A global assessment of the Raunkiæran shortfall in plants: geographic biases in our knowledge of plant traits

The functional traits (measured attributes) of organisms result from interactions with their biotic and abiotic environment. Traits allow us to understand both how individuals and the communities they form will respond to environmental change and how these changes will impact ecosystem services and processes (Lavorel & Garnier, 2002). Plants constitute most of the biomass on Earth (c. 82%; Bar-On et al., 2018), and their traits are the predominant drivers of terrestrial ecosystem functioning (Migliavacca et al., 2021; Fricke et al., 2022). Thus, to a first-order approximation, understanding the traits of plants means understanding terrestrial ecosystems.

There remains a sustained interest in both trait-based ecology (e.g. Lavorel & Garnier, 2002; McGill et al., 2006; Violle et al., 2007; Mouillot et al., 2021) and Open Science (Cheruvelil & Soranno, 2018; Gallagher et al., 2020b; Geange et al., 2021), both of which have contributed to the creation and sharing of large compilations of plant traits constituting millions of observations (e.g. Kattge et al., 2011; Maitner et al., 2017; Sauquet et al., 2017; Weigelt et al., 2020; Falster et al., 2021). However, despite this growing wealth of data, our knowledge of plant traits remains far from complete (the ‘Raunkiæran shortfall’; Hortal et al., 2015).

In addition to trait data being incomplete, recent work by Cornwell et al. (2019) suggests our knowledge of plant traits is also spatially biased, with marked latitudinal variation in coverage. The causes of these biases have not been rigorously tested, but may be driven by: wealthier countries being able to collect and disseminate more data (Meyer et al., 2015); smaller and more-accessible countries being able to sample proportionally more species (Hijmans et al., 2000; Kadmon et al., 2004; Hughes et al., 2021); and countries with few species and low endemism reaching higher completeness more easily. These spatial biases in turn may limit our ability to respond to urgent global changes, particularly if there are discrepancies between where the data are most urgently needed (e.g. where changes are highly uncertain or projected to be severe) and where they are being collected.

Cornwell et al. (2019) examined the coverage of a range of attributes, including traits, in the global flora with a focus on assessing the completeness (fraction of plant species with data available) of information using The Plant List as a taxonomic backbone. Here, we expand on this work by mapping trait completeness (the fraction of species with freely available trait data for a given trait) globally, using the geographic and taxonomic information in the recently completed World Checklist of Vascular Plants (WCVP; Govaerts et al., 2021) and trait information from the widely used TRY database (the most commonly used global plant trait database; Kattge et al., 2011). We test hypothesized drivers of variation in plant trait data availability across the globe, including factors related to: wealth, research expenditure, and educational expenditure; region size and accessibility; and biogeography. We compare spatial patterns of trait data completeness with those of phylogenetic and distributional data completeness (Rudbeck et al., 2022). We test for correlations between trait data completeness and impacts of global change. Finally, we identify solutions for filling significant regional gaps in trait data completeness using Open Science approaches, highlighting the model developed for the AusTraits database (Falster et al., 2021).

The set of public trait data we received from TRY contained 1.5 million unique species × trait combinations across 10.4 million trait observations. Of the 2027 unique traits, the most complete was plant growth form (32.5%; 113 620 species). Most traits had poor coverage (Fig. S1), with a mean global completeness across all traits of 0.21% and a median of 0.0051%. Surprisingly, some traits calculated from multiple measurements had higher coverage than their components (e.g. SLA was available for 14 127 species and leaf dry mass for 6064). After excluding traits with data for < 1% of species (3500), which could not be applied to all species, or which were not properties of individuals, our focal dataset included 5.1 million records across 122 230 species (34.9% of vascular plants) and 53 traits (Table S1). Trait coverage ranged between 1.01% and 32.5% global coverage (mean = 3.48%, median = 1.88%). For a single trait, within a single botanical country, trait completeness varied widely, ranging between 0 and 100% (mean = 19.4%, median = 12.7%), with entirely complete or incomplete data found on Antarctica and small islands. Mean completeness across traits ranged between 2.8% (New Guinea) and 58.7% (Føroyar) across countries (mean = 19.4%, median = 17.3%, Fig. 1). Correlations in completeness across botanical countries were positive and significant for the majority of focal traits, with the exception of four leaf morphological traits (width, length, margin type, and venation type; Fig. S2).

We found that the effects of four of our predictor variables had 95% confidence intervals that excluded zero (Table S2). Trait completeness was positively associated with mean species range size (0.41, 95% CI [0.37, 0.45]) and research expenditure (0.06, 95% CI [0.03, 0.09]). Trait completeness was negatively associated with endemism (−0.13, 95% CI [−0.15, −0.10]) and vascular plant species richness (−0.09, 95% CI [−0.12, −0.06]).

Despite the massive amount of trait data collated to date, we are a long way from fully capturing some of the simplest traits for the majority of plant species across the world's botanical countries. All 53 focal traits examined here were below the 40% coverage threshold noted by Penone et al. (2014) as being needed to impute missing traits with confidence, suggesting that imputation is not yet an option at a global level (but may be of use within certain regions). Our mapping shows that what information we do have is spatially biased, with trait coverage being higher in the Global North (in the socioeconomic sense, which includes Australia). Completeness is generally consistent across traits (see also Notes S1 for flower, wood, and seed traits), showing that we have a general lack of trait data in the Global South, as also noted by Cornwell et al. (2019), rather than simply a different set of traits being measured. However, we note that 19% of traits show a negative correlation with mean focal trait completeness (Fig. S7), but only four of these were above our 1% threshold for inclusion: leaf width, length, margin type, and venation type. These exceptions were driven by regional data aggregation efforts focused on Africa and China (Kirkup et al., 2005; Prentice et al., 2011; Dressler et al., 2014). The geographic biases we observe are similar to those observed for phylogenetic data by Rudbeck et al. (2022), with the factors we identified as driving trait data completeness (range size, endemism, species richness, and research expenditure) being a subset of the factors they identified as driving phylogenetic data completeness, suggesting similar mechanisms underlie the acquisition of both data types. Unfortunately, this correspondence between phylogenetic and trait data completeness (Figs 2, S8) likely means that regions with low trait completeness have relatively less to gain via phylogenetic trait imputation.

Due to the positive correlations between trait data availability and global changes related to temperature and human footprint, many of the regions undergoing the most severe changes may be best positioned to predict those changes. However, this correlation also means that trait data have been disproportionately collected from anthropogenically disturbed regions, which may bias inferences. Conversely, the negative correlation between trait data availability and predicted changes and uncertainty in precipitation may hinder our ability to predict plant responses to altered precipitation. We also note that trait coverage is low in tropical regions which may be approaching climatic thresholds, beyond which irreversible changes in ecosystems may occur (e.g. conversion of rainforest to fire-dominated forests; Malhi et al., 2009).

Our main analysis, which included traits measured anywhere, and our analysis focusing only on traits known to have been measured within particular botanical countries, differed substantially in the magnitude, direction, and significance of socioeconomic and biogeographic predictors. We argue that by removing the confounding influence of shared trait data, the georeferenced data provide a better picture of the drivers of trait data availability at a botanical country level. As we expected, we found trait data completeness is positively associated with national wealth and spending, with wealthier regions that spend more on research and education typically having better coverage. Also as expected, we found that trait coverage is negatively associated with endemism, likely driven by endemics having relatively small ranges and occurring in less-accessible areas, making them less likely to be sampled (Steinbauer et al., 2016; Enquist et al., 2019). Contrary to our expectations, however, we found that larger, less accessible, less secure, and more species-rich regions tended to have higher trait coverage. One potential reason for these unexpected relationships may be that scientists disproportionately choose to work in regions with these characteristics because of the relatively high species richness they harbor.

While we identify several factors that are associated with trait data completeness, we caution that relationships with socioeconomic factors can often be inconsistent and context-dependent (Rydén et al., 2020; Zizka et al., 2021). Furthermore, there are many socioeconomic factors beyond these which may be relevant, and different regions may differ in which factors are relevant (Meyer et al., 2015; Zizka et al., 2021). While much previous work has focused on the correlates of geographic data availability (e.g. Meyer et al., 2015; Hughes et al., 2021), less emphasis has been placed on the availability of trait data (but see Cornwell et al., 2019). We found no significant correlation between trait completeness and geographic completeness, suggesting that the drivers of these two may differ, although some variables will likely be relevant for both (e.g. research funding; Meyer et al., 2015).

The accuracy of our completeness estimates depends on our estimates of country-level species pools, which are hindered by both the Linnean shortfall (number of undescribed species) and the Wallacean shortfall, which are spatially biased (Meyer et al., 2015; Freeman & Pennell, 2021; Hughes et al., 2021). Thus, our analyses will overestimate the coverage for regions with many undescribed or unrecorded species. This provides a potential alternative explanation for our unexpected finding that relatively inaccessible, large, and insecure regions have higher trait coverage: We may be underestimating the species richness in these regions, thereby increasing completeness estimates. We also caution that our estimates of species range sizes are biased by the available data. Species were assumed to occupy the entirety of the regions they occur in, which will lead to overestimates of range sizes, particularly in large regions.

We note that in our main analysis (Fig. 1), some countries with high levels of wealth and spending appear relatively data poor (e.g. Australia and New Zealand). However, when we combine AusTraits with TRY, the level of trait completeness in Australia falls in line with levels seen elsewhere in the Global North (Fig. 3), suggesting that a lack of data integration may underlie some perceived knowledge gaps. In addition to relatively high rates of funding, countries in the temperate and polar regions of the Northern Hemisphere also have relatively low species diversity and large species ranges, allowing them to share data across boundaries. By contrast, many tropical countries have low rates of funding, high endemism, and high species diversity, factors which work against trait completeness. Although the focus of this study was botanical countries, the bias we observe will extend to other geographic classifications (e.g. biogeographic realms, biomes, and ecosystems; Olson et al., 2001) such that our knowledge of those occurring predominantly in the Global South will tend to be relatively poor.

The analyses presented here focus on the most widely used global plant trait database, TRY (Kattge et al., 2011, 2020). However, numerous other plant trait databases and datasets exist, some of which will not have been incorporated in TRY yet, and others which may not be able to be incorporated due to licensing issues. However, by integrating TRY with one open access resource (AusTraits), we were able to more than double trait data completeness for all Australian states (Fig. 3). This supports recent work by Feng et al. (2022) showing that database integration can lead to rapid gains in available data. The AusTraits model demonstrates how the creation and integration of regional databases can rapidly expand trait completeness. With a more limited geographic scope, AusTraits could target its data collection, allowing it to achieve similar completeness to the northern hemisphere despite high endemism and a low population density. This includes personally interacting with researchers to create a sense of community, developing a workflow where a database manager leads the input of datasets, reducing contributor effort, and contacting researchers and repositories with known large datasets. In particular, AusTraits makes use of expertise held within the systematics community, mining data embedded in taxonomic descriptions. This approach provides depth of coverage for key traits such as growth, leaf dimensions, and plant height aiding in the filling of regional gaps for modeling and conservation management. Due to socioeconomic limitations, regional efforts in some areas (particularly in the Global South) may only be feasible with South–South and South–North collaborative efforts. Data source integration is currently hindered by variation in data structure, data availability, taxonomy, and even trait names (Gallagher et al., 2020b), so new, regional efforts would benefit from adopting existing tools, taxonomies, and data structures (e.g. Boyle et al., 2013; Schneider et al., 2019; Falster et al., 2021).

In this study, we quantified trait data completeness relative to a subset of traits provided in the TRY database, but there are an infinite number of possible traits: Measurements can be taken at any point, organ, or developmental stage on an individual, at multiple levels of organization, and combined in any way (e.g. dry leaf mass per leaf area and above ground biomass divided by belowground biomass). A lack of standard trait definitions makes the integration of different databases challenging (Garnier et al., 2016). Thankfully, plant traits are often strongly correlated (Westoby et al., 2002; Wright et al., 2004; Díaz et al., 2016; Zeballos et al., 2017), and even sparse trait coverage may allow us to say something about the overall phenotypes of species (Mouillot et al., 2021), particularly when combined with phylogenetic information (Penone et al., 2014) or geographic information (Sandel et al., 2021). Imputation methods based on trait and phylogenetic correlations also provide estimates of uncertainty which can be used in sampling prioritization. Thus, global collection efforts focused on key traits representing known trait spectra (e.g. Westoby et al., 2002; Wright et al., 2004; Díaz et al., 2016; Zeballos et al., 2017), particularly in taxa or regions of high uncertainty, may be a reasonable path forward. For such large-scale efforts, traits that may be especially important to focus on are those that are: broadly relevant and apply to most or all plant species; of importance for many ecological processes; and fast and affordable to measure. For example, plant height, leaf mass per area, and leaf dry matter content are all strong candidates. However, we also acknowledge that the choice of which traits to measure is ultimately driven by the research question, and a diversity of research questions necessitates a diversity of traits.

While plant traits are critically important across disciplines and are urgently needed to allow us to predict responses to global change, the current state of our knowledge is both incomplete and geographically biased. Given the current state of available data, large scale (e.g. country, biome, and global) analyses of plant traits must be interpreted with caution and should attempt to quantify uncertainty caused by these massive data gaps. Moving forward, researchers can help by publishing their data and metadata openly, including the full set of raw and derived trait data (Keller et al., 2023). At larger scales, efforts to mobilize and integrate existing datasets, particularly those focused on particular geographic regions (e.g. Tavşanoğlu & Pausas, 2018; Falster et al., 2021; Báez et al., 2022) or types of traits (e.g. Iversen et al., 2017; LeBauer et al., 2018; Guerrero-Ramírez et al., 2021), hold promise to rapidly advance the state of our knowledge (Feng et al., 2022). However, what is ultimately needed is more data collection and research, particularly in the Global South, which requires funding, infrastructure, and capacity building.

None declared.

BSM and WLE conceived the initial idea. BSM, WLE, RG, J-CS and MT planned and designed the research. RG, J-CS, MT, EHW and WLE provided data. BSM conducted analyses with feedback from all authors. BSM led the writing with contributions from all authors.