TRY plant trait database – enhanced coverage and open access

This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. © 2019 The Authors. Global Change Biology published by John Wiley & Sons Ltd A list of authors and their affiliations appears in the Appendix. Abstract Plant traits—the morphological, anatomical, physiological, biochemical and phenological characteristics of plants—determine how plants respond to environmental factors, affect other trophic levels, and influence ecosystem properties and their benefits and detriments to people. Plant trait data thus represent the basis for a vast area of research spanning from evolutionary biology, community and functional ecology, to biodiversity conservation, ecosystem and landscape management, restoration, biogeography and earth system modelling. Since its foundation in 2007, the TRY database of plant traits has grown continuously. It now provides unprecedented data coverage under an open access data policy and is the main plant trait database used by the research community worldwide. Increasingly, the TRY database also supports new frontiers of trait-based plant research, including the identification of data gaps and the subsequent mobilization or measurement of new data. To support this development, in this article we evaluate the extent of the trait data compiled in TRY and analyse emerging patterns of data coverage and representativeness. Best species coverage is achieved for categorical traits—almost complete coverage for ‘plant growth form’. However, most traits relevant for ecology and vegetation modelling are characterized by continuous intraspecific variation and trait–environmental relationships. These traits have to be measured on individual plants in their respective environment. Despite unprecedented data coverage, we observe a humbling lack of completeness and representativeness of these continuous traits in many aspects. We, therefore, conclude that reducing data gaps and biases in the TRY database remains a key challenge and requires a coordinated approach to data mobilization and trait measurements. This can only be achieved in collaboration with other initiatives.


| INTRODUC TI ON
Plant traits-the morphological, anatomical, physiological, biochemical and phenological characteristics of plants measurable at the individual plant level (Violle et al., 2007)-reflect the outcome of evolutionary and community assembly processes responding to abiotic and biotic environmental constraints (Valladares, Gianoli, & Gomez, 2007). Traits and trait syndromes (recurrent coordinated expressions of multiple traits) determine how plants perform and respond to environmental factors (Grime, 1974;Wright et al., 2017), affect other trophic levels Loranger et al., 2012Loranger et al., , 2013, and provide a link from species richness to functional diversity, which influences ecosystem properties and derived benefits and detriments to people (Aerts & Chapin, 2000;Díaz et al., 2004Díaz et al., , 2007Garnier & Navas, 2012;Grime, 2001Grime, , 2006Lavorel et al., 2015;Lavorel & Garnier, 2002). In the context of the Global Earth Observation Biodiversity Observation Network (GEO BON) species traits are considered an Essential Biodiversity Variable to inform policy about biodiversity change (Kissling et al., 2018;Pereira et al., 2013). A focus on traits and trait syndromes, therefore, provides a crucial basis for quantitative and predictive ecology, ecologically informed landscape conservation and the global change sciencepolicy interface (Díaz et al., 2016;McGill, Enquist, Weiher, & Westoby, 2006;Westoby & Wright, 2006). To fully realize this potential, plant trait data not only need to be available and accessible in appropriate quantity and quality but also representative for the scales of inference and research questions . Here we analyse where the TRY plant trait database stands with respect to coverage and representativeness after 12 years of operation. We further review the mechanisms and emergent dynamics helping to increase both.
However, these databases were either focused on particular regions (BiolFlor, LEDA, BIOPOP, BROT, USDA Plants, Ecological Flora of the British Islands, BRIDGE) or specific traits (GLOPNET, SID). A 'database of databases' was in discussion for some time, but it had been impossible to secure long-term funding for such a project. Finally, at a joint workshop of the International categorical traits relevant to attribute species to plant functional types (PFTs), and provides the opportunity to publish plant trait data sets and receive a DOI.  Figure 1a). Currently, TRY is working on version 6. As of July 2019, the TRY database comprised 588 data sets from 765 data contributors (Table A1).
The dynamics of the number of data sets in TRY indicates an increasing success of calls to the scientific community for data contribution in 2007, 2013 and 2019. When the manuscript was submitted, data contributions responding to the call in 2019 were not yet fully integrated into the TRY database. Therefore all analyses presented in this paper are based on versions 1-5 of the TRY database (Table 1). TRY version 5, released on 26 March 2019, contains 387 data sets providing 11.8 million trait records, accompanied by 35 million ancillary data, for 2,091 traits and 280,000 plant taxa, mostly at the species level (Table 2). Data coverage is still driven by a few large (often integrated) databases, but increasingly small data sets (mostly primary data) contribute to the overall coverage ( Figure 3a). Plant trait data in TRY can be traced to >10,000 original references. This highlights the breadth of data integrated in the TRY database and its nature as database of databases, a 'second generation of data pooling' (M. Westoby, personal communication, August 24, 2009).
We now observe a tendency that new trait-based research is increasingly planned against the background of the TRY database.
Coverage and availability of trait data in TRY stimulate trait-based research, which then often leads to the identification of unexpected data gaps. This motivates data mobilization and/or new measurements, which improve the availability of plant trait data for the research community, and-if contributed to TRY-help the database grow. Examples for such a 'feed-forward data integration loop' are provided in Box 1.
To support this process, in this article, we take stock of the data compiled in the TRY database and present emerging patterns of data coverage and representativeness with a focus on the identification of principal and systematic gaps. Finally, we discuss ways forward and the potential future role of the TRY initiative for the research community.

| Plant trait data in the TRY database
Plant traits can be classified as categorical (qualitative and ordinal) or quantitative (continuous) traits . Some traits are rather stable within species (mostly categorical traits), and some of these can be systematically compiled from species checklists and floras (e.g. . These measurements of quantitative traits are single sampling events for particular individuals at certain locations and times, which preserve relevant information on intraspecific variation and provide the necessary detail to address questions at the level of populations or communities. Within individual field campaigns or experiments, researchers often aim to measure complete sets of these data: all traits of interest for all individuals or species in the analyses. However, across studies and data sets and at large scales, the coverage of these data shows major gaps, which provide major challenges concerning data completeness and representativeness .

| Data coverage
Compared to TRY database version 1 and the state reported in , TRY version 5 has substantially grown with respect to the number of trait records, traits, species, entities, georeferenced measurement sites and ancillary data (Table 2).

| Trait records and entities
The numbers of trait records (individual trait measurements) and entities (individual plants or plant organs on which the measurements have been taken) increased by a factor of about 6 for trait BOX 1 Examples for the 'feed-forward data integration loop' observed in the context of the TRY database • Iversen et al. (2017) indicated that in the TRY database only 1% of trait records were related to roots. This motivated the development of the Fine-Root Ecology Database (FRED) specializing in the mobilization of fine-root trait records from the literature (Iversen et al., 2017). In the meantime, the first versions of the FRED database have been contributed to the TRY database. The improved number and availability of trait data on roots allowed for a project on root trait functionality in a whole-plant context (https ://www.idiv.de/en/sdiv/worki ng_group s/wg_pool/sroot.html), which motivated additional mobilization of root trait data.
• The promising coverage of plant trait data from tundra regions in the TRY database encouraged the inclusion of plant traits in an analysis of tundra ecosystem change, scaling shrub expansion from site to biome (Bjorkman, Myers-Smith, Elmendorf, Normand, Rüger, et al., 2018). In the context of the project, a large number of additional trait data were mobilized by the Tundra Traits Team (Bjorkman, Myers-Smith, Elmendorf, Normand, Thomas, et al., 2018), which have recently been contributed to the TRY database.
• Moreno-Martínez et al. (2018) estimated the worldwide variation of several leaf traits to improve the parameterization of global vegetation models and remote sensing approaches predicting, for example, gross primary productivity. Due to the low representation of traits for crop species in the TRY database, they could not provide estimates for major agricultural regions (see white spots in figure 5 of Moreno-Martínez et al., 2018). The identification of these gaps motivated mobilization of trait data for crop plants and agro-ecosystems (Engemann et al., 2016;Martin, Hale, et al., 2018;Martin & Isaac, 2015), which were then contributed to the TRY database.
• Trait data on plant growth form (tree, shrub, herb, etc.) were compiled by TRY, extended and consolidated in the context of the BIEN initiative (Engemann et al., 2016;Enquist, Condit, Peet, Schildhauer, & Thiers, 2016) and then contributed to the development of the GIFT database . The upgraded plant growth form data were contributed again to TRY.
• Plant species richness is unequally distributed across the globe, with the highest species richness observed in the tropics (von Humboldt, 1817). The highest numbers of species with measurements in TRY are also found in the tropics, but as well the largest gap relative to reported species richness: less than 1% of estimated species richness is represented in TRY (Jetz et al., 2016). This principal and systematic mismatch of data coverage and representativeness has contributed to motivate the development of a 'global biodiversity observatory' of in situ measurements and space-borne remote sensing that tracks temporal changes in plant functional traits around the globe to fill critical knowledge gaps, aid in the assessment of global environmental change and improve predictions of future change (Jetz et al., 2016).
• TRY is involved in the sPlot initiative to establish a global vegetation-plot database (www.idiv.de/en/splot.html). sPlot supports the analysis of plant communities across the world's biomes by combining vegetation-plot data with traits from the TRY database (Bruelheide et al., 2019). This has resulted, for the first time, in global analyses of plant functional community data (Bruelheide et al., 2018). In contrast to single species measurements or trait values aggregated in grid cells, using vegetation-plot data allows understanding the role of traits for biotic interactions and community assembly processes. In turn, trait data measured in the context of sPlot are contributed to TRY. records and 5 for entities from TRY version 1 (2.1 million trait records measured on 1.1 million entities) to TRY version 5 (11.8 million trait records measured on 5.0 million entities). The average number of trait records per entity increased from 1.9 to 2.4 (Table 2).

| Traits
The number of traits has grown steadily from TRY version 1 to 5, apart from a steep step from TRY version 3 to 4 ( Table 2). This step was caused by the contribution of the FRED database, which added about 700 new traits for roots. Data coverage across traits is characterized by long-tail distributions: a small number of traits is well covered by records and species, while the majority of traits has only very low coverage of records and species (Figure 3). However, the number of continuous traits with more than 1,000 records (which are subject to intense data quality assurance during integration) has increased from about 200 in TRY version 1 to 600 in TRY version 5 ( Figure 3b). The number of traits with data for more than 100 species has increased from 300 to 700 (Figure 3c). In parallel, the number of records per trait and species ('intraspecific retakes') has increased from almost zero traits with on average more than 10 records per

Data acquisition
In the context of the TRY initiative, data acquisition so far relies on active contributions by the community-data sets need to be sent by email or uploaded at the TRY website (https ://www.try-db.org/TryWe b/Submi ssion.php). From time to time (2007, 2013 and 2019), TRY sends out calls for data contributions to the community. However, so far there has been no systematic screening of public data repositories like DRYAD or PANGAEA for plant trait data.

Data integration
The basic principle of data integration in the TRY database is to preserve the original trait and ancillary data and annotate these with complementing and consolidated information. Data integration consists of three major components: data consolidation, complementation and quality assurance. We here provide a brief overview; a detailed description can be found in Supporting Information and on the TRY website (https ://www.try-db.org/TryWe b/Datab ase.php).

Data consolidation
The data structure is transformed into the entity-attribute-value (EAV) model and the OBOE schema (Madin et al., 2007)

Data complementation
After consolidation, additional trait values are derived from contributed trait data where possible; for example, leaf nitrogen content per area from leaf nitrogen content per dry mass and specific leaf area (SLA) if both were measured on the same entity.

Data quality assurance
Continuous traits with >1,000 records in the database are subject to a three-step process: (a) Systematic errors, like a wrong unit for a given trait for all records of a specific data set, are identified across data sets with semi-automated procedures and cor- After a data set has been integrated into the TRY database, the data set custodian is asked for feedback; that is, whether consolidated trait names are appropriate and consolidated values correct. Data are reformatted for data release and format errors (i.e. tabs and line breaks in database cells) are corrected. Finally, the original and consolidated data (including flags for outliers and duplicates) are released on request as tab-delimited text files.
trait-species combination in TRY version 1 to more than 500 in TRY version 5 (Figure 3f).
The traits with the best species coverage in TRY version 5 are mostly categorical (Table 3). The categorical traits used for the classification of PFTs-plant woodiness, plant growth form, leaf type (broadleaved vs. needle-leaved), leaf phenology type (deciduous vs. evergreen), leaf photosynthesis pathway (C3, C4, CAM)-are still among the best covered. However, the number of species characterized for each of these traits has substantially increased from TRY version 1 to 5, most significantly for plant growth form from 31,327 to 263,357 species, supported by the contribution from the GIFT database . The 30 traits that were most often requested (Table 4) are dominated by continuous traits related to the global spectrum of plant form and function (Díaz et al., 2016), the leaf economics spectrum (Wright et al., 2004) and rooting depth. Only seven categorical traits are among these 30 traits. This indicates a switch between well covered-categorical-traits (Table 3) and most frequently requested-continuous-traits (Table 4). The first five most documented traits are categorical whereas among the 10 most requested traits, only one is categorical. However, within continuous traits, there is, in general, a good match between traits characterized for most species and traits most often requested. To some extent this may be influenced by the amount of available data for the individual traits. However, a noteworthy exception is rooting depth, as 10% of requests ask for this trait, while it is 'only' covered for 3,886 species, mostly contributed via the Global Dataset of Maximum Rooting Depth (Fan, Miguez-Macho, Jobbágy, Jackson, & Otero-Casal, 2017). This mismatch indicates a demand for more data on the most relevant belowground traits.

| Species
From TRY version 1 to 4, the number of species increased slowly, but almost doubled to version 5 due to the contribution of plant growth form data from the GIFT database, which added about 100,000 new species. As in the case of traits, the data coverage for species is characterized by long-tail distributions: few species are covered well by measurements and traits, while the majority TA B L E 3 Traits with best species coverage. The 30 traits covering the highest number of species in the TRY database version 5 and the number of species represented for these traits in TRY version 1. Data type: cat = categorical; con = continuous. Sorted by the number of species in TRY 5  Leaf C mass (mg/g) Leaf P mass (mg/g)

| Entity × trait and species × trait matrices
The trait data in the TRY database can be represented by two two-dimensional matrices: the entity × trait matrix, with entities in rows and traits in columns; and the species × trait matrix, with species in rows and traits in columns. Both matrices are characterized as large but sparse: high numbers of entities, species and traits in TRY make the two matrices large, but many cells in the matrices are empty. From TRY version 1 to 5 the size of the matrices has grown by a factor of 15, but at the same time the number of trait records to fill

| Ancillary data
The numbers of ancillary data, geo-referenced trait records and trait records with measurement date increased by a factor of almost 10 from TRY version 1 to TRY version 5 ( Table 2). The ratio of ancillary data to trait records, therefore, increased from TRY version 1 to 5 from 2:1 to 3:1 and the fraction of geo-referenced trait records from about 33% to 42% ( Table 2). The number of geo-referenced trait records with information on measurement date that could be standardized to year, month and day increased from 290,000 in TRY version 1 (15% of all trait records) to 2.5 million records in TRY version 5 (20%).
The increasing ratio of ancillary data to trait records indicates growing awareness for the relevance of environmental conditions during plant growth and trait measurements. In this context, geo-references (and date) are crucial, as they allow trait records to be related to information on climate, soil or biome type from external sources.  Díaz et al. (2016); red: the best covered quantitative traits representing each of the six plant parts (see Figure 6): shoot (plant height vegetative), reproductive organs (seed dry mass), whole plant (plant lifespan), leaves (SLA), roots (rooting depth) and dead material (litter decomposition rate)

(a) (b)
3,320 1° × 1° grid-cells; Table 2; Figure 6). Europe still has the highest density of measurement sites, but TRY version 5 also provides good coverage for the United States and China. The number of measurement sites has substantially improved for several other regions as well, for example Central America, Russia, Asia and parts of central Africa. However, there are still obvious gaps in boreal regions (Canada, East Russia) and some parts of the tropics and subtropics, particularly in Africa ( Figure 6).

| Data completeness and representativeness
To progress from a description of data coverage towards an analysis of representativeness, we need a baseline for comparison. At the global scale, this information has been lacking. Reference data sets have become available only recently for plant growth form  and phylogeny (Smith & Brown, 2018)  trait records to the different parts of the plants has only changed little from TRY version 1 to 5. However, the fraction of records for root traits has substantially increased (from 0.7% to 2.0%), due to the contribution of the FRED database.

| Plant growth form and PFTs
The In GIFT, about 50% of species are currently assigned to herbs, 30% to trees and 20% to shrubs (Figure 8). This distribution is well reflected by the species in TRY (excluding data from the GIFT database). However, the six best covered continuous traits in TRY indicate that this distribution is very much trait dependent, with a bias towards trees versus herbs, while the fraction of shrubs is surprisingly constant and close to the fraction in the GIFT database ( Figure 8). The overrepresentation of trees is most obvious for SSD, which is not surprising because SSD is a more general concept derived from wood density, a trait relevant for forestry, timber industry and estimates of forest vegetation biomass. However, the tendency of relatively more data for trees compared to other growth forms is also obvious for SLA, leaf nitrogen content per dry mass and leaf area, but opposite for root length per root dry mass (specific root length), which is frequently reported for herbs.

| Phylogeny
Smith and Brown (2018)  Visually, the 208,000 species with data in TRY are well distributed across the 350,000 species represented in the phylogeny of seed plants (Figure 9). An ancestral state reconstruction (ASR) of species trait number confirms that the long-tail distribution previously seen at the species level also holds in a phylogenetic context: some clades are covered very well (bright colours), while most clades have lower data coverage (dark colours). The ASR additionally shows how deep in phylogeny data gaps are rooted. This indicates the potential for, and limits to, phylogenetically or taxonometically informed gap-filling (Schrodt et al., 2015). Examples of high-coverage clades are (parts of) the Pinales, Poales and Asterales. When looking at the six best-covered continuous traits individually, we find these too are well distributed across the phylogeny (Figure 9). Jetz et al. (2016) reported a latitudinal gradient in disparities between plant species with regional measurements in TRY and estimated species richness, with the largest gap observed in the tropics, because these are especially rich in species. To address this in more detail, we   Kier et al. (2005). This approach accounts to some extent for intraspecific trait variation, as it counts only species with at least one trait measurement in the given ecoregion.  Smith and Brown (2018). Rings surrounding the phylogeny indicate from inside outwards: (i) number of traits per species (innermost ring); (ii) presence of data for six of the best covered continuous traits, specifically: leaf area, leaf area per leaf dry mass (SLA), leaf nitrogen content per dry mass (LeafN), seed dry mass, plant height and stem specific density (SSD). Colours of phylogeny branches represent an ancestral state reconstruction of the number of traits per species. White and grey circles indicate periods of 50 million years. For visibility, only 5% of species (randomly selected) are presented up to 1,400 species for some ecoregions in Europe (Alps conifer and mixed forests) and tropical South America (Napo moist forests, Tapajos-Xingu moist forests). In general, high absolute numbers of species with trait measurements for ecoregions are found in Europe, East Asia, Oceania, Australia, tropical South America and the United States (Figure 10a). East Asia, Oceania and tropical South America are also the regions with the highest numbers of species per ecoregions estimated by Kier et al. (2005; Figure 10b). contributed by the mycorrhizal intensity database (Akhmetzhanova et al., 2012). Some other ecoregions are also well covered with data for more than 50% of estimated species (Southeast Australia temperate savanna, Qaidam Basin semi-desert, Córdoba forests and mountain grasslands). Apart from these individual ecoregions spread across the world, large parts of Europe are well covered, with trait data for about 30% of the species number estimated by Kier et al. (2005). Some ecoregions in East Asia, Australia, tropical South America, the Sahara, and the United States are also well covered, providing data for about 20% of estimated species richness. Very low relative coverage (<2%) is observed for major parts of Canada, Africa, western Asia (Iran, Iraq, Pakistan, Afghanistan) and major parts of India. for an analysis of 100 species, and nine traits allow for the analysis of intraspecific variation of more than 10 species. However, the numbers are more humbling if the environmental context is taken into account ( Figure 11b). If we assume that trait records from a minimum number of 50 sites per species are necessary to represent intraspecific variation, four traits are sufficiently covered for about 100 species. If 100 sites are necessary, no trait is covered by data for more than 10 species.

| D ISCUSS I ON
Plant trait data provide a wealth of information directly relevant in several scientific contexts, from conservation, ecology and evolution to earth system sciences. To fully realize this potential, the TRY (entity × trait and species × trait) has increased faster than the number of trait records to fill the matrices. Therefore the sparseness of the matrices has increased from TRY version 1 to 5 (the fractional coverage declined). Rather than converging in a small number of traits, the scientific community continues to measure a large, diverse number of traits, following equally diverse motivations.
However, given the number of species has a natural limit and assuming the number of traits will continue to grow, but more slowly, once the most obvious ones have been covered, we expect that the sparseness of the entity × trait matrix will become stable: new data adding new rows for entities, but not many new columns for traits.
In comparison, the sparseness of the species × trait matrix should decline in the future; new data will mostly contribute to filling the matrix and increasing the number of species with data per trait. This reduced sparseness of the species × traits matrix will systematically improve the applicability of trait data for macroecology and earth system modelling and will facilitate multivariate analyses for an increasing number of traits. In parallel, the number of records per species-trait combination is increasing: between TRY 1 and 5 it already doubled and will further increase in the future. This increasing number of records per species-trait combination will improve data coverage for analyses of intraspecific trait variation and trait-environment relationships accounting for intraspecific variation. It is noteworthy that the matrix will not only become more complete, but the traits will increasingly be able to inform each other. The 'usual suspects' (i.e. the best covered continuous traits) might not be masters of all traits, but they surely will be very useful as baseline traits and provide a background against which other-maybe more influential-traits can be analysed for coverage, representativeness, orthogonality, etc.

| Data completeness and representativeness
Despite unprecedented and continuously growing data coverage, we observe a humbling lack of completeness and representativeness in many aspects. The best species coverage is achieved for categorical traits relevant to determine PFTs commonly used in global vegetation models. For the traits 'woodiness' and 'plant growth form', even full species coverage is within reach, due to the contribution of data from the GIFT database. For the first time, this provides a global baseline for these traits, which are relevant to understand basic patterns of variation for several other traits (Díaz et al., 2016). With this baseline, future analyses will be able to address representativeness in addition to coverage, which will per species, which is far above average (see Figure 3e). We therefore conclude that the coverage of trait data in TRY is biased towards the more abundant species in the respective ecoregions-which is reasonable and welcome for many kinds of analyses.
We have reported that intraspecific variation in space is increasingly well covered, but variation in time is hard to estimate.
Nevertheless, intraspecific variation in time is relevant for several traits to characterize the seasonal variation of plant and ecosystem function (Xu & Baldocchi, 2003;Xu & Griffin, 2006) and long-term trends to inform policy about biodiversity change (Kissling et al., 2018). About half of the geo-referenced trait records have information on the sampling date that could be standardized to year, month and day, but systematic replicates over time ('time series ') are rare (but see e.g. the 'Photosynthesis Traits Database'; Xu & Baldocchi, 2003). In principle, 'non-time series' data allow detection of trait changes over time , but these analyses are very challenging, as most traits demonstrate stronger variation in geographic space along climate and soil gradients than over time.
In addition, the variations of traits on different time scales (diurnal, seasonal, inter-annual variation and long-term trends) are superimposed and hard to disentangle. Apart from this, there is a need to collect and report repeated trait measurements from the same location or population to monitor biodiversity change and inform policy, for example in the context of GEO BON (Kissling et al., 2018). Figure 1 shows the most obvious way to mobilize additional trait data: the TRY initiative should regularly send calls for data contribution to the wider scientific community, that is the network of more than 6,000 researchers contributing and using trait data via TRY.

| Ways forward
These calls should be combined with regular publications of respective reference papers. This can be combined with (a) a systematic collection of data sets from public data repositories, which is becoming more effective with the general move by many journals to require that authors make their data open access; and (b) systematic extraction of trait data from the ecological literature, floras and herbarium specimens, which is a promising task, especially for its potential to open a window into the past. In parallel, TRY should further support the 'feed-forward data integration loop' outlined above: using trait data via TRY, identifying gaps, mobilizing and/or measuring new data, contributing additional data to TRY. This has proven very effective for focused data mobilization. If relevant gaps are detected, TRY can also send specific calls to the community.
As TRY has been designed as a community cyber-infrastructure based on the idea of incentive-driven data sharing (Kattge, Díaz, & Wirth, 2014), the collaboration and data exchange with other plant trait databases will continue to be the key to achieve a comprehensive representation of plant traits. TRY is, therefore, collaborating with many more recent trait database initiatives, such as, for example, FRED, GIFT, BIEN and the Tundra Trait Team, and since the early days of TRY-GLOPNET, LEDA, SID, BiolFlor, BIOPOP, BROT, the Ecological Flora of the British Isles, eHALOPH, USDA PLANTSdata, BRIDGE and many others. Importantly, these collaborations need to provide mutual benefit. Based on these collaborations, the TRY database may serve as a central node for plant traits in an overarching network of trait databases, currently emerging in the context of the Open Traits Network (Gallagher et al., in press). Finally, new techniques and approaches are gradually becoming available, which may substantially change how plant trait data are collected: remote sensing, citizen sciences, microbiological and molecular screening, etc.

| Towards a third generation of plant trait data integration and sharing
We expect that the combination of (a) systematic involvement of the

| CON CLUS ION
TRY has received institutional support since 2007 and is still growing considerably in quantity and quality. While TRY may be considered a success and potentially a role model for database initiatives, it is important to realize that this development needed time and patience.
It took until 2011 for the first TRY publications to appear because the early years of TRY were mostly devoted to the development of the database, organizing the community process towards a joint data sharing policy and building trust. This process involved initially dozens and later hundreds of scientist when it came to agree on moving towards open access. These dynamics do not fit into 3 year funding cycles as typically offered by national funding agencies. A key lesson of TRY is that the development of a database that is trusted by the community and accepted for its service and quality also needs the trust of the funders, that is long-term support, at the scale of decades rather than years. It also needs journals that are willing to accept long author lists and extended references lists to adequately acknowledge the original contributions that are the building blocks of communal databases.  Bauters, M., Vercleyen, O., Vanlauwe, B., Six, J., Bonyoma, B., Badjoko, H., … Boeckx, P. (2019). Long-term recovery of the functional community assembly and carbon pools in an African tropical forest succession. Biotropica, 51 (3), 319-329. https ://doi.org/10.1111/btp. 12647 Beckmann, M., Hock, M., Bruelheide, H., & Erfmeier, A. (2012). The role of UV-B radiation in the invasion of Hieracium pilosella-A comparison of German and New Zealand plants. Environmental and Experimental Botany, 75, 173-180. https ://doi.org/10.1016/j.envex pbot.2011.09.010 Belluau, M., & Shipley, B. (2017. Predicting habitat affinities of herbaceous dicots to soil wetness based on physiological traits of drought tolerance. Annals of Botany, 119 (6) (3) Estimation of photosynthesis traits from leaf reflectance spectra: Correlation to nitrogen content as the dominant mechanism. Remote Sensing of Environment,196,[279][280][281][282][283][284][285][286][287][288][289][290][291][292]. https ://doi.org/10.1016/j. rse.2017.05.019 Delpierre, N., Berveiller, D., Granda, E., & Dufrêne, E. (2015). Wood phenology, not carbon input, controls the interannual variability of wood growth in a temperate oak forest. New Phytologist, 210 (2)