In our final model set testing mismatch, winter weather, and land cover metrics on survival, the model with mismatch varying by winter season and proportion of Open habitat had the most support with 32% model weight (Table 3). All competitive models contained mismatch effects varying by winter season and collectively held 96% of the cumulative model weight (Table S4). In our top-ranked model, phenotypic mismatch lowered survival (primarily during Winter Season 2) (Table 4). During Winter Season 2, weekly survival from the top-ranked model varied by color morph throughout the season (Fig. 3A). When snow cover was present on the landscape (week 5 - 19), gray morphs had higher average weekly survival of 0.98 compared to 0.92 for red morphs. Conversely, when no snow was present, red morph individuals experienced lower mismatch and had higher weekly survival estimates (Fig. 3A). The top ranked model also indicated that high proportions of Open habitat negatively impacted survival throughout the winter season (Table 4). Across the entire 21-week period each winter season, derived survival estimates varied across the years (Figure 3B) with lowest during Season 2 (S = 0.41; 95% CI: 0.24 – 0.59) followed by Season 4 (S = 0.55; 95% CI: 0.35 – 0.73), Season 5 (S = 0.65, 95% CI: 0.48 – 0.79), and was highest during Seasons 3 (S = 0.85; 95% CI: 0.61 – 0.95) and 1 (S = 0.79; 95% CI: 0.58 – 0. 91).

Table 3. Model selection results for the effects of phenotypic mismatch, temperature, and land cover on ruffed grouse overwinter survival. Phenotypic mismatch (MM) with snow cover is included in all top-ranked models suggesting it is an important driver of winter grouse survival. Variables below include time-varying covariates of MM1-MM5 (weekly time-varying individual covariates of mismatch for each field season, the proportion of time in a week an individual is phenotypically mismatched with the snow cover in its winter habitat), and constant covariates for % Open and % Mature Forest and (proportion of Open or Mature Forest cover, respectively). All values were averaged spatially over individual winter use areas. Models ranked by AIC_c(Akaike’s Information Criterion, corrected for small sample size). ΔAIC_c represents the difference between the top-ranked model in the candidate set, w is the model weight, and kis the number of parameters. Only models within 2 ΔAIC_cof the top-ranked model and the intercept-only model are shown.