Survival and growth of woody plant seedlings in the understorey of floodplain forests in South Carolina


Robert H. Jones, Department of Biology, Virginia Polytechnic Institute and State University, Blacksburg, VA 24061-0406, USA (fax 540 231-9307; e-mail


1 Survival and height growth of permanently tagged understorey seedlings in four river floodplain forests in South Carolina were monitored for 8 years. Regression models were used to determine if a common survival pattern emerged and if the pattern varied according to species, spatial location, time, seedling size and seedling growth.

2 For most of the populations, per capita survival was initially poor but increased steadily with age. A few populations, however, had relatively high survival rates for 1–3 years after establishment, or brief periods (1–2 years) with much lower survival. Although the underlying survival function was best described by a negative power curve, a log-logistic model also fit the data and provided more versatility for fitting individual populations.

3 Significant differences in general survival pattern were found among species, sites (forests), years of establishment (cohorts), and all two-way interactions of species, site and cohort. Species differences were large. Spikes in mortality for individual species and cohorts were synchronized within sites and to a lesser degree among sites. In some sites, weak relationships were noted between mortality and peak river discharge during summer (negatively correlated) and winter (positively correlated).

4 Species differed strongly in their initial height, but effects of cohort and site on height were also significant as were all two-way interactions of species, site and cohort. During the 8 years of this study, very few seedlings grew taller than 30 cm. For most species, taller seedlings had greater per capita survival. Height growth, however, was not consistently related to survival. Since mean size of survivors changed little with time, age may be a better variable than size to use in models of survival.

5 Simulation models could probably be developed using a log-logistic function that includes (in order of importance): seedling age, species, intensity of winter floods (some sites) and occurrence of summer droughts (some sites).


Many aspects of forest succession are well known and have been successfully incorporated into predictive models (Huston & Smith 1987; Botkin 1993; Pacala et al. 1996). The establishment phase, however, remains one of the least understood and most often ignored components of forest dynamics (Canham 1989; Connell 1989; Lieberman et al. 1989). Botkin (1993) considers the modelling of seedling germination to be ‘computationally unwieldy’ and the prediction of seedling recruitment to be difficult due to the lack of general patterns among species. Models of forest dynamics often ignore small seedlings (Shugart 1984; Botkin 1993; Pacala et al. 1996), make very simplified assumptions about their spatial distribution and recruitment (Pacala et al. 1996), or start from known seedling populations without modelling new recruitment (Mou et al. 1993).

In addition to computational difficulties, a scarcity of long-term demographic data limits the incorporation of small seedlings into forest dynamics models. Most published studies focus on survival of planted seedlings (Ashton 1995; Ashton et al. 1995; McKee 1995), establishment in gaps (Denslow 1980; Augspurger 1984; Schupp et al. 1989; Tappeiner & Zasada 1993) or relatively short-term (<3 years) survival (Augspurger 1984; Turner 1990a,b; Houle 1992). Few studies have examined natural establishment and long-term survival in forest understories away from gaps (Hett & Loucks 1971, 1976; Woods 1984; Clebsch & Busing 1989; Busing 1994; De Steven 1994; Hart 1995; Boerner & Brinkman 1996). Such non-gap seedlings, via a process known as advance regeneration (Smith 1962), can however contribute to, or even dominate, the canopy replacement process (Smith 1962; Woods 1984; Connell 1989; Silvertown & Lovett Doust 1993).

Advance regeneration seedling pools are dynamic. Rates of seed input and seedling mortality vary among species, and also vary within species across space and time (Augspurger 1984; Canham 1989; Streng et al. 1989; Jones et al. 1994). Potential causes of mortality include abiotic stresses such as shade, drought and flooding (Walker et al. 1986; Streng et al. 1989; Walters et al. 1993) and biotic influences such as herbivory, disease or root competition (Janzen 1970; Woods 1979; Hubbell 1980; Maguire & Forman 1983; Augspurger 1984; Walker et al. 1986; Janzen & Hodges 1987; Streng et al. 1989; Jones et al. 1994). Microtopography, which often influences both seed dispersal and seedling stress, can also be important (Harcombe et al. 1982; Augspurger 1984; Beatty 1984), especially in floodplain forests (Huenneke & Sharitz 1986).

If responses of understorey seedlings to these various factors can be reliably predicted, forest succession models may be better able to incorporate seedling dynamics. However, most of our current knowledge of advance regeneration ecology is derived from anecdotal observation of seedlings in natural systems or laboratory experiments (Hook 1984; Burns & Honkala 1990a,b; Walters et al. 1993; Walters & Reich 1996). Few studies have used detailed, long-term (i.e. >5 years) observations of natural populations. We are aware of only two published studies on long-term demography of advance regeneration in floodplain forests (Streng et al. 1989; Jones et al. 1994).

In addition to their utility for modelling community dynamics, long-term demographic studies can test hypotheses concerning corollaries (and mechanisms) of survival. For example, several studies have suggested that tree seedling survival (either before or after disturbance) is more closely linked to size than to age (Johnson 1977; Marquis 1982; Streng et al. 1989; Loftis 1990). This makes sense for floodplain forests where taller seedlings may be able to avoid the often lethal effects of leaf inundation (Hosner 1958; Hook 1984; Jones et al. 1989). However, some understorey seedlings in floodplains grow very slowly (Streng et al. 1989; Jones et al. 1994) and it is therefore possible that growth rates may be a key to survival, as has been found for other tree seedlings (Walters & Reich 1996).

In this paper we examine long-term demographic patterns for understorey tree seedlings in four floodplain forests to answer two questions. First, is there a common pattern of seedling survival in forest understories that can be exploited to model density and species composition (assuming that seed input and germination are known)? Secondly, are there key aspects of the seedlings and of the understorey environment that may cause variations in the general pattern? Based on past studies (Streng et al. 1989; Jones et al. 1994), we hypothesize that survival differences among species will be large, that soil moisture (including flooding and drought) will affect survival in most species, and that taller seedlings will survive better.



The four forests sampled are located within the Savannah River Site (33°15′N, 81°38′W), a US Department of Energy facility in the Coastal Plain of South Carolina. They were chosen to provide contrasting flooding regimes (two frequently vs. two rarely flooded forests). They also contrast in summer drought regimes because two are in the floodplain of the Savannah River, a large river where soils can be subject to drought in summer, whereas the others are in the floodplain of Upper Three Runs Creek, a smaller river (and tributary of the Savannah River) where soil moisture is always plentiful. We refer to the four forests as flooded large (i.e. frequently flooded site in large river floodplain), flooded small, unflooded large, and unflooded small. When the Savannah River is at flood stage, the lower portion of Upper Three Runs Creek (including the small flooded site) is also flooded. About 60 woody species, many of them common to two or more of our sites, occur in the four forests. The most abundant are Liquidambar styraciflua, Quercus laurifolia, Nyssa sylvatica var. biflora, Q. nigra and Acer rubrum. More details on the study location and study design can be found in Jones et al. (1994). Plant names follow Radford et al. (1968).


In summer 1986 and spring 1987, 1 × 1 m plots were set out randomly within a 1-ha area in each forest to monitor demography of seedlings (stems less than 1 m tall). A total of 220 plots were established, 75 per unflooded forest where seedling densities are relatively low, and 35 per flooded site where densities are greater. All tree seedlings within each plot were marked with 2 × 3.5 cm aluminium tags loosely attached by copper wire to the stem base. Tags rarely disappeared (between one and two per plot per survey), and were easy to replace owing to the frequency of our sampling and careful noting of physical condition and size of each stem.

To measure seedling status, we visited plots once in ‘spring’ (March–June) and once in ‘autumn’ (August–November) from spring 1987 to spring 1995. Seedlings were tagged as they emerged in 1987 and 1988 but new seedlings were thereafter tagged only at the autumn survey. The data set included: (i) 2265 woody plant seedlings and sprouts that emerged prior to 1987 (referred to as the 1986 cohort); and (ii) 5574 seedlings that emerged during the 1987 through 1994 growing seasons (referred to as the 1987 through 1994 cohorts, respectively).

During autumn surveys, tagged seedlings were identified to species and the height of the tallest apical meristem above ground level was measured. During spring surveys, only the status (dead or alive) of each seedling was noted. There were two exceptions to this pattern of measurement: (i) heights were measured in springs of 1987 and 1988; and (ii) no heights were measured in autumn of 1987. Seedlings were considered dead only if all of the stem was obviously desiccated with no apparent chance for resprouting. If a stem was missing for two survey dates in a row, it was declared dead as of the first date of ‘missing’ status. On rare occasions (most commonly in oaks), stems resprouted after being declared dead. Resprouted stems were reclassified as alive.


Differences in survival related to three covariates (species, forest and annual cohort) were assessed by accelerated failure time analysis (Fox 1993). In this analysis, the individual seedling (i.e. not plot or forest) was considered an independent observation or replicate. Survival time in days was computed for each seedling as the difference between the date of tagging (i.e. in autumn) and the survey date at which mortality was noted. These data include interval censored and right censored observations (Lee 1980; Fox 1993). The 1986 cohort was deleted prior to analysis because it included seedlings and sprouts of unknown age, which were left censored and therefore difficult to analyse (Fox 1993). The 1994 cohort was also excluded because survival in this cohort was overestimated; some of the 1994 cohort that were missing in spring 1995 (i.e. the first sample date after tagging) could not yet be declared dead. Finally, uncommon species that had too few observations (i.e. less than 75) for reasonably precise estimates of survival were excluded, resulting in an analysis of the 13 most common species. We used the proc lifereg module of SAS (SAS Institute Inc. 1988) to analyse the data. We tested whether a normal, log-normal, Weibull, exponential, power or log-logistic function provided the best explanation of underlying survival. We selected the function that gave the best log likelihood statistic (indicating goodness-of-fit) with the fewest number of parameters, and which had the ability to fit individual populations (i.e. one species of one cohort at one site). We then used a single analysis to determine main effects of species, forest and cohort. To test for significance of each two-way interaction, we altered the model to fit one main effect and the interaction of the remaining two effects. Then, log likelihood ratios and degrees of freedom in the altered model were subtracted from those in the full main effects model for a log likelihood ratio test.

Seasonal variation in mortality and its relationship with precipitation and flooding were explored by correlation analysis. First, periodic mortality (i.e. over-winter and over-summer) was calculated for each combination of species and cohort with sample sizes greater than 50, and for all seedlings combined. Correlation coefficients were then calculated to compare periodic mortality in each site with the following: (i) periodic mortality in other sites; (ii) precipitation during the growing season (i.e. April–September); (iii) mean monthly discharge in the Savannah River during the growing and non-growing season; and (iv) peak daily discharge during the two seasons. Precipitation data were obtained from the US National Weather Service station closest to the research sites (Bush Field airport, Augusta, Georgia) and monthly river discharge data were obtained from the US Geological Survey gauge on the Savannah River at Augusta, Georgia.

Analysis of variance was used to assess effects of species, site and cohort on first-year height of the 13 most common species (i.e. the same species used in survival analysis). The relationship between height and seedling survival, and the potential effect of site on this relationship, were examined by logistic regression (Lowell et al. 1987; Trexler & Travis 1993) using proc catmod in SAS (SAS Institute Inc. 1988). Survival for one year at a time (transformed into the logit of mortality = log[proportion dying/proportion living]) was modelled as a dependent variable with height, site and the height × site interaction as independent variables. We also used a multiple logistic regression to examine independent effects of height and height growth on survival.



In most seedling populations, per capita survival rates increased over time. Survival curves for individual cohorts showed this clearly; they declined at decreasing rates (Fig. 1). Some of the curves, however, had an inverse S-shape reflecting an initially high per capita survival followed by decreasing and then increasing survival rates. When analysed without considering covariates, the data were most precisely fit using the Weibull function because it had the log likelihood closest to zero (Table 1). A power function provided nearly the same precision with one less parameter, and the log-logistic was able to fit the data but with less precision (Table 1). All three functions fit the data when the covariates of species, site and cohort were included in the analysis; however, only the log-logistic fit individual populations (Fig. 2). Therefore, we chose the log-logistic as our base model. It has the following form:

Figure 1.

Survival of annual species cohorts (all species combined) plotted by site. Time zero for each curve represents density of seedlings tagged in autumn standardized to 1000.

Table 1.  Functions used to model overall seedling survival patterns

Number ofparametersLoglikelihood
Figure 2.

An example of how the log-logistic model (developed using covariates of species, sites and cohorts) matched actual data from several populations in the large flooded site. Open symbols are actual and filled symbols are predicted values. Fits for populations at other sites had similar precision.

S(t) = (1 + exp ((ln (t) - μ)/σ))

where S(t) is proportional survival at time t (in days), and μ and σ are intercept and shape parameters, respectively. Without considering the covariates of species, site and cohort, parameter estimates were μ = 6.0284 (SE 0.01861) and σ = 0.6961 (SE 0.0125).


Seedling survival was significantly influenced by species, site, cohort and all two-way interactions of these three main effects (Table 2). Particularly large survival differences were found among species. Of the 13 most common species, four had relatively high long-term survival rates (Persea borbonia, Quercus laurifolia, Q. nigra and Ulmus spp.), and one (Betula nigra) had very low rates (Fig. 3). The remaining eight species were intermediate, and had very similar survival curves.

Table 2.  Significance of main and interaction effects on survival; n = 4495 seedlings for the 13 most common species measured during 1987–94

−2 × loglikelihoodd.f.P
Species × site76.514<0.001
Species × cohort162.465<0.001
Site × cohort89.118<0.001
Figure 3.

Mean survival curves for the 13 most abundant species in floodplain forest understories. Curves are based on the log-logistic model fit of the 1987 through 1993 cohorts in four river sites. A. rubrum and Vitis spp. have the same curve.

Differences among annual cohorts were also relatively large (Fig. 4). Seedlings established in 1987 and 1993 had relatively high survival rates, those established in 1988, 1989 and 1990 had lower rates, and those in 1991 and 1992 were intermediate (Fig. 4). According to Spearman rank correlation, long-term survival was unrelated to cohort size (rs = 0.64, P = 0.119).

Figure 4.

Mean survival curves for sites and annual cohorts, based on the log-logistic model fit of the 13 most abundant species.

Mortality was sometimes synchronized across species, cohorts and sites (Fig. 5). Periodic mortality was most synchronized between 1987 and 1991 and much less so thereafter. A particularly large, synchronized mortality spike occurred between 1990 and 1992 in the small flooded site. In both large river sites, relatively low mortality rates occurred in the winter of 1988–89. When all species and cohorts were combined for among-site comparisons, a significant correlation between the two large river sites was detected (r = 0.79; n = 15 periods; P < 0.001). Although coefficients were positive for all other site comparisons, none was greater than 0.50 (range 0.22–0.45) and none was significant (P > 0.09).

Figure 5.

Over-winter and over-summer (periodic) mortality for all cohort and species combinations where n > 50. Each line is a unique population (i.e. one cohort of one species at one site).

Only two correlations greater than 0.50 (or less than −0.50) were found when periodic mortality rates at each site were correlated with summer rainfall, mean monthly river flow or peak monthly river flow data. In the large river sites, summer mortality declined when peak summer flow increased, and in all sites except the small flooded, winter mortality increased when peak winter flows increased (Fig. 6). However, none of the correlations was significant, probably because of the small sample size (n = 7 or 8) and weakness of the relationship.

Figure 6.

Relationship between seasonal mortality and peak flow of the Savannah River.


First-year height was significantly related to species, site and cohort either directly or as an interaction effect (Table 3). When mean heights were inspected, however, the largest and most consistent differences were associated with species rather than sites or cohorts. Median height ranged from slightly more than 10 cm in Persea borbonia and Nyssa sylvatica var. biflora, to less than 4 cm in Ilex opaca and Betula nigra (Fig. 7). Only three of the 13 most common species had median heights greater than 10 cm and virtually all seedlings were less than 30 cm tall. Each species had considerable variation in height resulting in overlap among all species (Fig. 7).

Table 3.  Analysis of variance for height of seedlings in autumn of the first year of establishment; n = 3532 seedlings representing the 13 most common species


Site32.03 0.701
Species × site2323.56<0.001
Species × cohort646.57 0.004
Site × cohort188.12 0.012
Species × site × cohort575.77 0.043
Figure 7.

Height growth in year of establishment for the 13 most common species; data for all sites combined; boxes show median, 25th and 75th percentiles; bars are 10th and 90th percentiles; points are all data outside of 10th and 90th percentiles.

Changes in height over time were minor (Fig. 8). Even after 8 years of growth, 90% of all seedlings were less than 25 cm tall. Furthermore, of the 7839 seedlings tagged during the 8 years of this study (plus thousands more tagged during the summers of 1987 and 1988 that did not live until autumn of the year they germinated), only 13 individuals, all of them in the pre-1987 cohort, grew to a height of 1 m or larger. Of these 13, six had died back to less than 1 m tall by 1994, and the remaining were all shrub or vine life-forms, not trees. This pattern of very slow growth occurred despite periodic single-tree gaps that occurred in our sample plots. Only in one part of one site have we observed a burst of seedling growth responding to the blowdown of several large overstorey oaks in 1989. Even in this blowdown, none of the seedlings in the measurement plots had grown taller than 1 m by autumn of 1994.

Figure 8.

Height of survivors at various ages; all species, sites and cohorts combined; boxes show median, 25th and 75th percentiles; bars are 10th and 90th percentiles; points are all data outside of 10th and 90th percentiles; numbers are sample size.


Our first analysis of height and survival relationships considered means for each species. A weak correlation occurred between mean height of the 13 most common species (Fig. 7) and their predicted long-term survival (Fig. 3). Taller species had greater survival (Spearman rank correlation rs = 0.60, P = 0.029). One exception was Nyssa sylvatica var. biflora which was tall and had high survival from year 1 to year 2, but relatively low long-term survival.

Our next analysis of survival and height considered individual seedlings grouped by species and sites and age. Here we found that the relationship between height and survival was not consistent across sites or seedling age. For example, 1-year-old Acer rubrum seedlings had essentially no relationship between height and survival in the small flooded site, yet the relationship was apparent in the large flooded site and for 2-year-old Acer rubrum in the two flooded sites (Fig. 9). When data were analysed by logistic regression, slopes estimated for one species differed in magnitude and sometimes in direction (i.e. positive vs. negative) across sites. However, significant height × site interactions were found in only four out of the 12 tests. In general, height was more often a significant factor for populations at the small flooded site (Fig. 9).

Figure 9.

Relationship between survival and height using height class intervals for common species; data from all cohorts combined. Left panels are height in year of establishment (year 1) and survival through the next year (year 2). Right panels are height in year 2 and survival to year 3.

Since we found little evidence for consistent site effects on the height–survival relationship, we combined the data from all sites to examine more closely the differences among species (Table 4). Several patterns emerged. For some species (Acer rubrum, Liquidambar styraciflua, Quercus laurifolia), height was a significant factor affecting survival in each year (Table 4 and Fig. 10). In others (Persea borbonia, Quercus nigra, Ulmus spp., Pinus spp., Ilex decidua, Rhus radicans and Vitis spp.) the relationship was never significant. Finally, some species had significant relationships in some but not all years (Crataegus marshallii, Ilex opaca and Berchemia scandens).

Table 4.  Logistic regression slope estimates for logit of one-year mortality (log [proportion dying/proportion surviving]) vs. height in prior year, each regression for an individual species combined across all sites (n > 30 seedlings)

  1. † Year of height measurement (1 = year of establishment); survival was measured from this year to next year.‡ Model failed goodness-of-fit test.* Significant slope at P < 0.05; ** at P < 0.01; *** at P < 0.001.

Acer rubrum−0.125**‡−0.208***−0.320***−0.139***
Betula nigra−0.851***
Crataegus marshallii 0.014−0.396*−0.388*
Ilex opaca−0.327−0.626**−0.544**
Liquidambar styraciflua−0.149**−0.173**‡−0.176**
Nyssa sylvatica var. biflora−0.219***‡
Persea borbonia−0.076‡−0.076‡−0.084
Quercus laurifolia−0.138**‡−0.102*
Quercus nigra 0.063
Ulmus spp. 0.284−0.120‡−0.089
Pinus spp.−0.015
Ilex decidua−0.160
Berchemia scandens−0.392***−0.186* 0.072
Rhus radicans 0.039−0.005−0.040‡−0.028‡
Vitis spp.−0.172−0.070
Figure 10.

Relationship between survival and height predicted from logistic regression model; data from all sites and cohorts combined. Filled symbols represent significant regressions (significant slope parameters), open symbols represent non-significant regressions.

Our attempts to test for independent effects of height and height growth on survival were partially thwarted by the correlation between these two variables. For the first year, height and height growth are the same thing, so we could not test their independent influences on survival. For many of the other years, height and height growth were highly correlated (for species with n Ð 30, range in correlation coefficients was 0.50–0.83) which made fitting the logistic regression model difficult. However, nine tests were conducted with reasonable model fits (i.e. non-significant log likelihoods). Of these, five showed that neither height nor height growth were significantly related to survival (P > 0.05), three showed significant height but not height growth effects (Acer rubrum in 2 years and Crataegus marshallii), and only one showed a significant growth effect without a height effect (Ulmus spp.).



Our first goal, to find a common pattern of survival, was met. A log-logistic survival function fit the entire data set as well as individual populations. If just the entire data set was to be modelled, however, the power and Weibull functions would be reasonable alternatives (Table 1), even if the covariates of species, site and cohort were included (data not shown).

Several functions, including the log-logistic, have been used successfully to model other natural populations. The power (Hett & Loucks 1971) and exponential functions (Hett & Loucks 1968; Jones & Raynal 1987) have been used for understorey tree seedlings. The Weibull has been used for aphid (Nowierski et al. 1995) and pine cone (de Groot & Fleming 1994) populations. In floodplain forests, logistic patterns were found for understorey tree seedlings within their first growing season after germination (Streng et al. 1989; Jones et al. 1994).

There are several possible explanations for a log-logistic survival pattern. Perhaps there is an inherent tendency for a power or exponential pattern, but the pattern is periodically punctuated by heavy mortality events, thereby resulting in a logistic curve. Streng et al. (1989) reported a single year of heavy drought-induced mortality that resulted in a logistic survival pattern for understorey tree seedlings. One might expect the logistic pattern wherever environmental extremes occur (e.g. floodplains and drought-prone uplands), and the exponential or power pattern where environments are more stable. Alternatively, the initial period of high per capita survival that we observed in many of the curves (Fig. 1) may reflect positive effects of stored seed reserves (i.e. carbohydrates, proteins and fats) on survival during the first year or two after germination. However, since reserves are often completely utilized within days of germination (Bewley & Black 1994), this explanation is more plausible for first-year than for later-year survival. Another explanation involves responses to cumulative environmental stresses. Under this scenario, stresses that occur after germination weaken but do not kill seedlings outright. Later, a single additional stress (weak or strong) causes the weakened individual to succumb. The effects of cumulative stress (predisposing factors) followed by an acute mortality event have been well documented for adult trees (Manion 1981), but not for seedlings in forest understories. Finally, the initially high rate of survival in some of the curves could be an artefact of seasonal changes in mortality rates if survival is typically greater in winter months (i.e. immediately after seedlings are tagged) than in summer months; however, we found that over-winter mortality generally exceeds over-summer mortality (Jones et al. 1994).

Although many survival curves were logistic, their predominant feature was an increase in survival rate over time (Figs 1 and 2). This increase may be the result of a microsite sorting process. After initial seedling establishment, cumulative and acute stresses (e.g. flooding and root competition) may begin selectively to remove seedlings located in unfavourable microsites. In floodplain forests, unfavourable microsites are common owing to the strong impacts of hydrology on seedling physiology. However, such microsites may have high seed deposition (Huenneke & Sharitz 1986) and germination when water levels are low. Streng et al. (1989) found that light-seeded species in particular were prone to establish in wet microsites during dry periods only to suffer massive mortality when floods occurred. In addition to microsite sorting, increases in seedling size (discussed below) may contribute to greater per capita survival as seedlings age.


Another goal of our study was to determine if the basic survival pattern varied owing to differences among species or effects of environmental conditions. Variations attributable to both species and environment were found. As predicted, differences among species were particularly large (compare Fig. 3 with Fig. 4). However, some species had very similar survival curves. They may be even more similar than Fig. 3 suggests because each curve in that figure is a mean of four site curves and seven annual cohort curves. For modelling purposes it may be best to lump the species into three survival groups: high survival (four species), intermediate (eight species) and low (one species).

Variation in environmental conditions also influenced survival. Strong but indirect evidence of this was provided by the significant and large effect of year of emergence (Fig. 4). The differences among cohorts may have been caused by age-specific sensitivity to severe environmental stress. During years with unusually stressful environmental conditions, older, better-established cohorts may have endured while younger cohorts suffered heavy mortality. We have already demonstrated that older seedlings have greater per capita survival. The greater variation among annual cohorts than among sites suggests that the influence of weather phenomena and flooding, which act in short time-frames, are stronger than the effects of inherent site conditions that may have more long-term impacts. Since cohorts interacted strongly with species and sites (Table 2), it is also possible that cohort differences were confounded by species and site effects, especially if some cohorts were missing observations from one or more sites or species. In fact, all cohorts included seedlings from each site, two cohorts included all species, five were missing observations from one species, and two were missing data from two species. Thus, confounding effects caused by differential distribution of cohort observations across species and sites were probably minor.

Further evidence that environmental stress affected seedlings was provided by the synchronization of mortality across species, cohorts and sites (Fig. 5). Streng et al. (1989) found that synchronized mortality corresponded to major flood events. However, our attempts to show a clear relationship between mortality and hydrology mostly failed. First, sites that had greatest mean flooding did not necessarily have lowest mean survival. In fact, mean survival was poorest in the small unflooded site and greatest in the small flooded site (Fig. 4). Secondly, the correlations between periodic mortality and various hydrological factors revealed only two, relatively weak correlations: (i) peak summer flow of the large river was inversely correlated with summer mortality in the floodplain of that river, and (ii) peak winter flow was positively correlated with winter mortality in all sites except the small flooded (Fig. 6). Relating these findings to flooding, however, is problematic because no floods were ever noted in most portions of the two unflooded sites.

If peak river flows are not indicators of flood effects on seedlings, perhaps they are correlated with other aspects of the hydrological cycle that affect seedlings, such as precipitation, evapotranspiration or soil water budgets. In the two large river sites, it is possible that summer soil moisture affected mortality. The water-table in the large river sites can drop to Ð1 m below the soil surface during summer (Jones et al. 1994), possibly resulting in severe water stress for small seedlings. Streng et al. (1989) found that a summer drought in 1980 greatly enhanced mortality among several cohorts and species. We found that summer rain was negatively correlated with mortality in the large river sites, although the relationships were weak (r = − 0.45 and −0.09, and P = 0.31 and 0.85, for the flooded and unflooded sites, respectively).

Environmental influences other than hydrology may have contributed to synchronized mortality peaks. Pigs have occasionally rooted up entire 1 × 1 m seedling plots, and most field surveys have noted some herbivore (mostly insect) damage to leaves. However, we failed to find a statistically significant influence of pig rooting on seedling survival during the first 3.5 years of this study (Jones et al. 1994), which is the time when patterns of mortality were most synchronized (Fig. 5). The causes of mortality, and their periodicity (or randomness), have been difficult to pin down using comparative field studies (Streng et al. 1989; Jones et al. 1994). Manipulations of environmental conditions in forest understories, or observations of permanent plots before and after disturbances (Hall 1993), will be needed to test hypothesized relationships between flooding, summer soil moisture, herbivory and seedling survival.


It is clear from this and other studies (Canham 1988; Streng et al. 1989; Denslow et al. 1990; Pacala et al. 1996) that growth of woody plant seedlings in forest understories is very slow. Virtually all of the seedlings that we measured were less than 30 cm tall, even if they were 8 or more years old. Pacala et al. (1996) estimated that seedlings of nine species would need 29–158 years (depending on species) to reach 3 m in height if growing in 1% light underneath northern hardwood forest canopies. Although none of our seedlings reached and maintained sapling size (i.e. sustained height >1 m), saplings exist in each of our four research sites (Jones et al. 1994). Hall (1993) found that saplings in the understorey of a Texas floodplain forest were primarily associated with canopy gaps. Without gaps, it seems highly improbable that a seedling in our study sites could reach sapling size.


Effects of height and height growth on survival were generally weak, and species specific. Taller species had greater long-term survival as we predicted; however, the correlation was weak (rs = 0.60). Taller plants had greater short-term (year-to-year) survival in only half of our statistical tests (Table 4 and Fig. 10), and height growth was related to survival in only one of nine tests.

Species differences in the height–survival response were partly related to growth form. For example, Berchemia scandens (a vine) remains erect for more than one year after establishment whereas Rhus radicans (another vine) is often decumbent by the second year. A short B. scandens is probably one that has died back, possibly because of low vigour, whereas a short R. radicans seedling may be a high-vigour stem that has sprawled out. Therefore it is not surprising that height is key to survival in B. scandens but not in R. radicans (Fig. 10).

A more accurate assessment of growth vs. survival might have been achieved if we had measured leaf area or biomass. In a greenhouse study, Walters & Reich (1996) found that second-year survival of northern hardwood seedlings was strongly related to relative biomass growth during the first year, and growth was directly related to leaf area ratio. For saplings in northern hardwood forests, Kobe et al. (1995) found that stem diameter growth was strongly correlated with survival probability.

Our analysis of height vs. survival may have been confounded by seed size effects. Tall species had large seeds (Spearman rank correlation rs = 0.63, P = 0.019) as well as high survival. Seed size is often correlated with survival (Streng et al. 1989; Westoby et al. 1992); however, the cause of high survival (large seed or large seedling) is difficult to identify.

We conclude from our analyses that height (i.e. seedling size) influences short-term and perhaps long-term survival in some but not all species. Height growth, however, is not strongly related to survival. Finally, since most species grow very slowly and remain quite small for many years (Fig. 7), it appears that survival could be best modelled as a function of species and age rather than species and size. Our final recommendation, then, is to model survival using separate equations for each species (Fig. 3) with mortality spikes introduced to reflect impacts of occasional droughts and floods (Fig. 6).


This study was supported by Financial Assistance Award Number DE-FC09-96SR18546 between the US Department of Energy and the University of Georgia. We thank Bruce Allen, Cathy King, Maria Kramer and numerous others for their help with field work and data management. Philip Dixon and Tom Philippi provided helpful guidance for statistical analyses. Anonymous referees provided helpful comments on our manuscript.

Received 19 May 1997
revision accepted 8 December 1997