We analyzed and modeled the demography of Eryngium cuneifolium, an herbaceous species endemic to the fire-prone Florida scrub, using 10 annual censuses (1990-1999) of 11 populations at Archbold Biological Station. Nearly every aspect of the demography of this plant is affected by time since fire. Year, time since fire, life history stage, and plant age affected survival, growth, and fecundity of E. cuneifolium, but time since fire and life history stage had the most consistent effects. Survival, flowering stem production, and early seedling survival were highest in recently burned sites. Long-term survival, growth, and fecundity were highest for yearling cohorts recruiting recently after fire, with the largest contrast between plants recruiting two years postfire and those recruiting more than a decade postfire. Prior (historical) stage also affected individual plant fates. For example, plants with prior stasis or regression in stage subsequently died in greater numbers than plants with prior advancement in stage. Historical analyses did not suggest any cost associated with the initiation of flowering. We used a matrix selection approach to explicitly model Eryngium cuneifolium population viability in relation to fire. This simulation strategy included preserving observed data and variances within projection matrices formed for individual combinations of population and year. We built 54 of these matrices, each with six stages (seed bank, yearlings, vegetative plants, and three reproductive stages). Each of these matrices also represented a specific time since fire. In building matrices, we minimized the use of pooled data while retaining specific matrices whenever possible. In this way, we preserved both the correlation structure within individual matrices (populations, years) and protected patterns among matrices across the time-since-fire gradient. To deal with less-detailed data on recruitment processes, we evaluated 13 fertility and seed bank scenarios that bracketed a range of outcomes. All scenarios were similar in showing the positive effects of fire on the demography of E. cuneifolium. The scenario with high seed bank survival (0.5) and low germination rates (0-0.005) was the best predictor of observed postfire years of peak aboveground population size (similar to8 yr) and aboveground population disappearance (30-34 yr), and also did a good job of reproducing observed population trajectories. Finite rates of increase (lambda) were > 1 only during the first decade postfire but then declined beyond a decade postfire. Although prior (historical) stage affected most individual demographic parameters, it did not significantly influence finite rates of increase. Elasticities were highest for stasis and germination from the seed bank. Elasticities for survival increased with time since fire, while growth and fertility elasticities decreased. In historical models (those with information on stage from the second-to-last year), the elasticities for stasis were higher and the elasticities for growth lower, compared to models without this history. Bootstrapping suggested small standard errors for several types of model output. Most matrix elements were positively correlated, suggesting that favorable conditions affect many life history stages similarly, and that simulations using element selection would provide a less conservative risk assessment than the matrix selection technique used. We used a stochastic simulation program to simulate changing demography with time since fire, with various fire-return intervals, and for various initial population sizes. We obtained estimates of extinction risk and probability of population decline. Even populations as large as thousands of individuals will become extinct in the absence of fire. Fire-return intervals of 15 yr or less are necessary for E. cuneifolium persistence at individual sites. Fires at intervals longer than 20 yr create substantial extinction risks, and intervals longer than 12 yr produce declining populations. Cycles of widely divergent, alternating short and long fire-return intervals caused slightly higher chances of extinction compared to regular fire-return intervals. Although shrub regrowth is implicated in the decreased viability of E. cuneifolium populations under regimes of infrequent fire, aboveground fuel increases are often too slow to allow frequent burning in Florida rosemary scrub. If E. cuneifolium's rosemary scrub habitat burns less often than every 20 yr, local extinctions and metapopulation dynamics may be the norm. Other rosemary scrub specialists (e.g., Hypericum cumulicola) thrive with less frequent fires and persist in smaller gaps among the regrowing shrubs. Therefore, we suggest that temporal variation in fire-return intervals and spatial variation in fire intensity and patchiness (pyrodiversity) will allow coexistence of all Florida scrub species and hedge against local extinctions of specialists like E. cuneifolium.
