>900 propagated feltleaf willow (Salix alaxensis) cuttings heading back to Fairbanks tomorrow from @toolikfieldstation.bsky.social to overwinter before planting two common gardens next spring. #EVOME #arcticresearch

One month later @toolikfieldstation.bsky.social #EVOME #arcticresearch

Excited to present recent work on genomic offsets -
@sse-evolution.bsky.social #Evol2024. Unfortunately I couldn't be there in person, so if you missed my recorded talk, here are QR codes to the manuscripts and here is a link to the talk bit.ly/3WJGVVp

5 - while metapopulations with high levels of LA produced generally well-performing models for predicting within-landscape fitness, performance across all datasets decreased when projecting to novel climates that are increasingly differentiated from landscape values (x-axis) 9/n

4- the choice of environmental variables used in model training can impact performance. For instance adding nuisance environmental variables unrelated to the sources of selection pressure often decrease performance, implicating potential limits of some methods vs others (8/n)

3 - the choice of genetic markers used has varied and minor effects on model performance, where adaptive markers provide some minimal advantages, but not universally or by great margins (7/n)

2 - the spatial arrangement of environmental variables to which the metapopulation adapts can also affect performance, even when contrasting landscapes have metapopulations that reach similar levels of local adaptation (6/n)

We explore the limits of five genomic offset methods across these scenarios. While there were performance difference across methods, we discuss the overarching findings as follows (note flipped y-axes!!): 1 - the degree of local adaptation (LA) drives model performance (5/n)

We train genomic offset methods using genomic and environmental data from these sims. We evaluate model predictions within in silico common gardens in each of the 100 home environments. Well-performing models should have negative relationships between offset and fitness (4/n)

We used simulation data from Lotterhos (2023) to better understand these limits. Described in detail in the manuscript, these simulations represent biological realism of empirical data as well as challenging cases in which model performance of may suffer. (3\n)

Very happy to announce the release of my newest manuscript with Katie Lotterhos! "The limits of predicting maladaptation to future environments using genomic data (1 / n)