Community Theory: Testing Environmental Stress Models

Protocol for Testing Environmental Stress Models

M&S1987 and M&O1990 were expressly proposed as landscape or meta-community models and briefly laid out protocols enabling tests of ESMs. Here I make those criteria more explicit. Thus, criteria for evaluating the consistency of literature studies with ESMs include:

1. CSM/PSM models apply at community, assemblage or subweb levels, not population levels. That is, like HSS1960, they assume that species within a trophic level respond similarly to environmental stress and interactions, and that measures of how abiotic and biotic factors affect community structure are “relative,” i.e., that proportionally, impacts are zero-sum, totaling to 1.0.

2. The models consider only negative consumer-prey interactions. The difference between CSMs and PSMs is that negative effects are either weaker or stronger with high environmental stress, respectively.

3. Experiments must be done at least at two (ideally more) different points along environmental stress gradients. Environmental stress gradients can occur across short (e.g., meters as in high to low intertidal) or long distances (e.g., 10s to 100s of km as from xeric to mesic terrestrial habitats). Here, “sites” indicate specific locations along most of the range of an environmental stress gradient. For example, high and low intertidal locations can each be a site because they occur at the end points of a steep vertical thermal/desiccation environmental stress gradient. In streams, pools and rapids are separate sites because they have different flow velocities. On land, alpine and lowland areas are separate sites because thermal conditions differ. 

4. As implied in (3), models focus on spatial, not temporal differences. Temporal responses to perturbations test community resilience and thus relate to stability, while ESMs were proposed to reflect temporally averaged conditions that characterize sites as differing in overall environmental stress. Thus, while wave-exposed intertidal sites might have calmer periods, on average they will have much stronger wave forces than more wave-protected sites. However, certain temporal studies may test ESMs, e.g., in cases where studies were done over several years/seasons differing in environmental stress levels. 

5. Ideally, focal environmental stress gradients capture most of the full range of conditions across which the system occurs. This is most easily done in habitats with steep environmental stress gradients, such as intertidal height, wave-beaten shores vs. sheltered coves or bays, salinity or sedimentation estuarine gradients, streams differing in flow rate and periodicity of flooding events, frequency of ice scour or fires, and pH stress in oceans or lakes, to name a few. Terrestrial environments with small changes in environmental conditions across large distances are far more challenging because of the great distances involved, but mountainous temperature and moisture environmental stress gradients, while still logistically difficult, are more feasible. 

6. Studies should provide evidence that environmental stress affects both consumers and prey. Lack of such information makes assessment of consistency with an ESM subjective.

7. Studies should examine community-level responses to environmental stress, including at least demonstrably strongly interacting consumers and the most abundant prey (sensu Power et al. 1996). 

8. Studies should be conducted in the field. This criterion excludes strictly laboratory-based experiments, whose relevance to field conditions is problematic. Similarly, mesocosm experiments (e.g., flow channels, outdoor tanks or seawater tables) qualify only if they incorporate a wide range of environmental stress (e.g., fast or slow flow rates, hot or cold, high or low salinity, wet or dry). 

Assembling The Datasets

The 111 papers assessed by S&H2018 included 174 tests which they felt enabled testing if studies were consistent with either antagonistic/CSM, synergistic/PSM models, or additive/no effect (the effect of consumer on prey was statistically insignificant or neutral). Papers were sorted by habitat (marine, freshwater, terrestrial), consumer-resource interaction type (predation, herbivory), plant prey type (algae, herbaceous, woody), predator thermal strategy (ectothermic or endothermic), performance measure (biomass, density, survival), stress factor (fire, thermal and desiccation, drought, salinity, and others), and stress type (temporal, spatial, experimental). 

In reassessing the S&H2018 dataset, I read all abstracts, and in most papers, methods, relevant results and discussion, and in some cases, the whole paper. To these, I added papers from Google Scholar and Web of Science searches for papers citing ESMs. These searches found earlier papers not included in S&H2018 and scanned papers published since 2015 when S&H2018 ended their search. I summarized each paper, listing each by its environmental stress gradient, the specific environmental stress examined, the biotic factor (predation or herbivory), the habitat, whether the study was in the field, mesocosm, or laboratory, trophic level number, specific consumers, specific resource (food or prey), and a summary of results. I then determined if the paper tested if all interactors were affected by environmental stress, if environmental stress affected consumer or prey abundance, and assessed if results were CSM, PSM, or No Effect. Spreadsheets with this information, i.e., the “Included” vs. the “Excluded” sets are in SI Tables S1 and S2, respectively. The original citations used by S&H2018 are in SI Document S1 and the ones I added are in SI Table S1 with S&H2018 references in the Included dataset.

Data analysis followed the same protocol as in S&H2018, i.e., assessment of significance of consumer and stress effects was based on analyses in each paper. I used the same categories as S&H2018 in data summaries and presented results as the percentage of each result in each category. I used the SigmaStat module of Sigmaplot (v. 13.0) to conduct χ2 tests to determine if frequencies in each category differed from the null case of equality (SI Table S3). Expected numbers were rounded to the nearest whole number to meet Chi-square assumptions.