Climate change impacts on population growth across a species’ range differ due to nonlinear responses of populations to climate and variation in rates of climate change

repeated this procedure on 3 additional populations, with 218, 309, and 383 individuals marked and mapped in the three populations in 2017. All together, across the 5 populations for the 2016–2018 period, we obtained data from 1162 individuals, including seedlings. We estimated 5 size-dependent vital rates from these data: annual survival, mean size after one year of growth, variance in size after one year of growth, probability of fruiting, and number of fruits given fruiting. In four of the populations, we also estimated seedlings (in the second sur- vey) per fruit (in the first survey) by counting all seedlings within a 15 cm radius in the second survey around individuals that fruited in the first survey. Our C populations were located in Chugach State Park, S in Kachemak Bay State Park, N in Denali National Park, and E1 and E2 in Wrangell-St. Elias National Park and Preserve, and we obtained permits from each of these entities. We used field-collected soil temperature data and climate projections from global climate models (GCMs) to fit vital rate functions for the 2016–2018 period, as well as to construct IPMs and project future population growth rate. In all analyses, climate data were summarized by month, the timescale of the available future climate projections. Years were defined by the August 1 to July 31 interval, determined by the timing of the demographic censuses. For exam- ple, to estimate temperature over the 2017–2018 interval, we used data from August 1, 2017 to July 31, 2018. To obtain soil temperature data, we buried 1 or 2 small temperature loggers

(iButtons; www.maximintegrated.com/en/products/ibutton/ibuttons/index.cfm) ~5 cm underground in each population. We used the iButton data to obtain mean monthly soil tem- perature over the 2016–2017 or 2017–2018 time period in each population. We obtained cur- rent precipitation values from 5 GCMs for the 2015–2018 time period for each population [18], hereafter, “SNAP” data. We used 2015–2018 soil temperature and precipitation data to fit climate-vital rate relationships over the 2016–2018 period (S1 Appendix). We obtained future precipitation and air temperature values from these same GCMs over the 2018–2099 time period for each population, which we used to project population growth rate, after correcting future temperature to reflect discrepancies between GCM air temperature projections and soil temperature measured by iButtons (S1 Appendix; note that this correction assumes that the difference between soil temperature and climate projections is consistent across years). Because we were missing data for July soil temperature, we excluded July soil temperature from our annual and seasonal soil temperature calculations and used air temperature from SNAP data for hottest month temperature (which was often July; see S1 Appendix). We synthesized the iButton temperature and SNAP data into ten climate variables describ- ing 2015–2018 climate in each population: 1. average annual soil temperature; 2. average snow-free season soil temperature; 3. average snow-covered season soil temperature; 4. coldest

month soil temperature; 5. warmest month air temperature; 6. total annual precipitation; 7. snow-free season precipitation; 8. snow-covered season precipitation; 9. coldest month precip- itation; and 10. warmest month precipitation. Climate variables 1–4 were estimated from iBut- ton data (and thus do not include July); 5–10 were estimated from SNAP data. We describe derivation of these climate variables in S1 Appendix. We assessed the impact of these climate variables on D. alaskana vital rates using a model selection framework [19] in R [20]. For mean log size after one year of growth, variance in log size after one year of growth, probability of fruiting, and log number of fruits per size, we tested all possible subsets of a global model with log size in previous time step, average annual tem- perature, cumulative annual precipitation, and the interaction of average annual temperature and cumulative annual precipitation as predictor variables using the dredge function in the R package MuMIn [21], which allows comparison of multiple models using information theo- retic approaches. We used AICc [22] to select a best-fit model. We repeated this model selec- tion process for four other global models that included log size, temperature, and precipitation PLOS ONE Nonlinear responses of population growth rate to climate change PLOS ONE | https://doi.org/10.1371/journal.pone.0247290 March 3, 2021 5 / 20