PREPRINT VERSION U.S. Forest Response to Projected Climate-Related Stress: a Tolerance Perspective J EAN L I ÉNARD ∗ , J OHN H ARRISON † AND N IKOLAY S TRIGUL ∗ Department of Mathematics & Statistics, Washington State University Vancouver, and † School of the Environment, Washington State University Vancouver ∗ jean.f.lienard@gmail.com / jean.lienard@wsu.edu / nick.strigul@wsu.edu Although it is widely recognized that climate change will require a major spatial reorganization of forests, our ability to predict exactly how and where forest characteristics and distributions will change has been rather limited. Current efforts to predict future distribution of forested ecosystems as a function of climate include species distribution models (for fine scale predictions) and potential vegetation climate envelope models (for coarse-grained, large scale predictions). Here we develop and apply an intermediate approach wherein we use stand-level tolerances of environmental stressors to understand forest distributions and vulnerabilities to anticipated climate change. In contrast to other existing models, this approach can be applied at a continental scale while maintaining a direct link to ecologically relevant, climate-related stressors. We first demonstrate that shade, drought, and waterlogging tolerances of forest stands are strongly correlated with climate and edaphic conditions in the conterminous US. This discovery allows the development of a Tolerance Distribution Model (TDM), a novel quantitative tool to assess landscape level impacts of climate change. We then focus on evaluating the implications of the drought TDM. Using an ensemble of 17 climate change models to drive this TDM, we estimate that 18% of US ecosystems are vulnerable to drought-related stress over the coming century. Vulnerable areas include mostly the Midwest US and Northeast US, as well as high elevation areas of the Rocky Mountains. We also infer stress incurred by shifting climate should create an opening for the establishment of forest types not currently seen in the conterminous US. Keywords: Tolerance Distribution Model | Drought, Shade and Waterlogging Tolerance Indices | Forested Ecosystems | Climate Change | Drought-related Stress Introduction Understanding and predicting how forest distributions will respond to ongoing and anticipated climate change is a challenge with great ecological, economic, and cultural implications (Levin, 1999). It is well established that environmental stressors increase mortality of intolerant trees (Hanson and Weltzin, 2000; Thomas et al., 2004; Engelbrecht et al., 2007; Adams et al., 2009; Choat et al., 2012; Anderegg et al., 2013). Mechanistic forest models such as individual-based forest simulators and their approximations have been used to model ecosystem responses by simulating changes in individual demography driven by changing environmental factors and resource competition (Shugart, 1984; Botkin, 1993; Pacala et al., 1993, 1996; Dube et al., 2001; Scheller and Mladenoff, 2007; Strigul et al., 2008; Strigul, 2012; Liénard and Strigul, 2016). Individual-based models can simulate forest dynam1 L I ÉNARD , H ARRISON & S TRIGUL D ROUGHT TOLERANCE DISTRIBUTION MODEL ics at large scales, but doing so requires enormous, often infeasible, computational resources, and detailed parametrization of each species morphological, allometric growth and demographic characteristic. Furthermore, detailed spatially-explicit forest census data is required as an initial condition (Moorcroft et al., 2001; Scheller and Mladenoff, 2005). Another, often more feasible, approach to predicting how vegetation distributions will change with climate at regional to continental scales is to associate biome types with climate envelopes (Olson et al., 2001; Woodward et al., 2004) and assume that vegetation will migrate to fill potential climate niches. To some degree this approach is appealing as biome spatial distributions are strongly correlated with climatic variables (Sowell, 1985; Stephenson, 1990; Prentice et al., 1992), particularly temperature and precipitation (Walter and Box, 1976; Olson et al., 2001; Woodward et al., 2004; Engelbrecht et al., 2007; Moncrieff et al., 2015). However, the biome approach defines biomes into discrete entities at the landscape scale based exclusively on present-day species distributions, without explicit consideration of plant physiological traits. This limits the utility of this approach to represent ecosystem transitions across space and time (Moncrieff et al., 2015) or to identify ecosystems under stress. Alternatively, Species Distribution Models, SDMs have been employed to study how plant communities respond to climate across various scales (local to continental; Pearson and Dawson, 2003; Araújo and Luoto, 2007; Iverson et al., 2008; Schloss et al., 2012; Lawler et al., 2013). One well-understood shortcoming of SDMs, however, is that they ignore biocomplexity and species interaction effects. In fact, species distributions depend not only on climatic factors, but also on biotic interactions within plant communities, disturbances and dispersal (Pearson and Dawson, 2003; Elith and Graham, 2009). Consequently, these models have been useful in predicting species presence and absence across environmental gradients, but they have not been employed to evaluate whole ecosystem vulnerability to climate change; rather, they are generally used to examine plant presence or absence across environmental gradients (Thomas et al., 2004; Guisan and PREPRINT OF GLOBAL CHANGE BIOLOGY, DOI : Thuiller, 2005; Schrag et al., 2008; Elith and Leathwick, 2009). In the sections that follow, we describe, evaluate and apply a new approach, a Tolerance Distribution Model (TDM), which we believe represents an advancement over past modeling efforts in that it can be applied at large spatial scales while still retaining information about climate-species interactions. In this approach, tolerances are treated as quantitative measures of plant ability to endure environmental and biotic stressors. In contrast with SDMs, we scale species tolerances, not abundance, resulting in community-scale descriptors that maintain a link with ecologically relevant stressors and which are not limited to discrete biome classifications. In our TDM, we determine relationships between tolerance traits and climate variables, and use the resulting best model to identify regions that are vulnerable to anticipated climate change. We first employ species-level rankings of shade, drought and waterlogging tolerance developed by foresters (Niinemets and Valladares, 2006; Valladares and Niinemets, 2008; Liénard et al., 2015) to examine whether and how major climate and edaphic factors in the conterminous US affect tree species distributions. We also test for relationships between these tolerance characteristics. We then focus specifically on drought tolerance and describe, evaluate, and apply a novel drought tolerance distribution model. We use it to address the following questions: 1) what climate variables best predict species and stand-level drought tolerance? And 2) what can a drought tolerance model tell us about US forest vulnerability to anticipated climate change? Materials and Methods Overview. Our overall approach was to: 1) merge information on tree species tolerance with species fractional basal area to calculate a tolerance index for each FIA plot in the US (n>600,000 plots in all), 2) analyze the relationships between climate and these community-level tolerance indices, 3) choose the optimal climatic variables to relate climate with drought tolerance, and 4) use climate change models to identify regions vulnerable to future climates, areas where the anticipated increase of drought tolerance index exceeds the current natural variability. 10.1111/ GCB .13291 2 L I ÉNARD , H ARRISON & S TRIGUL D ROUGHT TOLERANCE DISTRIBUTION MODEL In the following sections we describe each of these steps in greater detail. Tolerance Rankings and Indices. Foresters, plant physiologists and agriculturalists have been studying tree tolerances for decades and have established a solid physiological basis for species-level tolerance rankings, numerical representations of tree species (e.g. Kramer 1969, Kozlowski 1997, Kozlowski and Pallardy 2002, Pallardy 2010, Scherrer et al. 2011, Eilmann and Rigling 2012, Zang et al. 2014). These rankings are determined based on a combination of tree physiology and species distributions. We have taken tolerance rankings from a range of sources and placed them into an internallyconsistent, five-level tolerance scale ranging from zero to one, with the following ranks: very intolerant = 0, intolerant = 0.25, intermediate = 0.5, tolerant = 0.75, and very tolerant = 1. Shade tolerance rankings were taken from Baker (1949); Burns and Honkala (1990); Niinemets and Valladares (2006); Humbert et al. (2007) and are described in detail in Liénard et al. (2015). Drought tolerance rankings were mainly taken from Niinemets and Valladares (2006), providing data for 92% of the trees surveyed in the FIA database. We further extended the dataset to cover all species with data coming from Burns and Honkala (1990) and references gathered from the TRY database (TRY-DB.org; see Supplementary Materials 1.1 for more detail). A substantial number of drought tolerance ranks for US trees were derived from Niinemets and Valladares (2006), who computed them based on expert opinion, site characteristics (total annual precipitation, ratio of precipitation to potential evapotranspiration and duration of the dry period), and plant physiology (minimum soil water potential that can be tolerated over the long term with 50% of foliage damage or dieback). Specifically, Niinemets and Valladares (2006) established an ad-hoc quantitative scale of site classifying characteristics and plant physiology based on expert knowledge, ranked all species from different datasets and continents, and finally cross-calibrated and rectified the rankings according to many metaanalyses and comparative studies (including notably Minore, 1979, Meerow and Norcini, 1997, Kuhns and Rupp, 2000, Cerny et al., 2002) and expert opinion (see Appendix B of Niinemets and Valladares, 2006, for details and more references). BePREPRINT OF GLOBAL CHANGE BIOLOGY, DOI : cause Niinemets and Valladares (2006), the source of many species-specific tolerance rank values in this study, utilize habitat and climate information as two (of many) factors to estimate species tolerance, one might argue that our effort to use climate information to re-project ecosystem tolerances back onto the landscape (described in details below) is somewhat circular. We argue, however, that plant physiology is also “baked into” tolerance indices, so that our drought tolerance model provides novel information about plant community distribution and vulnerability under anticipated climate change. Furthermore, even a simple re-projection of current plant-climate associations with the specificity provided by tolerance rankings constitutes a significant advance in our understanding of the vulnerability of plant communities to climate change. In addition to shade tolerance and drought tolerance, we also estimated waterlogging tolerance, which indicates the degree to which species are able to survive in water-saturated soils. Waterlogging stress is relevant to only a small fraction of all US forests, and we adopt here the rankings from Niinemets and Valladares (2006) without extending it to all species of the FIA database. Individual species tolerance traits were scaled up to the stand level by taking the weighted average of relative abundance of tree species based on basal area (Eq. 1 in Supplementary Materials 1.2). We refer to these up-scaled tolerance traits as ”tolerance indices” throughout the remainder of this paper. Basal area was taken from the USDA Forest Inventory and Analysis database (FIADB, Woudenberg et al., 2010, which includes data from n > 600, 000 forest stands across the conterminous US). We also extracted land ownership and soil moisture information from the FIA database. Additional information on how variables were computed is available in section 1.2 of Supplementary Materials. Model sensitivity and uncertainty. To quantify the degree to which errors in tolerance rank values might influence our results, we conducted a sensitivity analysis. In this analysis we added a random component to each species ranking of ±0.1 points in the [0, 1] scale and another random component to each individual tree ranking of ±0.1, and subsequently computed the tolerance indices and drought TDM model using these tree-specific rankings. We 10.1111/ GCB .13291 3 L I ÉNARD , H ARRISON & S TRIGUL D ROUGHT TOLERANCE DISTRIBUTION MODEL performed the above procedure 100 times and report here the width of the bootstrapped 95% confidence intervals for (a) species-level rankings, (b) plot-level index, and (c) TDM mean values in the climatic space. Eastern forests are over-represented in the FIA dataset due to historical and administrative reasons (mainly due to an earlier start of FIA programs on the East coast compared to the West coast). This sampling bias in the database leads to an overrepresentation of some combinations of average temperature and precipitation. In the computation of correlation between tolerance indices, we corrected this bias with a bootstrapping strategy (Efron and Tibshirani, 1994). The procedure consists of the repeated selection of random subsample of plots that tile the climatic space evenly, and allows the computation of 95% confidence intervals. This methodology is described and illustrated in more detail in Supplementary Materials 1.3. Climatic variables. Two databases were independently employed to link vegetation patterns and climate (Supplementary Materials 1.5): Worldclim (with normals computed from 1950 to 2000 using a spatial resolution of 30 seconds Hijmans et al., 2005) and PRISM (with normals computed from 1981 to 2010 using a spatial resolution of 800 meters, Prism Group,2015). In the main text we present results obtained with Worldclim database, and PRISM-based results are included as Supplementary Materials. Drought tolerance model. A drought tolerance distribution model was developed that predicts forest drought tolerance as a function of average annual temperature and precipitation (Fig. 3a). To construct the drought TDM, we first binned FIA plots according to their climate characteristics (0.5°C temperature bins and 60 mm/month precipitation bins) and then computed average drought tolerance for each bin. Climate bins containing fewer than 10 forest inventory plots were discarded, and drought tolerance values were averaged in all other cells. The method of spatial extrapolation we employed had very little impact on our results (Supplementary Materials 1.6). In addition, FIA plots occurring in wetland areas (6% of the FIA plots; Supplementary Materials 1.4) were excluded from this analysis because PREPRINT OF GLOBAL CHANGE BIOLOGY, DOI : in these plots water supply is presumably decoupled from the local precipitation regime, an assumption supported by a lack of strong link between waterlogging tolerance and annual precipitation (Fig. 1f; Supplementary Results S3 and Supplementary Materials 1.4). To evaluate which climate variables were most strongly correlated with drought tolerance, we evaluated a suite of 88 models, each of which used a pair of bioclimatic variables to predict drought tolerance. Temperature-related variables included: mean annual temperature and seasonality, mean diurnal range, isothermality, temperature range, max/min monthly temperatures, and mean temperature of the wettest/driest/warmest/coldest month (Supplementary Materials 1.6; Fig. S6). Precipitation-related variables included: annual precipitation and seasonality, max/min monthly precipitation, and precipitation of the wettest/driest/warmest/coldest quarter (Supplementary Materials 1.6; Fig. S6). A randomly selected subset of FIA plots (90% of the entire dataset) was used to calibrate the model, and remaining 10% of sites were used to evaluate model skill with greatest model skill defined as lowest mean squared error of model predictions. This crossvalidation strategy is useful to avoid the possibility of overfitting when choosing the optimal climatic variables to build the TDM. Of all models tested, the best model used average annual temperature and average annual precipitation as independent variables. To further evaluate this result we compared the drought tolerance index computed from the FIADB (current period) with the model’s prediction based on current temperature and precipitation. We computed standard deviation as an indicator of the natural variability of drought tolerance index and standard error of the mean as an indicator of model fit quality. We also did an out-of-sample validation by training the model using a variable fraction (from 10% up to 95%) of FIA plots, and subsequently comparing predictions with non-calibration data (Supplementary Results S8). Climate change scenarios. To calculate drought tolerance index values under future climate conditions we relied on projected climatic data using two Representative Concentration Pathways (RCP) adopted on the Fifth Assessment Report from the Intergovernmental Panel on Climate Change (Collins 10.1111/ GCB .13291 4 L I ÉNARD , H ARRISON & S TRIGUL D ROUGHT TOLERANCE DISTRIBUTION MODEL et al., 2013, see also Supplementary Materials 1.5): moderate forcing (RCP4.5) and severe forcing (RCP8.5). The climate change models that we consider are from Worldclim (worldclim.org, Hijmans et al., 2005). In this collection of model output, highresolution spatial layers are obtained by computing the difference between current and future conditions, and interpolating the differences to a 30 arc-second (roughly 1 km × 1km) grid. These differences are then bias-corrected with a baseline extracted from the current conditions of the Worldclim dataset. In the main text of this paper we computed median predicted drought tolerance using median values from the 17 available climate model predictions. Predictions using each individual climate change model are presented in Supplementary Results S9. Drought tolerance indices vary within each climate bin, which is expected due to the patch-mosaic patterns of forested ecosystems (resulting from the specific history of plots and including forest succession, disturbance, management regime, differences in seasonality and other high dimensional climate axes, different regional species pools). Projecting forest vulnerability to future climates requires first quantifying the departure from normal variability under present climatic conditions, and then computing whether projected climatic conditions move systems outside of their present-day variability. We calculate present-day variation in drought tolerance as a function of climate by deriving three models that predict the 90, 95 and 99% upper bounds of the drought tolerance distribution, respectively (reported in Supplementary Results S9). These models are then used to estimate the empirical probability that drought tolerance is lower than these thresholds under projected climate change (reported in main text at Fig. 5e-f). In some regions projected climatic conditions exceed the range of the TDM, implying a shift to conditions not currently found in the mainland US. To make predictions for these areas, we relied on an approach inspired by potential vegetation models; we cross-linked current biomes with future climatic conditions. Specifically, we relied on the worldwide biome classification reported in FAO (2012) and on the current worldwide climatic conditions from Worldclim (Hijmans et al., 2005), to match current climate with current biomes. We then matched future projected climate in the US with current word- wide climate and current biomes, which allowed us to label the putative forests types (Fig. 5 in main text). Putative forest types with a drought tolerance index outside the current range of US values were then labeled using current global ecological zones (FAO 2012). Software. All databases and computations were developed, manipulated, and analyzed using R 3.2.2 (R Core Team, 2014) using the packages rgdal 1.04 (Bivand et al., 2015) and raster 2.4-15 (Hijmans, 2015). The code necessary to reproduce the analyses is available in an online repository1 . Results Overview of current forest tolerance patterns. Shade, drought and waterlogging tolerance indices show distinct spatial patterns (Fig. 1a), demonstrating that these stand-level indicators can effectively describe forested ecosystems in the US. Shade and drought tolerance were also strongly correlated with mean annual temperature and precipitation (Fig. 1d-e), while waterlogging tolerance displayed no clear relationship with climate parameters except for demonstrating very low values when the mean annual temperature was higher than 20°C (Fig. 1f). In addition, the inverse relationship between shade and drought tolerances that has been reported for temperate tree species (Niinemets and Valladares, 2006; Valladares and Niinemets, 2008) clearly scales up to the landscape level (R between shade and drought tolerance: -0.794; 95% bootstrapped confidence interval [-0.771, -0.815]; Fig. 1c,d-e; cf. Supplementary Results S3). Regions where drought tolerance was high but shade and waterlogging tolerance were low included much of the US Intermountain West and much of the southeastern US (Fig. 1a). Regions where shade tolerance was high but drought and waterlogging tolerance were low included portions of the moist Pacific Northwest US, the Upper Midwest US, and the Northeast US (Fig. 1a). FIA plots with high waterlogging tolerance but low drought and shade tolerance were located primarily along the Mississippi River or its tributaries, and along the southwestern US coast (Fig. 1a). 1 https://github.com/jealie/TDM PREPRINT OF GLOBAL CHANGE BIOLOGY, DOI : 10.1111/ GCB .13291 5 L I ÉNARD , H ARRISON & S TRIGUL D ROUGHT TOLERANCE DISTRIBUTION MODEL (a) (b) (c) Shade Tolerance Index Shade Tolerance Index Drought Tolerance Index Waterlogging Tolerance Index (e) Drought tolerance index Precipitation (mm/month) 350 (f) Shade tolerance index 350 Cumulative Distribution Waterlogging tolerance index 350 1.0 1.0 300 0% 20% 40% 60% 80% 100% Waterlogging Tolerance Index Map Color Key (d) Drought Tolerance Index 300 5% high outliers (< 0.45) 300 250 0.8 250 0.8 250 200 0.6 200 0.6 200 0.4 0.3 150 0.4 100 150 0.4 100 0.2 100 0.2 0.2 50 150 50 0 0 -5 0 5 10 15 Temperature (⁰C) 20 25 30 0.1 50 0.0 0.0 0.0 0 -5 0 5 10 15 20 25 30 Temperature (⁰C) -5 0 5 10 15 20 25 30 Temperature (⁰C) Fig. 1: Overview of the tolerance indices in the conterminous USA. (a) Visualization of forest stands tolerance index mapped onto the hue–saturation–value color space. In the color triangle key (b), colors at the vertices correspond to forests where one tolerance index is high while others are low. Colors inside the triangle correspond to forests with intermediate tolerances (for example, yellow indicates forests resilient both shade and drought). (c) Cumulative distribution of the plot tolerances within the triangle key, showing that most plots implement a tradeoff between shade and drought tolerance, while plots with high waterlogging tolerance are infrequent. Separate maps of tolerance indices to drought, shade, and waterlogging are available in Supplementary Results S2. (d–f) Drought, shade, and waterlogging tolerances of forest stands plotted in climate space with mean annual temperature (x axis) and mean annual precipitation (y axis). The 5% high outliers for waterlogging tolerance index are represented by purple crosses. A drought tolerance model. Whereas waterlog- ging and shade tolerance are only indirectly linked to climate drivers (Liénard et al., 2015; Liénard and Strigul, 2015), drought tolerance can be directly linked to climate. We utilized this direct linkage to examine the vulnerability of forests to changing climate in the US, and focus the following analysis on drought tolerance modeling. Overall, the drought TDM reproduces current drought tolerance patterns in the continental US (compare Figs 2c and 2d). Areas where the drought TDM does not do as well include the lower Mississippi River Basin (a wetlanddominated region where climate and vegetation are likely decoupled by the overland supply of water) and some of the montane areas of the West coast (Fig. 10 in Supplementary Results S8). PREPRINT OF GLOBAL CHANGE BIOLOGY, DOI : Although one might reasonably expect climate variables such as drought duration or intensity to best predict drought tolerance, our statistical analysis of 19 bioclimatic variables reveals that mean annual temperature and mean annual precipitation were the best variables to minimize model error (Supplementary Results S6). Extreme annual temperature/precipitation (such as the Precipitation of the wettest quarter with the Max temperature of the warmest month) also did reasonably well as predictor variables but not as well as mean annual temperature and precipitation. Bioclimatic variables linked to sub-annual variability of climate (such as the precipitation seasonality or the annual range of temperature) consistently resulted in low model performances, and thus appear nonessential for the characterization of drought tolerance (Supplementary Results S6). 10.1111/ GCB .13291 6 L I ÉNARD , H ARRISON & S TRIGUL D ROUGHT TOLERANCE DISTRIBUTION MODEL Model of drought tolerance index (a) 350 Precipitation (mm/month) 350 Precipitation (mm/month) Model standard deviation (b) 1.0 300 0.8 250 0.6 200 0.4 150 0.2 100 0.0 50 0.5 300 0.4 250 0.3 200 0.2 150 0.1 100 0.0 50 0 0 -5 0 5 10 15 20 25 -5 30 0 5 15 20 25 30 Temperature (⁰C) Temperature (⁰C) (c) 10 (d) Current drought tolerance Modeled drought tolerance (based on current climatic conditions) 1.0 1.0 0.8 0.8 0.6 0.6 0.4 0.4 0.2 0.2 0.0 0.0 Fig. 2: Drought TDM based on FIA forest plots and Worldclim’s current annual precipitation and temperatures. (a) Representation of the drought tolerance predicted by the model as a function of temperature and precipitation. (b) Standard error of the mean of these predictions relative to the drought tolerance index measured in FIA plots. (c) Geographical distribution current drought tolerance index across surveyed plots. (d) modeled drought tolerance from the TDM based on current climatic conditions. Supplementary Results S8 show that the best model had overall low standard errors of the mean (less than 5%), indicating that data are adequately constraining the model. The analysis of errors reveals a symmetric, non-skewed profile that follows an exponential decrease around the mean, consistent with a high predictive power of the TDM (Supplementary Results S8). An out-of-sample validation showed that model performance remained similar when only about 70% of FIA plots were used to train the model, meaning that the data at hand is sufficient to constrain the model (Supplementary Results S8). Applying alternative regression schemes based on inverse distance weighting and kriging are also presented in Supplementary Results S7, and did not substantively change model predictions or error magnitude and distribution. PREPRINT OF GLOBAL CHANGE BIOLOGY, DOI : In addition to the error analyses summarized above, we conducted an extra sensitivity analysis to evaluate how errors propagate from rankings of individual tree, to the plot-level drought tolerance index, and to the drought TDM model. These indicated a roughly two-fold decrease in uncertainty at each up-scaling step (Supplementary Results S8). Collectively, these analyses suggest that the climate-level drought TDM is robust to small errors and approximations introduced at the level of drought tolerance rankings of individual trees. Using a drought TDM to understand forest vulnerability to climate change. Mean annual precipitation and temperature are both expected to change over the coming century (Fig. 3a-b), and the geographic distribution of drought tolerance will need to shift to accommodate this change (Supplementary Results 1.7). Projected climate trajectories 10.1111/ GCB .13291 7 L I ÉNARD , H ARRISON & S TRIGUL Current climate of US forested ecosystem and their projections in 2070 (a) Precipitation (mm/month) D ROUGHT TOLERANCE DISTRIBUTION MODEL Model HE Current Model IN Model AC Model IP Model BC Model MI Model CC Model MR Model CN Model MC Model GF Model MP Model GS Model MG Model HD Model NO Model HG Moderate scenario Severe scenario 350 300 250 Projected climatic trajectory of a subset of US forested plots (b) 350 300 250 200 200 150 150 100 100 50 50 0 0 -5 0 5 10 15 20 25 -5 30 0 5 Temperature (⁰C) (c) 10 15 20 25 Magnitude of projected drought tolerance change (with RCP 4.5) (d) Magnitude of projected drought tolerance change (with RCP 8.5) > 0.4 > 0.4 0.4 0.3 0.2 0.1 0.0 -0.1 -0.2 0.4 0.3 0.2 0.1 0.0 -0.1 -0.2 < 0.2 (e) 30 Temperature (⁰C) < 0.2 (f) Confidence in drought tolerance shift (with RCP 4.5) Confidence in drought tolerance shift (with RCP 4.5) Within expected range > 90% > 95% > 99% Within expected range > 90% > 95% > 99% Fig. 3: Application of TDM to predict future drought tolerance using multiple climate change models. (a) Overview of the predictions of 17 climate models for US forested ecosystems, under two scenarios: “moderate” forcing (RCP4.5) and “severe” forcing (RCP8.5). The current span of climatic conditions is represented by the grey shade and the future climatic boundaries by dotted and thick lines. (b) The trajectories of a subset of plots are shown. Models predict overall consistent increases of temperatures, while precipitation changes display a higher variability. (c and d) Median projected tolerance change across climatic models, with RCP4.5 (c) and RCP8.5 (d). (e and f) Confidence levels of increasing drought tolerance according to the TDM, in RCP4.5 (e) and RCP8.5 (f). Color indicates the likelihood that the median projected required drought tolerance will be higher than current drought tolerance (the probability thresholds of 90%, 95%, and 99% are relative to the tolerance variability displayed in Fig. 2b). In (c–f), the purple areas indicate regions where future climatic conditions fall outside the predictive range of the TDM (i.e., which are outside the current climate extent in panels a–b), and grey areas to climatic conditions that do not currently sustain forests in the USA, and where the TDM can thus not be applied. for forested plots in climate space (Fig. 3b) can be coupled with the drought tolerance model (Fig. 2a) to calculate the drought tolerance necessary to accommodate modeled future conditions. Extrapola- PREPRINT OF GLOBAL CHANGE BIOLOGY, DOI : tion of the model to future conditions using an ensemble of 17 climate models revealed a consistent progression toward greater required drought tolerance. This progression was geographically ubiqui- 10.1111/ GCB .13291 8 L I ÉNARD , H ARRISON & S TRIGUL D ROUGHT TOLERANCE DISTRIBUTION MODEL tous and consistent across scenarios (from RCP4.5 to RCP8.5, cf. Supplementary Results S9). Model predictions presented in this section were robust with respect to the source of current climatic data (two databases used, Worldclim in Fig. 1c-f and PRISM in Supplementary Results S5), to the regression method (binning, inverse distance weighting, and kriging, Fig. 2 and Supplementary Results S7), and climate change model choice (17 models considered in Fig. 3, see Supplementary Results S9 for their individual predictions). We report both magnitude of expected shifts in drought tolerance (Fig. 3c-d) and the probability that this change will require forest ecosystem change (Fig. 3e-f). Increases in drought tolerance are much more common than decreases (located in sparsely forested areas of the western US; Fig. 3c-d), and are widespread over forested ecosystems of the US in both climate scenarios investigated (Fig. 3c-d). The greatest predicted increases in drought tolerance occur in the upper Great Plains temperate steppe biome (in North Dakota and western Wisconsin) and in western mountain systems. Smaller, but still substantial, changes in drought tolerance are predicted in the northeastern US and along the West Coast (Fig. 3c-d). One must take caution in interpreting these absolute changes, however, as the drought tolerance scale is not linear (e.g. a 0-0.25 shift is not equivalent to a 0.75-1.00 shift). In many regions predicted increases in drought tolerance will not necessarily require significant shifts for forested ecosystems because in these areas increases in drought tolerance are likely to be small compared to natural variability (Fig. 3e-f). In some areas, however, drought tolerance distributions will likely move beyond the range of present-day variability. These atrisk regions include parts of the northeastern US and Northern Great Plains, as well as, to a lower extent, higher elevation areas in the Rocky Mountains (Fig. 4e-f). In the northeastern US, where the risks are the most pronounced, at-risk forest types include Maple/Beech/Birch, Spruce/Fir and White/Red/Jack Pine combination. Red pines (Pinus resinosa) in particular, a species with medium resistance to drought, has a distribution overlapping the most vulnerable areas identified in Fig. 3e-f and is already in the list of endangered species in three northeastern US states. Most (81.9%) of the FIA plots considered in PREPRINT OF GLOBAL CHANGE BIOLOGY, DOI : this study were located on private lands, with substantially smaller numbers occurring on public lands (6.6% state and 11.5% federal). We estimate that climate change over the next century will affect all three categories of land, but in our analysis private lands appear to be located in vulnerable areas more frequently (87.1% of total privately held FIA plots) than public lands (4.7% and 6.7% of state and federal lands classified as vulnerable, respectively). Putative forest types outside the range of the TDM. Some of the climate conditions (and hence drought-tolerance characteristics) anticipated to occur over the next several decades do not currently exist in the conterminous US (Williams et al., 2007; Ackerly et al., 2010). For regions with these climate conditions (purple regions Fig. 3) one can look to other continents and find appropriate forest types (Fig. 4). For example, low-altitude areas of Texas may eventually host tropical dry forests, with potential migrants coming from tropical dry forests of Eastern Mexico. The Gulf coastal plains of the southeastern US could host tropical moist deciduous forests, such as the Cuban moist forests. Finally, the tropical desert/shrublands of Mojave, Sonoran and Chihuahuan deserts may well extend northward. The Pacific Northwest’s Cascades are anticipated to have a climate similar to China’s Fujian province or Southern Brazil’s Paraná and Santa Catarina. As all analogs for the projected US Pacific Northwest climate area are far away, it is uncertain what species will migrate to fill the ecological niches created by a changing climate. Discussion In this work, we map vegetation tolerance patterns for the first time and study the relationship between these tolerance characteristics and climatic variables. The drought TDM establishes a geography of drought tolerance based on present-day climatic data. Once the relationship between drought tolerance and climate is quantified via a TDM, it is possible to use this relationship to examine how drought tolerance will likely change as climate shifts. It is also possible to identify “at-risk” regions: places where future drought tolerance of plants and plant communities is likely to fall outside the current range of variation. Somewhat surprisingly, areas with large predicted changes in drought toler- 10.1111/ GCB .13291 9 L I ÉNARD , H ARRISON & S TRIGUL D ROUGHT TOLERANCE DISTRIBUTION MODEL Putative forest types for projected climates currently absent in the US Subtropical humid forest Tropical desert/shrubland Tropical dry forest Tropical moist deciduous forest Fig. 4: Putative forest types for which projected future climatic conditions are outside the TDM range in the United States (areas in purple in d and f) using RCP 8.5 as a forcing scenario. ance are not necessarily the regions that are most at-risk. This is because some of these regions have high variance in drought tolerance, which may buffer them against future climate change (e.g. the northern Great Plains). In contrast, regions with relatively uniform drought tolerance can be at risk in the face of comparatively small changes in climate (e.g. the Northeastern US). Such an approach has similarity with the identification of so-called “no-go” areas for functional traits, which were deduced by taking empirical percentiles of observed distribution Stahl et al. (2014). More broadly, the identification of at-risk areas can also be seen as the geographical equivalent of the temporal departure from normality studied in Mora et al. (2013), which predicts when climatic variations will differ significantly from historical ranges. Relationship with other modeling approaches. TDMs, biome envelope models, SDMs, functional trait models, and climate-based aridity indices can all be used to understand how climate controls spatial distributions of species and communities. The drought tolerance index introduced in this study is a characteristic of forest communities based on species abundance. This quantity is different from traditionally used climatic aridity indices such as the Palmer Drought Severity Index, the Surface Water Supply Index or the Standardized Precipitation Index, which assess drought frequency by summarizPREPRINT OF GLOBAL CHANGE BIOLOGY, DOI : ing climatic variables (Hayes et al., 2006, Maliva and Missimer 2012). In addition, these modeling approaches operate at different levels of biological organization than a TDM. Whereas SDMs predict the distributions of species and populations, and biome models predict the presence or absence of whole communities, our TDM indicates the vulnerability of current vegetation to future conditions. Models predicting the dynamics of a single functional trait (such as maximal tree height, maximal seed mass or maximal wood density in Stahl et al, 2014) have a close relationship to the TDM, which analyzes tolerance indices instead of functional traits. The TDM approach can be seen as more general in scope compared to single functional trait analyses. Indeed, tolerance rankings summarize many functional traits; for example, drought tolerance ranking is linked to N content, photosynthetic capacity, leaf life span and lead dry mass per unit area (Hallik et al., 2009), seedling size (Westoby et al., 2002), deciduoudness and overall biomass allocation to roots (Poorter and Markesteijn, 2008). Whereas previous studies have shown that single traits are only weakly connected to climatic gradients (Reich et al., 2003, Stahl et al., 2014), vegetation cover types (Swenson et al., 2012, van Bodegom et al., 2014) and growth patterns (Paine et al., 2015), by encompassing several traits at once, tolerance indices can efficiently describe spatial and climatic patterns (Figs 1 and 2). 10.1111/ GCB .13291 10 L I ÉNARD , H ARRISON & S TRIGUL D ROUGHT TOLERANCE DISTRIBUTION MODEL Major ecological domains of the USA (a) Temperate desert Temperate mountain system Temperate oceanic forest Temperate steppe Tropical moist deciduous forest Subtropical desert Subtropical dry forest Subtropical humid forest Subtropical mountain system Subtropical steppe Temperate continental forest (b) (c) 350 Precipitation (mm/month) Precipitation (mm/month) 350 300 250 200 150 100 50 0 300 250 200 150 100 50 0 -5 0 5 10 15 20 25 30 Temperature (⁰C) -5 0 5 10 15 20 25 30 Temperature (⁰C) Fig. 5: Biomes and their climatic envelopes in the United States. (a) Forested plots of the FIA database split according to their major ecological domains (FAO, 2012). (b, c) Confidence ellipses encompassing 95% of the FIA plots for each biome, shown in the climatic space of annual temperature (x axis) and precipitation (y axis). Despite operating at a different scale, TDM-based predictions are broadly consistent with those of species and biome models. For example, the northward shift of ecosystem types in the eastern and northeastern US predicted by our TDM is also predicted using SDM and process-based approaches (Iverson et al., 2008; Morin et al., 2008, Rehfeldt et al., 2009). We highlight this region as one that is particularly vulnerable to climate change (Fig. 4), a result consistent with the findings of others (e.g. PREPRINT OF GLOBAL CHANGE BIOLOGY, DOI : Gonzalez et al., 2010). In the western US, some of the regions identified here as at risk to drought were also identified by Worrall et al. (2013), who used a SDM to study the ongoing decline of the droughtintolerant trembling aspen (Populus tremuloides). In contrast with the drought TDM, biome models do not highlight significant risks for mountain areas in western US, possibly owing to their coarser resolution (Gonzalez et al., 2010, Scholze et al., 2006). Our TDM approach also identifies mountainous ar- 10.1111/ GCB .13291 11 L I ÉNARD , H ARRISON & S TRIGUL D ROUGHT TOLERANCE DISTRIBUTION MODEL eas as at particular risk, which is consistent with a recent study of 6 species across the western US (Bell et al., 2014). In fact, we predict that in the western US high-elevation areas will likely be affected much more by climate change than the lower elevation areas, perhaps owing to the fact that high-elevation vegetation in this region is quite fine-tuned to withstand harsh, high-altitude conditions. One can also gain insight by contrasting the results of biome and TDM approaches. Overall, biomes show much less specificity with respect to climatic variables than shade and drought tolerance index, which display consistent and clear gradients across the whole climatic space (Contrast Fig. 1d-e with 5b-c). Non-mountain biomes encompass relatively narrow bands in climatic space, and they display a high overlap at intermediate temperatures (5 – 20 0 C) across a broad range of rainfall rates (50 – 150 mm/month; Fig. 5b). The overlap is even greater with mountain biomes, which encompass an exceptionally large range of temperature and precipitation (Fig. 5c). The significant overlap between ecotones in climate space somewhat limits the utility of a biome envelope approach. This problem persists regardless of the biome classification scheme used (cf. Supplementary Results S1). Model limitations and future directions. Al- though our drought TDM model effectively predicts drought tolerance, some unexplained variance remains. Discrepancies between modeled and measured drought tolerance may result from multiple sources, including: errors in initial assignment of drought tolerance, errors in climate data or models, a disconnect between climate and drought tolerance (e.g. due to overland water transfer such as occurs in floodplain wetlands), inaccuracies in FIA plot locations (which are often intentionally obscured to preserve land-owner privacy), or stochastic disturbances (e.g. harvest or fires). To evaluate the potential impact of these uncertainties on TDM predictions we estimated model robustness at the level of individual trees and species, at the plot-level of a forested patch, and at the landscape level. This exercise indicated that the uncertainties in tolerance rankings decreased markedly with upscaling, falling by 50% when individual species tolerances were used to calculate stand-level tolerance PREPRINT OF GLOBAL CHANGE BIOLOGY, DOI : indices, and decreasing by an additional 50% when stand-level tolerance indices were scaled up to compute the TDM. While we cannot rule out the presence of errors and imprecision in the tolerance rankings, this analysis shows that their impact on the final model is likely to be quite small. To address concerns regarding FIA plot locations, we relied on the fact that errors in coordinates are not generally large enough to negatively affect our analysis. FIA indicates (Woudenberg et al., 2010) that some plot coordinate data are only slightly altered, with only about 0-25% of plots swapped, and those within the same county. Furthermore, FIA states that in order to allow landscape level modeling the ecological similarity is preserved when plot coordinates are altered and swapped. The net result of this strategy is that error caused by FIA plot coordinate alteration is negligible in SDMs (Gibson et al. 2014), and the same should be true for the TDMs presented here. To address uncertainty associated with projected climate, we used 17 climate changes models separately (Supplementary Results S9) and together (Fig. 4); these models mostly agree on the areas of drought vulnerability in the US. Nonetheless, future research could investigate how fine-scale differences in projected climate propagate to the TDM predictions. At the landscape level our TDM focuses on forest tolerance distributions as a function of climate, while other forest change drivers in conterminous US are largely ignored in this iteration of the model. In particular, forest distributions in some areas are affected by fire (Scholl and Taylor, 2010). Phenology can also play an important role in forest climate change response (Chuine and Beaubien, 2001). Future versions of the TDMs presented here can certainly be modified to include these factors. In addition to these uncertainties, there is some potential for circularity in the TDMs presented here. Although habitat and climatic factors were used in formulating species tolerance rankings, this information was only one of many types of information used. Future studies comparing our results with individual-based tolerance rankings grounded exclusively in plant physiology could determine the degree of circularity in the establishment of tolerance rankings. However, we argue that because plant species physiological characteristics were used in 10.1111/ GCB .13291 12 L I ÉNARD , H ARRISON & S TRIGUL D ROUGHT TOLERANCE DISTRIBUTION MODEL determining tolerance rankings the TDM approach provides novel insight about ecosystem climate vulnerability. Furthermore, community-level tolerance may be an emergent property giving novel insight beyond the species rankings, much as communityweighted trait values have been used in functional trait studies (Ackerly and Cornwell, 2007). Although the concept of potential vegetation mapping has been broadly implemented using climate envelope models and SDMs, these previously developed models cannot be used to predict forest vulnerability to specific climate-related stressors. Our study presents an alternative approach, which considers community (as opposed to species or biome) weighted tolerances to environmental stresses. Output from our drought TDM has important management implications in that it highlights regions likely to experience significant climate-related vegetation stress. Acknowledgement We are grateful to Marlène Cassar, Kendra Lynn and the anonymous reviewers who helped improve the manuscript with their thoughtful comments. This work was partially supported by a grant from the Simons Foundation (283770 to N.S.) and a Washington State University SEED grant. References Ackerly, D. D., and Cornwell, W. K. (2007). A trait-based approach to community assembly: partitioning of species trait values into within-and amongcommunity components. Ecology letters, 10(2), 135-145. Ackerly, D. D., Loarie, S., Cornwell, W., Weiss, S., Hamilton, H., Branciforte, R., and Kraft, N. (2010). The geography of climate change: implications for conservation biogeography. Diversity and Distributions, 16(3):476–487. Adams, H. D., Guardiola-Claramonte, M., Barron-Gafford, G. A., Villegas, J. C., Breshears, D. D., Zou, C. B., Troch, P. A., and Huxman, T. E. (2009). Temperature sensitivity of drought-induced tree mortality portends increased regional die-off under global-change-type drought. Proceedings of the national academy of sciences, 106(17):7063–7066. Anderegg, L. D., Anderegg, W. R., and Berry, J. A. (2013). Not all droughts are created equal: translating meteorological drought into woody plant mortality. Tree physiology, 33(7):672–683. Araújo, M. B. and Luoto, M. (2007). The importance of biotic interactions for modelling species distributions under climate change. Global Ecology and Biogeography, 16(6):743–753. Baker, F. S. (1949). A revised tolerance table. Journal of Forestry, 47(3):179–181. Bell, D. M., Bradford, J. B., and Lauenroth, W. K. (2014). Early indicators of change: divergent climate envelopes between tree life stages imply range shifts in the western United States. Global Ecology and Biogeography, 23(2):168–180. Bivand, R., Keitt, T., and Rowlingson, B. (2015). rgdal: Bindings for the Geospatial Data Abstraction Library. R package version 1.0-4. Botkin, D. B. (1993). Forest Dynamics: An Ecological Model. Burns, R. M. and Honkala, B. H. (1990). Silvics of North America. Volumes 1, 2. US Dept. of Agriculture, Forest Service. PREPRINT OF GLOBAL CHANGE BIOLOGY, DOI : Cerny, T. A., M. Kuhns, K. L. Kopp, and M. Johnson (2002). Efficient irrigation of trees and shrubs. Utah State University Extension, Logan, Utah, USA. Electronic Publishing, HG-523. Choat, B., Jansen, S., Brodribb, T. J., Cochard, H., Delzon, S., Bhaskar, R., Bucci, S. J., Feild, T. S., Gleason, S. M., Hacke, U. G., et al. (2012). Global convergence in the vulnerability of forests to drought. Nature, 491(7426):752–755. Chuine, I. and Beaubien, E. G. (2001). Phenology is a major determinant of tree species range. Ecology Letters, 4(5):500–510. Collins, M., Knutti, R., Arblaster, J., Dufresne, J.-L., Fichefet, T., Friedlingstein, P., Gao, X., Gutowski, W., Johns, T., Krinner, G., et al. (2013). Longterm climate change: projections, commitments and irreversibility. In Stocker, T.F., e. a., editor, Climate Change 2013: The Physical Science Basis. Contribution of Working Group I to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge University Press. Dube, P., Fortin, M., Canham, C., and Marceau, D. (2001). Quantifying gap dynamics at the patch mosaic level using a spatially-explicit model of a northern hardwood forest ecosystem. Ecological Modelling, 142(1):39–60. Efron, B. and Tibshirani, R. J. (1994). An introduction to the bootstrap. CRC press. Eilmann, B., and Rigling, A. (2012). Tree-growth analyses to estimate tree species’ drought tolerance. Tree Physiology, 32(2), 178-187. Elith, J. and Graham, C. H. (2009). Do they? how do they? why do they differ? on finding reasons for differing performances of species distribution models. Ecography, 32(1):66–77. Elith, J. and Leathwick, J. R. (2009). Species distribution models: ecological explanation and prediction across space and time. Annual Review of Ecology, Evolution, and Systematics, 40(1):677. Engelbrecht, B. M., Comita, L. S., Condit, R., Kursar, T. A., Tyree, M. T., Turner, B. L., and Hubbell, S. P. (2007). Drought sensitivity shapes species distribution patterns in tropical forests. Nature, 447(7140):80–82. FAO (2012). Global Ecological Zones for FAO Forest Reporting: 2010 Update. Working Paper 179. Gibson, J., Moisen, G., Frescino, T., and Edwards Jr, T. C. (2014). Using publicly available forest inventory data in climate-based models of tree species distribution: Examining effects of true versus altered location coordinates. Ecosystems, 17(1):43–53. Gonzalez, P., Neilson, R. P., Lenihan, J. M., and Drapek, R. J. (2010). Global patterns in the vulnerability of ecosystems to vegetation shifts due to climate change. Global Ecology and Biogeography, 19(6):755–768. Guisan, A. and Thuiller, W. (2005). Predicting species distribution: offering more than simple habitat models. Ecology Letters, 8(9):993–1009. Hallik, L., Niinemets, U, and Wright, I. J. (2009). Are species shade and drought tolerance reflected in leaf-level structural and functional differentiation in northern hemisphere temperate woody flora? New Phytologist, 184(1):257–274. Hanson, P. J. and Weltzin, J. F. (2000). Drought disturbance from climate change: response of united states forests. Science of the Total Environment, 262(3):205–220. Hijmans, R. J. (2015). raster: Geographic Data Analysis and Modeling. R package version 2.4-15. Hijmans, R. J., Cameron, S. E., Parra, J. L., Jones, P. G., and Jarvis, A. (2005). Very high resolution interpolated climate surfaces for global land areas. International Journal of Climatology, 25(15):1965–1978. Humbert, L., Gagnon, D., Kneeshaw, D., and Messier, C. (2007). A shade tolerance index for common understory species of northeastern North America. Ecological Indicators, 7(1):195– 207. Iverson, L., Prasad, A., and Matthews, S. (2008). Modeling potential climate change impacts on the trees of the northeastern United States. Mitigation and Adaptation Strategies for Global Change, 13(5-6):487–516. Kozlowski, T. T. (1997). Responses of woody plants to flooding and salinity. Tree physiology monograph, 1(1), 1-29. Kozlowski, T. T., and Pallardy, S. G. (2002). Acclimation and adaptive responses of woody plants to environmental stresses. The Botanical Review, 68(2), 270-334. Kramer, P. J. (1969). Plant and soil water relationships: a modern synthesis. McGraw-Hill Book Co. Kuhns, M., and L. Rupp (2000). Selecting and planting landscape trees. Electronic Publishing, NR-460. Utah State University Extension, Logan, Utah, USA. Lawler, J., Ruesch, A., Olden, J., and McRae, B. (2013). Projected climatedriven faunal movement routes. Ecology Letters, 16(8):1014–1022. Levin, S. A. (1999). Fragile dominion: complexity and the commons. Perseus Publishing, Cambridge, MA. 10.1111/ GCB .13291 13 L I ÉNARD , H ARRISON & S TRIGUL D ROUGHT TOLERANCE DISTRIBUTION MODEL Liénard, J., Florescu, I., and Strigul, N. (2015). An appraisal of the classic forest succession paradigm with the shade tolerance index. PLOS ONE, 10:e0117138, http://dx.doi:10.1371/journal.pone.0117138. Liénard, J. and Strigul, N. (2015). Linking forest shade tolerance and soil moisture in North America. Ecological Indicators, 58:332–334. Liénard, J. and Strigul, N. (2016). An individual-based forest model links canopy dynamics and shade tolerances along a soil moisture gradient. R. Soc. open sci. 3 150589; DOI: 10.1098/rsos.150589. Meerow, A. W., and J. G. Norcini (1997). Native trees for North Florida. Department of Environmental Horticulture, Florida Cooperative Extension Service, Institute of Food and Agricultural Sciences, University of Florida, Gainesville, Florida, USA. Minore, D. (1979). Comparative autecological characteristics of northwestern tree species: a literature review. USDA Forest Service Technical Report, PNW-87. Pacific Northwest Forest and Range Experiment Station, Portland, Oregon, USA. Moncrieff, G. R., Hickler, T., and Higgins, S. I. (2015). Intercontinental divergence in the climate envelope of major plant biomes. Global Ecology and Biogeography, 24(3):324–334. Moorcroft, P., Hurtt, G., and Pacala, S. W. (2001). A method for scaling vegetation dynamics: the ecosystem demography model (ED). Ecological Monographs, 71(4):557–586. Mora, C., Frazier, A. G., Longman, R. J., Dacks, R. S., Walton, M. M., Tong, E. J., Sanchez, J. J., Kaiser, L. R., Stender, Y. O., Anderson, J. M., et al. (2013). The projected timing of climate departure from recent variability. Nature, 502(7470):183–187. Morin, X., Viner, D., and Chuine, I. (2008). Tree species range shifts at a continental scale: new predictive insights from a process-based model. Journal of Ecology, 96(4):784–794. Niinemets, Ü. and Valladares, F. (2006). Tolerance to shade, drought, and waterlogging of temperate northern hemisphere trees and shrubs. Ecological Monographs, 76(4):521–547. Olson, D. M., Dinerstein, E., Wikramanayake, E. D., Burgess, N. D., Powell, G. V., Underwood, E. C., D’amico, J. A., Itoua, I., Strand, H. E., Morrison, J. C., et al. (2001). Terrestrial ecoregions of the world: A new map of life on earth a new global map of terrestrial ecoregions provides an innovative tool for conserving biodiversity. BioScience, 51(11):933–938. Pacala, S. W., Canham, C. D., Saponara, J., Silander Jr., J. A., Kobe, R. K., and Ribbens, E. (1996). Forest models defined by field measurements: estimation, error analysis and dynamics. Ecological Monographs, 66(1):1–43. Pacala, S. W., Canham, C. D., and Silander Jr, J. A. (1993). Forest models defined by field measurements: I. the design of a northeastern forest simulator. Canadian Journal of Forest Research, 23(10):1980–1988. Paine, C.E., Amissah, L., Auge, H., Baraloto, C., Baruffol, M., Bourland, N., Bruelheide, H., Daı̈nou, K., Gouvenain, R.C., Doucet, J.L. and Doust, S. (2015). Globally, functional traits are weak predictors of juvenile tree growth, and we do not know why. Journal of Ecology, 103(4):978-989. Pallardy, S. G. (2010). Physiology of woody plants. Academic Press. Pearson, R. G. and Dawson, T. P. (2003). Predicting the impacts of climate change on the distribution of species: are bioclimate envelope models useful? Global Ecology and Biogeography, 12(5):361–371. Poorter, L. and Markesteijn, L. (2008). Seedling traits determine drought tolerance of tropical tree species. Biotropica, 40(3):321–331. Prentice, I. C., Cramer, W., Harrison, S. P., Leemans, R., Monserud, R. A., and Solomon, A. M. (1992). A global biome model based on plant physiology and dominance, soil properties and climate. Journal of Biogeography, 19(2):117–134. Prism Group (2015). Prism Climate Group, Oregon State University. http://prism.oregonstate.edu. R Core Team (2014). R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria. Rehfeldt, G. E., Ferguson, D. E., and Crookston, N. L. (2009). Aspen, climate, and sudden decline in western USA. Forest Ecology and Management, 258(11):2353–2364. Reich, P., Wright, I., Cavender-Bares, J., Craine, J., Oleksyn, J., Westoby, M., and Walters, M. (2003). The evolution of plant functional variation: traits, spectra, and strategies. International Journal of Plant Sciences, 164(S3):S143–S164. Scheller, R. M. and Mladenoff, D. J. (2005). A spatially interactive simulation of climate change, harvesting, wind, and tree species migration and projected changes to forest composition and biomass in northern Wisconsin, USA. Global Change Biology, 11(2):307–321. PREPRINT OF GLOBAL CHANGE BIOLOGY, DOI : Scheller, R. M. and Mladenoff, D. J. (2007). An ecological classification of forest landscape simulation models: tools and strategies for understanding broad-scale forested ecosystems. Landscape Ecology, 22(4):491–505. Scherrer, D., Bader, M. K. F., and Körner, C. (2011). Drought-sensitivity ranking of deciduous tree species based on thermal imaging of forest canopies. Agricultural and forest meteorology, 151(12), 1632-1640. Schloss, C. A., Nunez, T. A., and Lawler, J. J. (2012). Dispersal will limit ability of mammals to track climate change in the western hemisphere. Proceedings of the National Academy of Sciences, 109(22):8606–8611. Scholl, A. E. and Taylor, A. H. (2010). Fire regimes, forest change, and selforganization in an old-growth mixed-conifer forest, Yosemite National Park, USA. Ecological Applications, 20(2):362–380. Scholze, M., Knorr, W., Arnell, N. W., and Prentice, I. C. (2006). A climate-change risk analysis for world ecosystems. Proceedings of the National Academy of Sciences, 103(35):13116–13120. Schrag, A. M., Bunn, A. G., and Graumlich, L. J. (2008). Influence of bioclimatic variables on tree-line conifer distribution in the greater Yellowstone ecosystem: implications for species of conservation concern. Journal of Biogeography, 35(4):698–710. Shugart, H. H. (1984). A theory of forest dynamics: The ecological implications of forest succession models. Sowell, J. B. (1985). A predictive model relating North American plant formations and climate. Vegetatio, 60(2):103–111. Stahl, U., Reu, B., and Wirth, C. (2014). Predicting species’ range limits from functional traits for the tree flora of north america. Proceedings of the National Academy of Sciences, 111(38):13739–13744. Stephenson, N. L. (1990). Climatic control of vegetation distribution: The role of the water balance. The American Naturalist, 135(5):649–670. Strigul, N. (2012). Individual-based models and scaling methods for ecological forestry: implications of tree phenotypic plasticity. In Garcia, J. and Casero, J., editors, Sustainable Forest Management, pages 359–384. InTech, Rijeka, Croatia. Strigul, N., Pristinski, D., Purves, D., Dushoff, J., and Pacala, S. (2008). Scaling from trees to forests: Tractable macroscopic equations for forest dynamics. Ecological Monographs, 78(4):523–545. Swenson, N. G., Enquist, B. J., Pither, J., Kerkhoff, A. J., Boyle, B., Weiser, M. D., Elser, J. J., Fagan, W. F., Forero-Montana, J., Fyllas, N., et al. (2012). The biogeography and filtering of woody plant functional diversity in North and South America. Global Ecology and Biogeography, 21(8):798–808. Thomas, C. D., Cameron, A., Green, R. E., Bakkenes, M., Beaumont, L. J., Collingham, Y. C., Erasmus, B. F., De Siqueira, M. F., Grainger, A., Hannah, L., et al. (2004). Extinction risk from climate change. Nature, 427(6970):145–148. Valladares, F. and Niinemets, Ü. (2008). Shade tolerance, a key plant feature of complex nature and consequences. Annual Review of Ecology, Evolution, and Systematics, 39(1):237–257. van Bodegom, P. M., Douma, J. C., and Verheijen, L. M. (2014). A fully traits-based approach to modeling global vegetation distribution. Proceedings of the National Academy of Sciences, 111(38):13733–13738. Walter, H. and Box, E. (1976). Global classification of natural terrestrial ecosystems. Vegetatio, 32(2):75–81. Westoby, M., Falster, D. S., Moles, A. T., Vesk, P. A., and Wright, I. J. (2002). Plant ecological strategies: some leading dimensions of variation between species. Annual Review of Ecology and Systematics, 33:125–159. Williams, J. W., Jackson, S. T., and Kutzbach, J. E. (2007). Projected distributions of novel and disappearing climates by 2100 AD. Proceedings of the National Academy of Sciences, 104(14):5738–5742. Woodward, F., Lomas, M., and Kelly, C. (2004). Global climate and the distribution of plant biomes. Philosophical Transactions of the Royal Society of London. Series B: Biological Sciences, 359(1450):1465–1476. Worrall, J. J., Rehfeldt, G. E., Hamann, A., Hogg, E. H., Marchetti, S. B., Michaelian, M., and Gray, L. K. (2013). Recent declines of Populus tremuloides in North America linked to climate. Forest Ecology and Management, 299:35–51. Woudenberg, S. W., Conkling, B. L., O’Connell, B. M., LaPoint, E. B., Turner, J. A., Waddell, K. L., et al. (2010). The forest inventory and analysis database: Database description and users manual version 4.0 for phase 2. Zang, C., Hartl-Meier, C., Dittmar, C., Rothe, A., and Menzel, A. (2014). Patterns of drought tolerance in major European temperate forest trees: climatic drivers and levels of variability. Global change biology, 20(12), 37673779. 10.1111/ GCB .13291 14