Can seed banking assist in conserving the highly endemic New Zealand indigenous flora?

Data sources

The New Zealand seed plant flora was defined according to the New Zealand indigenous vascular plant checklist (de Lange and Rolfe 2010). For these species, we then obtained seed storage behaviour and seed mass from the Royal Botanic Gardens, Kew’s Seed Information Database (SID; Royal Botanic Gardens Kew 2017), accessible via the TRY trait database (Kattge et al. 2020). From SID we were able to obtain records of seed storage behaviour for 217 species. We then supplemented these data with additional records obtained from published material via extensive literature searching (completed in August 2021), adding records of seed storage behaviour for a further 203 species (these latter records are available as a Supplementary file). We searched individual species names alongside the terms ‘seed’, ‘storage’, ‘recalcitran*’, ‘intermediate’, and ‘orthodox’ using Google Scholar, Scopus, and Web of Science, and checked all search results for potential storage behaviour information. In addition, national library databases, including the National Forestry Library and the New Zealand Alpine Garden Society, were checked for grey literature relating to seed storage.

Growth form, plant height, and dispersal mode data were obtained from the dataset for the New Zealand angiosperm flora compiled by McGlone and Richardson (2022). In their dataset, McGlone and Richardson (2022) defined woody species as those with significant lignification of above ground parts, including trees (self-supporting and at least 5 m in height), shrubs (self-supporting and less than 5 m in height), and lianas (species reliant on other plants for support; includes true lianas, climbers, and scramblers). In contrast, herbaceous species had no or insignificant lignification of above ground parts and included herbs (self-supporting; definition includes sub-shrubs, which have herbaceous canopies but woody root stocks and lower stems) and vines (non-lignified species reliant on other plants for support). Plant height was the maximum height at maturity and taken as the upper bound (but not extreme maximum) of the range given for a species in flora compilations (http://www.nzflora.info/). McGlone and Richardson (2022) inferred dispersal mode from fruit characteristics described in flora compilations and supplemented by Thorsen et al. (2009), as species with fleshy fruits are ingested and dispersed by animals (biotic means; endozoochory), while those without fleshy fruits are dispersed by abiotic means such as wind, water, or gravity. We then supplemented the McGlone and Richardson (2022) angiosperm dataset with growth form, plant height, and dispersal mode data for the New Zealand gymnosperm flora (20 species) using their definitions and from the same data source, the Flora of New Zealand series (specifically Allan (1961) for the gymnosperm species). The most recent conservation status of the New Zealand seed plant flora was obtained from de Lange et al. (2017).

These different datasets were all based upon slightly different taxonomies, varying in spelling of species names, definitions of species, and placements of species in genera and families. We combined these datasets initially via fuzzy matching using the agrep function in R ver. 4.0.5 (R Core Team 2021). Species that were unable to be matched between datasets were then manually checked, with taxonomy edited following the New Zealand Plant Conservation Network (https://www.nzpcn.org.nz/) taxonomic treatments. Any remaining species that were unable to be satisfactorily matched were removed from further analysis.

Habitat environmental data

Given that the relative abundances of species with different seed storage behaviours varies considerably by habitat (Wyse and Dickie 2017), we obtained environmental variables describing the habitats in which each woody species typically occurs at the centre of its ecological niche. Species’ distribution data were acquired from the Global Biodiversity Information Facility (GBIF; GBIF.org 2022) limiting the data obtained to occurrences within New Zealand recorded as having reliable geographic locations, and for which a species could be matched taxonomically with at least 95% confidence. We extracted the annual precipitation, mean annual temperature, diurnal temperature range, and elevation at ~1 km resolution (Fick and Hijmans 2017). To reduce the potential effects of bias within the GBIF data, we used environmental filtering (Varela et al. 2014); only including occurrences in environmental space that were at least 0.1°C apart for temperature variables, were 25 mm apart for the precipitation variable, and 50 m apart for the elevation variable. For those species with at least 10 post-filtered occurrences, we calculated the centre of the ecological niche via Mahalanobis distance using a minimum covariance determinant approach (with k = 0.95); an approach that has been demonstrated to be robust to small samples sizes, errors, and bias (Etherington 2021). Analysis was done using R software (R Core Team 2021) with the packages rgbif (Chamberlain and Boettiger 2017), MASS (Venables and Ripley 2002), terra (Hijmans 2002), and sf (Pebesma 2018).

Data analysis

In the seed storage behaviour datasets, the following character states were used: ‘orthodox’; ‘orthodox?’; ‘intermediate’; ‘intermediate?’; ’sub-orthodox’; ‘recalcitrant’; and ‘recalcitrant?’. For our analyses, we treated all questionable classifications as being of that character state (i.e. ‘orthodox?’ became ‘orthodox’), and considered ‘sub-orthodox’ to be synonymous with ‘intermediate’. Thus, our seed storage behaviour dataset consisted of four character states: ‘orthodox’, ‘intermediate’, ‘recalcitrant’, and ‘unknown’. Where multiple conflicting records existed for a single species, we favoured orthodox over intermediate or recalcitrant (and intermediate over recalcitrant) following the rationale that an orthodox seed treated incorrectly could appear recalcitrant; for example, if harvested too soon, whilst a truly recalcitrant seed could not possibly exhibit orthodox behaviour.

Using the seed storage behaviour data, we first assessed the proportions of species where seed storage behaviour was known (i.e. ‘orthodox’, ‘intermediate’, or ‘recalcitrant’) by growth form and, for woody species, by plant family. Due to the ecological importance of canopy forming forest trees, which form and define forest habitats, we also compiled a list of dominant forest canopy trees to explicitly examine the seed storage behaviours of these species (Appendix 1). Dominant forest canopy trees were identified using data derived from New Zealand’s Land Use and Carbon Analysis System (LUCAS), a national network of permanent plots covering over 7.8 million ha that systematically samples natural forests established pre-1990 for the purpose of monitoring carbon sequestration rates (Coomes et al. 2002). Plots within the LUCAS network were each established on an 8-km grid across the main islands of New Zealand, and were 0.04 ha (20 m × 20 m) in area (Bellingham et al. 2020). Of the 1040 plots established initially, 874 randomly selected plots were remeasured from 2009 to 2014 and analysed by McGlone et al. (2024). Data from these re-measured plots were used here to identify dominant canopy tree species by calculating the mean basal areas of all tree species (as defined previously) across all plots in which they were present. Dominant canopy species were then defined as those tree species with mean basal areas of at least 3 m² ha⁻¹. This resulted in a list of 43 tree species, approximately 30% of the 140 indigenous tree species in the LUCAS plot network.

As a result of the low proportion of seed behaviour information available for herbaceous species, and the high level of taxonomic discrepancies within this group of plants, after initial data summaries further analyses were only conducted on woody plants. Globally, species that produce non-orthodox seeds that are unable to be conserved in a seed bank following conventional methodologies are typically woody plants (Tweddle et al. 2003; Wyse and Dickie 2017).

For the woody seed plant flora, we then examined the seed storage behaviour character states with respect to conservation status. Globally, the incidence of non-orthodox seeds (‘intermediate’ and ‘recalcitrant’) increases with increasing threat, as non-orthodox species are most prevalent in habitat types (i.e. tropical moist forests) facing highest levels of destruction as a result of human activities (Wyse et al. 2018). Thus, in a global context, the species facing the greatest in situ conservation threats will pose the greatest challenges for ex situ conservation.

Finally, we examined the incidence of species with different seed storage behaviours, including that of ‘unknown’ seed storage behaviour, with respect to the plant traits and environmental variables compiled here. For the woody species of known seed storage behaviour, we undertook a boosted regression trees (BRT) analysis (Elith et al. 2008) to examine the ability of the plant trait and environmental predictor variables to predict the likelihood that a species produces non-orthodox seeds, and assess which variables are the strongest predictors of this trait as well as potential interactions between variables. Models were constructed and evaluated within the gbm.step function from the dismo package (Hijmans et al. 2017) using 10-fold cross-validation. Following testing, modelling was undertaken using a tree complexity (number of interactions) of 2, and learning rate (the contribution of each individual tree to the developing model) of 0.001, and a bag fraction (the proportion of observations used in each step) of 0.6. The number of trees used was 2950. Functionality within the gbm.step function was used to calculate the relative contributions (relative importance) of each predictor variable, the area under the receiver operating curve (AUC), and the correlation of the data with predicted values. These latter two metrics provided an assessment of model performance.

Predicting seed storage behaviour for entire New Zealand woody flora

For the New Zealand woody seed plant flora (620 species), we were able to obtain seed storage information for approximately 30% of species (Table 1). To understand the potential extent of the woody seed plant flora that is likely to be able to be conserved in an ex situ seed bank using conventional methodologies, we used existing global models to predict the likely seed storage behaviour for the unknown species (Wyse and Dickie 2018). These models were developed to provide decision-making support for ex situ plant conservation and use data on species seed traits, habitat environmental variables, and the seed storage behaviour of known related species to predict the likelihood of a species producing non-orthodox seeds.

Using the combined datasets of known storage behaviour and predicted storage behaviour for the remaining species, we examined how the relative proportions of orthodox versus non-orthodox species varied with geographical region for all 620 woody species. Using our GBIF data and a similar workflow to Brandt et al. (2021), we determined the number of orthodox species, recalcitrant and intermediate species, and species with insufficient information to predict storage behaviour within each of the 16 mainland political regions across New Zealand (Stats NZ 2019). To guard against outliers or errors we only included those records on the main islands of New Zealand (Land Information New Zealand 2011) and required at least five occurrences for a species to be counted within a region.

Seed ecology and conservation of the New Zealand flora

Here, we have compiled all available seed storage data for the highly endemic New Zealand flora, to our knowledge an unprecedented exercise for a national flora (although see Crawford et al. (2007) for a compilation of data on the survival of Australian species within a seed bank). The proportion of recalcitrant species in the New Zealand flora (11% of all known species) is considerably higher than the estimated value of 3% of all plants for the global biome ‘temperate broadleaf and mixed forests’ (Wyse and Dickie 2017), in which much of the New Zealand landscape occurs (Olson et al. 2001). The temperate broadleaf and mixed forests biome is predominantly distributed in the Northern Hemisphere and data from it are biased towards deciduous forests, including those of Great Britain and Ireland (Olson et al. 2001), where recalcitrant species may make up just 1% of the flora (Wyse and Dickie 2018). In contrast, the New Zealand woody flora occur in evergreen rain forests (Grubb et al. 2013; McGlone et al. 2016).

The higher incidence of orthodox seed behaviour among New Zealand indigenous herbaceous species compared to woody species was in line with global trends (Tweddle et al. 2003; Wyse and Dickie 2017). However, the incidence of recalcitrant and intermediate herbaceous species found here in the New Zealand flora was much higher than a previous global estimate based on the available data at the time. In their study, Tweddle et al. (2003) found non-orthodox seeds to be exceedingly rare amongst herbaceous species, as 99.8% of the 517 herbaceous species in their dataset were orthodox, and just 0.2% of species (equates to a single species) were recalcitrant. Tweddle et al. (2003) discuss that their finding was likely due to recalcitrant seeds being incompatible with the regeneration of annual and biennial species. Data are not available on the proportion of their herbaceous species that were annuals, however in the New Zealand flora the majority of herbs are perennial, with just 5.7% comprising annual species (data from McGlone and Richardson 2022). For comparison 37% of herbaceous species in the indigenous flora of California (Carl Freeman et al. 1979), 25% of the herbaceous species from North and South Carolina (Conn et al. 1980), and 27% of the herbaceous species of Britain and Ireland (Hill et al. 2004), are annuals. This low proportion of annual herbs may explain the higher incidence of recalcitrant herbs observed here.

The New Zealand flora largely originated from post-25 Ma immigration either down the northern Pacific island chains, or via dispersal from Australia, with few of the vicariant Gondwanan species surviving the Oligocene marine transgression (McGlone and Richardson 2022). The flora is comparatively young, with 90% of extant lineages younger than 15 million years and 50% of lineages evolving during the last 5 Ma (Heenan and McGlone 2019). Structurally, New Zealand forests are complex, resembling tropical forests and with high numbers of biotically-dispersed species (Cockayne 1921; Dawson and Sneddon 1969; McGlone et al. 2016). Of the lowland woody genera, 60% also have taxa in tropical or subtropical regions (McGlone and Richardson 2022). The high proportion of species with tropical affinities, particularly in warm temperate ‘podocarp-broadleaf’ forests (forests with a broadleaved-angiosperm canopy and emergent Podocarpaceae trees) in the north of the country, may explain the comparatively high proportion of recalcitrant species in the New Zealand flora relative to other temperate broadleaf and mixed forests globally. We found a higher incidence of recalcitrant species in the warm temperate northern regions, consistent with the finding that temperature was the strongest environmental predictor of seed storage behaviour as non-orthodox species were more likely to occur in habitats with lower diurnal temperature ranges and higher mean annual temperatures. The dominant tree species of New Zealand have different southern range limits depending on their individual environmental tolerances (Wardle 1991), which may explain the gradual rather than abrupt differences in the incidence of recalcitrant species among regions moving south. The relationship between the incidence of recalcitrant species and temperature follows patterns observed globally (Tweddle et al. 2003; Wyse and Dickie 2018), however, unlike global trends, rainfall was not an important predictor of seed storage behaviour despite the strong rainfall gradients within New Zealand. Thus, the primary environmental correlate for seed storage behaviour in the New Zealand woody flora appears to be temperature, via latitude and elevation gradients.

Plant traits were more important correlates of seed storage behaviour than environmental variables. While previous work indicates that tree species globally have a higher incidence of recalcitrant seeds relative to all plants, potentially driven by the high diversity of tree species in tropical moist forest and mangrove habitats (Wyse et al. 2018), this is the first study to explicitly detect a trend of increasing likelihood for recalcitrant seeds with plant height. The seed trait relationships observed here were consistent with global trends, with the finding that recalcitrant species had seeds that were larger, and more likely to be animal-dispersed (Daws et al. 2005, 2006; Wyse and Dickie 2018). We do not have the data to examine whether the seed coat ratio, previously reported to predict seed storage behaviour alongside seed mass (Daws et al. 2006), relates to seed storage behaviour in the New Zealand flora. However, contrary to the predictive model derived from seed traits of the flora of Panamá indicating that recalcitrant seeds are large and with thin seed coats, which also successfully predicted seed storage behaviours of British species, numerous large-seeded New Zealand trees have unusually thick-walled (2–3 mm) endocarps (Carpenter et al. 2018b) including the recalcitrant Elaeocarpus dentatus (Elaeocarpaceae). Daws et al. (2005) suggest that, for large-seeded species, the recalcitrant seed trait is advantageous as the swift germination correlated with the trait may confer a selective advantage by reducing the amount of time during which the seed is vulnerable to predation. Yet, the recalcitrant E. dentatus can take up to 7 years to germinate (Carpenter et al. 2018b). Before faunal changes associated with human colonisation of New Zealand, vertebrate seed predators were avian rather than mammalian (Carpenter et al. 2020). Thus, seed trait relationships could potentially differ to those from other flora where mammalian seed predators predominate. Further research should examine these relationships within the New Zealand flora. However, it is certainly evident that some dominant, non-orthodox trees such as Agathis australis (Araucariaceae) and Podocarpaceae species germinate rapidly after being shed, forming a seedling bank (Ogden 1985).

Our finding of a prevalence of recalcitrant species among bird-dispersed, lowland, dominant canopy trees raises concerns for conservation. Lowland habitats have seen a disproportionate level of habitat loss due to clearance for agriculture (Ewers et al. 2006), and with the exception of the West Coast region (South Island) remaining lowland forest largely occurs on private land, where there is less legal protection (Pannell et al. 2021). In situ conservation options for species in these forests are therefore more limited than they are in higher elevation forests. Beech (Nothofagaceae) forests are most common at high elevation and are likely to contain a high proportion of species suitable for conservation in a conventional seed bank, whilst the more diverse conifer-angiosperm forests (including podocarp-broadleaf forest and kauri forest) present more challenges. However, in contrast to global trends (Wyse et al. 2018) we did not find a relationship between threat status and seed storage behaviour, suggesting that the species most threatened in situ are no less likely to be conserved ex situ within a conventional seed bank than non-threatened species; a promising finding with respect to their conservation. However, it is certainly evident that further research is required for many species.

Future directions

Our compilation of seed storage information for the New Zealand flora has highlighted many knowledge gaps that future research must aim to fill. Herbs and shrubs are the plant growth forms with the most limited seed storage data, and should therefore be a focus of further research. Given the low proportion of annuals in the New Zealand herbaceous flora relative to other regions, and the higher incidence of recalcitrant species in the New Zealand herbaceous flora than amongst herbs globally as indicated by the currently available evidence, deepening our understanding of seed storage behaviour of New Zealand herbaceous species should be a priority. Models derived from global data may fail to accurately predict the seed storage traits of these species. Further, our results illustrate the woody plant families for which very limited, or no, information is available in the New Zealand flora. The seed storage behaviour of a given species has been shown to be strongly correlated with those of its close relatives (Wyse and Dickie 2018). Therefore, to better understand seed storage behaviour of the New Zealand flora it will be more immediately beneficial to survey species from families and genera for which there is currently little information, rather than gaining deeper insights into genera were there are already many species with known storage information. The exceptions are genera such as Pittosporum and Coprosma, or families such as Podocarpaceae, in which there is variability in seed storage behaviour among species. For such taxa, it would be useful to gain further insights to understand plant or habitat traits that may predict seed storage traits. Loranthaceae and Santalaceae are families with no seed storage data for New Zealand species, and globally very little additional information. The Loranthaceae, in particular, is a family of concern in New Zealand as many of these mistletoe species are classed as ‘at risk – declining’ due to threats including mammalian herbivores (Australian brushtail possum; Sweetapple et al. 2002), and reductions in avian pollinators (Kelly et al. 2007). We have been unable to find any data on seed storage behaviour for Loranthaceae globally, whilst the available information for Santalaceae is only marginally better. Within Santalaceae, many species are intermediate (Thomson 2006), while Osyris lanceolata is recalcitrant (Mwang’ingo et al. 2004).

Another substantial knowledge gap in New Zealand seed conservation is that of seed longevity in an ex situ seed bank. This refers to the length of time an orthodox seed can remain deep-frozen in storage before viability is significantly lost. Information on seed longevity in an ex situ seed bank is critical for effective collection management, as it will set the seed viability re-test intervals for the collections, determine re-collection strategies (Liu et al. 2020), and inform any adaptation of current conventional protocols that may be required to suit short-lived seeds. Accelerated ageing tests have been found to correlate with real time data of seed survival in storage, and can augment the real time data to aid understanding of potential seed lifespan in storage (Davies et al. 2020). Data suggest that there is considerable variability among, and within, orthodox species in their longevity in an ex situ seed bank, with indications that species from hot dry environments may store longer than those from cool, wet habitats (Probert et al. 2009; Davies et al. 2020; Liu et al. 2020). For example, among the Hawaiian indigenous flora, 63% of 295 species examined by Chau et al. (2019) retained at least 70% viability after 10 years in storage, while 23% declined to less than 70% of the initial viability within 5 years of storage.

Finally, urgent work is also required on developing protocols for alternative ex situ conservation methods for threatened recalcitrant species in the New Zealand flora. To date, extensive work has been undertaken for two rain forest trees, Didymocheton spectabilis and the critically threatened Syzygium maire, to determine appropriate cryopreservation protocols (Park 2013; van der Walt et al. 2021a, 2021b, 2022). For both species, research has examined optimal processing of excised embryos and tissue culture methodologies; however, further work is required to establish viable approaches. We suggest that, in addition to continuing work on these species, at risk recalcitrant tree species such as Meryta sinclairii and Pseudopanax chathamicus should be future priorities for research. Developing seed storage protocols for recalcitrant, dominant canopy tree species such as Podocarpus totara, Dacrycarpus dacrydioides, Beilschmiedia tawa, and Elaeocarpus dentatus would also aid in the supply of seedlings of these species for restoration programmes, particularly species such as B. tawa, P. totara, and D. dacrydioides that undergo mast seeding (Webb and Kelly 1993; Yukich Clendon et al. 2023), and thus for which seed availability fluctuates annually.