When a foundation crumbles: forecasting forest community dynamics following the decline of the foundation species Tsuga canadensis

In the forests of northeastern North America, invasive insects and pathogens are causing major declines in some tree species and a subsequent reorganization of associated forest communities. Using observations and experiments to investigate the consequences of such declines are hampered because trees are long-lived. Simulation models can provide a means to forecast possible futures based on different scenarios of tree species decline, death, and removal. Such modeling is particularly urgent for species such as eastern hemlock (Tsuga canadensis), a foundation species in many northeast forest regions that is declining due to the hemlock woolly adelgid (Adelges tsugae). Here, we used an individual-based forest simulator, SORTIE-ND, to forecast changes in forest communities in central Massachusetts over the next 200 years under a range of scenarios: a no-adelgid, status-quo scenario; partial resistance of hemlock to the adelgid; adelgid irruption and total hemlock decline over 25 years, adelgid irruption and salvage logging of hemlock trees; and two scenarios of preemptive logging of hemlock and hemlock/white pine.We applied the model to six study plots comprising a range of initial species mixtures, abundances, and levels of hemlock dominance. Simulations indicated that eastern white pine, and to a lesser extent black birch and American beech, would gain most in relative abundance and basal area following hemlock decline. The relative dominance of these species depended on initial conditions and the amount of hemlock mortality, and their combined effect on neighborhood-scale community dynamics. Simulated outcomes were little different whether hemlock died out gradually due to the adelgid or disappeared rapidly following logging. However, if eastern hemlock were to become partially resistant to the adelgid, hemlock would be able to retain its dominance despite substantial losses of basal area. Our modeling highlights the complexities associated with secondary forest succession due to ongoing hemlock decline and loss. We emphasize the need both for a precautionary approach in deciding between management intervention or simply doing nothing in these declining hemlock forests, and for clear aims and understanding regarding desired community- and ecosystem-level outcomes.


INTRODUCTION
The functions that determine recruitment of tree seedlings in SORTIE-ND are derived 142 from field data describing seed production, dispersal, and establishment in a variety of soils species-specific light transmission coefficients (Canham 1988, Canham et al. 1994, 1999. 164 Species-specific seedling mortality was modeled as a function of the recent growth of each 165 seeding, its shade tolerance, and within-cell light availability (Kobe et al. 1995). We used 166 seedling growth and mortality parameters provided by C. Canham (personal communication) 167 based on previous studies of New England forests (e.g., Pacala et al. 1993, Pacala et al. 1996. 168 The growth and mortality of saplings and adults were modeled using neighborhood competition hemlock control plots numbers 3 and 6 (hereafter, "HeRE3-hemlock" and "HeRE6-hemlock") 185 and the two experimentally logged plots, plots 2 and 4 (hereafter, "HeRE2-logged" and "HeRE4-   (Table 2). The first scenario, using the core parameterization of SORTIE 196 described above, included only background levels of natural mortality of eastern hemlock.

201
For initial inputs to the model under all six scenarios, and to encompass a range of initial 202 conditions typical of forests of this region, we used data from a total of six study plots (Fig. 1).

203
The first three plots were from the aforementioned HF Hemlock Removal Experiment, including 204 the two HF-HeRE hemlock control plots used for model evaluation (HeRE3-hemlock and 205 HeRE6-hemlock) and one hardwood control plot (hereafter, "HeRE7-hardwood"). These plots network of plots. For our modeling, we used data from the full ForestGEO plot (hereafter, "FG-212 full") and two, 2-ha sub-regions of the plot that currently are dominated either by eastern 213 hemlock ("FG-hemlock") or red oak ("FG-hardwood").

214
In all simulations, we initialized the models using data for the ten tree species that 215 comprised > 95% of the total basal area of the plots (Table 3). Initial model input data for all 216 plots comprised species identities, sizes (DBH) and x-y locations (tree maps) of field-measured 217 trees. For the HF-HeRE plots, all trees were measured and located for individuals > 5 cm DBH 218 and 1.3 m in height; saplings (DBH < 5 cm and height < 1.3 m) were counted but not located; at 219 the HF-ForestGEO plots, all trees and saplings with DBH > 1-cm and height > 1.3-m were 220 measured and located. For the HF-HeRE plots, we used means of ten years of seedling plot data 221 as initial inputs (Table 4), while for the ForestGEO plots, we used a common seedling density of 250 seedlings per hectare for all species because seedling data were unavailable. Raw data for all Stochastic behavior in SORTIE-ND results from differences in initial conditions and 246 from using random draws from specified distributions for dispersal, growth, and mortality 247 parameters. We used field data from six plots as our initial conditions and did not vary these 248 among runs. However, comparisons of simulations initialized by different plots did allow us to 249 examine how variation in initial community structure could result in different outcomes. We did 250 use random draws from probability distributions for dispersal, growth, and mortality parameters.

251
For each plot × scenario combination, we ran 10 stochastic simulations.

287
Under a scenario of partial hemlock resistance to adelgid impacts (Scenario 2), SORTIE 288 simulated a variety of impacts on species composition and structure that depended largely on the 289 initial composition of the plot. In the hemlock-dominant plots, American beech and black birch 290 recruited as saplings to high densities into the gaps created by dying hemlock trees (Fig. 7), while hemlock concomitantly retained its overall dominance and eventually began to recover 292 over time. The effect of this is that by year 200, there was an overall dampening of the size 293 structure of these plots and a reduction of up to one-half the original total basal area (Fig. 6). By 294 contrast, in the two hardwood-dominated plots, existing adult white pine and yellow birch also 295 took advantage of these gaps, increasing their dominance in the canopy over time, resulting in 296 only minor changes in the size structure of these plots relative to starting profiles (Figs. 7 and 6).  Under a scenario of both hemlock and merchantable pine logging by year 25 (Scenario 313 6), differences in simulated community trajectories among the six plots depended on which species took most advantage of the increased light created by tree removal, and whether or not 315 white pine was able to recover and gain dominance over time (Fig. 11). Across the full FG plot 316 and within the FG-Hardwood subplot, American beech and yellow birch increased considerably 317 in abundance in the sapling and adult tiers, respectively, preventing white pine from recovering 318 in these plots. In the two HeRE hemlock plots (3 and 6), black birch became the most abundant 319 species, although the white pine individuals remaining after logging were able to grow and 320 become dominant in terms of basal area by the end of the simulation. Forecasts -changes in diameter growth rates upon canopy removal 323 We examined projected species' growth rates before (year 10) and shortly after (year 17) 324 preemptive hemlock logging in Scenario 5 simulations to determine how the different species 325 would respond to major increases in light and space after disturbance. These results indicated 326 that all species would experience an increase in diameter growth upon the creation of canopy 327 openings, at all growth stages (Fig. 12). However, growth rates, and the relative increases in 328 growth after hemlock removal, differed significantly by species and study plot. White pine and 329 red oak had the highest adult and sapling growth rates at the two times and across all plots; black 330 birch also showed noticeable increases in year-17 growth rates in the two HRE-hemlock plots. hardwood-dominated plots, white pine will increase as the oak and red maple components of 363 these stands mature and decline naturally over time.

364
Simulations also suggest that black birch would be competitive with white pine in 365 hemlock gaps, but that its eventual relative abundance depends on some threshold of initial 366 abundance and adequate seed production, as well as enough hemlock removed to provide ample 367 light conditions for this relatively light-demanding species. This was particularly evident in the 368 HeRE3 and HeRE6 hemlock plots, where black birch seedlings had relatively higher growth 369 rates after hemlock harvesting than white pine (e.g., Fig. 12), but where slight differences in 370 initial relative abundances and canopy gap sizes in these plots likely mediated the ability for 371 black birch to take advantage of these better growth rates. In other locations in northeastern and define a new future reality for some stands in our study areas over the longer term.

394
On the whole, our results indicate that community trajectories and future forest structure 395 will be qualitatively similar whether hemlock dies slowly from the adelgid, or is removed rapidly             Table 2). For observed data, values are total basal areas per dbh class; for