Latitudinal gradients in population growth do not reflect demographic responses to climate

Abstract

Spatial gradients in population growth, such as across latitudinal or elevational gradients, are often assumed to primarily be driven by variation in climate, and are frequently used to infer species' responses to climate change. Here, we use a novel demographic, mixed-model approach to dissect the contributions of climate variables vs. other latitudinal or local site effects on spatiotemporal variation in population performance in three perennial bunchgrasses. For all three species, we find that performance of local populations decreases with warmer and drier conditions, despite latitudinal trends of decreasing population growth toward the cooler and wetter northern portion of each species' range. Thus, latitudinal gradients in performance are not predictive of either local or species-wide responses to climate. This pattern could be common, as many environmental drivers, such as habitat quality or species' interactions, are likely to vary with latitude or elevation, and thus influence or oppose climate responses.

Keywords: Achnatherum lemmonii; Danthonia californica; Festuca roemeri; climate change; demography; distribution; integral projection model; latitude; population growth; space-for-time.

Fig. 1

Conceptual diagram showing how latitudinal variation in environmental drivers can alter inference of climate responses. In the top row, population growth λ increases linearly with colder and wetter conditions (gray solid lines), but other environmental drivers (soils, competitors, enemies, etc.) may also vary with latitude and affect population growth in ways that reinforce or oppose climate patterns with latitude (black dashed lines). In the bottom row, the joint effects of climate and other environmental factors can alter inferred patterns of climate responses either when sampling variation across sites (black dashed lines) or when sampling variation across years within a site (gray solid lines). In the extreme, opposing climate responses within sites vs. latitudinal trends across sites can generate an example of Simpson’s Paradox (panel D), in which the direction of inference depends on the level of study (in this case, within vs. across sites).

Fig. 2

(A) Map of the study sites used for demographic monitoring of three perennial bunchgrasses in Oregon and Washington, USA (see also Appendix S1: Table S1). Several sites contained study populations for more than one species (divided circles). In total, we monitored five, six, and seven populations, respectively, of Achnatherum lemmonii, Danthonia californica, and Festuca roemeri from 2015 to 2018 along a latitudinal gradient. Lines show the estimated latitudinal range limits for each species in the study area (i.e., prairies west of the Cascade and Sierra mountain divides [darker gray shading shows the extent of primary ecoregions in which prairies are located], see Appendix S1 for details). Note that the northern range limit for Achnatherum is driven by scattered occurrences in coastal British Columbia, but this species is exceedingly rare north of the Willamette Valley in Oregon (~45.5° N). Across all study populations and years, (B) spring climatic water deficit (CWD_Spring) and (C) mean temperature (T _Spring) decrease with latitude. Conditions tended to be warmer during the 2015–2016 annual transition, and cooler and wetter in the 2016–2017 transition, compared to other years in the study. Other climate variables (see Methods) were not significantly correlated with latitude (Appendix S1: Fig. S2s

Fig. 3

Spatiotemporal patterns of population growth for three perennial bunchgrasses. The top row shows annual population growth rates from 2015 to 2018 along a latitudinal gradient, with 95% bias‐corrected confidence intervals from 1,000 bootstrap replicates including parameter uncertainty. The bottom row shows mean and 95% confidence intervals for transient stochastic population growth rates λ_s at each site from random sampling of each annual matrix for 100 time steps and 1,000 replicates including parameter uncertainty.

Fig. 4

Fitted survival and growth responses to climate drivers across three perennial bunchgrasses (A‐I). Lines give the predicted vital rate responses from the best‐supported climate models (Table 1), and points show the distribution of climate values observed across sites and years. Vital rates are shown for average‐sized individuals and the mean latitude across study sites for each species (black solid lines). Where there were significant latitude by climate interactions, we also show vital rates for the minimum (red dashed) and maximum (blue dotted) latitudes across study sites for each species, and highlight the observed climate values for these sites (red, minimum, blue, maximum). For Achnatherum, site intercepts for growth mean and variance were significantly correlated with latitude (see Results), so we also show growth rates predicted for the lowest latitude (red dashed) and highest latitude (blue dotted) sites. Climate variables are standardized to mean 0 and variance 1. Note that growth mean and variance are shown on a transformed scale as a proportion of the range of possible sizes based on a species‐ and size‐specific minimum and maximum value (see Appendix S1). Survival and growth are shown here because they have the greatest effects on population growth rates; see Appendix S1: Fig. S4 for additional vital rate climate relationships.

Fig. 5

Contour plots of population growth rate as a function of climate and latitude for three perennial bunchgrasses. Climate here is shown as the first principal component of all climate variables, reflecting variation from warmer/drier to cooler/wetter conditions. Contour lines show changes in population growth rate in 0.1 increments. Predictions are from the best‐supported models allowing interactions between latitude and climate as well as a random site effect (top, A–C), the best‐supported models that exclude latitude but include a random site effect (center, D–F), and the best‐supported models that include climate but exclude latitude or site effects (bottom, G–I). Points show the distribution of latitude and climate conditions across sites and annual transitions (squares, 2015–2016; circles, 2016–2017; triangles, 2017–2018). Note that contour lines may be slightly curved even in the absence of a latitude effect due to sampling variation and the effects of aggregating different climate variables into a common PC variable on the y‐axis.