Detecting demographic patterns of old coast redwood forests in northern California

Forest demography is difficult to describe accurately. Tree recruitment and mortality are uncommon events, so that rates based on these observations often require long survey periods using large numbers of trees. Demographic rates can scale with tree size, so challenges describing demography are pronounced in coast redwood forests that are dominated by extremely large Sequoia sempervirens. Here, we present 10 years of observations from 15 1-ha plots in unmanaged primary coast redwood forests in northern California. The size distribution of live stems followed a “reverse-J” shape, with many small trees relative to large trees. Sequoia dominated these plots, particularly in terms of stem area. Mortality and recruitment rates varied among common species, with Sequoia having much lower mortality relative to co-occurring conifers and hardwoods. Mortality varied substantially among size classes and taxonomic groups, while little variability in recruitment occurred among taxonomic groups. Mortality rates were consistently greater than recruitment across taxonomic groups during our 10-year observation period. Our area experienced heavy winter storms in 2022–2023 resulting in numerous blowdowns, although mortality still outpaced recruitment when we excluded years affected by the storms. Simulations using current plot data found low statistical power to detect potential future changes in the mortality rate of Sequoia. We consider how these results could inform conservation of old Sequoia forests.

Tree demography is a critical determinant of forest structure and function. For example, small sustained changes to tree mortality rates (e.g., from 1 to 2% year −1) in a forest would eventually halve average tree age, potentially reducing expected average tree size and carbon storage (van Mantgem et al., 2009; McMahon et al., 2019). The accurate description of forest demographics hinges on detection of uncommon events (i.e., the birth and death of individual trees), so that observations from large numbers of trees over long periods of time are needed to accurately estimate rates. Tree mortality, seed production, and growth can scale either positively or negatively with tree size (Greene and Johnson, 1994; Stephenson et al., 2014; Das et al., 2016), which introduces further difficulties identifying demographic differences among size-based subpopulations.

Challenges describing demography are pronounced in old coast redwood forests, as these forests are dominated by extremely large Sequoia sempervirens (D. Don) Endl. (coast redwood). Old forests dominated by S. sempervirens (hereafter Sequoia) are also of high conservation value, having less than 5% of their area remaining following a history of extensive logging beginning in the 19th century (Burns et al., 2018). The great ages and large sizes attained by Sequoia translate to record-holding amounts of per-area aboveground biomass (van Pelt et al., 2016; Sillett et al., 2020, 2024). In spite of the interest in Sequoia, there are relatively few published estimates of demographic rates for this species, with most studies based on a small number of plots or mortality proxies. Busing and Fujimori (2002) estimated annual mortality and recruitment rates of 0.29% year −1 for Sequoia ≥ 10 cm stem diameter at 1.7 m over 29 years in a single 1.44 ha plot (Sequoia mortality rates dropped to 0.21% year −1 for trees > 40 cm diameter and 0.17% year −1 for trees > 100 cm diameter). Busing and Fujimori (2002) found other tree species at the plot had mortality rate of 1.03% year −1 and recruitment rate of 1.09% year −1 for stems ≥ 10 cm diameter at 1.7 m. Benda et al. (2002) used the recruitment of downed wood in streams to estimate average conifer mortality rate of 0.04% year −1 in Sequoia dominated forests in northern California. A recent analysis found increasing numbers of Sequoia > 100 cm DBH (diameter at breast height, 1.37 m), which may be related to recruitment into this size class due to lengthening harvest intervals (Chisholm and Gray, 2025). The Sequoia demographic rates reported by Busing and Fujimori (2002) are less than mortality rate of 0.82% year −1 found in the related giant sequoia (Sequoiadendron giganteum (Lindl.) J. Buchholz) in unburned forests at Sequoia and Kings Canyon national parks (the Sequoiadendron mortality rate was 0.10% year −1 for stems approximately > 30 cm DBH) (Lambert and Stohlgren, 1988). Historically both Sequoia and Sequoiadendron lacked major pests and pathogens (Olson et al., 1990; Weatherspoon, 1990), which may partially explain low mortality in these species. The reported demographic rates for Sequoia are low relative to rates published in a global study of forest tree demography (Stephenson and van Mantgem, 2005).

Describing and contextualizing mortality and recruitment in coast redwood forests is complicated by the resprouting habit of Sequoia and co-occurring hardwoods (e.g., Notholithocarpus densiflorus (Hook. & Arn.) Manos, C. H. Cannon & S. Oh, tanoak) (Olson et al., 1990; Tappeiner et al., 1990; O’Hara et al., 2017; Sillett et al., 2020). Unusually for conifers, Sequoia resprouts vigorously following damage to stems, branches or roots, with sprouts developing even in cases where the main stem is toppled. Notholithocarpus resprouts from buds at the stem base following mechanical damage. Mortality for resprouting species often represents the death of the aboveground structural unit (i.e., the ramet, or stem) rather than the genetic individual (i.e., the genet). Resprouts may have large root-to-shoot ratios with the ability to draw on stored resources from remaining tissues, allowing the individual genet to persist and potentially reoccupy available growing space (Pausas et al., 2016; Peltier et al., 2023). The ability of major species in coast redwood forests to resprout confers resilience to disturbance (Falk et al., 2022), although major disturbances (e.g., logging) may require centuries for natural recovery of some important features of old forests (e.g., dominance of Sequoia, development of large trees, Veirs and Lennox, 1981; Russell and Michels, 2010; Iberle et al., 2020). Recovery from acute disturbances may be rapid for resprouting species, but consequences of chronic changes to mortality still apply (i.e., changes to average tree—or stem—age and size, McMahon et al., 2019). While some features of coast redwood forests will be buffered if changes to mortality are limited to small trees (e.g., stand-level biomass), other features may be affected (e.g., expected regeneration in canopy gaps, Busing and Fujimori, 2002; van Pelt et al., 2016).

Sequoia stem mortality is often caused by windthrow (blowdown) following heavy winter rains (Lorimer et al., 2009). Windthrow can affect individual trees directly, or trees can be damaged or toppled by being struck by other windthrown stems. Importantly, windthrow can affect large trees, which influences stand-level forest characteristics such as aboveground biomass and the presence of canopy gaps. Sequoia is resistant and resilient to even high-severity fire; thick bark and high crowns confer fire resistance, while vigorous post-fire resprouting allows Sequoia to rapidly regenerate fire-damaged foliage (Peltier et al., 2023; Kane and Carrasco, 2025; Keeley and Pausas, 2025). Sequoia can also develop adventitious roots to withstand flooding sedimentation on alluvial flats (Stone and Vasey, 1968). Mortality in coast redwood forests is also associated with landslides and slope failures on unstable soils (Hunter and Parker, 1993; Benda et al., 2002; Lorimer et al., 2009).

Several studies have documented chronically increasing tree mortality in boreal, temperate and tropical forests (van Mantgem and Stephenson, 2007; van Mantgem et al., 2009; Peng et al., 2011; Brienen et al., 2015; Hember et al., 2017; Senf et al., 2018; McDowell et al., 2020; Bauman et al., 2022; but see Acker et al., 2023) and increasing occurrence of abrupt forest die-back (Allen et al., 2015; Hammond et al., 2022). These patterns are often linked to warming and consequent changes in the hydrologic cycle and disturbance regime. Within the region occupied by coast redwood forests, precipitation volatility is expected to increase (Swain et al., 2018, 2025) along with a compression of the rainy season (Luković et al., 2021). These changes portend increasing frequencies of droughts, storm activity, and wildfire, all of which may affect Sequoia forests. Recent studies show tree growth is sensitive to drought for species in Sequoia forests (Sillett et al., 2022; Robinson et al., 2023; Lalemand et al., 2024), suggesting that recurring droughts may affect tree growth and competitive relationships, which may in turn influence stem mortality probabilities (Cailleret et al., 2017). Changes in fog distribution (Johnstone and Dawson, 2010), if realized, would be expected to affect Sequoia populations. The recent arrival of the exotic forest pathogen Phytophthora ramorum, the causal agent of Sudden Oak Death (SOD), has the potential to alter forest species composition, dynamics, and fire hazards (Forrestel et al., 2015; Sillett et al., 2020), particularly where stands have heavy representation of sensitive species such as Notholithocarpus. These changes are occurring in a forest type that has generally experienced more than a century of fire exclusion (Lorimer et al., 2009).

While rapid demographic changes (e.g., at decadal scales) of any major species in coast redwood forests may serve as an ecological indicator of changing conditions, we are particularly interested in changes to Sequoia populations, as Sequoia dominates many forests where it occurs and is of great conservation and societal value. We compiled information over 10 years from 15 1-ha plots in old unmanaged forests without a history of logging (i.e., primary or old-growth) in the northern portion of the range Sequoia within Redwood National and State Parks (RNSP). We were interested in assessing how demographic rates varied among species and tree sizes in this forest type. We extended our results to conduct a simulation study to determine if current monitoring efforts are sufficient to detect potential future changes in mortality for Sequoia.

Study area

RNSP is located in far northern coastal California (Figure 1). The climate is Mediterranean, with dry, warm summer seasons alternating with cool, rainy winters. Dry summer conditions are ameliorated by fog and high stratus clouds, providing considerable moisture to coastal locations (Torregrosa et al., 2016). Fire was historically frequent in upland areas but has been effectively excluded over the past century (Lorimer et al., 2009). Besides Sequoia, coniferous forests at RNSP typically contain Pseudotsuga menziesii ([Mirbel] Franco var. menziesii, Douglas-fir) and western hemlock (Tsuga heterophylla (Raf.) Sarg.). Grand fir (Abies grandis (Douglas ex D. Don) Lindl.), Sitka spruce (Picea sitchensis (Bong.) Carrière) are also present in these forests. Common hardwoods include tanoak and Pacific madrone (Arbutus menziesii Pursh). Recent events at RNSP that could have broadly influenced forest conditions include the appearance of sudden oak death (Morgan, 2017; Sillett et al., 2020), and a series of large storms with high winds in the winter of 2022–2023 (DeFlorio et al., 2024).

Data sources

Field data

We capitalized on several ongoing plot-based forest monitoring efforts at RNSP. We considered large plots (1-ha) in old forests dominated by Sequoia with multiple complete measurements. The plots are located in areas that do not have records of logging or recent management. There were 15 monitoring plots at RNSP that met these criteria (Table 1 and Figure 1). The plots were located with specific site requirements (see below). We used the centroid of each plot to extract elevation, slope, and aspect from a 10 m digital elevation model. We focused on surveys from 2014 to 2024, an observation interval within a few years of plot establishment and representing the longest common interval to all plots (avoiding potential artifacts calculating demographic rates with unequal sample intervals) (Sheil and May, 1996; Lewis et al., 2004).

Each plot has complete censuses of individual trees at or above a minimum stem diameter, which varied among programs. Trees have been tagged, tracked for status (live or dead), and repeatedly measured for stem diameter at the tag, typically at or near breast height (1.37 m). During field surveys crews searched for new trees reaching the diameter threshold to be considered trees, which varied among programs. The following subsections describe the monitoring programs.

Cussins plots

This is a collection of six 1-ha plots established along a ridgetop site to serve as reference conditions for a demonstration second-growth forest (logged) restoration project (Teraoka and Keyes, 2011). The plots were stem mapped and trees were measured in 1995. In 2010, we returned to these sites, placing plot markers and tagging trees, and remeasuring individual trees ≥ 20 cm DBH. Since 2010 all Cussins plots were resurveyed in 2014, 2015, 2023, and 2024 (with stem diameter measurements occurring in 2010, 2015, and 2023). In addition to these surveys, in 2011 the minimum stem diameter measured at the Cussins1 plot was lowered to ≥ 5 cm DBH (i.e., trees ≥ 5 to < 20 cm DBH were measured and tagged). Field crews assigned newly dead trees to categories for proximate causes of mortality; biotic (i.e., heavy pest or pathogen infestation evident), mechanical (i.e., stem breakage, windthrow), or unknown (standing dead without obvious damage).

Redwoods and climate change initiative plots (RCCI)

This program surveys six 1-ha monitoring plots at RNSP in areas with locally tall Sequoia (with plot-level maximum tree heights approaching or exceeding 100 m) supporting analyses that demonstrated how coast redwood forests achieve global maxima for biomass and leaf area index (van Pelt et al., 2016; Sillett et al., 2020). The RCCI plots are remeasured annually for stem mortality and recruitment for stems ≥ 5 cm DBH (stems < 5 cm DBH are tracked in subplots, which were not considered here). The RCCI track stem diameters above the basal buttress for large trees, which can be well above 1.37 m in height. For the RCCI plots at RNSP stem diameter measurements occurred between 2009 to 2011, and again between 2014 to 2016. Intensive measurements of individual tree architecture and growth have been done at the RCCI plots (Sillett et al., 2015, 2020), but are not used in this study. Within the RCCI plots at RNSP, sudden oak death has caused extensive mortality of Notholithocarpus beginning in 2017 at a single plot (RNP low) (Sillett et al., 2020).

Sudden oak death plots (SOD)

This is a group of three 1-ha plots established in 2010, randomly located in areas of heavy representation of Notholithocarpus at Redwood National Park. To date, sudden oak death has not been detected in these plots (van Mantgem and Sarr, 2015), though it is found in other areas of the park (Morgan, 2017; Sillett et al., 2020). The SOD plots have been remeasured annually for mortality and recruitment of stems ≥ 5 cm DBH, with stem diameter remeasurements on a 5-year schedule (diameter survey years include 2010, 2015, and 2020). A single SOD plot (LostMan) has two missing surveys in 2021 and 2023 due to site accessibility restrictions. Field crews assigned newly dead trees to proximate cause categories as in the Cussins plots. The data for the Cussins and SOD plots appears in van Mantgem and Wright (2026).

Analyses

We considered differences among taxonomic groups, defined as follows; Sequoia, other conifers (Abies, Picea, Tsuga, Pseudotsuga), and hardwoods (primarily Notholithocarpus, but also including Acer macrophyllum Pursh, Alnus rubra Bong., Arbutus, Chrysolepis chrysophylla (Hook.) Hjelmq., Cornus nuttallii Audubon, Frangula purshiana (DC.) A. Gray ex J. G. Cooper, Umbellularia californica (Hook. & Arn.) Nutt.). For each taxonomic group we calculated stem diameter distributions, plot-level stem density (count of live stems per unit area), and plot-level stem area (summed cross-sectional area of live stems per unit area). Stem diameters for individual trees were taken from measurements at 2014. If stems were not measured in 2014, we used earlier and later measurements to interpolate stem diameters to expected sizes in 2014. We calculated the annualized mortality and recruitment rate (Nakagawa et al., 2000; Kohyama et al., 2018) for common species (defined as ≥ 100 live stems in 2014) for plots with a minimum stem diameter was ≥ 5 cm:

where m and r are, respectively, the annualized mortality and recruitment rate, n live 0 is the count of live trees at the start of the observation period, n died is the count of trees that died by the end of the observation period, n recruit is the count of trees that recruited by the end of the observation period, and t is the observation period length in years.

To gain a better understanding of how tree size influences demographic rates in these forests, we calculated demographic rates using the full set of plots for trees with a minimum diameter of ≥ 20 cm. We also calculated demographic rates for Sequoia with stem diameters ≥ 100 cm. Our most common hardwood, Notholithocarpus, experienced elevated mortality from sudden oak death at a single plot (Sillett et al., 2020); to estimate demographic rates without the effects of this exotic pathogen we calculated demographic rates for this species excluding this plot. Separately, we gauged the impacts of the winter storms of 2022–2023 by calculating demographic rates excluding observations following these storms in 2023 and 2024.

Statistical models

We found that mortality was consistently greater than recruitment rates (see Results), so we created a statistical model of differences in plot-level mortality and recruitment rates between 2014 and 2024 for three broad taxonomic groups (Sequoia, other conifers, and hardwoods). We considered new recruits as stems that grew to ≥ 5 cm between 2014 and 2024, and therefore only included plots where the minimum stem diameter was ≥ 5 cm. To describe differences in mortality among tree sizes, we modeled how morality rate varied by taxonomic group and stem diameter (binned into three stem diameter classes, 5 to < 20 cm, 20 to 50 cm, and > 50 cm). We used all plots for this model. We did not attempt a size-based model of recruitment, as we could not determine which potential parent trees contributed to recruitment. We had only nine plots with annual resolution data, and at most 15 years of data at any of these plots. Given these limitations, we did not pursue analyses of annual trends in demographic rates.

We modeled counts of mortality and recruitment using a negative binomial regression, with plot identity as a random effect:

where μ ij and θ are the mean and dispersion parameter for the negative binomial distribution. The expected mean count μ ij of mortality or recruitment for the j th observation in the i th plot was modeled using a mixed-effects model with a log link function:

where α is the global intercept, β are regression coefficients, Typ e ij are covariates for observation type (mortality or recruitment), TaxG r ij are covariates for taxonomic group, log (Ntree ij) is an offset for the number of live trees in 2014, and u i is a normally distributed random intercept for plot identity. We fit an additional model that included covariates for elevation, aspect, and slope, but the parameter estimates for these effects overlapped 0 and were not considered further.

We used a similar approach for a size-based mortality model, but considered the effects of size class, taxonomic group, and their interactions on mortality only:

where μ ij is expected mean count of mortality for the j th observation in the i th plot, α is the global intercept, β k are regression coefficients, Siz e ij are covariates for stem size class, TaxG r ij are covariates for taxonomic group, log (Ntree ij) is an offset for the number of live trees in 2014, and u i is a normally distributed random intercept for plot identity.

We fit the models using Stan via the brms package (Bürkner, 2017) in R (R Core Team, 2025). Beginning with weakly informative priors, we ran models with four chains of 5,000 iterations each, including 3,000 warmup iterations which were discarded. We determined model convergence using visual assessments of traceplots and ensured that the r ̂ convergence statistic was less than 1.01 for all variables (Vehtari et al., 2021). We present posterior medians and 95% credible intervals (CIs) for model parameter estimates. For selected parameter estimates we also present the probability of direction (PD), which can be defined as the proportion of the posterior samples that have the same sign (i.e., positive or negative) as the posterior median.

Sequoia mortality power analysis

We tested the ability of the current plot networks to detect a 10-year chronic change in Sequoia mortality rate, creating simulations similar to a power analysis. First, we used the observed plot data for Sequoia from 2014 to 2024 to describe size-based differences in stem mortality similar to the model outlined above, but without a term for taxonomic group. We used the standard deviation of u i from this model to generate new, normally distributed random intercepts for each simulated plot. We also included the posterior median from the fitted α parameter as a constant in each simulation. We created a set of scenarios varying the number of plots (15, 30 or 45 plots) and varying values for a change in the mortality rate suggesting an approximate increase of 10, 25, 50, 75 and 100% in mortality probability, resulting in 15 total scenarios. For each iteration we created a starting value for the number of trees in each size class per plot determined from a negative binomial distribution parameterized from the observed plot data. We ran 500 iterations for each of the 15 scenarios, for a total of 7,500 iterations.

We then refit the model to each simulated dataset including an additional parameter for the change in the mortality rate (β change parameter) using 5,000 iterations, including 3,000 warmup iterations. We used these model fits to caculate the 95% credible intervals for β change. We removed from consideration any models that had divergent transitions or r ̂ values > 1.01 (5% of models). We calculated power by determining the proportion of simulations where the 95% credible intervals for the β change parameter exceeded zero. We used the conventional power threshold of 0.8. The simulations were run using the USGS Hovenweep Supercomputer (Falgout et al., 2023).

Forest structure

The stem size distribution of live trees across taxonomic groups in 2014 followed a “reverse-J” shape, with many small trees relative to large trees (Figure 2). The stem size distribution for Sequoia showed greater frequencies of large trees relative to other conifers and hardwoods. Mean live stem density for all species in 2014 was 272 (1 SD = 137) trees ha −1, while mean live stem area was 217 (1 SD = 66) m 2 ha −1. Stem density showed substantial variation in the representation of the taxonomic groups across the plots, while stem area was heavily dominated by Sequoia (mean stem area representation of Sequoia was 75.5%). Compared to 2014, stem densities in 2024 were slightly lower for all but one plot (2024 average stem density = 244 trees ha −1, 1 SD = 113).

Demographic rates

Mortality surpassed recruitment from 2014 to 2024 for each species where we had > 100 starting individuals across the plots (Table 2). We found low rates of stem recruitment and mortality for Sequoia, with the mortality rate of Sequoia nearly half that of Pseudotsuga (though with low numbers of Pseudotsuga to calculate mortality). The plot data did not record any instances of recruitment for Pseudotsuga. Tsuga had high rates of stem recruitment and mortality relative to Sequoia and Pseudotsuga. Notholithocarpus had relatively high recruitment and extremely high mortality, partially reflecting the effects of sudden oak death in one of the RNSP RCCI plots (removing the RNP low plot resulted in Notholithocarpus starting counts of 638, with 39 recruits and 123 mortality counts by 2024, giving an annual recruitment rate of 0.6% year −1 and annual mortality rate of 2.12% year −1). Excluding recruitment and mortality observations following the winter storms of 2022–2023 removed approximately a third of the observed mortalities, but the imbalance between recruitment and mortality rates remained (e.g., from 2014 to 2022 Sequoia stem recruitment rate was 0.18% year −1 and stem mortality rate was 0.35% year −1) (Supplementary Table S1). Within the Cussins and SOD plots where proximate mortality causes were recorded, most mortalities were associated with unknown causes (n = 140 of a total of 311 proximate causes recorded), though there were many dead trees coded with mechanical proximate causes of mortality in 2023 (n = 103). For stems ≥ 20 cm DBH we found similar patterns for demographic rates among species, but with lower rates relative to plots with a minimum DBH of ≥ 5 cm (Supplementary Table S2). For Sequoia ≥ 100 cm diameter, we had a count of 574 live stems in 2014 with 6 observations of mortality by 2024, giving a stem mortality rate of 0.11% year −1.

Our statistical model of differences between mortality and recruitment showed that between 2014 and 2024 recruitment was substantially less than mortality (Figure 3). Relative to the non- Sequoia conifers taxonomic group, Sequoia had fewer counts of mortality and recruitment, while counts were higher for the hardwoods group (though 95% CIs slightly overlapped zero). The pattern of mortality outpacing recruitment was common to all taxonomic groups, without evidence for interaction terms between taxonomic groups and observation type.

Our size-based model stem mortality showed substantial variability among taxonomic groups and stem size classes (Figure 4). For taxonomic groups, Sequoia again had lower stem mortality relative to other conifers, while we had less evidence for higher stem mortality of hardwoods relative to non- Sequoia conifers (the 95% CI of estimate overlapped 0, PD = 94%). Mortality was less in the larger two stem diameter size classes (20 to 50 cm and > 50 cm diameter classes) relative to the smallest size class (5 to < 20 cm). We found evidence for interactions between taxonomic group identity and stem diameter size class, showing attenuation of differences for Sequoia at > 50 cm and intensification of differences for hardwoods at > 50 cm. Sequoia at > 50 cm had lower predicted stem mortality than other taxonomic group and size combinations, and hardwoods at > 50 cm had greater predicted mortality than other taxonomic group and size class combinations (although we had a low sample size for this group).

Sequoia mortality power analysis

Given the observed variability in Sequoia stem mortality at our forest plots, simulations showed that increasing magnitudes of stem mortality probabilities and number of plots increased the likelihood that a change mortality would be detected (Figure 5). Relatively small increases in Sequoia stem mortality (10 and 25%) would be difficult to detect with even triple the number of 1-ha plots currently monitored at RNSP. When Sequoia mortality was set to levels approximating ≥ 75% increase, we found that a collection of ≥ 30 1-ha plots may be needed to discern this change. With 15 plots being monitored, it is questionable if we can expect to detect even a doubling of Sequoia mortality probabilities (100% increase) over 10 years.

Our data provide a rare estimate of current conditions and dynamics of old coast redwood forests in the northern portion of its range. Stem diameter size distributions followed a “reverse-J” pattern, which is often viewed as a characteristic of stable forest tree populations (Westphal et al., 2006). Sequoia size distributions had high frequencies of large trees relative to co-occurring conifers and hardwoods. Demographic rates were also variable among species, with Sequoia having lower stem recruitment and mortality rates relative to common co-occurring conifers and hardwood species. Demographic differences among taxa were consistent with Busing and Fujimori (2002), who found Sequoia stem recruitment and mortality rates were more than three times lower than that of co-occurring tree species. This may be a consequence of plant traits (e.g., Sequoia ’s heavy investment in rot-resistant heartwood and ability to recover from major mechanical damage by resprouting) and differing effects among taxa of lethal pests and pathogens. The low mortality of Sequoia may partially explain how it is able to attain extremely large sizes and is congruent with the great ages stems of this species can attain, with some individuals reaching ages that exceed 2,000 years (Brown, 1996). Higher mortality for co-occurring species mirrors their shorter maximum recorded lifespans relative to Sequoia (Brown, 1996; Biondi et al., 2023). Our estimates of annualized recruitment and mortality rates for Sequoia stems were comparable to an earlier report for Sequoia (Busing and Fujimori, 2002), but low relative to reports from other forest types (including other temperate gymnosperm forests) presented in a global analysis of tree demographic rates (Stephenson and van Mantgem, 2005). These results show that current demographic rates for Sequoia are exceptionally low on a global scale.

In the absence of major disturbances, there is an expectation that mortality and recruitment should be approximately balanced in old forests (i.e., a dynamic equilibrium between small-scale mortality and recruitment) (Lutz et al., 2014). For common species and across broad taxonomic groups we found mortality was consistently greater than recruitment from 2014 to 2024 (e.g., mortality of Sequoia stems was slightly more than twice as great as recruitment). This contrasts with Busing and Fujimori (2002), who found identical stem recruitment and mortality rates in Sequoia. It is possible that in future years recruitment may increase to exploit growing space following mortalities at our plots, particularly for the aggressively resprouting and reiterating Sequoia. However, if this demographic imbalance is maintained into the future, it could result in changes to fundamental forest metrics such as forest age structure and average tree stem size (van Mantgem et al., 2009; McMahon et al., 2019).

Although these findings suggest parallels with reports of increasing mortality in otherwise undisturbed old forests (e.g., van Mantgem et al., 2009), we are uncertain that stem mortality will continue to outpace recruitment at RNSP. We recognize that a decade of observations is short relative to the potential > 2,000-year maximum lifespan of Sequoia. Our observations of high mortality relative to recruitment could arise from three non-mutually exclusive sources; poor data quality, sampling error, or recent changes in the underlying environmental or biological conditions. We believe that poor data quality is unlikely, as the data were taken from repeated measurements of permanent forest plots. Tree status (live or dead) is typically straightforward to observe and repeated surveys provide the opportunity to correct observation errors (e.g., early observations of mortality can be corrected if an individual stem is later found to be alive). Increasing mortality may simply be the result of sampling error, where by chance our 10-year observation period at RNSP found more frequent observations of mortality relative to recruitment. An additional partial explanation of the imbalance between stem recruitment and mortality is that the winter of 2022–2023 saw a series of large storms with strong winds in northern California (DeFlorio et al., 2024). However, mortality still exceeded recruitment when excluding observations following the 2022–2023 winter storms. While the 2022–2023 winter storm activity was anomalous for our observation period, hydroclimatic volatility may increase in coming years in northern California (Swain et al., 2018, 2025), which could lead to greater frequencies windthrow mortality.

Other potential causes of the demographic imbalance have less a priori support; RNSP has so far not experienced large increases in temperature seen in other coast redwood protected areas (Gonzalez, 2015, 2016). Decreasing coastal fog (Johnstone and Dawson, 2010) could affect Sequoia, but potentially with less impact on co-occurring conifers relative to Sequoia (which is adapted to take advantage of fog moisture, Burgess and Dawson, 2004). Our forest plot surveys did not capture targeted observations of pests and pathogens, though Sequoia has very few enemies that result in mortality (Olson et al., 1990). For hardwoods, the presence of sudden oak death clearly affected Notholithocarpus at one plot, although excluding this plot still produced imbalanced stem recruitment and mortality rates for this species. At present we have insufficient information to support or refute any of these potential causes. Expanding the duration, extent, and scope of plot-based field surveys could help determine if imbalanced mortality and recruitment is merely a sampling anomaly or an early symptom of changes to the physical or biological conditions where northern coast redwood forests occur.

We caution, however, that simply expanding current plot-based observations may not be sufficient to detect future demographic changes in Sequoia. Our simulations showed that even tripling the number of plots currently monitored may fail to detect increasing Sequoia mortality unless changes were large (e.g., ≥ 50% increase in mortality probabilities). This implies that our ability to uncover demographic changes from plot-based information alone may be limited. How then to improve our ability to detect changes in Sequoia populations?

Plot-based sampling could be augmented with censuses of randomly selected individuals across the landscape, which may efficiently capture mortality rates (McMahon et al., 2019). However, this approach is not likely to be feasible at RNSP, an area with steep topography and heavy vegetation. Censusing random individual trees also fails to capture other elements of interest, such as stand structure and information concerning co-occurring species. Reducing plot areas to < 1 ha may allow researchers to sample more plots across a broader area, though at the sacrifice of accuracy in estimating demographic rates. Indeed, it may be desirable to use plot areas substantially larger than 1-ha to capture Sequoia demography, as the number and spatial arrangement of trees may influence sampling accuracy (Williams et al., 2001; Gray, 2003). Obtaining data across a wider range of topographies and climates may help to determine the influences of these factors on coast redwood forest demographics, as has been found analyses of other forest types at regional and global scales (Stephenson and van Mantgem, 2005). Additional years of observations across additional plots could also help determine if there are demographic trends in old coast redwood forests, as in other forests (e.g., van Mantgem et al., 2009).

Coupling ground-based observations with remote sensing could complement current efforts (International Tree Mortality Network, 2025). Reflectance-based remotely sensed data are particularly suited to capture changes in canopy cover for patches of large trees (i.e., canopy gaps) over large spatial scales. These data may be especially useful when overstory mortality is spatially clustered (Khatri-Chhetri et al., 2024) but may be incomplete because understory trees can escape detection using these methods. Airborne laser scanning (ALS, or lidar) can identify individual overstory trees (or approximations of individual trees, Jeronimo et al., 2018). Multi-temporal ALS could provide a landscape-scale depiction overstory changes, which could then be used to approximate mortality and recruitment of large trees. ALS assessments of canopy change could in turn be related changes in reflectance using high temporal resolution data (e.g., Sentinel-2, https://sentinel.esa.int/web/sentinel/missions/sentinel-2) to provide annual estimates of overstory tree mortality and recruitment. A fusion of remotely sensed products and plot data could be used to develop a low-cost, passive monitoring system for large Sequoia and co-occurring species.

While there are ongoing efforts to track Sequoia populations at RNSP, many of these plot-based networks lack programmatic support. If these networks are to be maintained, or even expanded, a targeted monitoring program for Sequoia would be useful. The design of such a monitoring program could focus on Sequoia growth, recruitment, and death as the basic processes that govern the nature and persistence of Sequoia populations. Once developed, this monitoring strategy could be expanded to areas outside of RNSP to provide range-wide assessments of Sequoia populations. Given the stressors that may affect Sequoia in the coming century, and Sequoia ’s global significance, such a monitoring strategy could represent an important component in preserving Sequoia into the future.