The need for mechanistic explanations in (seed) ecology

This is the authors' abstract. We don't add key points for this paper.

‘I would rather discover one cause than gain the kingdom of Persia’. Democritus (460–370 bc) ‘Data do not understand causes and effects; humans do’. Pearl & Mackenzie (2018) The increasing availability of global-scale data on plant traits (e.g. Kattge et al., 2020), species distribution (e.g. GBIF.org), climate variables (e.g. Fick & Hijmans, 2017), sophisticated numerical methods (e.g. machine learning tools, R packages) and computing power (e.g. cloud computing) has enabled researchers to understand our biosphere in an unprecedented manner (Farley et al., 2018). However, these techno-scientific advances come with a cost. Researchers with sufficient technical skills in data management can now study global patterns and produce numerically sophisticated and apparently robust papers, without a clear hypothesis to test nor attempt to interpret any patterns from a mechanistic perspective. In addition, these broad-scale analyses tend to use the most readily available data rather than necessarily the most relevant data. This is further fuelled by the growing culture that values ‘fast’ science over research that may take years to complete (the publish-or-perish culture; Sarewitz, 2016). As a consequence, there is an increase in research based on correlating ‘everything’ to see if any patterns emerge, instead of a hypothesis-driven approach (see Calude & Longo, 2017 for examples of spurious correlations). An outcome for plant ecology is that key factors in determining plant fitness, such as fire regime, light availability, herbivory, pollinator availability and other biotic interactions, are underconsidered in broad-scale studies, as they are less available than climate information, in particular. This is exacerbated by the long-standing belief that climate is the major factor shaping ecological patterns (Pausas & Bond, 2019; Brown et al., 2023). Studying global-scale patterns also tends to hide biological mechanisms, as these act at local scales and may vary across environments; thus, broad-brush approaches may mask key local processes. In this letter, we highlight the potential for broad-scale correlative studies that ignore mechanisms to hinder progress in ecology. We specifically focus on seed dormancy, but we also provide a few other recent examples to illustrate that this is currently a general problem in ecological studies. Recent studies using thousands of species and millions of records have concluded that seed dormancy is an adaptive strategy for plants living under seasonal climates (Rubio de Casas et al., 2017; Wyse & Dickie, 2018; Zhang et al., 2022; Rosbakh et al., 2023 – all published in New Phytologist). Note here that the information available in large databases is for seed dormancy types that reflect prolonged (multi-year) dormancy (inherent or true dormancy; Baskin & Baskin, 2014) and not for temporary dormancy that may be imposed by within-year adverse environmental conditions. These studies have consistently emphasized that physical dormancy (hard seeds) is linked to strong seasonal fluctuations. This conclusion is derived from correlations between species-specific seed dormancy records from global databases and annual climatic data (precipitation and temperature seasonality from global climate maps) using species locations (from GBIF.org). However, there are problems in expecting dormancy to be selected for as an adaptive response solely to climate seasonality (i.e. predictable intra-annual variation). The question arises as to why seeds would adapt to survive for many years in the soil seed bank (seed dormancy) if there is predictable intra-annual variation? They could simply germinate once mild temperatures and soil moisture were restored in the next favourable season (environmentally induced dormancy; Pausas et al., 2022; Pausas & Lamont, 2022) – but they do not. Seed dormancy is not an adaptation to intra-annual variability but to the presence of strong and unpredictable inter-annual variability (Cohen, 1966; Philippi, 1993; Venable, 2007). In this case, dormancy functions as a bet-hedging mechanism that allows seeds to remain dormant over a number of years, reducing year-to-year variation in fitness should germination be induced every year, and instead taking advantage of exceptionally good years for seedling recruitment. Since the correlation between seed dormancy and seasonal climates has no clear causal basis, one must consider what other factors might drive this relationship. For example, many ecosystems with seasonal climates are fire-prone. Fire provides both a mechanism for dormancy release (via heat or smoke) and creates conditions (postfire) that are optimal for germination and establishment (low competition, high resource availability, minimal predation, low pathogen load). There is much experimental evidence to show that fire-type temperatures (i.e. greatly exceeding summer temperatures) increase germination of many species with physical dormancy, whereas hot summers do not (Fig. 1). Experimental evidence also shows that scarification (simulating fire heat) greatly increases germination compared with changes in season (Ruiz-Talonia et al., 2023; Supporting Information Fig. S1). Thus, fire-induced dormancy release increases plant fitness as it is uniquely synchronized with optimal germination conditions (best-bet strategy; Pausas et al., 2022). Therefore, the underlying driver of the correlation of seed dormancy with a seasonal climate may be because the vegetation in strongly seasonal climates is also the most fire-prone (Keeley et al., 2012; Lamont, 2022); thus, fire provides a more plausible mechanism to select for prolonged seed dormancy than climate seasonality. By contrast, a recent correlative study concluded that the fire regime has little to do with seed dormancy and that macroclimate variables were the most (statistically) significant (Rosbakh et al., 2023). However, this study has several conceptual drawbacks. Specifically, for the fire regime variable they considered a single parameter, potential fire season length (PFSL, defined as the average number of months with climatic conditions prone to fire, obtained from Senande-Rivera et al., 2022), a variable not considered as a standard descriptor of the fire regime as it is indirect. It is unclear which of the components of the fire regime this variable is supposed to represent; furthermore, fire regime components are not necessarily correlated, preventing use of a single descriptor (Archibald et al., 2013; Pausas, 2022). Potential fire season length provides no clues about such important components as fire frequency and season in which it occurs, nor about postfire conditions and seedling fitness. The authors do not provide a mechanism linking PFSL with dormancy release. Thus, the use of the term ‘fire regime’ in the title of Rosbakh et al. (2023) is misleading; they have not considered the range of fire variables that are known to describe the fire regime, whereas they used a wide range of variables to describe the climate regime. The analysis by Rosbakh et al. (2023) is then developed from their contention that seed dormancy is most adaptive in areas with the longest fire season (i.e. tropical ecosystems, Fig. S2). However, fires in tropical grasslands are so frequent that fire-released seed dormancy is almost redundant, whereas it is common in Mediterranean shrublands where fire is highly stochastic but guaranteed within the lifespan of the seeds (Keeley, 1991; Lamont, 2022; Pausas & Lamont, 2022). Fire type (crown or surface) and season of the fire (e.g. after the dry winter in savannas; after the wet spring in Mediterranean regions) are likely to be more important than the length of the fire season, as they are related to both fire frequency and intensity; and fire season matches with the germination phenology of the corresponding floras (Lamont et al., 2022; Pausas & Lamont, 2022). It is also important to note that different species in fire-prone vegetation may have different strategies for surviving and reproducing at a given site. For instance, the fact that physical seed dormancy is strongly linked (causally, not just statistically) to high-intensity fires does not mean that all or most species in fire-prone ecosystems must possess physical seed dormancy. Only a subset of Mediterranean species has this trait, whereas many have other strategies for dealing with fire (e.g. smoke-released dormancy, resprouting, thick bark, serotiny). Therefore, physical seed dormancy does not need to be dominant nor the only trait adapted to the historical fire regime. Of course, high-intensity fires are not the only driver selecting for physical seed dormancy as it increases seed longevity generally (e.g. for ensuring survival of frost, exceptional summer heat, digestion), and global analyses may well fail to capture fire regime as a key variable. But what is important is that there is a mechanism that explains the selection for seed dormancy in seasonal ecosystems with intense crown fires: in ecosystems with prolonged hot, dry summers and intense crown fires that create huge swathes of colonizable patches, physical dormancy of seeds whose dormancy can be broken by heat is clearly adaptive (Pausas et al., 2022; Pausas & Lamont, 2022). We are not against broad-scale analyses and synthesis, nor against correlative studies as an essential first step in research, but when the results lack a biological basis (e.g. inherent seed dormancy is adaptive in seasonal climates) then they may impede the identification of key ecological processes. The overemphasis on climate and lack of testing alternative hypotheses to environmental factors is not exclusive to seed ecology. For instance, global patterns of species diversity are often based on correlative analyses with climate (Currie et al., 2004; Peters et al., 2016; Coelho et al., 2023). This is despite the presence of other possible candidates, that may or may not be correlated with climate, that directly affect the fitness of species (e.g. stochastic disturbances, soil fertility, physiographic features, mutualists and herbivores) and contribute to explaining species richness at both local (Huston, 1994; He et al., 2019) and broad scales (Pausas & Ribeiro, 2017; Brundrett, 2021; Moritz et al., 2023). Trait ecology is also prone to these pitfalls (Fig. S3). For instance, wood density in trees is a trait likely shaped by multiple factors; increased wood strength as a response to disturbances, such as cyclonic winds, could be a key factor in some regions (Curran et al., 2008; Dantas & Pausas, 2022), and the role of dense wood in fire tolerance and epicormic bud insulation appears important among some fire-prone lineages (Burrows, 2002). However, a recent global analysis (based on 36 000 samples and sophisticated machine learning methods) failed to consider any disturbance-related variable among the 79 examined (Yang et al., 2024). Species distributions are also typically explained at the macroclimate scale only, despite the existence of alternative drivers (Schwilk & Keeley, 2012; Pausas & Bond, 2021). A recent example is the prediction of African vegetation using a plant growth model applied to species distribution and climate variables (Higgins et al., 2023). While the model may correctly predict some current vegetation, it fails to incorporate the most important ecological mechanisms that shaped African vegetation, such as large herbivores and frequent fires (Bond, 2019; Owen-Smith, 2021). Exclusion experiments show that these massive plant consumers drastically change vegetation composition and structure (Venter et al., 2018; Pellegrini et al., 2021; Beckett et al., 2022). We need mechanistic models that incorporate understanding of how ecosystems work in order for such models to be useful predictive tools. Global analyses are an ambitious endeavour to find universal rules, but it needs to be appreciated that such rules may fail at identifying mechanisms that create broad-scale patterns if likely causal variables are not included in the first place, and when they are defined by multiple factors. Most critically, such a broad-scale approach may even hide key local ecological processes (Fig. S3); more integration between broad-scale description and hypothesis-based studies is needed. Furthermore, hypothesis-driven science cannot be replaced by computer mining of immense databases; the scientific method can be enriched by the use of large databases but not replaced by it. If ecology aims to be a predictive science, we should focus more on a mechanistic understanding than on describing correlations with vast amounts of data. And this is important as technologies become more accessible to everyone. The current rise of artificial intelligence (AI) tools may exacerbate this problem through the illusion of understanding (Messeri & Crockett, 2024). Alternatively, AI may help to investigate critical variables that differ from climate and soils at broad scales and to overcome some of the research biases. In addition, carefully crafted, comparatively based and prolonged field observations are the key to understanding natural phenomena and need to be fully promoted (Nanglu et al., 2023). We thank Robin Stewart and Tom Parker for comments on the manuscript. This research has been performed under the framework of the projects FocScales (Promteo/2021/040, Generalitat Valenciana) and DISTEPIC (PID2022-141530NB-C21, Ministerio de Ciencia e Innovación, Spain). Any use of trade, product or firm names is for descriptive purposes only and does not imply endorsement by the USA Government. None declared. JGP initiated the project, analysed the data, and wrote the first version of the manuscript. BBL, JEK and WJB contributed ideas, examples and edits to the final text. Fig. S1 Germination of six hard-seeded Australian species under control conditions and after scarification. Fig. S2 Relationship between the potential fire season length and latitude for the Northern and Southern hemispheres. Fig. S3 Many ecological patterns depend on the scale. Please note: Wiley is not responsible for the content or functionality of any Supporting Information supplied by the authors. Any queries (other than missing material) should be directed to the New Phytologist Central Office. Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.