California forests show early indications of both range shifts and local persistence under climate change

Study region and study species

Our study region was the US portion of California Floristic Province (CAFP; Fig. S1 in Supporting Information), a hotspot of plant biodiversity in a Mediterranean climate region (Myers & Mittermeier, 2000). Long-term, regional changes in tree species distributions and abundances in this region have also been associated with changes in land use (Thorne et al., 2008) and fire regimes (Dolanc et al., 2014). We selected 13 species from the pool of tree species that are either endemic to (entire range within) or that have most of their range within the latitude and elevation of the CAFP, as well as the more widespread Abies concolor. This selection (Table 1) ensured that the analysis of global and local signals was based on the majority or entirety of the species range and that there were sufficient observations to implement the interpolation techniques used.

Forest sampling and climate data

We analysed geolocated field measurements from the Forest Inventory and Analysis (FIA) programme of the US Department of Agriculture (USDA) Forest Service (2008) to examine the distribution of juveniles (regeneration) versus conspecific adults in the CAFP. The FIA data yield a large, well-distributed sample of forest conditions (Appendix S1). A FIA plot consists of four circular, contiguous 168-m² macroplots where trees > 12.5 cm diameter are measured and four nested microplots where trees of 2.5–12.5 cm are measured; a plot fits within an area of c. 65 m × 65 m. Some of us (W.W.D., R.K.M.) had access to true plot locations – as opposed to coordinates intentionally altered for privacy (Food Security Act of 1985, 7 U.S.C. 2276(d)) – as part of ongoing studies of the spread and impacts of the forest disease sudden oak death (Rizzo et al., 2005; Meentemeyer et al., 2011). This high spatial precision allowed us to overlay FIA data with high-resolution downscaled climate data. We used data from 2001−07 for 7879 FIA plots in the CAFP (Appendix S1).

The FIA sampling procedure distinguishes different tree size classes. In this study, we defined regeneration as those plots with individuals < 12.7 cm diameter at breast height (d.b.h.) present; this approximately matched regeneration to establishment conditions in the second half of the 20th century (12.7 cm d.b.h. corresponded to 48.42 ± 7.37 years on average across our species; see Appendix S2 for size–age regressions), while trees > 12.7 cm d.b.h. are likely to have established under prior conditions. This approach was used to detect effects of climate change and land-use change (e.g. fire management, logging practices) where changes occurred or accelerated in the second half of the 20th century (e.g. Keeley et al., 1999; Boisvenue & Running, 2006).

We obtained estimates of mean annual temperature (MAT) and precipitation (PRECIP) within the CAFP for historical climate (1950−2000) from the PRISM dataset (Daly et al., 1994) downscaled to 90 m using a spatial gradient and inverse distance squared weighting approach (Flint & Flint, 2012; see Franklin et al., 2013, for a full description of the dataset and methods). These downscaled values were extracted for the true FIA plot locations.

Global indicators: regeneration distribution in climate and geographical space

We defined forest regeneration for a given species as the presence of any juvenile size class (< 12.7 cm d.b.h.) along with conspecific adults in a FIA plot. Conversely, a plot with adults present but juveniles absent would be considered non-regenerating forest. By only considering regeneration where there were adults in the overstorey we ensured that a seed pool was present and that regeneration was not confounded with range shifts dependent on processes such as dispersal. While dispersal-driven range shifts are also important, the focus of our study was forest persistence. We applied a framework originally developed to characterize niche dynamics in space and time for invasive species (Guisan et al., 2014) to investigate displacement in forest regeneration in environmental and geographical space. We estimated regeneration dynamics using the techniques provided by Broennimann et al. (2012) for niche estimation and niche overlap, and the methodology developed by Petitpierre et al. (2012) to identify niche dynamics. All analyses were conducted using R (R Development Core Team, 2014).

Estimating global regeneration dynamics: persistence, expansion and loss

Regeneration and non-regeneration density of occurrence was estimated in a two-step process (full methods and code are available at Broennimann et al., 2012). First, forest plots were placed in 1000 × 1000 gridded bidimensional spaces defining either the climate (MAT, °C; PRECIP, mm) or the geographical space (ELEVATION, m; LAT_Transformed, a unit-less transformation of latitude derived to avoid publishing true plot locations). Climate niche space provided information on regeneration displacement along climate gradients, whereas geographical space helped evaluate geographical indicators expected from a warming fingerprint (higher elevation and latitude). Mean annual temperature and precipitation have been used in previous analysis of the turnover hypothesis (Zhu et al., 2014) and are highly correlated with the group of 19 bioclimatic variables (Appendix S2) widely used in species distribution studies (Booth et al., 2014). Second, kernel density estimation was applied to the gridded environmental space using forest plot occurrences. The result was a normalized density distribution in environmental space (Fig. 2a, b). These steps were carried out for regenerating and non-regenerating forests. This approach enabled direct comparison of densities without depending on different sample sizes (see Broennimann et al., 2012, for an extended methodological explanation).

Local indicators

Are there signals of regeneration displacement?

We detected differences between forests with and without regeneration in both climate and geographical space (Fig. 3a), although the main signal was the overlap in both. Regeneration expansion was as prevalent as loss in climate space; whereas regeneration loss was clearly more prevalent in geographical space. These results were somewhat surprising because we expected high levels of regeneration loss in both climate and geographical spaces given that lagged tree responses to changes in climate are expected to lead to an extinction debt (Bertrand et al., 2011). Our results suggest that higher-elevation species are experiencing larger loss than lower-elevation species, especially in geographical space (Fig. 3b). Nevertheless, the dominance of regeneration persistence, in both climate and geographical dimensions, may be indicative of buffering mechanisms to climate change that may slow rates of species distribution shifts (Lloret et al. 2012; Dobrowski et al., 2015). However, this overlap could also indicate that other factors may have had a recent, strong influence on regeneration dynamics (see below).

Generally, regeneration dynamics in both geographical and climate space agreed, but some species did exhibit opposite dynamics (Fig. 3b). The regeneration dynamics for these cases showed that regeneration expansion in climate space exceeded regeneration loss in geographical space (e.g. tanoak, grey pine, red fir). Our interpretation is that these species are occupying new portions of climate space while their geographical range shrinks. This distinction is important, because it warns of simultaneous range reduction and shifting, unlike an overall range reduction that would tend to diminish both climate and geographical space or a range shift that would imply comparable regeneration expansions in climate and geographical space. While examining a more complete geographical space (latitude, longitude and elevation) might further disentangle these dynamics, our results highlight the importance of analysing both climate and geography. In areas of complex terrain, for example mountain ranges oriented in different directions with regard to prevailing winds and a strong rainshadow effect, typical correlations of climate and geography can be decoupled. For example, Loarie et al. (2008) shown that in California movement to higher elevations often means taking a southward path.

It would be misleading to interpret our results of large areas of regeneration persistence as meaning high levels of regeneration in forests (Table 1). The methods used here aim to describe shifts in regeneration niche. For example, coast live oak shows a large regeneration expansion index but fewer than half of the total plots recorded any regeneration in the last c. 50 years (Table 1).

Are global indicators of forest regeneration consistent with a climate change fingerprint?

We found slightly more evidence for the turnover (6 species of 13 studied, 46%) than the climate change migration (five species, 39%) hypothesis. Similarly, mixed results were also found by Zhu et al. (2014), although they found even higher support for the turnover hypothesis in eastern forests of the United States (77%). Nevertheless, three of our studied species did not clearly fall into either of the two categories since they were regenerating at higher temperatures but lower precipitation, thereby making a straightforward attribution to higher turnover rates difficult. We argue, however, that these species could also be included in the turnover hypothesis because TWI analysis suggested higher potential availability of water via topographically mediated hydrology (Fig. 1c).

Our results concur with findings from other studies that for some species juveniles seem to be concentrating at lower elevations (Lenoir et al., 2010; Crimmins et al., 2011; Rabasa et al., 2013). Regeneration tended to occur at lower elevations for the majority of our study species as well (9 out of 14, 65%; Fig. 4a). Our analysis of TWI revealed that counter-intuitive responses in regeneration, such as shifts to lower-elevation warmer and/or drier macroclimatic conditions could be compensated via topoclimatic effects on temperature such as cold air pooling and topographic shading (Dingman et al., 2013; Curtis et al., 2014) and/or water balance effects (Lenoir et al., 2010; Crimmins et al., 2011).

For the interior western United States, with its water-limited semi-arid to arid climates, Bell et al. (2014) reported regeneration loss at the northernmost part of their study area. Our results concur with this finding because the majority of our species were also showing more regeneration at lower latitudes (Fig. 4b). Our results partially agree with Monleon & Lintz (2015) for those species that were analysed in both studies. On the other hand, our studies disagree with respect to California bay laurel, where they predicted regeneration at lower temperatures and we detected regeneration at higher temperatures. In addition, previous work from Lenihan et al. 2008 and Dobrowski et al. (2015) projected expansion of hardwoods of the mixed hardwood conifer belt (tanoak, California black oak, bay laurel, canyon live oak and interior live oak) under climate change – species we found to be regenerating at higher elevations. Nevertheless, direct comparisons are difficult due to different methods used. We restricted our analysis to regeneration in the presence of adults and we used a different characterization for juveniles.

All in all, results show that a single common signal of range dynamics in a broad-scale analysis is largely absent (Rabasa et al., 2013). Such species-specific patterns were also found by Monleon & Lintz (2015), although their signal of regeneration at colder temperatures is more evident for the northernmost species of the US Pacific coast. Interestingly, some of the tree species that have historically occurred together (coast live oak–tanoak; incense cedar–sugar pine–black oak; blue oak–grey pine) are showing distributional shifts in opposite directions, suggesting the potential for reorganization of biota into novel assemblages, in agreement with the shifts in vegetation composition during the 20th century that have been described in the region (Thorne et al., 2008; Dolanc et al., 2014). Indeed, expecting a common fingerprint when rates of change in climate are heterogeneous (Flint & Flint, 2012; Dobrowski et al., 2013) may be a rather oversimplified hypothesis (VanDerWal et al., 2012) and potentially misleading for management purposes at the species-specific level. Recent analyses of California plants and animals also suggests individualistic upslope and downslope shifts in species elevation in response to climate change (Rapacciuolo et al., 2014), and Serra-Diaz et al. (2014) projected different bioclimatic velocities – the pace at which species and populations are exposed to climate change – among some of the species of this study.

Local indicators and the heterogeneity of regeneration

The existence of regeneration hotspots reveals a complex regeneration landscape across current species ranges and may add information to the signal of the range-wide global dynamics. Such multimodal behaviour has also been identified in patterns of seedling density and adult density (Fig. 6 in Monleon & Lintz, 2015; to a lesser extent in Figs 2 & 3 in Rabasa et al., 2013). Indeed, the complex landscape of regeneration patterns may be due to many drivers, including legacy effects (Nowacki & Abrams, 2014), land-use dynamics (Syphard et al., 2011; Dolanc et al., 2014), change in successional trajectories resulting from altered severity and frequency of disturbance (Syphard et al., 2009), and local adaptation (Sork et al., 2010). Also, spatially structured biotic interactions can overshadow the fingerprint of climate change and have a strong effect on species distributions (Lenoir et al., 2009; Serra-Diaz et al., 2015). For example, Gworek et al. (2007) found a significant effect of seed-caching rodents on survival of Jeffrey pine seedlings, obscuring a clear pattern of species migrating northwards, which is consistent with the results of this study. Overall, species-specific abilities to track climate change are diverse, and therefore the main drivers may shift and interact depending on the context (Bertrand et al., 2011; McLaughlin & Zavaleta, 2013). Nevertheless, we found that disturbance and silvicultural treatments were related to the number of local regeneration hotspots (Table 2), emphasizing that we need to better understand the synergies between human-mediated natural disturbances, land management and climate change in forested landscapes.

Our methods should be understood as a first step in picturing the varying local responses of regeneration. Indeed, the study of local maxima and minima using kernel density estimation of regeneration allowed us to detect nuanced differences between populations within species ranges. However, further work is necessary to distinguish signal and noise in regeneration using such techniques. For instance, it is possible that regeneration was not detected in the plots. In addition, spurious effects may be found at the edges of the distribution, which are also are very interesting locations in terms of climate change. Improved methods for detecting regeneration hotspots could also benefit from the potential connections between regeneration hotspots by dispersal. Incorporating more detailed demographic variables for hotspot detection (such as basal area or recruitment intensity as in Zhu et al., 2014) could help identify drivers along different gradients, especially if interactions between soil and climate variability were included in the analysis (Bertrand et al., 2012; Ibáñez et al., 2014). Land managers could further benefit from hotspot analyses carried out in latitude–longitude space so that regeneration hotspots could be identified explicitly (see Appendix S3).

Arguably, analysis of the distribution of regeneration hotspots reflects the distribution of the plots, the orientation of predominant mountain ranges, the species ranges and other attributes of their distributions. We found no systematic bias in the methods since there is a low correlation between the number of plots, regeneration, regeneration dynamics and local regeneration hotspots, and one would expect that any bias in the distribution of plots should be similar for both regeneration and no-regeneration samples.

Deepening our understanding of early signals: the timing and the drivers

Both regeneration dynamics and the relationship between adults and juveniles may provide an indicator of species range resilience or shift (Lenoir et al., 2009; Zhu et al., 2012; Woodall et al., 2013; Bell et al., 2014; Monleon & Lintz, 2015), but several limitations of our approach should be acknowledged explicitly. First, the age of juveniles was identified based on a size–age relationship (< 12.7 cm d.b.h. in our case). Presumably, the time spent in the juvenile size class depends on species and site productivity, and thus we may expect smaller individuals classified as juveniles to be more prevalent in drier sites than wetter sites. Also, most studies classify juveniles using the same criterion across species, despite different species having different growth rates. Instead, Monleon & Lintz (2015) applied a statistical threshold to separate seedlings and adults. We chose a threshold that could generally distinguish individuals that established during the last 50 years of rapid climate change; but making the link between the exact age of juveniles and microclimatic site conditions remains a challenge.

For this study we adapted relatively simple, although extensively used, climate descriptors, namely mean annual temperature and mean annual precipitation (see Garcia et al., 2014, for further dimensions of climate change). Patterns of regeneration are more proximally driven by bioclimatic variables other than average temperature and precipitation, for example water deficit, seasonality of temperature and moisture (Flint & Flint, 2012). We envision that future work will test the overlap of juveniles and adults using multiple variables related to climate change (see Blonder et al., 2014, for an n-dimensional technique), including extreme events. Indeed, the large overlap of regeneration and non-regeneration spaces suggests that other factors shape regeneration dynamics, but could also point to relatively stability of the range. Interestingly, Benavides et al. (2015) found trade-offs between survival and growth of juveniles for several species: recruits were abundant in the upper elevations of the range but had higher growth rates at lower elevations. This apparent trade-off could eventually stabilize species ranges (Benavides et al., 2015).

Our results suggests that signals of regeneration shifts are present but not widespread, and importantly the directions of such changes are species specific and do not always concur with the upslope or northward migration expected due to recent warming. Regeneration distributions for tree species also show a direct anthropogenic signal of forest management and disturbance. Conservation managers are likely to need to be able to adapt large-scale mitigation plans to observed local-scale patterns rather than assuming a uniform pattern of how climate change may affect species biogeography. The results of this study suggest that macroecology in the Anthropocene will have to synthesize spatially varying drivers of those factors shaping the future distribution of species, rather than using a global fit to detect range change.