Atmospheric and Oceanic General Circulating Models (AOGCMs) are often used by evolutionary ecologists to predict how biological populations might respond to global climate change. Although these models are capable of recreating historical environmental conditions at the continental scale, their accuracy at the regional scale is considerably lower. This deficiency is of critical importance to evolutionary ecologists for two reasons. First, the entire distribution of many species is substantially less than an entire continent. As such, applying these broad scale predictions to narrowly distributed species may not be appropriate. Second, species that have relatively broad distributions frequently exhibit geographic variation for a variety of traits. How this geographic variation will respond to global climate change cannot be evaluated using only continental scale predictions.

Here, I evaluated the ability of an AOGCM to recreate historical season lengths at the regional geographic scale for the striped ground cricket, Allonemobius socius/fasciatus and suggest a simple correction factor that greatly improves its performance. Using Allen’s double sine wave approximation I compared the season lengths produced by an AOGCM to the season lengths observed at multiple weather stations (N = 1,445) within 150 regions located across the continental United States from 1961-2000.

Results/Conclusions

I found that the AOGCM underestimated season length 91% of the time (+/- 47%, stdev) by an average of average of 42% (+/- 18%) across all regions examined. This underestimation was primarily due to lower daily maximum temperatures, which were 5.9C (+/- 1.4C) lower in the model than the historical observations; rather than daily minimum temperatures, which were 2.6C (+/- 1.0C) lower. A simple adjustment based on the differences in the simulated and observed average daily maximum and minimum temperatures, improved the model’s ability to recreate historical season lengths. After this adjustment, the model overestimated season lengths by an average of only 2% (+/- 8%).

Although more sophisticated adjustments could further improve the model’s ability to recreate historic season lengths, the simulated season lengths produced by this adjustment were well within the coefficient of variation for season length across weather stations (COV = 11.2% +/- 3.2).