Being John Harper: Using evolutionary ideas to improve understanding of global patterns in plant traits

2.1 Seed mass

Globally, seed mass varies over more than ten orders of magnitude (Harper et al., 1970). The dust-like seeds of orchids and some hemi-parasitic plants weigh less than a millionth of a gram and do not contain sufficient resources to support a seedling without external assistance (e.g. from mycorrhizae, Harper et al., 1970). At the other end of the spectrum is the double coconut (Lodoicea maldivica), whose spectacular seeds can weigh 27 kg.

Seed size has known effects on almost all aspects of plant regeneration, including seed production (smaller-seeded species make more seeds per unit canopy area per year; Jakobsson & Eriksson, 2000; Shipley & Dion, 1992; and per unit total above-ground plant dry mass per lifetime; Aarssen & Jordan, 2001), dormancy (bigger seeds are less likely to have dormancy; Rubio de Casas et al., 2017), seed survival in the soil (smaller seeds have higher survival in the soil; Thompson, Band, & Hodgson, 1993), seedling size (large seeds produce larger seedlings; Jurado & Westoby, 1992) and seedling survival (seedlings from larger seeds have higher rates of survival; Bergholz et al., 2015; Grime & Jeffrey, 1965; Lebrija-Trejos et al., 2016; Leishman & Westoby, 1994; Leishman, Wright, Moles, & Westoby, 2000; Metz et al., 2010; Walters & Reich, 2000; Westoby, Leishman, & Lord, 1996). In many cases, increasing the amount of data available has simply confirmed existing ideas about the ecology of seed size. However, there have been some surprises. I highlight two of these surprises below: the relationship between seed size and seed dispersal distance and the relationship between seed size and seed predation.

Until recently, it was widely believed that smaller seeds achieved greater seed dispersal distances (Greene & Johnson, 1993; Venable & Brown, 1988). Small-seeded species do have lower terminal velocities than do large-seeded species (Augspurger, 1986; Greene & Johnson, 1993) and for wind-dispersed species, lower terminal velocities are associated with greater dispersal distances (Nathan, Safriel, & Noy-Meir, 2001; Tackenberg, 2003). However, bringing together data from around the world revealed a positive correlation between seed mass and seed dispersal distance (Thomson, Moles, Auld, & Kingsford, 2011). That is, large seeds actually disperse further than do small seeds, even in wind-dispersed species! Analysis of global datasets reveals that plant height, growth form, terminal velocity and dispersal syndrome are better predictors of species’ dispersal distance than is seed mass (Tamme et al., 2014; Thomson et al., 2011). Once these factors are accounted for, there is no significant relationship between seed mass and dispersal distance (Thomson et al., 2011), and adding seed size to models that predict seed dispersal distance from dispersal syndrome and growth form has little effect on predictive power (Tamme et al., 2014). Consistent with these global findings, a detailed study of 12 wind-dispersed neotropical species found that wing loading was not significantly related to mean dispersal distance and that maximum plant height was a substantially better predictor of dispersal distance than was the rate of diaspore descent in the natural environment (Augspurger, Franson, & Cushman, 2017).

It was thought that species with larger seeds would suffer higher rates of predation because large seeds have a higher energetic reward, higher apparency and require less energy to crack per gram of nutritive tissue (Charnov, 1976; Feeny, 1976; Fricke & Wright, 2016). Much of the support for this idea came from studies of the effect of caging on seedling recruitment (which confound seed predation with survival through germination and early establishment, including exposure to herbivory) and from cafeteria trials on captive animals (which are not necessarily representative of feeding choices in natural situations). The interspecific relationship between seed size and predation rates on seeds exposed to the full suite of seed predators in natural ecosystems varies considerably across both studies and habitat types (Radtke, 2011), and most large studies find non-significant or even negative relationships between seed size and predation rate (Blate, Peart, & Leighton, 1998; Gong, Tang, & Wang, 2015; Kollmann, Coomes &, White, 1998). Recent studies offer some insights into these surprising results. First, even though animals often display preferences for particular sizes of seeds, the relationship between animal size and the size of the seeds they ingest under natural conditions is extremely weak (Chen & Moles, 2015). Second, there appears to be an upper limit to the size of seeds that a given size of seed predator can efficiently handle (Muñoz & Bonal, 2008), leading large seed predators to select large seeds and small seed predators to select smaller seeds (Abbott & Van Heurk, 1985; Honek, Martinkova, Saska, & Pekar, 2007; Rosin & Poulsen, 2016, 2017). As most ecosystems support a spectrum of seed predators ranging from tiny insects to large mammals, there is likely to be a willing predator regardless of the size of the seed.

Seed mass varies substantially across climatic gradients. Species in warmer environments tend to make larger seeds (Amatangelo, Johnson, Rogers, & Waller, 2014; Dubuis et al., 2013; Gallagher & Leishman, 2012; Malhado et al., 2015; Moles et al., 2014; Šímová et al., 2015; Simpson, Richardson, & Laughlin 2016; Swenson et al., 2012; Swenson & Weiser, 2010). Consistent with this temperature gradient, tropical seeds tend to be larger and less likely to have dormancy than seeds from higher latitudes (De Frenne et al., 2013; Gallagher & Leishman, 2012; Graae et al., 2009; Murray, Brown, Dickman, & Crowther, 2004; Pakeman et al., 2008; Rubio de Casas et al., 2017). Studies with large geographic and taxonomic scope generally find a strong positive relationship between seed mass and rainfall (Gallagher & Leishman, 2012; Moles et al., 2005, 2014; Pakeman et al., 2008; Sandel et al., 2010; Swenson et al., 2012; Swenson & Weiser, 2010). However, this pattern is not always detected at smaller scales, where some studies report negative (Dubuis et al., 2013; Simpson et al., 2016; Stromberg & Boudell, 2013) or null relationships between seed mass and moisture (Cornwell & Ackerly, 2009; Costa-Saura, Martínez-Vilalta, Trabucco, Spano, & Mereu, 2016; Malhado et al., 2015; Šímová et al., 2015). Seasonality in both rainfall and temperature is associated with lower seed mass (Malhado et al., 2015; Swenson et al., 2012; Swenson & Weiser, 2010; but see Šímová et al., 2015).

2.2 Plant height

Plants range in height from a single millimetre (e.g. Wolffia arrhiza and Azolla microphylla; Díaz et al., 2016) to the world's tallest angiosperm, Eucalyptus regnans (greatest verified height 99.8 m, with a plausible historic record of 114 m; Koch, Sillett, Antoine, & Williams, 2015; Mifsud, 2003) and the world's tallest living tree, a Sequoia sempervirens (115.6 m; Sillett et al., 2010). There is a long history of debate over the relative importance of factors including hydraulic limitation, mechanical stress and carbon balance in limiting the maximum height of plants (Becker, Meinzer, & Wullschleger, 2000; Givnish, Wong, Stuart-Williams, Holloway-Phillips, & Farquhar, 2014; Niklas, 1994; Ryan & Yoder, 1997). Current consensus tends towards a synthesis of these ideas, in which hydraulic constraints reduce photosynthesis in taller plants, thus limiting height growth (Ambrose et al., 2010; Koch, Sillett, Antoine, & Williams, 2015; Koch, Sillett, Jennings, & Davis, 2004).

Plant height is a key component of a species’ ecological strategy. First, plant size affects a plant's ability to access resources. Taller plant species have better access to light (Falster & Westoby, 2003; Poorter, Bongers, Sterck, & Woll, 2005) and tend to have larger canopies and bigger leaves (Cornelissen, 1999; Díaz et al., 2016; Kleiman & Aarssen, 2007). Taller plant species also tend to have deeper and longer root systems, which allow them greater access to water and nutrients in the soil (Li & Bao, 2015; Padilla & Pugnaire, 2007; Sperry, Hacke, Oren, & Comstock, 2002). Second, plant growth changes with height. The rate (mg/year) of carbon accumulation of individual plants generally increases with increasing tree size (Stephenson et al., 2014). However, the cost of maintaining non-photosynthetic support tissues increases with height (King, 1990). Thus, the rate of carbon accumulation per unit leaf area or per unit plant mass tends to decrease with increasing height (Metcalf, Rose, & Rees, 2003; Stephenson et al., 2014). Consistent with these findings, plant metabolic rates scale with mass^0.75 (Niklas & Enquist, 2001). That is, larger plant species tend to have higher absolute metabolic rates, but lower metabolic rates per unit mass. Plant height is negatively associated with species’ ability to respond to disturbance (Kühner & Kleyer, 2008; Lavorel, McIntyre, & Grigulis, 1999), and generally increases along successional gradients (Tilman, 1987), although recent data suggest that plant height has no significant correlation with a species’ competitive effect on neighbouring plants (Kunstler et al., 2016). Taller plant species tend to have greater seed dispersal distances (Tamme et al., 2014; Thomson et al., 2011) and lower rates of molecular evolution than do short plant species (Lanfear et al., 2013).

One of the most interesting recent developments in our understanding of plant size is an appreciation of the importance of the small, suppressed individuals in plant communities. In the past, it was thought that large plants were responsible for most seed production and biomass accumulation (both within populations of a species and across different species in a community), because large individuals produce more seeds and more biomass per plant than do small individuals. However, the sheer numerical dominance of small individuals both within and across species (Chambers & Aarssen, 2008; Tracey & Aarssen, 2014) means that small individuals actually produce the majority of the biomass (Tracey, Irwin, Mcdonald, & Aarssen, 2017) and seeds (Chambers & Aarssen, 2008). This body of work overturns the traditional size-advantage hypothesis (Aarssen, 2015) and provides a nice example of the insights that can be gleaned by combining trait ecology with a more demographic approach in which the fitness of all individuals is considered rather than selectively focussing on a few “representative” mature plants.

The height of plant species is correlated with the abiotic conditions under which they grow. Plant height increases with increasing mean annual precipitation (de la Riva et al., 2017; Fonseca, Overton, Collins, & Westoby, 2000; Moles et al., 2014, 2009; Šímová et al., 2015; Swenson et al., 2012; Swenson & Weiser 2010). Similar patterns are found with other indices of moisture availability (Givnish et al., 2014; Ordonez et al., 2009; Zhang, Nielsen, Mao, Chen, & Svenning, 2016), although this pattern is not always seen at smaller spatial scales (Amatangelo et al., 2014; Cornwell & Ackerly, 2009; Guittar, Goldberg, Klanderud, Telford, & Vandvik, 2016; Muscarella et al., 2016). Plant height tends to decrease with increasing rainfall seasonality, but the strength of this correlation is very low (R² = .05–.06; Moles et al., 2009; Swenson et al., 2012; Swenson & Weiser 2010).

Plant height increases with mean annual temperature and decreases with temperature seasonality (Amatangelo et al., 2014; Dubuis et al., 2013; Guittar et al., 2016; Hulshof, Swenson, & Weiser 2015; Moles et al., 2014, 2009; Šímová et al., 2015; Swenson & Weiser 2010). Consistent with this temperature gradient, plant height decreases towards the poles at the cross-species and community scales (Moles et al., 2009; Zhang et al., 2016), but not at the within-species level (De Frenne et al., 2013). There are exceptions to the positive relationship between plant height and temperature, but most of these exceptions are underpinned by substantial differences in water availability (such as shifts from short-statured dry forest to taller pine/oak forest up elevational gradients in dry areas, e.g. Hernández-Calderón, Méndez-Alonzo, Martínez-Cruz, González-Rodríguez, & Oyama, 2014). Interestingly, although there has been much interest in the importance of freezing temperatures in shaping plant strategies (Zanne et al., 2014), there is little difference in the predictive power of minimum vs. mean temperature in large-scale patterns in plant height (Moles et al., 2009; Swenson & Weiser, 2010).

2.3 Wood density

Wood density (dry mass divided by fresh volume, Preston, Cornwell, & DeNoyer, 2006) varies from 0.08 g/cm³ in Homalium celebicum and 0.1 g/cm³ in Ochroma pyramidale (balsa), to woods that sink in water, such as Krugiodendron ferreum (leadwood; 1.35 g/cm³) and Caesalpinia sclerocarpa (1.39 g/cm³; Zanne et al., 2009). Species with higher wood densities tend to have lower mortality rates and lower growth rates (Hietz, Rosner, Hietz-Seifert, & Wright, 2017; Kraft, Metz, Condit, & Chave, 2010; Kunstler et al., 2016; Poorter, 2008; Poorter et al., 2008), but higher ability to tolerate competition (Kunstler et al., 2016). Consistent with these findings, later successional, shade-tolerant species tend to have higher wood density (Falster & Westoby, 2005; Poorter et al., 2010).

Species with high wood density tend to have high mechanical strength (Chave et al., 2009; van Gelder, Poorter, & Sterck, 2006). High wood density also tends to be associated with a higher density of smaller vessels (Hietz et al., 2017; Preston et al., 2006) and a lower risk of cavitation (Chave et al., 2009; Reich, 2014). However, evidence for a relationship between wood density and hydraulic conductivity has been mixed (Chave et al., 2009; Poorter et al., 2010; Zhang, Cao, Fan, Zhang, 2013). One possibility is that this relationship is obscured by differences in traits such as plant size (Hietz et al., 2017). Another possibility is that the relationship is masked by the considerable variation in wood composition between species (Ziemińska, Westoby, & Wright, 2015) and the fact that the different functional properties of wood depend on different stem traits. A stem's mechanical strength is largely determined by fibre cell walls (Chave et al., 2009; van Gelder et al., 2006), while hydraulic conductivity depends on the size and density of its vessels (Hietz et al., 2017). In short, it is not clear whether wood density alone is sufficient to capture the trade-offs associated with the competing demands of strength, storage and hydraulic conductivity.

Attempts to integrate wood density into broader strategy schemes (e.g. Reich, 2014) have been hindered by surprisingly weak evidence for correlations between wood density and other plant traits. Several studies have found negative relationships between wood density and plant height (Hietz et al., 2017; Poorter et al., 2010; Preston et al., 2006), but others have found no significant correlation (Poorter, 2008; Wright et al., 2007) or mixed results (Falster & Westoby, 2005). Díaz et al. (2016) brought together data for stem specific density for over eleven thousand species from around the world. Stem density was positively correlated with height across the full dataset, but this relationship was driven by the difference between woody and herbaceous species; relationships within woody (slope = −0.34, R² = .01) and herbaceous (slope = 0.37, R² = .002) species were extremely weak and ran in opposite directions. Similarly, while there is some evidence that wood density is negatively related to leaf size (Tozer, Rice, & Westoby, 2015; Wright et al., 2007), other studies show no significant relationship (Ackerly, 2004; Baraloto et al., 2010). Díaz et al. (2016) revealed a positive relationship between leaf area and stem density across their full dataset, but weak negative relationships within both woody and herbaceous species (both R² = .03). Some studies have found positive correlations between wood density and leaf mass per unit area (Bucci et al., 2004; Pivovaroff, Sack, & Santiago, 2014), but others show no significant relationship (Ackerly, 2004; Baraloto et al., 2010; Wright et al., 2007). Díaz et al. (2016) found a significant positive correlation between leaf mass per unit area and stem density across a global dataset including 5,381 species (slope = 0.99, R² = .36). However, this relationship was largely driven by the difference between woody and herbaceous species, and the strength of the relationship across the 2,870 woody species was low (R² = .06). Overall, the lack of consistent, strong relationships between wood density and other functional traits is counter to the idea that wood density is part of a fast–slow plant economic spectrum (Reich, 2014) and suggests we still have much to learn about stem traits and the way they relate to plant ecological strategy.

Plants growing in warm areas tend to have higher wood density than do plants growing in cooler regions (Šímová et al., 2015; Swenson et al., 2012; Swenson & Weiser, 2010; Wiemann & Williamson, 2002; Zhang, Slik, Zhang, Cao, 2011, but see Zhang et al., 2013). Consistent with this temperature gradient, wood density is higher at lower latitudes (Swenson et al., 2012; Swenson & Enquist, 2007; Wiemann & Williamson, 2002; Zhang et al., 2011). There is also some evidence that wood density is negatively related to temperature variation (Swenson & Enquist, 2007; Swenson et al., 2012; but see Šímová et al., 2015). The largest study of relationships between climatic variables and wood density to date spanned 4,667 seed plant species from North and South America (Swenson & Enquist, 2007). In this study, maximum temperature was the strongest correlate of wood density, but the R² of this relationship was only .06.

Evidence for a relationship between wood density and rainfall or moisture availability is mixed. Some studies have found that wood density is highest in more arid regions (Cornwell & Ackerly 2009; Pfautsch et al., 2016; Preston et al., 2006; Swenson & Enquist, 2007), and species with lower wood density have been found to be more susceptible to drought-induced mortality (Greenwood et al., 2017). However, other studies have found the highest wood densities in areas with high rainfall or moisture availability (Ordonez et al., 2009; Swenson et al., 2012; Swenson & Weiser, 2010; Zhang et al., 2011). Non-significant results are also common (Costa-Saura et al., 2016; de la Riva et al., 2017; Šímová et al., 2015; Wiemann & Williamson, 2002; Zhang et al., 2013). There is also no consistent relationship between rainfall seasonality and wood density (Šímová et al., 2015; Swenson et al., 2012; Swenson & Weiser, 2010). However, most of the existing studies of biogeographic patterns in wood density are relatively small in taxonomic and/or geographic scope, so the field is ripe for a global synthesis.

2.4 Leaf mass per unit area and the leaf economics spectrum

Leaf mass per unit area (LMA, the inverse of specific leaf area, SLA) is a measure of the amount of biomass invested in building a given area of leaf. The leaf economic spectrum neatly captures the trade-offs, costs and benefits associated with building leaves with high vs. low LMA. At one end of the spectrum, we have species like Oxalis stricta (LMA = 17.3 g/m²; Shipley, 1995) or Lupinus perennis (LMA = 65 g/m²; Wright et al., 2004) that have thin, flimsy leaves. Low LMA leaves generally have a high nitrogen content (Hikosaka & Shigeno, 2009; Reich, Walters, & Ellsworth, 1992; Wright et al., 2004), and about half of leaf nitrogen is allocated to machinery for photosynthesis (including RuBisCO; Hikosaka & Shigeno, 2009). Greater nitrogen content, combined with high gas exchange rates (Reich et al., 1999) through the low density and/or thin leaf tissue, allows low LMA species to achieve an increased rate of photosynthesis (Hikosaka & Shigeno, 2009; Reich et al., 1999, 1992; Wright et al., 2004). The major downside of low LMA leaves is that they tend to have short leaf life spans (e.g. the leaf life span of L. perennis is only 29 days; Wright et al., 2004), which results in part because their low leaf toughness renders them vulnerable to mechanical damage and herbivory (Coley, 1988; Onoda et al., 2011; Reich, Uhl, Walters, & Ellsworth, 1991; Wright & Cannon, 2001). At the high LMA end of the spectrum are species like Hakea recurva subsp. recurva (LMA = 1,509 g/m²; Wright et al., 2004) or Picea mariana (LMA = 390 g/m²; Chapin, Bretharte, Hobbie, & Zhong, 1996), which have thick and/or dense leaf tissue. High LMA leaves have low nitrogen content and photosynthesize slowly (Reich et al., 1992; Wright et al., 2004). The upside is that they have very long leaf life spans (e.g. P. mariana leaves can live for 20 years; Chapin et al., 1996). In summary, the leaf economic spectrum runs from low LMA leaves that are cheap to build and yield a high photosynthetic income for a short time, to high LMA leaves that require a higher initial investment of biomass and return a slow, steady photosynthetic revenue over many years.

The success of the leaf economics spectrum has stimulated ecologists to bring together leaf economic traits with wood and root traits (Baraloto et al., 2010; de la Riva et al., 2016; Freschet, Cornelissen, Van Logtestijn, & Aerts, 2010; Laughlin, Leppert, Moore, & Sieg, 2010; Reich, 2014) and to quantify the way that plant traits fill multidimensional space (de la Riva et al., 2016; Díaz et al., 2016). However, current evidence suggests that LMA is not strongly or consistently correlated with other traits that reflect plant strategy, such as seed mass, plant height or wood density (Baraloto et al., 2010; Díaz et al., 2016; Niinemets & Kull, 1994; Wright et al., 2007).

High LMA, with associated low rates of leaf turnover, is thought to be particularly beneficial in low-nitrogen environments, because plants are only able to resorb about half of the nitrogen invested in leaves during senescence (Aerts, 1996; Wright et al., 2004). Consistent with this, many species with very high LMA are from regions with low soil fertility such as Australia, South Africa and tropical rainforests (Wright et al., 2004). LMA was negatively related to soil N and P in the Netherlands (Ordonez et al., 2009) and in Australian woodlands (Dwyer, Hobbs, & Mayfield, 2014). Leaf mass per unit area was also negatively correlated with soil P in New Zealand forests (Simpson et al., 2016) and with an index of soil fertility in Wisconsin forests (Amatangelo et al., 2014). However, negative relationships between soil fertility and LMA are not universal. Cornwell and Ackerly (2009) found no significant relationship between LMA and soil available nitrate and a positive relationship between LMA and soil available ammonium across 54 species in California. LMA was negatively related to soil total nitrogen (R² = .13) and soil organic carbon (R² = .21) along an 1,100 km transect across the Mongolian Plateau, but not significantly related to soil total P, available N or available P (Liu et al., 2012). A global study spanning 400 species across 99 sites found no significant relationship between LMA and soil N or P (Ordoñez et al., 2009). The largest study to date included 1,509 species from 288 sites worldwide (Maire et al., 2015) and found that soil organic carbon (R² = .04) and total N (R² = .01) were very weakly correlated with LMA, while available P was not significantly related to LMA. In short, the evidence for higher LMA in lower fertility sites is not nearly as clear-cut as the theory predicted.

Leaf mass per unit area is not strongly related to precipitation. Some studies have found higher LMA at sites with lower precipitation (Liu et al., 2012; Šímová et al., 2015; Swenson et al., 2012), while others have found the reverse (Gallagher & Leishman, 2012; Santiago et al., 2004). The most common result is for no significant relationship between LMA and precipitation (Dwyer et al., 2014; Hernández-Calderón et al., 2014; Onoda et al., 2011; Ordoñez et al., 2009), results that vary through space and time (Amatangelo et al., 2014; Muscarella et al., 2016) or for significant results with R² values so low as to have limited biological relevance. The two biggest studies to date (spanning 2,370 and 10,792 species, respectively) both found extremely weak (R² of .002 and .001) negative relationships between LMA and mean annual precipitation (Moles et al., 2014; Wright et al., 2005). Precipitation seasonality also has little effect on LMA (Šímová et al., 2015; Swenson et al., 2012).

Some studies have found that variables such as vapour pressure deficit or potential evapotranspiration are more closely related to LMA than is mean annual precipitation (both R² = .14 across 2,370 species worldwide, cf R² of .002 for precipitation, Wright et al., 2005). Several studies using indices of aridity or water availability have reported higher LMA at more arid sites (Bertiller et al., 2006; Cornwell & Ackerly, 2009; Costa-Saura et al., 2016; Wright et al., 2005), and species with lower LMA are more susceptible to drought-induced mortality (Greenwood et al., 2017). However, non-significant relationships between LMA and measures of aridity are also common (Maire et al., 2015; Moreno, Bertiller, & Carrera, 2010; Ordonez et al., 2009; Ordoñez et al., 2009).

Evidence for a relationship between LMA and mean annual temperature is mixed. Many studies report declining LMA with increasing temperature (Amatangelo et al., 2014; Dubuis et al., 2013; Guittar et al., 2016; Hernández-Calderón et al., 2014; Read et al., 2014; Simpson et al., 2016; Swenson et al., 2012). However, null results are also frequent (Liu et al., 2012; Ordoñez et al., 2009; Šímová et al., 2015), and some of the largest studies have found positive correlations between LMA and mean annual temperature, including a study spanning 1,092 climbing plant species from 34 countries (R² = .12; Gallagher & Leishman, 2012), a study of 2,370 species from 163 sites worldwide (R² = .10; Wright et al., 2005) and a global study of 10,792 species (R² = .06; Moles et al., 2014). Not only is the relationship between temperature and LMA not consistent across studies, it also differs across functional guilds. Evergreen species have lower LMA and shorter leaf life spans at higher temperatures, while deciduous species have higher LMA and longer leaf life spans at higher temperatures (Kikuzawa et al., 2013; Wright et al., 2005). Evidence for a latitudinal gradient in LMA is also mixed (Gallagher & Leishman, 2012; Swenson et al., 2012).

In summary, although the leaf economics spectrum has been a highly successful conceptual advance in trait ecology, the relationships between leaf economic traits and both other traits and abiotic factors are either extremely weak or not consistent across studies.

2.5 The role of soil fertility in shaping plant ecological strategy

None of the plant traits reviewed above are consistently related to soil fertility (as indicated by measures of nitrogen, phosphorus and organic matter). Some studies have found positive relationships between soil fertility and plant height (Amatangelo et al., 2014; Ordonez et al., 2009), but others find the reverse (Simpson et al., 2016). Evidence for a relationship between soil fertility and seed mass is also mixed, with studies reporting positive (Amatangelo et al., 2014; Cornwell & Ackerly, 2009; Simpson et al., 2016) and negative (Dainese & Sitzia, 2013; ter Steege et al., 2006) relationships. Some studies have found larger leaves at sites with higher fertility (Ordonez et al., 2009; Tozer et al., 2015), while others find smaller leaves at higher fertility sites (Cornwell & Ackerly, 2009). Leaf mass per unit area is sometimes positively related to soil fertility, sometimes negatively related to soil fertility (Amatangelo et al., 2014; Dwyer et al., 2014; Ordonez et al., 2009; Simpson et al., 2016), and some studies yield null or mixed results, or R² values so low as to be ecologically unimportant (Cornwell & Ackerly 2009; Liu et al., 2012; Maire et al., 2015; Ordoñez et al., 2009). Finally, some studies have found that areas with poorer soils have species with higher wood density (Chave et al., 2009; ter Steege et al., 2006), but positive or mixed results are also common (Cornwell & Ackerly, 2009; Ordonez et al., 2009). We know that soil fertility has enormous impacts on the growth of individual plants and crops, and there are many reasons to expect plant traits to change along fertility gradients. So, why are large-scale patterns in the relationships between traits and soil fertility so elusive?

One possible explanation for the absence of consistent relationships between soil fertility and plant traits is that differences in soil pH might confound the results. Soil pH has important effects on soil nutrient availability (Lucas & Davis, 1961; Maire et al., 2015), and soil pH was the strongest correlate of LMA, leaf N and photosynthetic rate in a global study of the effects of soil fertility on leaf traits (Maire et al., 2015). Similarly, soil pH was a better predictor of seed mass, specific root length, root tissue density, bark thickness, leaf dry matter content, plant height and LMA in New Zealand forests than was soil P (Simpson et al., 2016). However, the direction of the relationship between soil pH and plant traits varies across studies (Dwyer et al., 2014; Maire et al., 2015; Simpson et al., 2016).

Factors such as disturbance might also obscure plant responses to large-scale gradients in soil fertility. However, most such confounding factors surely also apply to large-scale gradients in temperature, precipitation and latitude, and we have little difficulty in retrieving trait relationships along these gradients. It is also possible that the lack of consistent relationships in the current literature might result from soil fertility being of secondary importance to large-scale gradients in climatic factors.

Despite the present lack of clarity around soil fertility, there is reason for optimism. Global datasets, such as the Harmonized World Soil Database, have improved dramatically in the last few years, and there is ever-increasing scope for rapid advances in our understanding of the ways that soil fertility shapes plant ecological strategy. Global databases containing relevant measures of soil fertility have been more difficult, and thus slower to develop than have datasets of climate variables. Part of the problem is the lack of an agreed set of analytical methods for quantifying soil fertility. There is also surprisingly little international agreement as to which soil variables are the most important to plants. For example, in Australia, dialogue about the importance of soil fertility focuses on nitrogen and phosphorus. By contrast, researchers in more fertile countries such as Britain tend to focus on soil pH. One interesting direction for the future will be to determine the extent to which these differing local foci reflect differences in the most important drivers of the region's ecology and to what extent the local foci simply reflect traditions formed before the availability of quantitative evidence.

2.6 Some statistical considerations

The massive datasets and wide array of environmental variables available to large-scale ecologists are a major source of strength in our field—but these factors can also lead to problems in interpreting statistical significance. I highlight two issues here. First, it is becoming increasingly common for ecologists to simply include as many abiotic factors as they can in their manuscripts. Papers using this “fishing expedition” approach run a high risk of yielding spurious results. In fact, using an alpha of 0.05, an ecologist who runs analyses of their focal trait against all 19 WorldClim variables has a 62% chance of recording at least one significant result, even in the absence of any real pattern. Second, statistical significance becomes problematic when we analyse global datasets including thousands of replicates. The problem here is the reverse of the power issues common in fields in which obtaining individual datapoints is very difficult, such as behavioural ecology. With the immense statistical power in many global-scale analyses, almost any effect comes out as statistically significant. An example from my own (currently unpublished) work combines 95,272 records for the presence of spinescence (from the TRY database) with data for rainfall in the driest month. GLMM yields a p-value of .000002 and an R² of .0002. That is, this highly statistically significant relationship has so little explanatory power that it is ecologically meaningless. The effects of vast power are not a problem as long as we focus on the magnitude of effects, rather than p-values alone. However, slopes and R² values can also be tricky to interpret.

Global-scale analyses at the interspecific scale seldom explain a high proportion of the variation in plant traits (R² values are almost always below .6). This low predictive power is sometimes taken as an indication that there is so much noise in global-scale studies that they are not worthwhile. However, an understanding of where the variance lies can quickly put the R² values of global relationships in perspective. Plant traits such as seed mass, plant height and leaf size typically vary over four to six orders of magnitude among the coexisting species at a single site (Leishman et al., 2000; Westoby et al., 2002). For most plant traits, a third to half of the global variation is at the within-site level (Moles et al., 2014). This necessarily puts an upper limit (of half to two-thirds, depending on the trait) of the variation that could possibly be explained by broad environmental factors such as latitude, temperature and precipitation, or even combinations of these factors. One way macro-ecologists could better explain this situation to our readers would be to routinely report the proportion of the variance that lies within vs. between sites and express the variance explained by broad environmental gradients as a proportion of the cross-site variation.

Finally, I suggest we take advantage of the advanced statistics and enormous power of global datasets to move away from the implicit assumption plants respond in a linear manner to climatic gradients. The vast majority of analyses in large-scale ecology currently fit straight lines to either log-transformed or raw trait data. However, many of the variables that underpin global patterns in plant traits could generate nonlinear trends. For instance, there could be low temperature thresholds beyond which cells experience freeze damage, or high temperature thresholds beyond which proteins begin to denature. The alternative stable states literature also shows that plant communities sometimes display abrupt state shifts in response to gradual changes in climate (Scheffer et al., 2001; Schröder, Persson, & De Roos, 2005). Determining whether plant responses to climate change are likely to be gradual or abrupt is a critical topic for future research.

3.1 The life-history spectrum: Bringing seed mass and plant height together via fitness

A species’ seed mass has historically been understood as a trade-off between producing many small seeds, each with a low chance of survival, vs. producing fewer larger seeds, each with a higher probability of survival (Geritz, van der Meijden, & Metz, 1999; Smith & Fretwell, 1974). There is excellent evidence for a trade-off between the size of seeds and the number of seeds that a plant can make for a given amount of energy (Aarssen & Jordan, 2001; Jakobsson & Eriksson, 2000; Shipley & Dion, 1992). There is also very strong evidence for the survival advantages associated with large seeds, particularly under stresses such as low light and competition (Bergholz et al., 2015; Grime & Jeffrey, 1965; Lebrija-Trejos et al., 2016; Leishman & Westoby, 1994; Leishman et al., 2000; Walters & Reich, 2000; Westoby et al., 1996). However, empirical data show that the survival advantage of large-seeded species is nowhere near large enough to counterbalance the advantage of small-seeded species during seed production (Moles & Westoby, 2004). That is, the trade-off between many small seeds each with low survival vs. few large seeds with higher survival is only part of picture. To fully understand the advantages of producing large vs. small seeds, we need to know the lifetime fitness associated with each strategy (Equation 1).

(1)

First, consider lifetime seed production per adult plant. We know that there is a negative isometric relationship (slope = −1) between seed mass and the number of seeds produced per unit of plant canopy area per year (Henery & Westoby, 2001; Figure 1a). However, species mean seed mass is positively related to maximum plant size (Díaz et al., 2016; Moles et al., 2004; Figure 1b). That is, large-seeded species have a greater area of canopy with which to produce seeds, which helps to offset the seed production advantage of small-seeded species. Accordingly, the slope of the relationship between seed mass and the number of seeds produced per individual plant per year is substantially shallower (slope = −0.3; Moles et al., 2004; Figure 1d) than the relationship between seed mass and the number of seeds produced per unit area of canopy per year. All else being equal, this would still give small-seeded species a substantial advantage during seed production. However, species with larger mean seed mass also tend to have longer maximum reproductive life spans (Moles et al., 2004; Figure 1e). That is, large-seeded species tend to produce seeds for many more years than do small-seeded species, which further offsets the seed production advantage of small-seeded species (Moles et al., 2004). In fact, the slope of the relationship between seed size and lifetime seed production is not significantly different to zero (Moles et al., 2004; Figure 1g). That is, even though small-seeded species can produce more seeds for a given amount of energy, their smaller size and shorter reproductive life span means that small-seeded species do not have a higher lifetime seed output than do large-seeded species.

Next, consider the survival advantage of large-seeded species. Both experiments and field studies show that large-seeded species have lower rates of early mortality than do small-seeded species (Dalling & Hubbell, 2002; Eriksson, 1997; Moles & Westoby, 2004; Westoby et al., 1996; Figure 1c). However, seed reserves are finite (White & Veneklaas, 2012), and the advantages of large seeds are transient. Later-stage seedlings from large-seeded species and small-seeded species experience similar levels of mortality (Dalling & Hubbell, 2002; Saverimuttu & Westoby, 1996; Verdú & Traveset, 2005). The early survival advantage of large-seeded species would carry through to a higher number surviving to adulthood, if all else were equal (Figure 1f, T1). However, large-seeded species tend to take longer to reach maturity (Moles et al., 2004; Figure 1h). That is, the seedlings of large-seeded species are exposed to a longer period of juvenile mortality than are the seedlings of small-seeded species, and thus might have similar levels of survival to adulthood as do small-seeded species, despite their higher early survival (Figure 1f). Measuring survival from seed production to adulthood is a non-trivial exercise, and existing data are too scarce to reach firm conclusions about the relationship between seed mass and survival from seed production to adulthood (Moles & Westoby, 2006). However, the data we do have suggest that seedlings from small-seeded species are at least as likely to survive to adulthood as are seedlings from large-seeded species (Moles & Westoby, 2004; Figure 1i).

Bringing together information about the correlations between seed size, plant size and plant longevity reveals a life-history spectrum in which seed mass and plant size act in opposing directions on both lifetime seed production and seed survival to adulthood. The upshot is a spectrum from “fast” to “slow” plant life-history strategies (Figure 2). At the “fast” end of the life-history spectrum, we have species such as Capsella bursa-pastoris, in which small seeds are produced on small plants that have high early mortality rates but reach reproductive maturity quickly and have a relatively short reproductive life span. At the “slow” end of the life-history spectrum are species such as Quercus robur—species with large canopies and long reproductive life spans which produce large seeds with relatively low rates of seedling mortality, but which take many years to reach reproductive maturity.

The life-history spectrum presented here is not consistent with the traditional understanding of species’ seed mass as the result of a trade-off between producing many poorly provisioned seeds each with a low survival probability vs. producing fewer larger seeds each with a higher probability of survival. It is also counter to the original formulation of Westoby's (1998) leaf–height–seed scheme, in which seed mass and plant size were thought to be relatively independent traits that reflected different aspects of plant strategy. However, the life-history spectrum emerged as the first principal component in two global quantifications of multivariate relationships between six major plant traits (Díaz et al., 2016; Sandel, Monnet, & Vorontsova, 2016). It has strong parallels with Pianka's (1970) r–K selection theory, but because it is not based on density dependence, the life-history spectrum is better able to explain the existence of many small-seeded, short-statured species that did not evolve in low-competition environments (Bonser, 2013). The life-history spectrum also complements and extends existing work such as Franco and Silvertown's (1996) fast–slow continuum, Reich's (2014) fast–slow plant economics spectrum, Grime's C-S-R triangle (Grime, 1974, 1977) and the pace-of-life syndrome (Ricklefs & Wikelski, 2002).

Considering trait effects on a species’ lifetime fitness gives a much more accurate indication of the overall advantages and disadvantages of each strategy than does the common practice of considering trait effects on one function one at a time. Applying this approach to wood density, LMA and leaf size seems likely to yield interesting results. Another promising direction for future studies would be to assess the fitness implications of different trait strategies in relation to climate variables.

3.2 Good times and bad times: Understanding the different mechanisms through which climate can influence plant fitness

Extreme climatic events can have profound impacts on plant populations (Bastos et al., 2014; Gutschick & BassiriRad, 2003; Jentsch & Beierkuhnlein, 2008; Saccone et al., 2009). However, there is a growing body of evidence suggesting that some plant traits are more closely correlated with conditions during the good parts of the growing season than with conditions during the harshest parts of the year. For example, a study of Australian species found the strongest associations between leaf width and precipitation in the wettest parts of the year (e.g. precipitation of the wettest quarter, R² = .12) and non-significant relationships between leaf width and precipitation in the harshest parts of the year (Tozer et al., 2015). Similarly, precipitation in the wettest month was the best predictor (R² = .26) of plant height across a global dataset including 5,784 species, while the maximum temperature in the warmest month and precipitation in the driest month explained almost no variation in plant height (Moles et al., 2009). The timing of flowering and leaf unfurling also seem to respond more to mean conditions than to extremes (reviewed in Reyer et al., 2013).

Climate can interact with plant traits to affect two critical components of fitness: survival and offspring production. Extreme events can cause differential mortality among individuals according to their trait values, potentially leading to truncation of trait distributions. Examples of this include the disproportionate mortality of small swallows during a cold snap in Nebraska (Brown & Brown, 1998) and of plant species with thin leaves during a heatwave in Australia (Groom et al., 2004). However, surviving is only part of the battle—fitness is also determined by an organism's ability to produce large numbers of successful offspring. The strength of the correlations between plant traits and conditions during the growing season might reflect the fitness advantage accrued by species whose traits are best suited to taking advantage of the good times.

Further quantification of which aspects of climate are most important in shaping plant traits and developing theory that incorporates the effects of both the good times and the bad are important goals for the future, particularly for ecologists who are interested in predicting the likely impact of climate change (including the forecast increase in the intensity and frequency of extreme events, Meehl & Tebaldi, 2004) on species’ traits, distributions, population persistence and community composition.

3.3 How does selection shape cross-species patterns in plant traits?

Much of our theory about cross-species patterns in plant traits is written as though selection acts on species’ mean trait values. While this has been a convenient shorthand, we know it is not true (Harper, 1967). We also know that there is substantial variation in traits within species, both across and within sites, and that this variation can have substantial impacts on population dynamics, interspecific competition and plant fitness (Bolnick et al., 2011; Messier, McGill, & Lechowicz, 2010). Cross-species studies of plant traits typically dismiss within-species variation, highlighting the small magnitude of within-species variation relative to cross-species variation. For example, while seed mass typically varies two- to fourfold within a species, this within-species variation is dwarfed by the 11 orders of magnitude variation in seed mass at the cross-species level (Leishman et al., 2000). However, this is not the case for all traits (Messier et al., 2010), and even when it is true, it leaves a disjunction between the cross-species variation being studied and the selective processes that shape plant traits.

The problematic disjunction between the level at which selection acts and our understanding of large-scale patterns in ecology is not unique to trait ecology. The 2017 Harper review outlined how population-level demography could be used to better understand ecosystem ecology (Fridley, 2017). There is also increasing evidence that the disjunction between the level at which selection acts and present trait theory is limiting our ability to understand present-day patterns in plant ecology. For example, experimental studies that change abiotic factors such as water availability or temperature often result in plant responses opposite to those found along large-scale gradients in those same abiotic factors (Sandel et al., 2010). Understanding how short-term responses interface with macroecological gradients and understanding how selection at the within-species level scales to global patterns in plant traits are pressing tasks for the future.

3.4 Using phylogenetic studies and palaeobotany to inform our understanding of present-day plant traits

Ecologists often use phylogenetic analyses to account for the fact that species are related and are therefore not independent replicates. However, the relationships between species, combined with palaeobotanical information, can tell us volumes about the times, places and circumstances under which traits evolved. That is, phylogenetic relationships between species are much more than a statistical inconvenience, and it seems tragic that the most common use of our unprecedented understanding of the relationships between species is to “control” for the effects of phylogeny rather than to use phylogenies as a source of information that can help us to understand the evolutionary history of the species in question. Phylogenetic analyses and palaeontological information can help us to understand the ways in which biotic and abiotic factors shaped plant evolution, present-day species distributions and community assemblages, and coordination between different traits (Barnosky et al., 2017; Eriksson, 2008; Wiens & Donoghue, 2004). For instance, Edwards et al. (2017) used phylogenetic analysis of Viburnum to address the origins of temperate deciduous forest, showing that being subjected to prolonged freezing has repeatedly and consistently led clades to evolve deciduous leaves. Another example is the evolution of fleshy fruit. While many ecologists assume that plants evolved fleshy fruit to facilitate seed dispersal by vertebrates, the fossil record shows that the first fleshy fruit arose several million years before the first potential vertebrate dispersers appeared on land (Tiffney, 2004), so the earliest function of fleshy fruit is much more likely to be defence against invertebrate seed predators than seed dispersal (Mack, 2000). Finally, historic information can provide critical perspective for research that attempts to predict how species and ecosystems might respond to changing climates and for decisions about how best to manage ecosystems in the face of global change (Barnosky et al., 2017). This is a field in which there is enormous scope for future progress.

3.5 Understanding the interacting suite of biotic and abiotic factors that shape plant fitness

Experiments play an absolutely critical role in elucidating the mechanisms underpinning patterns in ecology. However, the recent changes in our understanding of seed predation and seed dispersal (described in the seed mass section, above) highlight the potential pitfalls associated with extrapolating from data collected under artificial conditions and/or focussing on a single component of a complex process. Quantifying ecological outcomes in the context of the complexities of real ecosystems is one of the next major challenges for ecology.

Most ecological studies to date have focused on a single type of interaction, such as invertebrate herbivory or seed dispersal, or on relationships between one type of trait and environmental conditions. However, to fully understand the forces shaping plant ecological strategy, ecologists will need to bring together data for soil fertility, pathogenic and beneficial microbes, climate, competition, herbivores, pollinators, seed predators and seed dispersers, and to assess the implications of all these interacting factors for lifetime fitness. This presently seems like an overwhelmingly difficult task and is made prohibitive by the breadth of the literature in each of these subfields. However, we live in an era of extraordinary data availability and of unprecedented connection and collaboration between ecologists worldwide. An integrated ecology would give us a way to address some of ecology's biggest outstanding questions, such as the impact of species’ extinctions on community function, and the role of biotic interactions in maintaining species diversity.