Amplitude and timescale of metacommunity trait-lag response to climate change

Climate change is altering the structure and functioning of communities 1. Trait-based approaches are powerful predictive tools that allow consideration of changes in structure and functioning simultaneously 2, 3. The realised biomass-weighted trait distribution of a community rests on the ecophysiology of individuals, but integrates local species interactions and spatial dynamics that feed back to ecosystem functioning. Consider a response trait that determines species performance (e.g. growth rate) as a function of an environmental variable (e.g. temperature). The change in this response trait’s distribution following directional environmental change integrates all factors contributing to the community’s response and directly reflects the community’s response capacity 3. Here we introduce the average regional community trait-lag (TLMC) as a novel measure of whole-metacommunity response to warming. We show that functional compensation (shifts in resident species relative abundances) confers initial response capacity to communities by reducing and delaying the initial development of a trait-lag. Metacommunity adaptive capacity in the long-term, however, was dependent on dispersal and species tracking of their climate niche by incremental traversal of the landscape. With increasing inter-patch distances, network properties of the functional connectivity network became increasingly more important, and may guide prioritisation of habitat for conservation.

A: Performance/growth curves in response to temperature of two species with different temperature optima. Blue bars and lines indicate their respective growth rates in two environments differing in temperature. B: Temperature increase due to climate change. C: During stable conditions, the community trait distribution converges around the current environmental optimum, indicated by the blue handle with round ends. The community weighted mean trait (CWMT) coincides with the current optimum trait value. Grey bars indicate relative biomasses of species with particular trait values of the response trait in question, here the temperature optimum T opt . D: Consider a change in environmental conditions as in panel B, indicated by the blue arrow, that shifts the current optimum to higher values of T opt . The new optimum is indicated by the blue handle with square ends. Due to the increase in abundance of species with trait values that match the new environment more closely, the mean of the trait distribution has shifted as well, tracking current conditions (indicated by the red arrow). Tracking has not been complete, however, and the CWMT has developed a lag, the trait-lag TL, behind the new optimum (indicated by the black two-headed arrow).
recovery (t Rec ), as well as the amplitude TL Max of the lag. The analysis focuses on a novel measure to say, metacommunities with the highest response capacity had the lowest lag amplitude TL Max , vailing effects. r max had a slightly larger effect at weaker competition levels. The 3way interaction 117 between distance, α ij and r max was due to a shift in the interaction at the greatest distance (16 km) 118 (Suppl Fig S1). 119 Network properties that characterised the functional connectivity of the landscape had overall 120 comparatively small effects on response capacity, but their effect became noticeable at the two 121 greatest distance levels (8 and 16 km) (interaction term, Table 1). The best combination differed 122 between distance levels (Suppl Fig S2). At both distance levels (8 and 16 km), combinations with a 123 higher degree of clustering and a shorter characteristic path length corresponded to higher response 124 capacity. Interestingly, at the greatest inter-patch distances (16 km), a larger leading eigenvalue of 125 the connectivity matrix appeared to confer higher response capacity (Fig 3). Table 1: Anova results of main effects and interactions of factors distance, competition (α ij ), maximum growth rate (r max ) and network combination (Network) on metacommunity response capacity (= 1/ T L M C ). The intention is not to examine statistical significance but to indicate the relative effect sizes of factors, calculated as η 2 . Commonly used benchmarks for small, medium and large effect sizes are 0.02, 0. 13 131 Local and Regional Components The realised regional response that mediated TL MC could be 132 decomposed into local and regional response mechanisms or components, the relative importance 133 and timescale of which depended on manipulated factors (above all inter-patch distance, Fig 4).

134
After the onset of climate change, shifts in the relative abundances of local resident species present,  Fig 4). By t Max , TL Disp contributed with 8-16% to lag reduction (Suppl Fig S3). monotonically with distance. Dispersal thus fuelled regional re-organisation and trait-lag recovery.
With increasing inter-patch distances, an increasingly greater part of dispersal could be at- persal (TL SS ) was independent of patch distance and landscape extent, but was affected by growth 163 and competition levels (Fig 5 A). Dispersal dynamics on the other hand began to counter lag de-164 velopment earlier at smaller distances and the magnitude of lag reduction due to dispersal was 165 much reduced at the largest distance level (Fig 5 B). The contribution of TL Trav to lag reduction of dispersal was greatest, dispersal also contributed most to recovery of that lag (Suppl Fig S3).

182
Due to competitive exclusion during stable conditions, these communities had the lowest initial 183 trait distribution variances (Suppl Fig S4), which hampered the initial response capacity of local 184 communities and depressed overall response capacity by causing a large initial lag. Eventhough We have proposed to follow and examine the development and recovery of the average regional

225
Faster growth rates on the other hand were overall beneficial for initial response capacity, 226 in spite of associated lower community trait variances. Because local response diversity was suf-227 ficiently high in all cases, the speed of the response was the decisive determinant. Faster growth 228 rates accelerated competitive displacement and species turnover rates, which allowed communities 229 to track changing conditions faster and delayed the development of a trait-lag.

230
Determinants of long-term adaptive capacity. Although local response capacity contributed 231 substantially to absolute lag reduction (orange area in Fig 3), it was limited to an initial shift 232 and a temporary tracking of temperature change by community trait distributions only. Regional network began to matter. Their effect was comparatively small because patch arrangement was not 300 manipulated directly, but selected for from essentially random landscapes. Landscape modification by humans and resulting habitat loss and fragmentation are rarely random, but follow existing 302 vegetation patterns (e.g. preferential conversion of areas with high primary productivity) as well 303 as jurisdictional and historical land use patterns 5,23 . In more structured landscapes, network prop-304 erties should influence regional adaptation even more.

305
Network and graph theoretic perspectives on landscape connectivity provide a valuable tool 306 for the representation and analysis of habitat patch arrangement in landscapes 24 . Where spatial 307 planning necessitates the prioritisation of one area over another, network measures may guide the 308 decision so as to maintain connectivity between increasingly isolated patches 25,26 . Removal of 309 patches with high betweenness-centrality, for example, has been shown to reduce connectivity 310 and impair spatial insurance effects that maintain metacommunity robustness to habitat loss 27 .

311
Preservation of patch connectivity over larger regions is particularly relevant in the light of climate 312 change that may necessitate range shifts for species persistence 28 . Because we here focused on 313 whole-metacommunity response capacity, we limited the analysis to network-wide metrics.

314
In our models, higher levels of clustering were beneficial for response capacity. In clusters shifts, and improving regional response capacity. Corridors of stepping stones facilitate and accel-325 erate species range shifts across larger spatio-temporal scales [28][29][30] .
High clustering and short characteristic path lengths appeared to allow for alternative re-327 sponse mechanisms, such that either high overall reachability but low clustering, or low character-328 istic path lengths but high clustering could maintain response capacity (Suppl Fig S2).

329
An intriguing result was that a larger leading eigenvalue of the connectivity network corre- forms of competition other than the generalised Lotka-Volterra formulation could be considered.

378
Introducing resource competition, for example, would allow to introduce a second habitat quality 379 factor (resource availability), such that the interplay of two traits (temperature optimum and re-380 source uptake rate) could be studied. Lastly, a simplified setup would relax computational and data 381 storage constraints and allow to track also local trait distributions and their variability within the 382 landscape. To better assess commonalities and differences with other models, local and regional 383 diversity measures could be recorded.

384
Conclusions. Trait distributions are a powerful and versatile conceptual tool to study community 385 responses to climate change. Recasting biodiversity theory in terms of traits integrates the mech-386 anisms that generate and maintain diversity and allows scaling from individual fitness through 387 community trait distributions to ecosystem properties 3 . We have applied theory that predicts a 388 trait-lag in response to directional environmental change 10,13 to the metacommunity level to pro-389 pose a simple metric of integrated regional response capacity, the inverse of the integrated lag.
where r max is equal for all species; T is the current ambient temperature; V T is the variance of 403 the temperature response curve (see Suppl Table S1 for parameter values). The model followed the where S is the total number of species in a community; carrying capacity K i = 1 for conve-409 nience; the interspecific competition coefficient α ij < 1, ensuring coexistence; m is background 410 mortality.

411
After each growing season, species allocated a fixed fraction (Fecundity) of their realised 412 local biomass to seed production. Seeds dispersed according to a Poisson random draw from the 413 2D dispersal probability distribution (kernel) where x is distance; c = 0.5 (leptokurtic); β determines mean dispersal distance. Success 415 of dispersal events were stochastic, hence actual functional connectivity was not fully determined.

416
For each propagule parcel sent out by a species, a random number was drawn from a Poisson dis-