A thermodynamic device placed outdoors, or a local ecosystem, is subject to a variety of different temperatures given by short-tem (daily) and long-term (seasonal) variations. In the long term a superstatistical description makes sense, with a suitable distribution function f(beta) of inverse temperature beta over which ordinary statistical mechanics is averaged. We show that f(beta) is very different at different geographic locations, and typically exhibits a double-peak structure for long-term data. For some of our data sets we also find a systematic drift due to global warming. For a simple superstatistical model system we show that the response to global warming is stronger if temperature fluctuations are taken into account.

37 figures. Significantly extended version, to appear in Physica A. Added new material in section 6 quantifying the stronger response to global warming if temperature fluctuations are taken into account. Concluding section 7 and several new references added