges, D., Basham, E.W., Zhang, J., 464 Zhang, Z., Géron, C., Fazlioglu, F., Candan, O., Sallo Bravo, J., Hrbacek, F., Laska, K., Cremonese, E., 465 Haase, P., Moyano, F.E., Rossi, C., Nijs, I., 2020. SoilTemp: A global database of near-surface 466 temperature. Glob. Chang. Biol. 26, 6616–6629. https://doi.org/10.1111/gcb.15123 467 Lembrechts, J.J., Hoogen, J. Van Den, Aalto, J., Ashcroft, M.B., De Frenne, P., Kemppinen, J., Kopecký, M., 468 2021a. Mismatches between soil and air temperature. EcoEvoRxiv. 469 https://doi.org/10.32942/osf.io/pksqw 470 Lembrechts, J.J., Lenoir, J., Frenne, P. De, Scheffers, B.R., 2021b. Designing countrywide and regional 471 microclimate networks 1–7. https://doi.org/10.1111/geb.13290 472 Lembrechts, J.J., Nijs, I., 2020. Microclimate shifts in a d ynamic world. Science (80-. ). 368, 711–712. 473 Lembrechts, J.J., Nijs, I., Lenoir, J., 2019. Incorporating microclimate into species distribution models. 474 Ecography (Cop.). 42, 1267–1279. https://doi.org/10.1111/ecog.03947 475 Lenoir, J., Hattab, T., Pierre, G., 2017. Climatic microrefugia under anthropogenic climate change: 476 implications for species redistribution. Ecography (Cop.). 40, 253–266. 477 https://doi.org/10.1111/ecog.02788 478 Li, Y., Zhao, M., Motesharrei, S., Mu, Q., Kalnay, E., Li, S., 2015. Local cooling and warming effects of forests 479 based on satellite observations. Nat. Commun. 6. https://doi.org/10.1038/ncomms7603 480 Macek, M., Kopecký, M., Wild, J., 2019. Maximum air temperature controlled by landscape topography 481 affects plant species composition in temperate forests. Landsc. Ecol. 34, 2541–2556. 482 https://doi.org/10.1007/s10980-019-00903-x 483 Maclean, I.M.D., Suggitt, A.J., Wilson, R.J., Duffy, J.P., Bennie, J.J., 2017. Fine-scale climate change: 484 modelling spatial variation in biologically meaningful rates of warming. Glob. Chang. Biol. 23, 256– 485 268. https://doi.org/10.1111/gcb.13343 486 Nakagawa, S., Schielzeth, H., 2013. A general and simple method for obtaining R2 from generalized linear 487 mixed-effects models. Methods Ecol. Evol. 4, 133–142. https://doi.org/10.1111/j.2041- 488 210x.2012.00261.x 489 Olson, D.M., Dinerstein, E., Wikramanayake, E.D., Burgess, N.D., Powell, G.V.N., Underwood, E.C., Amico, 490 J.A.D., Itoua, I., Strand, H.E., Morrison, J.C., Loucks, J., Allnutt, T.F., Ricketts, T.H., Kura, Y., Lamoreux, 491 J.F., Wesley, W., Hedao, P., Kassem, K.R., 2001. Terrestrial Ecoregions of the World : A New Map of 492 Life on Earth 51, 933–938. 493 Paradis, E., Schliep, K., 2019. Ape 5.0: An environment for modern phylogenetics and evolutionary analyses 494 in R. Bioinformatics 35, 526–528. https://doi.org/10.1093/bioinformatics/bty633 495 R Core Team, 2021. R: A language and environment for statistical computing. 496 Randin, C.F., Ashcroft, M.B., Bolliger, J., Cavender-Bares, J., Coops, N.C., Dullinger, S., Dirnböck, T., Eckert, 497 21 S., Ellis, E., Fernández, N., Giuliani, G., Guisan, A., Jetz, W., Joost, S., Karger, D., Lembrechts, J., Lenoir, 498 J., Luoto, M., Morin, X., Price, B., Rocchini, D., Schaepman, M., Schmid, B., Verburg, P., Wilson, A., 499 Woodcock, P., Yoccoz, N., Payne, D., 2020. Monitoring biodiversity in the Anthropocene using remote 500 sensing in species distribution models. Remote Sens. Environ. 239, 111626. 501 https://doi.org/10.1016/j.rse.2019.111626 502 Richard, B., Dupouey, J.L., Corcket, E., Alard, D., Archaux, F., Aubert, M., Boulanger, V., Gillet, F., Langlois, E., 503 Macé, S., Montpied, P., Beaufils, T., Begeot, C., Behr, P., Boissier, J.M., Camaret, S., Chevalier, R., 504 Decocq, G., Dumas, Y., Eynard-Machet, R., Gégout, J.C., Huet, S., Malécot, V., Margerie, P., Mouly, A., 505 Paul, T., Renaux, B., Ruffaldi, P., Spicher, F., Thirion, E., Ulrich, E., Nicolas, M., Lenoir, J., 2021. The 506 climatic debt is growing in the understorey of temperate forests: Stand characteristics matter. Glob. 507 Ecol. Biogeogr. 30, 1474–1487. https://doi.org/10.1111/geb.13312 508 Sanderson, B.M., Knutti, R., Caldwell, P., 2015. A representative democracy to reduce interdependency in a 509 multimodel ensemble. J. Clim. 28, 5171–5194. https://doi.org/10.1175/JCLI-D-14-00362.1 510 Senf, C., Sebald, J., Seidl, R., 2021. Increasing canopy mortality affects the future demographic structure of 511 Europe’s forests. One Earth 1–7. https://doi.org/10.1016/j.oneear.2021.04.008 512 Senf, C., Seidl, R., 2020. Mapping the forest disturbance regimes of Europe. Nat. Sustain. 513 https://doi.org/10.1038/s41893-020-00609-y 514 Simard, M., Pinto, N., Fisher, J.B., Baccini, A., 2011. Mapping forest canopy height globally with spaceborne 515 lidar. J. Geophys. Res. Biogeosciences 116, 1–12. https://doi.org/10.1029/2011JG001708 516 Stevens, J.T., Safford, H.D., Harrison, S., Latimer, A.M., 2015. Forest disturbance accelerates 517 thermophilization of understory plant communities. J. Ecol. 103, 1253–1263. 518 https://doi.org/10.1111/1365-2745.12426 519 Tennekes, M., 2018. Tmap: Thematic maps in R. J. Stat. Softw. 84. https://doi.org/10.18637/jss.v084.i06 520 Thrippleton, T., Bugmann, H., Kramer-Priewasser, K., Snell, R.S., 2016. Herbaceous Understorey : An 521 Overlooked Player in Forest Landscape Dynamics? Ecosystems 19, 1240–1254. 522 https://doi.org/10.1007/s10021-016-9999-5 523 Valavi, R., Elith, J., Lahoz-Monfort, J.J., Guillera-Arroita, G., 2019. blockCV: An r package for generating 524 spatially or environmentally separated folds for k-fold cross-validation of species distribution models. 525 Methods Ecol. Evol. 10, 225–232. https://doi.org/10.1111/2041-210X.13107 526 Von Arx, G., Graf Pannatier, E., Thimonier, A., Rebetez, M., 2013. Microclimate in forests with varying leaf 527 area index and soil moisture: Potential implications for seedling establishment in a changing climate. J. 528 Ecol. 101, 1201–1213. https://doi.org/10.1111/1365-2745.12121 529 Wickham, H., 2016. ggplot2: elegant graphics for data analysis. Springer, New York, USA. 530 Wolf, C., Bell, D.M., Kim, H., Paul, M., Schulze, M., Betts, M.G., Northwest, P., Service, U.F., Way, S.W.J., Or, 531 C., 2021. Temporal consistency of undercanopy thermal refugia in old-growth forest. Agric. For. 532 Meteorol. 307, 108520. https://doi.org/10.1016/j.agrformet.2021.108520 533 Wood, S., Scheipl, F., 2014. gamm4: Generalized additive mixed models using mgcv and lme4. 534 World Meteorological Organization, 2018. Guide to meteorological instruments and methods of 535 observation, 2018 Editi. ed. WMO. 536 Zellweger, F., Coomes, D., Lenoir, J., Depauw, L., Maes, S.L., Wulf, M., Kirby, K.J., Brunet, J., Kopecký, M., 537 Máliš, F., Schmidt, W., Heinrichs, S., den Ouden, J., Jaroszewicz, B., Buyse, G., Spicher, F., Verheyen, K., 538 De Frenne, P., 2019. Seasonal drivers of understorey temperature buffering in temperate deciduous 539 forests across Europe. Glob. Ecol. Biogeogr. 28, 1774–1786. https://doi.org/10.1111/geb.12991 540 Zellweger, F., De Frenne, P., Lenoir, J., Rocchini, D., Coomes, D., 2018. Advances in microclimate ecology 541 22 arising from remote sensing. Trends Ecol. Evol. xx, 1–15. https://doi.org/10.1016/j.tree.2018.12.012 542 Zellweger, F., De Frenne, P., Lenoir, J., Vangansbeke, P., Verheyen, K., Bernhardt-römermann, M., Baeten, 543 L., Hédl, R., Berki, I., Brunet, J., Van Calster, H., Chudomelov, 2020. Forest microclimate dynamics drive 544 plant responses to warming. Science (80-. ). 368, 772–775. 545 Zuur, A.F., Ieno, E.N., Elphick, C.S., 2010. A protocol for data exploration to avoid common statistical 546 problems 3–14. https://doi.org/10.1111/j.2041-210X.2009.00001.x 547 548