Comparison of model predictions (gray area) and experimental observations (white area) of endemic prevalence at the thermal optimum, T_opt = 20°C.

Estimates under constant environmental conditions fall on the vertical dashed line, representing endemic prevalence at Topt. We used the high temperature variability sub-treatment (sd* = 3.3) to generate endemic prevalence predictions using nonlinear averaging (Eq 5), stochastic simulations of the model (Eqs 1.1–1.4), and stochastic simulations with acclimation (Eqs 1.1–1.4 and 2–4). Temperature fluctuations were modeled using an autocorrelated temperature model (S1) and were restricted to occur within a +/− 6 range around 20°C. The acclimation models represent conditions where time to host acclimation is fast (ψH = 6 days) and the effect of beneficial acclimation is strong (slope = 0.4, Eq 4). Additionally, the results of the acclimation models depicted in this figure differ in whether they assume the infection rate, η(T) (Eq S2.2 and Fig 2B), and equilibrium within-host parasite intensity, ω(T) (Eq S2.3 and Fig 2D), acclimate as a host trait, H, or as a parasite trait, P (S2 Table). All versions of our model simulations predict that temperature variability should suppress endemic prevalence and therefore shift left of the dashed line. Our endemic phase experimental observations illustrate the opposite effect and show that temperature variability leads to higher endemic disease prevalence. Uncertainty around model predictions and experimental results represent 95% credible intervals. The data in the gray portion of this figure is generated by Eqs S2.1–S2.6 and 1–5. Parameterization of these models is found in Tables 1, S2, and S3. The experimental data depicted in the white portion of this figure can be found in S1 Data.