Thermal variability is a key driver of ecological processes, affecting organisms and populations across multiple temporal scales. Despite the ubiquity of variation, biologists lack a quantitative synthesis of the observed ecological consequences of thermal variability across a wide range of taxa, phenotypic traits, and experimental designs. Here, we conduct a meta-analysis to investigate how properties of organisms, their experienced thermal regime, and whether thermal variability is experienced in either the past (prior to an assay) or present (during the assay) affect performance, relative to the performance of organisms experiencing constant thermal environments. Our results – which draw upon 1,712 effect sizes from 75 studies – indicate that the effects of thermal variability are not unidirectional and become more negative as mean temperature and fluctuation range increase. Exposure to variation in the past decreases performance to a greater extent than variation experienced in the present and increases the costs to performance more than diminishing benefits across a broad set of empirical studies. Further, we identify life history attributes that predictably modify the ecological response to variation. Our findings demonstrate that effects of thermal variability on performance are context-dependent, yet negative outcomes may be heightened in warmer, more variable climates.

Systematic literature review

To understand how thermal variability affects performance, defined as physiological or demographic rates or states, we conducted two systematic literature searches of the effects of thermal variation during acclimation and acute conditions. Our first search, conducted on 14 November 2020 using the ISI Web of Science (WOS) database with the search terms: AK=((temperature OR thermal) NEAR (vari* OR fluc*)) AND SU=(Life Sciences & Biomedicine) yielded 176 results. To increase sample size and decrease publication bias, we conducted a second systematic literature search on 3 June 2021 using the SCOPUS database with the search terms: KEY ("thermal performance curve" OR "thermal fluct*" OR "thermal vari*" OR "temperature vari*" OR "fluctuating temperatures" OR "thermal regime" AND ("ecology" OR "physiology")), which yielded 405 results. There were 43 papers returned in both WOS and SCOPUS searches.

Inclusion criteria

We screened abstracts and titles from both searches for inclusion using the 0.4.1 version of the revtools R package (Westgate, MJ, 2019) and excluded 189 studies (Figure 2). We then assessed eligible studies (n=306) and excluded studies that lacked a constant and fluctuating treatment (n=115), did not feature a consistent, controlled fluctuation pattern (e.g. pulse press, multiple stochastic cold exposures, etc.) (n=64), were reviews, commentaries, or perspectives (n=33), were theoretical or modelling studies (n=24), were not biologically relevant (e.g. engineering, chemical studies, etc.) (n=19), lacked reported error measurements (n=4), lacked extractable or comparable data (n=4), and featured more than 1°C difference between the mean temperatures in constant and fluctuating treatments (n=13). For studies to meet these inclusion criteria, the experimental design had to be explicitly focused on thermal variability. Subsequently, we conducted a cited reference search from the remaining eligible studies and included an additional 49 studies. In total, we included 75 studies with 1,712 effect sizes (Figure 2) (see Table S2 for a list and description of studies included). All studies included involved ectothermic organisms. We excluded any population or community-level responses and species with unresolved phylogenies or that were not identified to the species level in the Open Tree of Life database.

Data extraction

From the studies that met our inclusion criteria, we extracted mean response values, any measure of variance (SD or SEM), and sample size from tables and figures using Webplotdigitizer (Webplotdigitizer, v4.5, 2021). Any studies that reported error as SEM were converted to SD by multiplying SEM by the square root of the sample size. Further, if studies reported findings using medians and the IQR, and we could confirm the data to be approximately normally distributed, we estimated the mean based on the reported median, and the SD to be the IQR divided by 1.5 (Higgins & Green, 2011). If any extracted values were missing sample sizes or variances, the points were automatically excluded via the meta-analysis software metafor (v3.0.2, Viechtbauer, 2010).  Additionally, we collected aspects of experimental design (experiment type, duration, etc.), thermal regime (mean temperature, fluctuation range, etc.) as well as life history traits (age, size) and response metrics (trait directionality, see Analysis and Hypothesis Testing for definition) to investigate potential mechanisms mediating responses to thermal variability.

R