Model Application

The validated model was used to predict broadscale patterns of climate response in A. thaliana . This requires inferring the spatial distribution of genetic variation and germination timing on a continental scale. Both components are crucial because they dictate the distribution of plant genotypes and the temperatures they experience, respectively.
Inferring the spatial distribution of Genetic Variation and Germination Date
We inferred the distribution of A. thaliana genetic variation using kriging (Oliver & Webster, 1990), a method of interpolation used in geostatistics for spatially autocorrelated data (Appendix S5). Kriging was considered suitable because spatial autocorrelation inA. thaliana’s genetic variation was observed in our dataset (average Moran’s I =0.146, P=0) and is consistent with isolation by distance previously reported in the species (Platt et al., 2010; Sharbel et al., 2000). We produced a kriged genetic landscape at 1°x1° resolution across Europe by kriging each column of the GSM using the autoKrige function from the automap package (Hiemstra, 2013) in R.
Across its European range, A. thaliana germinates at different times of the year (Donohue, 2002). In order to determine the most likely growing season of different sites, we used data from Exposito-Alonso (2020) which identified k= 4 germination strategies defining coherent climate regions. We smoothed boundaries by replacing the value of outlier cells (those assigned to a different cluster from all its neighbours) with the most common value in the 8 neighbouring cells. The four regions (Central Europe CEUR, South Mediterranean SMER, North Mediterranean NMER, Scandinavia SCAN) corresponded to three germination seasons (spring, summer, fall). We assumed all plants germinated on a single date for each season. These dates were February 27 for spring, 25 May for summer, and 3 October for fall and were chosen based on the transplant dates of our plantings. These three dates were used to predict trait changes across years within regions. For landscape predictions in a given year, we allowed germination dates to differ across individual cells (1 minute spatial resolution). We predicted individual dates using the climate region, longitude, and latitude as predictors in a linear model and transplant dates as the response variable.

Projected Climate Response

We first predicted climate responses across Europe to identify sites that are susceptible to future decline under the RCP2.6 and RCP8.5 climate change scenarios (van Vuuren et al., 2011), with RCP8.5 being a worse scenario. We obtained daily minimum/maximum temperature projections for RCP2.6 from CCSM4 ensemble r1i1p1 (Gent et al., 2011) and for RCP8.5 from CMCC-CM ensemble r1i1p1 (Scoccimarro et al., 2011). Temperature rasters were resampled to 1°x1° to match the resolution of the kriged genetic landscape using the R\raster package (Hijmans et al., 2020).
We predicted DTB and SP from 2041 to 2099 using the RCP projections and in 2006 using temperature records from E-OBS v19.0eHOM (Cornes et al., 2018). We assumed a single genotype present in each cell (inferred through kriging) and a constant germination date across years. Negative values of SP were set to zero.
Finally, we emulate revegetation trials by using our model to predict the performance of specific genotypes across Europe under climate change. This allowed us to determine whether known genotypes could be used as a source of climate-resilient genetic variation at sites where the fitness of local populations was predicted to decline. As a proof-of-concept, we focused on predicting the fecundity of the Eden-2 and Ll-2 genotypes in 2006 and 2099 under RCP8.5. Eden-2 is a Swedish genotype that must be exposed to prolonged chill before flowering (‘vernalization’,https://www.arabidopsis.org), while Ll-2 originates from Spain and shows a low expression of the key flowering repressor FLC (Rosloski et al., 2013); the two genotypes were predicted to be the latest- and earliest-bolting of the 2029 genotypes, respectively.
Testing additional scenarios
The high levels of genetic diversity observed in A. thalianameans Ll-2 and Eden-2 are unlikely to represent the diversity of climate responses in the species. The performance of other genotypes may be of interest, and land managers may want to predict the effect of different germination dates on trait values. In order to aid in the visualisation of these multiple alternative scenarios, we have developed AraCast (https://adaptive-evolution.biosciences.unimelb.edu.au/shiny/AraCast2/), a shiny app that generates trait predictions for different genotypes, germination dates, and climate change scenarios.
Results