Thinking big: the case for experimental macroecology

Abstract

Ecology has long faced a productive but unresolved tension between the scales at which patterns are observed and the scales at which mechanisms are understood. The distribution of life across the planet is reorganizing rapidly under interacting pressures (climate change, habitat loss, resource overexploitation, biological invasions, and pollution) that individual studies are rarely designed to meaningfully attribute, address, or disentangle. The availability of species distribution, environmental, phylogenetic, and functional trait data has expanded dramatically over recent decades (Heberling et al. 2021), yielding increasingly detailed descriptions of how species richness, abundance, and range limits shift across space and time. Yet predictive capacity remains limited, particularly under novel levels and combinations of drivers (Urban et al. 2016). The bottleneck is no longer restricted to data availability, but also to a more articulate understanding of causation pathways, i.e. the capacity to attribute large-scale patterns to specific processes in ways that generalize reliably across ecological contexts.

The gap between pattern and mechanism reflects a deeper structural feature of ecological inference: observational data, however extensive, cannot readily distinguish among multiple processes that can generate the same large-scale pattern (Lawton 1999). For example, the latitudinal diversity gradient can be produced by variation in energy availability, evolutionary time, climatic stability, biotic interaction intensity, or some combination thereof. Range limits may reflect physiological tolerance, competitive exclusion, dispersal limitation, enemy release, or demographic lag, and frequently several of these simultaneously. Without experimental intervention, it is difficult to assign causal priority among alternatives. The problem is compounded under global change, which typically involves shifts in multiple interacting drivers simultaneously: predictions based on single-factor correlations are likely to fail under conditions with no historical analogue (Urban et al. 2016). Closing this gap requires not simply more data, but differently designed data: observations embedded in experimental or quasi-experimental designs that test causal hypotheses at the scales where ecological patterns emerge and where the consequences of global environmental changes for biodiversity and ecosystems are realized.

Macroecology is often introduced as the ‘ecology of large scales', but this definition can be misleading. Lawton (1999) defined macroecology as ‘the search for major, statistical patterns in the types, distributions, abundances, and richness of species, from local to global scales, and the development and testing of underlying theoretical explanations for these patterns.' Likewise, Brown (1999) argued that what makes a study macroecological is not its absolute geographic extent, but the kind of question it asks and the level of aggregation at which inference is sought. A stone, pond, island, forest chronosequence, continent, or global database may each support macroecological inference if the study asks how general patterns emerge from variation among many ecological units (May 1994), and if those patterns are linked to hypotheses about process. Conversely, a geographically extensive study might not necessarily qualify as macroecological if it merely catalogues local particularities without seeking general statistical regularities or their causes (McGill 2019).

Experiments have rarely occupied the centre of the macroecological research programme, partly because macroecology developed around observational data, and partly because many of its canonical questions concern spatial or temporal extents that are difficult to manipulate directly. Yet there is nothing in the logic of macroecology that excludes experimental inference. On the contrary, if macroecology is concerned not only with detecting patterns but also with explaining them, experiments and quasi-experiments become essential whenever alternative mechanisms can generate the same statistical regularity. Experimental macroecology should therefore be understood not as a departure from macroecology, but as an extension of it: the deliberate use of manipulative, quasi-experimental, distributed, or synthetic designs to test whether the mechanisms proposed to explain macroecological patterns actually operate, and whether they generalize across ecological contexts.

This mode of inference has yielded important advances. Recurring statistical regularities, including species–area relationships, body size distributions, latitudinal diversity gradients, and range size distributions, are replicable, often large in effect size, and evidently reflect something fundamental about how ecological systems are organized. Yet their mechanistic interpretation remains persistently contested. The difficulty is that macroecological patterns are often mechanistically undetermined. Similar statistical regularities can be generated by different combinations of variables and observational data cannot firmly adjudicate among them (Araújo et al. 2007) because it cannot hold the relevant variables constant. This limitation has direct practical consequences. Forecasts of biodiversity change depend not only on whether a pattern has been detected but on which mechanisms generate them, because different mechanisms imply different sensitivities to the specific changes underway. The persistent failure of many range-shift projections likely reflects, in part, the use of correlative models that capture patterns without identifying the underlying mechanisms (Pearson et al. 2006, Urban et al. 2016). Closing this gap between pattern and mechanism will directly affect which species are expected to be winners and losers under future conditions, and on what timescales.

The experimental tradition in ecology has a distinguished record of revealing mechanisms. Foundational manipulative experiments demonstrated competitive exclusion (Connell 1961), the structuring role of keystone predators (Paine 1974), and the effects of disturbance (Sousa 1979), helping to transform ecology from a primarily descriptive into a more explicitly mechanistic science. Whole-ecosystem experiments later extended this logic to nutrient cycling, trophic cascades, and the early detection of regime shifts (Carpenter et al. 2011). The strength of experiments lie in their capacity to distinguish causal effects from statistical association by manipulating putative drivers, controlling background variation, and testing counterfactual expectations. Their limited role in macroecology has not arisen because experiments are conceptually inappropriate, but because macroecological questions often require inference across contexts that are difficult to manipulate directly. A single-site or short-term experiment may uncover a mechanism, but it does not automatically establish the generality, context-dependence, or emergent statistical consequences of that mechanism The relevant question is, therefore, not whether experiments belong to macroecology, but how manipulative, quasi-experimental, distributed, or synthetic designs can be constructed so that they carry macroecological inference.

This special issue, ‘Towards Experimental Macroecology', is organized around that premise. Closer integration between the theoretical framework of macroecology and the causal logic of experimental ecology is needed if we are to improve predictions of biodiversity and ecosystem change under non-analogue conditions – a world where both the combination of drivers and the rates at which changes occur exceeds what has previously been observed. The contributions span marine, freshwater, and terrestrial systems, and organisms ranging from soil fungi and freshwater invertebrates to boreal trees and migratory vertebrates. They also operate across multiple, partly orthogonal dimensions of scale: from fine-grained local manipulations to distributed experiments and synthetic analysis spanning broad environmental gradients. Rather than offering a catalogue of case studies, the contributions converge around a shared ambition: to design studies that simultaneously satisfy the inferential demands of macroecology and the mechanistic demands of experimental biology. In what follows, we characterize the intellectual foundation of experimental macroecology, distinguish the methodological approaches it encompasses, and synthesize how the contributions to this issue collectively advance understanding of biodiversity and ecosystem change.

Experimental macroecology can be defined as the deliberate use of experimental and quasi-experimental designs at spatial or temporal extents sufficient to address the emergent large-scale patterns that define macroecology, while explicitly linking empirical results to theoretical and process-based models. The word deliberate is important: studies qualify not simply by being large, but by being designed, or synthesized post hoc, with macroecological inference as an explicit objective. The potential for making inferences using experimental macroecology rests on two complementary principles. By manipulating conditions and randomizing treatments, experiments can establish process-level mechanisms that observational data alone cannot confirm. Likewise, the scope of individual experiments can be broadened, either by explicitly distributing replicated experiments across space and time, or by synthesizing results across sites, systems, and contexts. Such synthesis can occur by design, through coordinated distributed experiments in which a standardized manipulation is replicated across sites explicitly selected to span environmental gradients (e.g. the International Tundra Experiment, ITEX; the Nutrient Network, NutNet; the International Drought Experiment, IDE; Fraser et al. 2012, Borer et al. 2014), or after the fact, through meta-analyses that aggregate independently conducted experiments to estimate average effect sizes and identify sources of heterogeneity (Gurevitch et al. 1992). By holding experimental design constant while allowing environmental context to vary, coordinated distributed experiments enable explicit quantification of the generality and context-dependence of ecological responses, which is precisely the information macroecological predictions require. As the global literature on ecological experiments grows, new opportunities of syntheses and meta-analyses emerge. Experimental macroecology can thus be defined by two criteria: 1) the use of experimental or quasi-experimental designs to isolate and quantify ecological processes, and 2) the implementation of such designs at extents sufficient to address macroecological questions, i.e. questions about variation in responses across experiments, rather than about the mean-field response to treatments within (or across) experiments.

The contributions to this issue do not each address a single isolated question. Read in aggregate, they converge on four recurring themes that together advance the predictive understanding of biodiversity change.