A North American Regional Climate Change Assessment Program (NARCCAP) climate model was used to generate 29 years of daily maximum temperature data at around 17,000 spatial locations in North America. Our ultimate goal is to develop accurate models for forecasting extreme weather conditions, e.g., heat waves. The goal of this talk is to develop a reasonable time series model for the temperature data at any given spatial location. Future work will also consider the spatial pattern. Since the data shows a strong seasonal variation, as well as long range correlations, it is not clear {\it a priori} whether to apply a model with long range dependence, or periodic stationarity, or both. It has been observed, e.g., in financial time series, that nonstationarity can mimic long range dependence. We will see that this is also the case for climate data. Hence a periodically stationary time series model may be indicated.