1We studied a cliff-face forest ecosystem dominated by a single long-lived tree species that has been previously shown to have slow, pulsed recruitment. We assessed the degree to which microsite and climatic variability over a long period of time control recruitment, morbidity and mortality of trees at a previously disturbed cliff site.
2We sampled more than 2000 Thuja occidentalis (eastern white cedar) over an 18-year period using a series of dynamic cohorts. We also examined a smaller area more intensively for 9 years.
3Microsite and climate both played a role in controlling emergence and survival. Seedlings emerged preferentially in horizontal microsites such as large ledges and shelves but survival there was poor, whereas crevices and smaller ledges had lower emergence but the best survival. Decaying logs, cliff edges, vertical cliff faces and the smallest ledges proved unsuitable for seedling recruitment. Very few seedlings survived for more than 5 years.
4While spring and summer climate influenced emergence and early survival, climate effects decreased with increasing plant size and mortality at the later stages of recruitment was independent of climate.
5Drought and pathogens were the most common causes of mortality in horizontal habitats, while drought and rockfall were important in vertical habitats.
6There appear to be a finite number of safe sites on cliff faces, and recruitment to those sites limits the demographic changes in tree populations over time.
7Long-term studies on long-lived species have the value of sorting real, but unimportant, short-term variation in plant response to climate and site conditions from the long-term trends that are principally responsible for moulding the structure of the ecosystem.
There is general agreement among plant ecologists that seedling recruitment is the most critically important stage for understanding the dynamic assembly of ecosystems. This is because the final configuration of the community is controlled by safe sites and barriers to survival that are species-specific and variable over space and time (Harper et al. 1965; Grubb 1977; Weiher & Keddy 1999; Cornett et al. 2000; Castro et al. 2004). Multiple requirements need to be met for a microsite to be considered a safe site, including the existence of suitable conditions for germination, favourable conditions for early growth and an absence of consumers or competitors that would prevent a seedling from reaching adulthood.
Despite this general agreement, the establishment phase remains one of the least understood and most often ignored components of forest dynamics (Jones & Sharitz 1998). In practical terms this is because ecosystems normally have a large number of species, each of which is filtered differently over time and space in a way that is challenging to follow. The ‘species’ and ‘space’ problems can be dealt with by undertaking very extensive sampling, but the ‘time’ component can only be addressed by costly long-term studies. The rarity of such studies is unfortunate because there is clear evidence that safe sites in stable ecosystems, especially forests, appear and disappear with enormous temporal variability (Boerner & Brinkman 1996; Jones & Sharitz 1998) and the limited duration of most studies provides only a small window of observation that may not be representative of long-term average conditions.
Sometimes, however, natural ecosystems offer opportunities to ask otherwise intractable questions. The limestone cliffs of the Niagara Escarpment, southern Ontario, Canada, support a stunted forest composed of a single tree species, Thuja occidentalis L. (eastern white cedar). This forest is the most ancient and least disturbed in eastern North America, and shows demographic stability (the same negative exponential age distribution for at least 200 years, Kelly & Larson 1997). Recruitment of new trees occurs in pulses separated by as much as 60 years (Larson & Kelly 1991), possibly explained by the ‘cork in the bottle’ hypothesis of Davis (1951). This hypothesis proposes that microsites suitable for colonization are claimed by slow-growing, long-lived organisms that exclude new colonists by their sheer physical presence. New colonists can only arrive when there is a physical disturbance that removes the residents.
The time-scale of replacements is long, meaning that testing the hypothesis on natural cliffs is almost impossible, but limestone extraction all along the Niagara Escarpment has created open cliff faces in quarries with abandonment dates from the present back to the mid-1850s. The younger quarries represent locations where the processes of early recruitment and mortality can be monitored over practical time frames so that the importance of spatial and temporally variable factors can be isolated. Quarry operations initiate primary succession by removing all biomass and natural substrate. The heterogenous rock surface that is left behind is recolonized by cryptogams, herbaceous higher plants and, finally, the tree T. occidentalis, resulting in rock faces indistinguishable from the natural escarpment within approximately 70 years (Ursic et al. 1997). We report the results of a long-term dynamic analysis of recruitment patterns of Thuja occidentalis on cliffs in an abandoned quarry along the Niagara Escarpment to determine the degree to which microsite and climatic variability over a long (18-year) period of time act to recruit, cause morbidity and cause mortality in cedar trees that eventually form the cliff-face forest.
The vertical cliffs at the study site, a quarry near Guelph, Ontario, Canada (43°33′ N, 80°13′ W), have remained undisturbed since quarry operations ceased in the mid-1940s and have undergone natural revegetation typical of cliffs in the region (Ursic et al. 1997). The age of the oldest trees at the site puts the initiation of the study at about 40 years after the arrival of the first colonists. We have used this site in previous studies of cliff ecology because in many important ways it is typical of the rest of the Niagara Escarpment ecosystem (Matthes-Sears & Larson 1995; Matthes-Sears et al. 1995; Booth & Larson 1998; Booth & Larson 2000a,b).
A longer-term and larger-scale study took place over 18 years (1986–2003) while a shorter-term and smaller-scale study was conducted for 9 years (1995–2003). For the 18-year study, all T. occidentalis growing within a 25 m wide section of cliff face were tagged in the autumn of 1986 and their heights were measured. The microsite of each plant was characterized in terms of orientation only (vertical, horizontal or cliff base; Table 1) because rockfall caused finer-scale characteristics to change over the course of the long-term study. All individuals were re-located in autumn 1987, autumn 1988 and each spring (early May) and each autumn (late October) from 1989 through to 2003. Plants were classified as healthy or unhealthy based on foliage colour. Mortality and its most likely cause, if evident, were recorded. Any newly emerged seedlings were tagged and measured. All living trees were remeasured at approximately 5-year intervals.
Table 1. Summary of microsite and climate variables used as explanatory variables for emergence and survival in the 18-year and 9-year studies
Categories were combined for the survival analyses.
Categories were omitted from the survival analyses.
Categories were combined for the survival analyses.
A. Microsite variables
Categories and description
Vertical (cliff face including ledges < 30 cm wide)a
Horizontal (shelves and large ledges wider than 30 cm)
Cliff base (horizontal microsite within 1 cm of a vertical rock face)
Loose rocks or gravel
Cedar or mixed cedar/deciduous
Cliff base (horizontal microsite within 1 cm of a vertical rock face)a
Large ledge (largest dimension > 10 cm, but width not exceeding 30 cm)
Horizontal shelf (width exceeding 30 cm)
Cliff edge (horizontal microsite within 5 cm of the top of a vertical rock face)
Wooden log (surface of woody debris on horizontal shelves)b
B. Climate variables
Time scale over which variable was calculated
Emergence, 18-year study
1. Total precipitation (mm)
Sum for March–May of each year
2. Mean maximum temperature (°C)
Average for March–May of each year
3. Mean minimum temperature (°C)
Average for March–May of each year
4. Extreme maximum temperature (°C)
Average for March–May of each year
5. Extreme minimum temperature (°C)
Average for March–May of each year
6. Total ppt of previous summer (mm)
Sum for March–October of previous year
7. Mean temperature of previous summer (°C)
Average for May–October of previous year
8. Mean temperature of previous winter (°C)
Average for November of previous year through April of current year
Survival, 18-year study
1. Total precipitation (mm)
Sum for 6-month period preceding each census
2. Mean max. temperature (°C)
Average for 6-month period preceding each census
3 Mean min. temperature (°C)
Average for 6-month period preceding each census
4. Extreme max. temperature (°C)
Average for 6-month period preceding each census
5. Extreme min. temperature (°C)
Average for 6-month period preceding each census
Emergence and survival, 9-year study
1. Deviation of total precipitation from normal (mm)
Sum for 7 days preceding each census date
2. Deviation of mean maximum temperature from normal (°C)
Average for 7 days preceding each census date
3. Deviation of mean minimum temperature from normal (°C)
Average for 7 days preceding each census date
The 9-year study was conducted to allow closer examination of the fate of new seedlings, many of which were missed in the larger-scale study because they emerged after the spring census and died before the autumn census. The second goal was to investigate in greater detail the effects of microsite and short-term climatic fluctuations on seedling emergence and first-year mortality. Ten vertical transects were laid out in random positions across the cliff face and permanently marked using paint. Transects were 30 cm wide and 6 m to 15 m long. The proportions of different microhabitats present were estimated by randomly selecting 35 locations along each of the 10 transects. An area 10 cm in diameter in the centre of the transect at each random location was characterized with respect to five physical microsite variables and assigned to one of nine microsite types (Table 1). Proportions were calculated as the number of random locations in each category divided by the 350 samples taken. The number of microsite classes was later reduced to six for the survival analyses due to low seedling numbers in some of the categories.
The fate of three cohorts of seedlings (1995–97) that emerged within the transects was followed from 1995 until 2003. Seedlings were surveyed weekly (1996 and 1997) or every 3 weeks (1995) between May and October of their first growing season, and every spring and autumn thereafter until 2003. The microsite of each seedling, defined as an area 10 cm in diameter surrounding it, was characterized at the time of emergence using the six microsite variables listed in Table 1. Changes in seedling microsite characteristics over the short-term study were exceedingly rare. Seedling death was assigned to the most likely cause if one was evident. Drought was assumed to be the cause if a seedling turned brown and dried out, while seedlings that fell over suddenly while still green and while soil moisture was available were assumed to have succumbed to pathogens.
Daily minimum and maximum temperatures, daily precipitation and the corresponding 31-year (1971–2002) daily climate normals were obtained from the nearest Environment Canada weather station (Waterloo-Wellington, located approximately 18 km south-west of the study site). These data were used to generate climate variables on two different scales to be used as predictors of seedling demography in the 18-year and 9-year studies (Table 1).
For annual emergence in the 18-year study, five climate variables characterizing the main germination period (March–May) were derived for each year from 1989 through to 2003. In addition, three variables characterizing the previous year's climate were used. For seedling survival in the 18-year study, values of five climate variables were calculated for the 6-month period (summer, May–October; and winter, November–April) preceding each of the two annual census dates.
Emergence and survival in the 9-year study were analysed using three short-term predictor variables representing the conditions of the 7 days preceding each census date (Table 1). The deviation of each climate variable from its corresponding normal was used to eliminate the confounding effect of seasonal climate fluctuations.
The survivorship data from the 18-year study, and both emergence and survivorship data from the 9-year study, were analysed using Cox regression (Allison 1995; Fox 2001) as implemented by PROC PHREG in SAS (version 9.1; SAS Institute 1999). This technique, a semi-parametric regression that uses the maximum likelihood estimation method, has been specifically developed to test the effects of covariables on time-to-event data (such as emergence or survival data) and is designed to make use of censored observations (i.e. observations where the event of interest, e.g. death, has not occurred by the end of the observation period). Covariables may include those whose values for an observation are fixed (in our case, microsite), as well as those whose values change over time (climate variables). Microsite-specific effects of climate were investigated by including interaction terms between climate and microsite variables.
For the 18-year study, the survival of all newly emerged seedlings (n = 2075) was analysed using the spring of each seedling's emergence as the time origin. One microsite variable (horizontal vs. vertical), five climate variables (Table 1) and a dummy variable marking each census period as either winter or summer were used as explanatory variables. Three-way interaction terms between these variables were used in the Cox regression model to investigate season- and microsite-specific effects of climate. To determine if there was a relationship between survival and growth rate and whether this relationship differed with microsite orientation, the analysis was then repeated after including interaction terms between microsite and an additional explanatory variable, relative growth rate (growth increment/height). In order for growth rate to be available, plants had to have survived at least two consecutive height measurements. For this reason, the subset of 305 individuals used in the second analysis had a median age of 8 years at death, compared with < 1 year in the original data set, and the analysis therefore examines survival at a slightly later life stage.
A third Cox regression was then performed on the survival of all trees present at the first census, using the covariables microsite and initial height in 1986 (four height classes: 1–5 cm, 6–25 cm, 26–100 cm, and > 100 cm). Climate could not be included in this analysis as all trees were exposed to the same climatic conditions.
For the 9-year study, both the emergence and first-year survival of seedlings emerging between 1995 and 1997 were analysed. The beginning of the year of emergence was used as the time origin for each seedling. Two parallel Cox regressions were performed for each response (emergence and survival). First, the combined effects of microsite (six categories, Table 1) and three climate variables (Table 1) were investigated. Secondly, the effects of the five physical microsite variables (litter, soil, rocks, stability and canopy type; Table 1) were tested alone. While the physical variables are likely to be correlated among themselves, testing for the effect of each independently was not the goal of this study.
The number of seedlings emerging each year of the 18-year study could not be analysed using Cox regression due to the lack of an appropriate zero point for the time variable. Instead, all possible subsets linear regression (Neter et al. 1983) was used to identify the combination of climate variables that best predicted annual seedling numbers. The adjusted R2 was used as the criterion to select the best model. A logarithmic transformation was applied to the dependent variable before the selection procedure, and residuals were examined to confirm that the assumptions of parametric analysis were met.
Chi-square tests were used to compare the frequency distribution of microhabitats where seedlings emerged in the 9-year study with the frequency distribution of available microhabitat types in the transects. They were also used to compare the frequencies of different types of morbidity among microsites.
The 2245 T. occidentalis trees surveyed were almost evenly distributed between vertical (48.8%) and horizontal (45.7%) microsites, with the remaining 5.6% growing at the cliff base. At the first census (autumn 1986), 321 plants were present in the study area, of which 109 survived through the entire study period while 212 died. The number of plants alive at any one census ranged from 248 (autumn 1988) to 891 (spring 1993), with 259 plants alive at the last census (autumn 2003). Despite this massive pulsation of total population size, there were only minor changes in the size distribution of all living plants at the site over the 18 years of the study (Fig. 1). These changes were restricted to the youngest age class, which fluctuated with the size of previous recruitment pulses but always remained dominant. The microsite composition of the different age classes was also unchanged between 1986 and 2002. While the smallest height classes were heavily dominated by plants in vertical microsites, the ‘tail’ of the size distribution (200 cm and up) consisted almost exclusively of plants in horizontal microsites. The vast majority (96.3%) of all plants surveyed were < 50 cm in height and grew < 5 cm a year, and most (92.2%) were < 20 cm tall with growth rates < 2 cm year−1 (Fig. 2). The tallest trees were not necessarily the fastest growing, and trees with the highest growth rates spanned a wide range in size. Trees in vertical and horizontal microsites had different distributions of growth rates and sizes: while most cliff-face trees were small and slow-growing, trees that occupied horizontal microsites were more evenly distributed among smaller and larger growth rates and sizes (Fig. 2). The largest trees in the study were all growing on horizontal surfaces while the cliff face supported very few large trees. In contrast, many more trees < 50 cm tall existed on the cliff face than on level ground (Fig. 2).
The emergence of 2075 seedlings was recorded over the 18 years of the study. Recruitment pulses alternated with years of little or no emergence and were about evenly distributed over the duration of the study (four occurring in the first half and four in the second; Fig. 3a). However, the three highest pulses occurred within the first 8 years and, as a result, 84% of all emergent seedlings were recorded in the first half of the study and only 16% in the second.
Over 37% of all seedlings emerged on the cliff face in vertical microhabitats that included rock faces, crevices and small ledges. Five per cent emerged at the cliff base and the remaining 57% on horizontal surfaces, including horizontal shelves and large ledges (Fig. 3a).
The number of seedlings emerging each spring of the 18-year study was best predicted by a combination of three annual climate variables: the total amount of precipitation during the germination period, the mean maximum temperature during the germination period, and the mean temperature of the previous summer (all-subsets linear regression, P = 0.028, F = 4.46, d.f. = 3,11, ), with emergence favoured by lower values of all three variables.
Mortality was generally higher during the summer than over the winter but fluctuated widely from year to year (Fig. 3b, c). These patterns were driven by the high mortality rate of newly emerged seedlings. High emergence in the spring was always followed by high mortality in the summer and the following winter, with newly emerged seedlings making up the majority of deaths (Fig. 3b).
Survival analysis of newly germinated seedlings showed significant microsite- and season-specific effects of climate (Table 2). In the case of microsite, the large estimated coefficients suggest a greater biological significance of this factor than indicated by the low or marginal statistical significance. In the summer, high precipitation and high minimum temperatures were favourable for seedling survival in both horizontal and vertical microhabitats. High maximum summer temperatures strongly increased mortality in horizontal sites, but did not affect survival in vertical sites. In the winter, high precipitation was unfavourable in both microhabitats, but there was no effect of winter temperatures on survival in either habitat.
Table 2. Results of survival analysis (Cox regression) showing the effects of microsite orientation, microsite-specific climate variables, and (for the subset of plants that lived long enough) relative growth rate on the survival of Thuja occidentalis seedlings that emerged in the 18-year study. Degrees of freedom = 1 for all tests. A positive sign of the estimated regression coefficient indicates that the hazard of mortality increases with an increase in the value of the parameter
Likelihood ratio test: chi-square = 798.85, d.f. = 13, P < 0.0001.
Likelihood ratio test: chi-square = 37.75, d.f. = 15, P = 0.001.
Results for the subset of slightly older plants for which growth rates were available showed much smaller effects of all season- and microsite-specific climate variables on survival (Table 2). Only the favourable effect of high summer precipitation on survival in horizontal microsites was significant. Growth rate was the most significant effect, but only for plants in vertical microsites where high growth rates increased the hazard of mortality. In horizontal microsites, in contrast, mortality was not significantly affected by growth rate.
Survival of the array of different-sized plants already present at the first census varied significantly with both microsite and plant size (likelihood ratio test: chi-square = 192.45, d.f. = 7, P < 0.0001; Fig. 4). Survival of the smallest plants was significantly better on the cliff face compared with horizontal sites, while there was the opposite trend for the larger size classes (although differences became smaller with increasing plant size and were no longer statistically significant). For plants in horizontal microsites, chances of survival improved consistently with increasing plant size. The same was not true for plants on the cliff face: here, the largest trees had a lower chance of survival than the next smallest size class.
Morbidity and causes of mortality in plants older than 1 year
Only the 599 plants that survived their first year and died later during the study were used in this analysis, as morbidity of first-year seedlings was rarely observed in the larger-scale study, where most seedlings found in the spring were gone without a trace by the next census in the autumn. Even for plants older than 1 year, death was usually sudden: 80% of the plants that died had appeared completely healthy at the previous census 6 months earlier, with only 20% having shown signs of morbidity such as yellowing, browning or partial loss of foliage. Sudden death was significantly more likely in horizontal microsites than in vertical ones (chi-square test: P = 0.0038; Fig. 5). Plants in horizontal microsites that showed signs of morbidity were also more likely to die and less likely to recover than those in vertical microsites (chi-square test: P = 0.0064; Fig. 5).
Actual external causes of death were observed for a total of 133 adult plants, 66 from vertical and 67 from horizontal microsites; 60.2% of them succumbed to rockfall, which either dislodged the entire plant or exposed enough of the root system that the plant subsequently wilted and died. The second most frequent cause of death (31.6%) was being covered by litter or fallen tree branches. Both of these factors affected mainly smaller plants and killed those in horizontal and vertical microsites in almost equal numbers. The remainder of known deaths had a variety of other causes, including herbivory, getting buried by falling rocks or debris from above, competition (shading out) by faster-growing herbaceous plants sharing the same microsite, disturbance by digging animals, or instability leading to seedlings being washed off small ledges.
Climatic extremes and demographic trends
While the multiple regression approach was considered to be the most powerful method of investigating the demographic response to climate because it considered all of its components simultaneously, we also thought it useful to try to see if extremes of the individual climate components over the 18 years of the project would show any pattern of association with demographic extremes. We therefore constructed tables of rank orders for various components of climate and demographic response variables that would allow us to search for obvious correlations (Appendices S1 and S2 in Supplementary Material). Over the 18 years of the study, total growing season precipitation varied almost threefold while mean temperature varied by three degrees, but no obvious links between any single climatic signal and seedling emergence or mortality were apparent. Years with high emergence (1989, 1992, 1993) had no clear association with extremes of any climatic variable. Two of the years with low emergence (1998, 2000) were associated with high temperatures during the germination period, but another (2002) was not. The years 1986–88 had the lowest seedling and adult mortality during periods that were cold and wet, but during a cold and wet period in 1992 there was high mortality. These results suggest that selection of different 2- to 4-year periods for a short-term study of the controls of emergence or mortality by climate could have led to almost any result being obtained.
The three seedling cohorts whose fate was followed over 9 years represented two moderate-sized recruitment pulses (1995 and 1997) and a year of very poor recruitment (1996) (Fig. 3a). In total, the 9-year study followed 571 seedlings from emergence to either their death or the autumn of 2003.
All three cohorts emerged between April and July, with the majority of seedlings emerging in May. Mortality occurred concurrently with emergence as early as the beginning of May and many seedlings were dead only a week after emergence. The number of living current-year seedlings peaked in the second half of May.
Despite the large fluctuations in the number of seedlings that emerged each year, their relative distribution over the different microhabitat types remained similar. Chi-square comparisons of the relative area occupied by each microhabitat type and the proportion of seedlings that emerged there showed that not all microsites were utilized equally (Fig. 6a). Significantly more seedlings than expected were found on horizontal shelves and large ledges (both P < 0.0001), while significantly fewer emerged on small ledges, the cliff edge (both P < 0.0001) and decaying logs (P = 0.003). Very few seedlings emerged on the featureless vertical cliff face, even though this habitat type comprised the largest proportion of the total area sampled (P < 0.0001).
While less than one-third of the sampling area was classified as stable, more than two-thirds of all seedlings were found there (Fig. 6b; P < 0.0001 for chi-square comparison). Sites with litter (Fig. 6c), loose rocks and gravel (Fig. 6d) or soil (Fig. 6e) also had significantly higher emergence than expected (all P < 0.0001). A tree canopy was present over 55% of the sampling area, while 30% of potential germination sites were completely open and a small fraction was under overhangs. Significantly fewer seedlings emerged under cedar or mixed cedar/deciduous canopy than expected (P = 0.006; Fig. 6f), but for all other canopy types, the number of seedlings recorded was not significantly different from expected (P > 0.05).
Cox regression of time to emergence (from the beginning of the growing season) as a function of microsite and climate showed significant microsite-specific effects of temperature and precipitation on the timing of seedling emergence (Table 3). In most microhabitats except the cliff face and smaller ledges, high maximum temperature in the preceding 7 days significantly decreased the chance of seedlings emerging. Minimum temperature had no effect in most habitats, except at the cliff base where low minimum temperature slightly decreased emergence. Except at the cliff edge, the chance of seedlings emerging was increased by high rainfall in the preceding week.
Table 3. Results of survival analysis (Cox regression) showing microsite-specific effects of the preceding week's climate on the emergence and survival of Thuja occidentalis seedlings in the 9-year study. The sign of the parameter estimate indicates whether the hazard of mortality increases or decreases with an increase in the parameter value. d.f. = 1 for all tests, n = 571
Likelihood ratio test: chi-square = 318.14, d.f. = 18, P < 0.0001.
Likelihood ratio test: chi-square = 147.45, d.f. = 18, P < 0.0001.
Factors controlling seedling survival
Ninety per cent of all seedlings that emerged each spring were dead by the end of July of the same year. No seedlings on the vertical face, on small ledges, on the cliff base and edge or on logs were alive after 5 years. After 9 years, only six plants survived: one in a crevice and five on medium-sized ledges.
Survival analysis showed that both climate and microsite factors significantly affected survivorship curves (Table 3). Seedlings on ledges, in crevices and on the cliff edge had an increased risk of death during periods of high maximum temperature. In contrast, high maximum temperatures did not increase the hazard for seedlings on horizontal surfaces and on the cliff base. Low minimum temperatures increased the hazard of mortality for seedlings in all microsites except the cliff edge. Rainfall had no effect on survival, except for marginally increasing the hazard for seedlings on smaller ledges.
In the Cox regression of survivorship against physical microsite variables, litter and soil were not significant (P = 0.335 and P = 0.639, respectively) but stability (P = 0.019) and the absence of loose rocks or gravel (P < 0.0001) significantly improved seedling survival. Ten times more seedlings were alive on stable sites compared with unstable ones after 5 years (Fig. 6b), and only seedlings on stable sites were alive after 9 years. Better survival was also seen under either type of tree canopy compared with open sites (both P < 0.0001).
Morbidity and causes of mortality in first-year seedlings
Periods of extended morbidity were rarely observed in seedlings. Apparently healthy seedlings usually died or disappeared completely from one weekly census to the next. Actual causes could be ascribed to 21% of seedling deaths, or 119 seedlings. Drought, pathogens and rockfall were the most frequent among them and together accounted for 84% of observed seedling deaths. Smaller numbers of seedlings succumbed to foliage loss due to injury or herbivory, litterfall or substrate instability. Causes of mortality differed markedly among microsites (Fig. 7). Drought, the most common cause overall (45% of known seedling deaths), caused the majority of observed deaths in crevices and on smaller ledges, but fewer than half of those on large ledges and horizontal shelves and none at all at the cliff base. More seedlings in vertical compared with horizontal microhabitats succumbed to rockfall, while the opposite was true for pathogens. Pathogens were, however, responsible for 100% of seedling deaths observed at the cliff base.
Long-term intensive studies offer glimpses of natural processes that cannot be provided by short-term studies (Berkowitz et al. 1989). In the current global debate over past and future impacts of changing climate regimes on the structure of vegetation communities or individual species, it is important to have access to evidence whose time-scale permits the isolation of long-term trends in climate and plant response. Long-term ecological research sites (LTERs) around the world might seem to offer consistent access to such information but even at these sites, truly long-term monitoring of plant response to microsite and climatic drivers, such as found in Jones & Sharitz (1998), Arévalo et al. (2000) and Dzwonko & Gawronski (2002), represents only a minuscule fraction of the literature.
Our site presented an opportunity to investigate the dynamical aspects of tree demography in response to microsite availability and climate over a long period of time. The site has already been well characterized ecologically, which minimized the risks normally associated with intensive long-term studies of small sites, such as lack of site replication and boundary properties (Berkowitz et al. 1989). The results of our 18-year study show that when a recruitment opportunity is opened up along the limestone cliffs of the Niagara Escarpment, climatic conditions and microsite properties and availability jointly control the growth, health, survival and mortality of T. occidentalis during the first growing season. After that, survival becomes increasingly independent of climatic fluctuations: there was enormous variation over the 18 years in climate variables such as the mean, maximum and minimum growing season temperature and the summer total precipitation, but the patterns of morbidity and mortality in adult plants showed little correlation with this. Microsite, in contrast, continued to exert an influence on survival, albeit indirectly, by controlling the growth rate and resulting size of the plants, but only on vertical surfaces.
size structure and patterns of growth
Between 1986 and 2003, two-thirds of the trees in the study were replaced and total population size on the cliff varied enormously. Despite this, the shape of the size distribution of individuals present remained largely unchanged except for variation in the number of individuals in the very youngest age class. Thus, the system was undergoing massive demographic change over the 18 years that was largely invisible to height-class analysis. This replacement appeared to be generally unrelated to climate trends. Even though the time-scale studied by Kelly & Larson (1997) was 10 times longer (> 200 years), they too concluded that mature cliffs of the Niagara Escarpment show the same size- and age-class distribution over time. This size-class and population-size constancy in turn supports the ‘cork in the bottle’ hypothesis of Davis (1951) and the concept of ‘safe sites’ as proposed by Harper et al. (1965) because new recruits able to survive their first year are apparently restricted to sites vacated by organisms that perish. Davis’ hypothesis gains further support by the lack of any significant control of long-term survival by the surrounding climate.
Growth rate could not be predicted by tree height for trees either on horizontal or vertical surfaces, suggesting that a variety of environmental forces constrain height growth even in well-established plants. The uncoupling of size from growth rate or size from age has been reported previously for trees growing on cliffs (Kelly et al. 1992; Larson et al. 2000) or other hostile sites (Schweingruber 1993). While most trees grew slowly and were very small, relatively large trees only grew on sites that were horizontal. In vertical sites fast growth was a disadvantage, as cliff-face plants over 1 m in height had an increased probability of dying. This is exactly the pattern that the ‘cork in the bottle’ hypothesis predicts: Davis (1951) argued that the capacity for slow growth should be selected for on cliffs, because otherwise growing plants would force themselves out of their own ‘bottle’ over time.
T. occidentalis produces seeds almost every year, with mast seed crops occurring every 3–5 years (Johnston 1990). We had shown previously (Bartlett et al. 1991; Booth & Larson 1998) that seed availability did not limit the recruitment of this species on cliffs, unlike Thuja forests where the annual number of germinants depends on the size of the seed crop (Cornett et al. 1997). However, the result of the present study that the temperature of the previous summer (i.e. during cone production) was one of the predictors of seedling emergence suggests that this may not always be true. Total rainfall during the germination period had no significant influence on the number of seedlings emerging each year, even though short-term fluctuations in emergence were correlated with the previous week's rainfall. An absence of extreme temperature fluctuations during the germination period (low maxima, high minima) was correlated with high annual seedling emergence, but in the short term only low maximum temperature during the preceding week significantly predicted high emergence in most microsites.
Vertical cliff face, rocky shelves (both large and small) and cliff edges represented the largest percentage of microsites available for colonization, but the vertical face, smallest ledges and the cliff edge were grossly under-represented in terms of new seedlings. Most emergence took place on wide ledges or shelves, probably because seed rain was greater and seed retention better on horizontal surfaces compared with vertical ones. The soil, litter, rocks or gravel that tend to cover such microsites not only add roughness that helps trap seeds, but also retain moisture necessary for germination (Cornett et al. 1997; Tsuyuzaki et al. 1997; Jumpponen et al. 1999; Elmarsdottir et al. 2003). Seeds germinate most successfully when the surrounding air is water-saturated, and insufficient hydration is often the reason for failure of living seeds to germinate (Harper & Benton 1966).
Ground surface stability was necessary for both emergence and survival, as demonstrated previously for other barren substrates (Tsuyuzaki et al. 1997). Getting crushed or covered by debris was a commonly observed cause of mortality, and few seedlings survived in microsites with loose rocks or gravel present, even though emergence was higher than expected in such sites. Other microsite factors also had different or even opposite effects at different life stages. The presence of a tree canopy decreased emergence but increased survival, and litter and soil favoured emergence but had no influence on survival.
For first-year seedling survival, the climatic conditions during the first growing season were vastly more important than those during the following winter. Only winter precipitation was significant in increasing mortality, possibly because rockfall, treefall or similar causes of accidental deaths are most likely to occur during winter storms. Mortality of the youngest seedlings was influenced by both microsite and climate, with low minimum temperature being most consistently unfavourable in virtually all habitats. Cool summer nights may produce damp conditions that favour pathogens, and pathogens were an important cause of mortality for young seedlings, especially at the cliff base, the microhabitat with probably the highest moisture availability. Mortality of saplings and adults, on the other hand, occurred with no apparent rhythm or pattern, with apparently random causes such as rockfall and tree- or litterfall. Declines in survivorship were always more profound for trees in horizontal microsites than on the cliff faces, but individual cohorts showed largely the same declines in survivorship over the census period.
Of the nine microsite types distinguished in the 9-year study, only medium-sized ledges and crevices were represented in greater proportion among 5-year survivors than among emergent seedlings and potentially colonizeable cliff area. Both these microhabitats can therefore be considered safe sites for the life stage up to 5 years. The best 5-year recruitment was on medium-sized ledges due to a combination of moderately high emergence and relatively good survival. Crevices, which were second best, provided only a limited number of safe emergence sites but seedling survival there was excellent. Crevices in the rock probably represent ideal seedbeds as they provide both water-saturated conditions and stability. The proportion of seedlings in all other microhabitats decreased over the 5 years; for example, the largest ledges and shelves had the highest seedling emergence but also the highest seedling mortality. These results demonstrate that the filters operating on this ecosystem vary not only with microsite but also strongly with stage of recruitment: the ‘safe site’ definition is stage-of-development dependent. Castro et al. (2004) came to the same conclusion in their study of Pinus sylvestris at its southern distribution limit. The end result for T. occidentalis was the same as that for P. sylvestris: due to different combinations of filters active at different developmental stages in different microhabitats, very few seedlings survived their first summer.
Studies of T. occidentalis regeneration in closed-canopy forests or wetlands have frequently shown decaying logs to be a favourable substrate for seedling recruitment (Scott & Murphy 1987; Cornett et al. 1997; Simard et al. 1998; Rooney et al. 2002) but in our study few seedlings emerged and none survived on woody debris. In the absence of a closed canopy and a moist organic forest floor, the logs on the horizontal shelves represented a dry, inhospitable substrate that was unsuitable for seedling recruitment.
causes of death and morbidity
Rockfall and drought were the main causes of seedling death in vertical habitats, while pathogens were a much more frequent cause of mortality in horizontal habitats. This may be due to a greater tendency of horizontal surfaces to be damp, favouring pathogen growth. In slightly older plants microsite controlled not only survival and mortality, but also morbidity. Many cliff-face plants experienced extended periods of morbidity and a substantial percentage of them subsequently recovered and lived until the end of the experiment. Plants in horizontal microsites were much more likely to die suddenly and if they showed signs of morbidity, they rarely recovered. This illustrates that the safe sites for regeneration on cliffs are not the larger horizontal surfaces as one might expect, but the smaller features located on the vertical cliff face such as medium-sized ledges and crevices.
Desiccation and high temperatures at the soil surface are commonly implicated in seedling mortality in exposed habitats (Cui & Smith 1991; Gray & Spies 1996), and Johnston (1990) and Cornett et al. (1997, 1998) found that moisture availability is the key variable in the survival of T. occidentalis over the first few years. Our results show evidence that both of these factors may be involved in causing seedling mortality. Both in the short-term and long-term survival analyses, high maximum temperature was significantly related to low seedling survival, especially on horizontal surfaces. Rainfall was only significant in the long-term study, but drought was frequently identified as a cause of death in the short-term study as well.
Vertical surfaces on cliffs have abundant moisture available below the surface because hydraulic pressures force percolating rainwater towards vertical cliff faces (Larson et al. 2000), and seedlings can be expected to gain access to this water as they increase in size. This may explain why the influence of rainfall disappeared for plants in vertical sites in the slightly older subgroup, and disappeared completely when adults found at the first census were analysed.
In contrast, cliff edges and the outer parts of large ledges are removed from this moisture source and represent local deserts where evaporation is rapid. While the shallow soil on these horizontal surfaces allows seed germination to take place, it does not generally retain sufficient water to support the emerging seedlings through the growing season.
Long-term demographic studies can give an enormously important perspective on the response of plant communities to changing environments. The 18-year study has allowed us to see that recruitment of a long-lived tree species in a low-productivity habitat is controlled by short-term climatic fluctuations, but not in a simple fashion. The particular microsites in which seeds or seedlings find themselves have a profound influence on their success, and the microsite–climate interaction turns out to be the most important control. Once establishment has taken place, the importance of climate fades. The same microsites and climatic conditions that initially favour establishment can later actually cause morbidity or mortality. We were able to conclude that if one looks at only one process over one small period of time or space, one will get an answer that is surely wrong. Only a long-term study can show the whole picture, which is complete but also complex.
Global ecosystems are confronted with changing climatic and other conditions at all time-scales and yet most published research studies are short-term in duration. Ecologists cannot often afford the time to conduct long-term projects and dependence on short-term studies, however undesirable, is likely to continue indefinitely. With a literature dominated by short-term studies of population or organismal responses to climate or climate change, there must nevertheless be an effort to conduct studies such as ours, and also to conduct meta-analyses of published data to detect long-term trends that the original authors were not able to see. Even the 18 years included here represent only a small sliver of time in a forest that is among the most long-lived and stable in the world.
This study was supported by the Natural Sciences and Engineering Research Council of Canada, the Ontario Ministry of Natural Resources (Environmental Youth Corps Program), and Human Resources Development Canada (Job Creation Partnerships Program). A large number of people helped with the fieldwork. We thank two anonymous reviewers for constructive comments on a previous version of this manuscript, and J. Horrocks for statistical advice.