Climatic sensitivity of species’ vegetative and reproductive phenology in a Hawaiian montane wet forest

Understanding the way tropical tree phenology (i.e., the timing and amount of seed and leaf production) responds to climate is vital for predicting how climate change may alter ecological functioning of tropical forests. We examined the effects of temperature, rainfall, and photosynthetically active radiation (PAR) on seed and leaf phenology in a montane wet forest on Hawaiʻi using monthly data collected over ∼6 years. We expected that species’ phenologies were more sensitive to temperature and PAR than to rainfall at this wet tropical site because rainfall is not limiting. Seed production declined with increasing temperatures for two foundational species in Hawaiian forests (Acacia koa and Metrosideros polymorpha). Seed production also declined with rainfall for two species, and greater PAR for one species. One species showed relatively flat responses to climate. Community-level leaf phenology was not strongly seasonal. Unlike seed phenology, we found no effect of temperature on leaf phenology. However, leaf fall increased with rainfall. Climatic factors explained a low to moderate proportion of variance for both seed and leaf litterfall, thus the impact of future climate change on this forest will depend on how climate change interacts with other factors such as daylength, biotic, and/or evolutionary constraints. Our results nonetheless provide insight into how climate change may differentially affect different species with potential consequences for shifts in species distributions and community composition.

Variability in temperature is not as great in the tropics compared to temperate or high latitude regions, thus tropical species may be more sensitive to warming because these organisms are adapted to a narrower range of temperatures and may be living closer to their upper thermal limits (Janzen 1988, Tewksbury et al. 2008, Wright et al. 2009). Tropical regions are not warming as much or as fast as high latitude regions (IPCC 2014), yet the tropics may experience novel climates outside of their historical range much sooner (Williams et al. 2007, Mahlstein et al. 2011, Mora et al. 2013. Consequently the physiological tolerance of tropical species combined with the pace of environmental change will determine their vulnerability to climate change (Tewksbury et al. 2008, Kingsolver 2009).
Although temperature has not received much attention in the search for abiotic drivers of tropical phenology, temperature is a fundamental constraint on numerous biological processes (Kingsolver 2009). Seasonal flowering patterns were linked to changes in temperature at two tropical sites with long-term data (Pau et al. 2013). In a lowland moist seasonal forest in Panama and a montane ever-wet forest in Puerto Rico, warmer months were associated with greater flowering activity (Wright & Calderón 2006). There are few physiological experiments that have examined the temperature sensitivity of tropical reproduction, but reproductive organs are known to be highly sensitive to temperature (Larcher & Winter 1981, Slot & Winter 2016. It is unclear if tropical leaf phenology is also sensitive to temperature fluctuations. There is remarkable diversity in patterns of tropical flowering (Newstrom et al. 1994, Sakai 2001, seed and leaf production (Frankie et al. 1974;Reich 1995;Leishman 2000).
Identifying species' phenological strategies should help us predict the potential 'winners' and 'losers' of climate change. For example, species that continuously reproduce throughout the year may be less strongly cued by climate and may thus be less sensitive to future climate change (Pau et al. 2011). The divergent responses of species' reproductive phenologies to climate change may alter plant community composition if species are recruitment limited (Drake et al. 1998, Hubbell et al. 1999, Inman-Narahari et al. 2013. Plant phenology also has important cascading effects throughout the community by structuring the timing of food availability for many organisms. The Hawaiian flora provides unique insight into the ecology and evolution of plant diversity, and is considered a model system due to its isolation, endemism and the relative simplicity of its processes due to low species diversity (Sakai et al. 1995, Wagner & Funk 1995, Price & Wagner 2004). Yet there are few published studies on the reproductive phenology of Hawaiian forests (e.g., van Riper III, 1980;Drake, 1992;Berlin et al., 2000). More work has examined monthly leaf litterfall, yet these studies have generally focused on leaf litter's role in nutrient cycling and contributions to aboveground net primary productivity (e.g., Vitousek et al. 1995, Raich 1998, Schuur & Matson 2001, Austin 2002, not seasonality and responses to climatic variability. This lack of understanding limits our knowledge of how climate change may alter Hawaiian forest phenology and associated ecosystem functions. Hawaiian forests have lower tree diversity, but are structurally similar to other continental tropical forests (Ostertag et al. 2014) with tropical phenological strategies represented. Their extreme isolation in the Pacific makes them a unique signal for climate change impacts on ecological communities, unlike forests such as the Amazon, which experience large local feedbacks between the canopy and atmosphere (Kooperman et al. 2018), complicating the climate signal. In this study we examine monthly seed production of the four dominant tree species and community-wide leaf litterfall from a montane wet forest on the Island of Hawaiʻi, and compare their sensitivities (i.e., direction and magnitude of response) to temperature, rainfall, and photosynthetically active radiation (PAR). We hypothesize that species' phenologies are more sensitive to temperature and PAR than to rainfall because tropical species should be especially sensitive to temperature and rainfall at this site is not limiting. Hawaiʻi. The forest is comprised of 18 woody flowering tree species, 3 tree fern species, and the vegetation is highly representative of montane wet forests in Hawaiʻi (see Ostertag et al., 2014 for detailed site information). Mean annual precipitation is 3440 mm and mean annual temperature is 16° C (Giambelluca et al. 2013).

METHODS
To monitor reproductive and leaf phenology, sixty-four litter traps within a 4-hectare plot were placed in a regular grid, 10-meters apart, and monitored as part of the Hawaiʻi Permanent Plot Network (HIPPNET). Reproductive (fruit and seeds) and leaf litterfall were censused each month following standard protocol (Wright et al. 2005). Fruits and seeds were identified to species and converted to number of seeds for each fruit, then summed across all traps for each month. Leaf litterfall, for all species combined, was summed across all traps, divided by the number of traps censused (because some months, traps were knocked over), and weighed each month. Sixty-five months of fruit/seed data were available between the months of October prohibiting year-to-year models). Thirty-two months of community leaf litterfall data were available. Because some collections could only be attributed to a month and year, leaf litterfall rates (e.g., g m -2 day -1 ) could not be determined accurately.
Climate stations at both sites are maintained by HIPPNET and record daily temperature (° C; HMP45C-L,Vaisala), rainfall (mm; tipping bucket rain gauge; TB3 CS700, Hydrological Services), and photosynthetically active radiation (PAR; mol s -1 m -2 ; Quantum sensor). All climate data were aggregated to monthly averages except for rainfall, which was summed each month.
Of the 21 plant species present at Laupāhoehoe, 12 were present at least once in the litter baskets and 4 dominant species were ultimately examined: Metrosideros polymorpha ('ōhi'a lehua; bird or insect pollinated, wind dispersed, Myrtaceae), Acacia koa (koa; insect pollinated, wind dispersed, Fabaceae), Coprosma rhynchocarpa (pilo; wind pollinated, bird dispersed, Rubiaceae), and Cheirodendron trigynum (ʻōlapa; bird or insect pollinated, bird dispersed, Araliaceae). These four dominant species comprise 57.5% of the relative abundance and 52.7% of the total basal area of the Laupāhoehoe FDP and were the trees that reached canopy and subcanopy levels (Ostertag et al. 2014). The other 8 species occurred too infrequently in the litter traps for statistical analysis. STATISICAL ANALYSES-To examine relationships between seed production (counts) and leaf litterfall (grams) with climatic predictors, we used generalized additive models (GAMs) to estimate flexible (potentially non-linear) and independent smoothing functions to predictors and response variables (Venables & Ripley 1999, Zuur et al. 2009). We used a Poisson log-link likelihood for seed production and a Gaussian log-link likelihood for leaf litterfall, and estimated relationships between seed production or leaf litterfall and climatic predictors in GAMs using a cubic regression spline. Using GAMs we estimated response curves (i.e., 'smooth terms') to climatic predictors, temperature, rainfall, and PAR as well as month (to account for monthly seasonality separate from climatic seasonality) and year (to account for any yearly trend). We also included daylength as a predictor, but large (> 0.80) concurvity values (i.e., a generalization of collinearity for smooth terms in GAMs) limited accurate interpretation (see Supporting Information; Fig. S1 and S2, Table S1). Because time-series data are often non-independent, we accounted for serial autocorrelation in the error term using an AR(1), i.e., an autoregressive term of 1 month. Model residuals were examined and showed no significant autocorrelation. Seed production was modeled separately for each species using concurrent monthly climate data because fruit maturation, dispersal, and germination should be timed to climate (van Schaik et al. 1993). Leaf litterfall, however, is not as clearly timed to climate as leaf production might be because leaves are longer lived on the canopy. Thus leaf litterfall was examined using collection dates lagged one month prior to collection (when leaves had not yet fallen and are still on the canopy) because Pearson's correlation coefficients were stronger with one-month lag compared to a two-or three-month lag.
In addition to examining each species' seed production and community-wide leaf litterfall responses to light, temperature, and rainfall (i.e., the full model), we compared all possible reduced models using the Akaike Information Criterion (AIC) (Burnham & Anderson 2010) to assess which combination of climatic factors resulted in the best-fit model.

RESULTS
SEED PRODUCTION-The most strongly seasonal species was C. rhynchocarpa, with protracted annual seed production from June-February and peak seed production in December ( Figure 1).
The other three species generally produced seeds year-round however some still exhibited seasonality in seed production. C. trigynum had a clear peak season of seed production from July-November, whereas seed production by A. koa and M. polymorpha were highly variable throughout the year. A. koa had the most variable monthly seed production with a coefficient of variation (CV) ranging each year from 0.85 -3.00 (examining only 5 years where every month was represented). The CV of C. trigynum ranged from 0.65 -1.35, M. polymorpha ranged from 0.77 -1.18, and C. rhynchocarpa ranged from 0.22 -1.05. Across years, A. koa seed production had the largest interannual variation indicated by the highest CV (1.24), followed by C.
All GAM smooth terms (i.e., response curves to climatic factors) were significant (p < 0.001) for seed production by all species (Figure 2 a-d). A. koa seed production was the most responsive species to monthly climatic variation, showing increasing seed production with warming temperatures to ~15 ° C, and increasing seed production with more rainfall and greater PAR up to ~350 mol s -1 m -2 ( Figure 2a). C. trigynum was relatively insensitive to changes in climate compared to the other species with small increases in seed production up to ~16 ° C, increases up to ~ 600 mm rainfall, and a flat response to PAR (Figure 2b). Seed production by C. rhynchocarpa showed small increases up to ~16 ° C, decreases in response to rainfall more than ~500 mm, and rapid decreases in response to PAR above 300 mol s -1 m -2 ( Figure 2c). Seed production from M. polymorpha decreased at temperatures above ~15 ° C, was generally not responsive to seasonal changes in rainfall, and increased slightly with increasing PAR ( Figure   2d). Although all climatic smooth terms were significant, the amount of variation in seed production explained by climatic variables was moderate to low depending on species (A. koa: Model comparisons based on AIC showed that the best-fit models for all species included all three climatic factors with close to 100% AIC weight (Table S2a-d). LEAF LITTERFALL-Leaf litterfall did not show strong seasonality (Figure 3). Only rainfall was significant (p < 0.001) whereas temperature and PAR were not (p > 0.05). Leaf litterfall increased strongly with more rainfall (Figure 4). However the model overall did not explain much variation in leaf litterfall (R 2 = 0.09). The proportion of variance explained improved when rainfall was the only climatic predictor in the model (R 2 = 0.14). There were four equivalent best-fit models (ΔAIC <2). The best-fit models each accounting for more than 30% of the AIC weight included either rainfall or PAR separately (Supplementary Table 1e). Two other models considered equivalent best-fit models include rainfall and PAR together, or temperature alone.

DISCUSSION
Because the low latitudes are thought to be climatically stable and many species are active yearround, variation in tropical species' phenologies have been understudied (Cook et al. 2012, Chambers et al. 2013. However, many tropical plant species are known exhibit distinct phenological patterns (Newstrom et al. 1994, Sakai 2001. These patterns may be linked to seasonal changes in the abiotic environment (van Schaik et al. 1993). But the degree of sensitivity to climate among tropical species to climatic variations is unknown in many regions, impeding our ability to understand their vulnerability to climate change. We examined seasonal relationships between climate and plant phenology, while accounting for variation due simply to month of the year (as well as changes in daylength; see Supplementary Information). These relationships with climate provide insight into favorable and unfavorable conditions for seed and leaf litterfall production. SEED PRODUCTION-Response curves showed that seed production was sensitive to changes in climate but no species responded similarly to all three climatic factors. With the exception of one species (C. trigynum) seed production was more sensitive to temperature and PAR than to rainfall, supporting our initial hypothesis. Seed production of A. koa and C. rhynocparpa appeared more sensitive to climatic variation and had higher interannual variability than M. polymorpha and C. trigynum. Greater interannual variability suggests that photoperiod is not a strong control of reproductive phenology because daylength does not vary each year, yet there was some support for the effect of daylength (Supplementary Information). M. polymorpha and C. trigynum had weaker responses to climate than A. koa and C. rhynocparpa. M. polymorpha, however, did show a discernable decline with temperatures above 15-16 ° C as did A. koa, both foundational species in Hawaiian forests. For all four species at Laupāhoehoe, responses to rainfall and PAR were in the same direction. This could be explained by a large diffuse component of PAR, which can scatter more light through the canopy, as opposed to casting strong shadows, increasing light availability (Roderick et al. 2001, Butt et al. 2009, 2010, Pau et al. 2013. The moderate to low R 2 , particularly for M. polymorpha, indicates that factors other than climate play a role in monthly seed production and/or seed production is not very predictable for this species. An unexamined factor that may explain the timing of seed production and reproductive phenology is conservatism within lineages (Wright & Calderón 1995, Davies et al. 2013 Although seed production should follow flower production, there may be different climatic cues or resource requirements for seed rather than flower production (Augspurger 1983, Wright & Calderón 2006, Slot & Winter 2016. A. koa is a species that is known to flower often but seed development does not always follow. While the presence of flowers was not recorded in censuses, PhenoCam (i.e., a tower-based digital camera; Richardson et al., 2009Richardson et al., , 2018 observations overlapped with censuses from 2017 and 2018, and seed production followed flowering both years (Fig. S3). Flowering of A. koa in PhenoCam images generally occurred December -March (although it began in February in 2018) while seed production peaked, on average, between February -April (Fig. 1).
Even when viable seeds are produced, conditions for establishment may limit recruitment (Inman-Narahari et al. 2013). For example, M. polymorpha and C. trigynum do not appear to be seed or dispersal limited, but instead limited by favorable sites for establishment (Drake 1992, Inman-Narahari et al. 2013. In other cases, seed limitation and dispersal failure may contribute to the decline of native species on Hawaii (Chimera & Drake 2010, Inman-Narahari et al. 2013. LEAF LITTERFALL-Community-level leaf litterfall was not well predicted by seasonal changes in climate. Unlike seed production, temperature appeared to have no effect on leaf litterfall. Leaf litterfall was responsive to increasing rainfall. A synthesis of leaf litterfall from tropical South America showed that across sites, litterfall seasonality was associated with rainfall seasonality, wherein sites that had more seasonal litterfall also had more seasonal rainfall and vice-versa (Chave et al. 2010) However, there was no relationship between leaf litterfall accumulation and total annual rainfall (Chave et al. 2010). Satellite and eddy-covariance measurements have shown positive greening or productivity responses to increases in light availability in tropical wet forests (Huete et al. 2006, Saleska et al. 2007. In contrast, water-limited sites have shown reduced photosynthesis during the dry season, thus different tropical forest types can exhibit asynchronous responses to climatic variability with water more limiting in dry sites and light more limiting in wet forests (Pau et al. 2010, Zhang, Xiao, et al. 2016. The dry season greening of tropical forests, even in wet sites, has been intensely debated (Samanta et al. 2010, Morton et al. 2014 (Parmesan & Yohe 2003, Root et al. 2003. Phenology is a key axis of functional traits and vital to understanding variation in life history strategies as well as niche differentiation. In Hawaii, a challenge for native plant restoration is the management of invasive species. Identifying differences in the timing of seed production and resource acquisition by native and invasive species, which may change with climate change, should provide insight into mechanisms for establishment and assist in the timing of management efforts (Drake et al. 1998, Funk 2013. Additionally, it has been pointed out that species-poor island flora may be accompanied by low functional redundancy, i.e., species perform unique roles in their communities (McConkey & Drake, 2015). Thus communities with strong species' dependencies are more vulnerable to shifts in the timing of seed or leaf production. Species or populations responding differently in magnitude or direction to climate change may result in phenological mismatches and novel ecological communities (Visser & Both 2005, Thackeray et al. 2016).
Shifts in phenology may therefore be viewed as having negative impacts on a community (i.e., phenological mismatch) or viewed positively as an ability to adapt to climate change (Visser & Both 2005). Some hypothesize that species that can track climate change by adjusting their phenologies should persist under a changing climate (Cleland et al. 2012). Indeed, research has shown that phenological sensitivity is associated with increased population abundance, however, most of this evidence is from temperate regions where phenological sensitivity is defined as the number of days a species' shifts a phenological event per degree temperature change (Cleland et al. 2012). These species can maintain optimal performance (e.g., flowering or fruiting at a different time), whereas species that do not track climate change may face unfavorable conditions (e.g., climate that is too warm for optimal seed production).
Here we do not examine phenological tracking given the length of our record (~ 6 years).
We show that three of the four dominant species at our site are sensitive to climatic variability, and the direction and magnitude of responses represent favorable or unfavorable periods of growth. End of the century climate projections show an average temperature increase of 2 -4 °C over the Hawaiian Islands with more warming at high elevations. Increased rainfall and more cloud cover is projected for the windward side of the islands whereas the leeward side is projected to get drier and with fewer clouds (Lauer et al. 2013. As the climate changes in the future, species' phenologies may shift earlier or later if particular months become unfavorable. Alternatively if species do not shift to more favorable months, fecundity and growth may decline. Ultimately the sensitivity, exposure, and adaptive capacity of species should be considered in determining species' vulnerabilities to climate change. At least one species (C. trigynum), and likely the eight others that reproduced too irregularly for statistical analysis, appear not highly sensitive to climate variability at this site.
Understanding climate change impacts on species and communities has focused on predicting species' range shifts using bioclimatic envelopes (i.e., correlative models of species presence and mean climate; Guisan & Thuiller 2005, Elith & Leathwick 2009). Shifts in some species' range and distribution but not others may result in the formation of novel communities (Williams et al. 2007). However species' range limits depend on their ability to complete their life cycle and each species' distinct reproductive niche is critical to consider (Chuine 2010, Bykova et al. 2012. Divergent responses to climate change could alter community composition if reproduction or growth declines for some species but not others. How the reproduction and growth of different species will respond to climate change has potential consequences for future shifts in species distributions and the persistence of biodiversity.

ACKNOWLEDGMENTS
We thank the National Geographic Committee on Exploration (Grant number #WW-235R-17).
Support for HIPPNET has included NSF EPSCoR (Grants No. 0554657 and 0903833) and the USDA Forest Service, the University of Hawai'i, and the University of California, Los Angeles (NSF Grant No. 0546784). We thank Kaikea Blakemore, Christa Nicholas, Leila Ayad, and Evan Blanchard for their help with data collection and processing.

DATA AVAILABILITY STATEMENT
The data used in this study are archived at the Knowledge Network for Biocomplexity (doi:10.5063/F1QR4VF7).

FIGURE LEGEND
FIGURE 1. Average monthly seed production of four dominant species in a Hawaiian montane wet forest. Each species is scaled on a different axis length because of large differences in species' seed production. Maximum seed totals are labeled for each species so relative differences each month can be compared.    forest. Each species is scaled on a different axis length because of large differences in species' seed production. Maximum seed totals are labeled for each species so relative differences each month can be compared.  Figure 4. Monthly leaf litterfall responses to climatic factors (R 2 = 0.09). GAM smooth terms for rainfall was significant (p < 0.001) whereas temperature and PAR were not (p > 0.05; grey shading shows two standard error bounds). Negative log values indicate values between 0-1.