Data from: Marcescence and prostrate growth in tree ferns are adaptations to cold tolerance
Data files
Feb 04, 2025 version files 28.50 MB
-
README.md
11.64 KB
-
treefern_skirts.zip
28.49 MB
Abstract
Cold tolerance strategies in plants vary from structural to biochemical. Many temperate tree fern taxa are marcescent—retaining whorls of dead fronds encircling the upper trunk—or develop short or prostrate trunks possibly to insulate against cold temperatures that might detrimentally affect their trunks and growing crowns. We asked the following questions: (1) do global growth patterns traits of temperate taxa relate to environmental factors associated with latitude (a proxy for seasonality and frost tolerance) and elevation (a proxy for temperature), (2) do growth patterns of tree ferns in New Zealand vary along a temperature related gradient, and (3) do marcescent tree fern skirts insulate the growing crown from sub-zero temperatures? We review the global and regional distributions of these structural and morphological traits within Cyatheales. Further, we assess the patterns of tree fern marcescence, and other traits potentially associated with cold tolerance (no trunk, prostrate, short-trunked), by comparing the ecological niches of nine taxa of the Cyatheales along environmental gradients across New Zealand. Finally, we conducted a field experiment to assess the thermal insulation properties of tree fern marcescent skirts. We identified significant trends among growth forms, marcescence, and envionmental gradients consistent with our hypothesis that these are adaptations to tolerate cold. In addition, our field experiments provide quantitative evidence that marcescent skirts have a strong insulating effect on tree fern trunks. The Cyatheales have evolved several strategies to protect the pith cores of their trunks from extreme cold temperatures in temperate forests allowing them to capture niche space in environments beyond the tropics.
README: Marcescence and prostrate growth in tree ferns are adaptations to cold tolerance
https://doi.org/10.5061/dryad.tx95x6b7f
Description of the data and file structure
The treefern_skirts.zip file contains an RStudio project with all the data and code required to reproduce our analyses. Open the treefern_skirts.rproj file and ensure that you have the required packages/versions.
Files and variables
File: treefern_skirts.zip
Description:
The ./data/
folder contains all data for the project.
The ./data/global.csv
contains the dataset put together for the global review of treefern growth forms, alongside temp.csv
which has all of the GBIF records of treefern occurrences globally. We could not supply these data, as they are protected by CC, and so requires direct download from GBIF. Initially, we reviewed the descriptive literature on tree ferns to identify, where possible (skirts are not a frequently described trait), the global proportion and distribution of tree ferns with marcescent skirts and/or prostrate taxa. We conducted a literature search using the Google Scholar (https://scholar.google.com), Scopus (www.scopus.com) and ISI Web of Science www.isiknowledge.com) databases with the search terms [“tree fern” OR “tree-fern” OR "treefern"] and [“skirt*” OR “collar” OR “marcescen* OR “attached necromass” OR “prostrate”]. To identify taxa from America we also used Google Scholar to search in Spanish and Portuguese using ["helecho* aborescente”, OR “helecho* arbóreos” AND “falda”] and [“samabaia” AND “saia”]. Finally we checked for the presence of skirts as a recorded trait in the Botanical Information and Ecology Network (BIEN; https://bien.nceas.ucsb.edu/bien/biendata/) database. The geographic distribution and elevation data of all extant species from the order Cyatheales were downloaded from the global biodiversity information facility (GBIF.org, 2019) and taxa cross-checked with those identified from the review as having a marcescent skirt, and a ‘skirt’ field (Boolean) was added to the database (Table 1 of manuscript). As taxonomic and descriptive information on tropical Cyatheales is sometimes poor, for species whose median values of GBIF records distributions occur in the temperate zones, their growth habit—trunk without skirt), prostrate, no trunk—and presence or absence of recorded skirt trait were collated separately for the northern and southern hemispheres. The structure of this dataset is as follows. The 'gbifID' column holds the GBIF observation numeric identifier. The 'species' column is the latin binomial of the species observation, while the 'scientificName' column has the latin binomial and taxonomic authority used for the observation. The countryCode column contains the two letter country code associated with the observation and the 'decimalLatitude' and 'decimalLongitude' columns hold the WGS84 coordinates of the observation. The 'elevation' column shows the elevation above mean sea level in metres of the observation (as extracted from the WorldClim elevation layer). The 'skirt' column is a binary (0,1) coding for whether the species exhibits a skirted growth form. Details for additional data on tree ferns are described in the methods section of our paper. For more information please contact the corresponding author.
The ./data/regionalPresences/
folder contained our three datasets of treefern occurrences in New Zealand (LOXcun.csv, NVS.csv and tfPres.csv), which was used to construct our ENMs. However, our regional occurrence data included sites from the New Zealand National Vegetation Survey and required multiple stakeholders consent for our study to use. One condition on it's use was to not share the raw occurrence data. As such, we have instead included a copy of our dataset after spatial thinning and environmental predictor extraction, but have included all our original code for inspection. The NVS data may be obtained by request from Manaaki Whenua LandCare Research (https://nvs.landcareresearch.co.nz/). The location-less data for reproducing our analyses are contained in our ./output/occdata/ folder.* The pred_ext_values_ls.rds contains the data to be used to fit our ENMs, while the treefern_niche_space.rds contains the cleaned data for creating our PCA and biplots. Both RDS files are nested lists containing the environmental values of our occurrences and absences (see the rasters description below). The observations type (presence or absence) is coded as either a "1" or "0" respectively in the 'type' column. The treefern_niche_space.rds does not contain any values for 'absences' and only includes environmental values for the four variables used in our analysis.
The ./data/rasters/NZTM2000/final/
folder contained our final set of rasters that we used to fit our ENMs, while the .../extras folder contains an elevation file which we used for plotting presences. These data are not supplied as they are protected by copyright. However, they can be freely downloaded from the Manaaki Whenua LandCare Research website (please see the Access Information section for more details). We selected four variables representing distinct abiotic environmental gradients important for tree fern survival, growth, and fecundity from NZEnvDS (Table SM-1). To these ends, we selected number of growing degree days above 5 degrees celcius (Celsius) mean annual solar radiation (measure of energy coming into the system; Mega-Joules / m^2 / day), precipitation in the driest quarter (potential limiting factor for growth; Millimetres) and topographic index (the influence of topography and impacts of soil fertility/quality; index based on neighbouring cells). Additionally, the elevation.tif contains the elevation above sea level in metres, which was plotted alongside our presences. The original raster files are not supplied here, as they are protected by copyright, but are freely available for download from the Manaaki Whenua LandCare Research website (see Access Information).
The subzero.csv
contains the information for creating figure 4, while the tf_skirt_insulation file contains the raw data initially collected, and is needed for the statistical analysis of temperature differences within and outside of tree fern skirts. The 'minute', 'hour', 'date' and 'month' hold the minute, hour, day of the month and numeric month of the corresponding temperature log respectively. The 'adj_hour' column rounds up the 'minute' and 'hour' columns to the nearest hour. The 'md', 'mdh' and 'md.ah' columns are the concatenated versions of the minute:date, minute:date:hour and minute:date:adj_hour columns respectively. 'dirurnal' is a binary column showing whether the observation was taken during the "Night" or "Day". The 'id' column holds an identifier of two paired sensors (one under the skirt and a nearby external sensor) for a given observation. The 'loc' column indicates whether the observation is from under the skirt ("tf") or external ("tf"). The 'val' column shows the temperature of a given observation in degrees Celcius. The 'genus' column holds the species latin binomial and the 'genus_loc' column has the concatenation of the 'genus' and 'loc' columns. The tf_skirt_insulation.csv
is a filtered version of our subzero.csv dataset, with all positive ambient temperatures and paired skirted temperatures filtered out - all columns are otherwise the same.
The ./scripts/
folder contains all of our code for executing our analyses.
These .R scripts are numbered and correspond to the three analyses outlined in our paper. The functions folder contains all the custom functions created for our study and have detailed descriptions of their purpose as well as verbose comments to explain decisions and outputs.
The./output/
folder contains all the raw data outputs from the MaxEnt models (see./output/replicated/42/...) for each species. The output in these folders contain the raw MaxEnt output files for each species and cross-validated replicate. There are more output files than we can reasonably describe here, but are thoroughly documented in the java program desciption that can be found on https://biodiversityinformatics.amnh.org/open_source/maxent/ website. However, the key files we used for our analysis are the "maxent.html" files contained in each species folders as these show the mean AUC scores and the standard deviation among CV replicates. Additionally, we used the 'species_#_variable.dat'
files contained in ".../plot/" folder to create our response curves - Note the "#" in the file name is a placeholder for the CV replicate and the "variable" is a placeholder for the names of each environmental variable. They contain the raw data of the marginal response curve plots from the individual replicated maxent models, which include predicted responses across the range of each variable. The insulationSummary.csv contains the summarised data for our tables in the insulation field experiment - this includes the mean temperature difference ('mean' - degrees Celcius), the standard deviation ('sd' - degrees Celcius) around these and three quantiles ('q0.05', 'q0.5', and 'q0.95' - degrees Celcius), and the sample size ('sampleSize') around these for each species ('species' - latin binomial). In addition, the "./output/occ_data/
" folder contains the processed occurrence data described above that was necessary for meeting our obligations for the use of NVS data.
The ./graphs/
folder stores all of our visual output and contains a folder for supplementary materials (./graphs/supplementary/) as well. We have not supplied these graphs, but they can be readily reproduced using the code supplied.
Code/software
To reproduce our analyses users will require R (Version 4.0.4, R Core Team, 2022), and RStudio (Version 1.4.1103; Allaire, 2012) if you wish to use our project. Data manipulations were done through a combination of dplyr (Version 1.0.10; Wickham et al., 2022), sf (Version 0.9-4; Pebesma, 2018), raster (Version 3.3-13; Hijmans, 2013), lubridate (Version 1.7.9; Grolemund & Wickham, 2011). ENMs were fitted with dismo (Version 1.1-4; Hijmans et al., 2017), while the linear models were fitted with lme4 (Version 1.1-23; Bates et al., 2015) and pairwise comparisons made with predictmeans (Version 1.0.4; Luo et al., 2014). Growth form and insulation (Figs 2, 3, 4) graphs were created with ggplot (Version 3.3.2; Wickham, 2011), gghalves (Version 0.1.0) or ggExtra (Version 0.9), and viridis (Version 0.5.1) colour palettes. To run dismo::maxent the user will additionally need to install Java, rjava, the maxent software (https://biodiversityinformatics.amnh.org/open_source/maxent/) and move the maxent.jar file (see dismo::maxent for more details).
Access information
Other publicly accessible locations of the data:
- GBIF (all datasets were CC BY) - https://www.gbif.org/
- NVS (all datasets can be accessed with approval from Manaaki Whenua LandCare and associated stakeholders) - https://nvs.landcareresearch.co.nz/
- NZEnvDS (CC BY 4.0) - https://datastore.landcareresearch.co.nz/dataset/nzenvds
Methods
Global distribution of growth forms
We reviewed the descriptive literature on tree ferns to identify, where possible (skirts are not a frequently described trait), the global proportion and distribution of tree ferns with marcescent skirts and/or prostrate taxa. We conducted a literature search using the Google Scholar (https://scholar.google.com), Scopus (www.scopus.com) and ISI Web of Science (www.isiknowledge.com) databases with the search terms [“tree fern” OR “tree-fern” OR "treefern"] and [“skirt*” OR “collar” OR “marcescen* OR “attached necromass” OR “prostrate”]. To identify taxa from America we also used Google Scholar to search in Spanish and Portuguese using ["helecho* aborescente”, OR “helecho* arbóreos” AND “falda”] and [“samabaia” AND “saia”]. Finally we checked for the presence of skirts as a recorded trait in the Botanical Information and Ecology Network (BIEN; https://bien.nceas.ucsb.edu/bien/biendata/) database.
The geographic distribution and elevation data of all extant species from the order Cyatheales were downloaded from the global biodiversity information facility (GBIF.org, 2019) and taxa cross-checked with those identified from the review as having a marcescent skirt, and a ‘skirt’ field (Boolean) was added to the database (Table 1). As taxonomic and descriptive information on tropical Cyatheales is sometimes poor, for species whose median values of GBIF records distributions occur in the temperate zones, their growth habit—trunk (with and without skirt), prostrate, no trunk—and presence or absence of recorded skirt trait were collated separately for the northern and southern hemispheres.
Environmental niche models: Regional correlates of growth form
New Zealand provides a useful case study region to investigate the distribution of traits within tree fern taxa in relation to environment. Forest ecosystems occur throughout the almost 13 degrees of latitude provided by the three main islands of the New Zealand archipelago, and various mixtures of the tree fern assemblage of nine species are abundant in most of these forests (Brock et al., 2016). As well, New Zealand forests have been relatively well described and studied compared to many other regions with tree fern taxa (e.g. Wardle, 1991).
Three species of erect trunked tree fern occur across New Zealand in environments that seasonally experience sub-zero temperatures: Cyathea smithii (Cyatheaceae), Dicksonia squarrosa, and Dicksonia fibrosa (both Dicksoniaceae). Dicksonia fibrosa in particular is commonly exposed to sustained sub-zero temperatures and has a recorded lethal temperature of −11°C (Warrington & Stanley, 1987) compared to −4.2°C for Cyathea smithii (Bannister, 2007). All three of these species also have marcescent skirts (Brock & Burns, 2021; Page & Brownsey, 1986). For both D. fibrosa and C. smithii the skirt trait is consistently expressed along a latitudinal gradient (Brock & Burns, 2021; Figure 1a,b). However, the expression of a skirt in D. squarrosa is highly variable; individuals are frequently without any skirt at all in the north of the country, whereas skirts are common and well-formed in the south (Brock & Burns, 2021). There are also a further two taxa of tree fern that occur at higher elevations only in the understorey: Dicksonia lanata subsp. lanata (Dicksoniaceae) and Cyathea colensoi (Cyatheaceae); these species either have no trunk (Dicksonia lanata subsp. lanata), or tend to produce a prostrate trunk although short (< 1m) erect trunks have occasionally been recorded (Brock et al., 2016).
To identify environmental patterns among the skirted, non-skirted, prostrate, and trunkless tree fern occurrences across New Zealand we created environmental niche models (ENM) (Townsend Peterson, 2006) of each of the nine native tree fern species. ENMs were created using presence-only data with a maximum entropy algorithm through the dismo package (Version 1.3-3; Hijmans et al. 2017). Each species was modelled individually and was evaluated using five-fold cross-validation to calculate standard errors around response curve estimates and variation in the area under the receiver operator curve metric.
Presence data for these taxa in New Zealand were collated from the national vegetation survey (NVS; http://nvs.landcareresearch.co.nz) and GBIF, resulting in 28,948 unique presence records across the nine native species of tree fern (Table 2). Insufficient distribution data was available to separate the two sub-species of D. lanata and as such these two sub-species (no trunk / prostrate to short erect trunk) were modelled as a single taxon. A potential bias in these data is that they are unlikely to reflect a complete natural distribution of each species, due to anthropogenic modifications to the landscape over the last c. 800 years (McGlone, 1989). However, suitable forest habitat for the Cyatheales in our study remains across the New Zealand landmass (and across the broad ranges of these readily-dispersed taxa) such that the current distribution data are notlikely significant constraints for comparisons of occurrence amongst the various taxa and across environmental gradients (McGlone, 1989; Mountier et al., 2018). We are not attempting to describe the precise niche of these tree fern species, but compare the environmental conditions in which they occur; therefore, we made the following two assumptions:
1. The available data are of species occurring naturally in the landscape, therefore we can assume that locations of species presence relative to one another reflects a subset of the former habitat for a given tree fern species, and thus adequately describe their relative ecological niches for the purposes of growth form and skirt trait comparisons.
2. To establish and compare the environmental conditions associated with different tree fern taxa and their growth forms and skirt trait expression, we can use ENMs to describe where these taxa occur along environmental gradients.
Environmental data for the ENMs were obtained from McCarthy et al. (2021) ‘New Zealand Environmental Data Stack’ (NZEnvDS). All environmental predictors and observation points used a NZGD2000 datum with a Transverse Mercator projection. Environmental predictors had a resolution of 100 m × 100 m and the extents included NZ’s three largest islands, as well as all near shore islands (McCarthy et al., 2021).
Candidate layers from the NZEnvDS were selected on potential ecological relevance to tree fern establishment, growth and reproduction. From these we determined four layers which cover distinct aspects of a tree fern’s abiotic environment that could be reasonably expected to have some influence on tree fern growth forms. As such, each layer represents a distinct hypothesis or explanation for the distribution of growth forms. For each species presence/pseudo-absence points we subsequently checked that the candidate layers had Pearson’s pairwise correlations < ~ |0.7| with each other, to ensure that the marginal response curves were interpretable (Table SM-1 in Supplementary Materials; Dormann et al., 2013).
The final model included mean precipitation in the driest quarter (mm), the summed mean temperature of days above 5°C (a.k.a. growing degree days above 5°C = GDD5), mean annual solar radiation (MJ/m2/day) and a topographic position index (ordinal factor) (McCarthy et al., 2021). These four layers in the model represent aspects of a tree fern’s abiotic environment: moisture availability, ambient temperature, solar input, and position along a catena representing soil conditions (Brock et al., 2018; Lehmann et al., 2002; Richardson et al., 2008).
We quantified each covariate’s predictive power using leave-one-out cross-validation (LOOCV) and reported it as the permutational importance of each predictor to the MaxEnt model’s fit (Phillips et al., 2017). The performance of each ENM was evaluated using the area under the receiver-operator curve (AUC) (Fielding & Bell, 1997) calculated and averaged over five-fold cross-validation and subsequently calculated for test data (Table SM-2). AUC was chosen as the evaluation metric because the presence-only data precluded the creation of true confusion matrices. Output from the models is displayed on a complementary log-log (clog-log) scale allowing for interpretation as probability of occurrence (Phillips et al., 2017). Along each environmental predictor gradient, we plotted the probabilities of occurrence to establish whether the different tree fern taxa responded differentially across New Zealand. The variability across the five replicates from the cross-validated runs was used to estimate standard deviations around the mean for each response curve.
In order to evaluate how niche space established in each dataset differed between tree fern growth forms we used a principal components analysis fitted on a random sample of 10,000 points from the environmental layers used for the MaxEnt models to sample the background environment of New Zealand. The observed environmental values for each growth form were added as supplementary points to the principal component space. Subsequently, 50%, 75% and 99% kernel density estimates were overlaid to show the concentration of points for each of these growth forms in New Zealand’s climate space. In addition, we added the environmental vectors for each variable to the PCA plot to show their associations with the first two principle components.
Field experiment: Insulation potential of skirts
To establish the insulation potential of tree fern marcescent skirts, fifteen DS1921G-F5 thermochron iButton® (Maxim Integrated Products) thermocouples were installed underneath the skirts of fifteen individual tree ferns, five individuals (>2 m tall) of each of the three skirted species (C. smithii, D. squarrosa, D. fibrosa) in the Lake Okataina Scenic Reserve (38°05′19″S, 176°25′38″E), west of Rotorua (Bay of Plenty, northern New Zealand) (Fig. 1). This forest was selected as it has an easily accessible frost flat heathland (Smale 1990) that creates a clearing in the centre of the reserve and supports all three species that consistently retain fronds. A further fifteen thermocouples were installed in exposed custom-made radiation screens no more than 2 m away from each of the tree ferns with a beneath skirt thermocouple (screens as in Morales et al. 2016). The thermocouples were installed in June 2019, set to record temperature data hourly, and collected in early September 2019: the coldest period of the southern hemisphere winter. We compared the differences between beneath skirt and exposed temperatures during periods when exposed temperatures were sub-zero using a linear model with random-effects accounting for repeated measures.