The thermal dependence of development rate for eggs, larvae and pupae was determined from the offspring of 10 field-collected females, raised individually on the grass Ehrharta erecta (Poaceae) from egg to emergence in glass vials (height 130 mm, diameter 25 mm) sealed with a moistened foam stopper. Temperature was controlled at 8°C, 12°C, 15°C, 25°C or 30°C with a 12 : 12 light : dark photoperiod. We observed the animals daily and recorded time to completion of each life cycle stage, the inverse of this being the developmental rate. We then fitted polynomial regressions to the relationship between development rate and temperature for each life stage, using the adjusted r² method (Quinn & Keough 2002).

We used historical weather data for 1945–2007 (Bureau of Meteorology, Australia) from Laverton (37.86° S, 144.76° E), a rural site close to Melbourne, to model the physiological response of H. merope to temperature. This station is a ‘high-quality’ site, unaffected by changes in exposure, urbanization, instrumentation, etc., during the study period. Weather records (mean monthly maximum and minimum air temperature, wind speed and cloud cover) were translated into microclimates experienced by immature H. merope using biophysical modelling software (Niche Mapper, http://www.zoology.wisc.edu/faculty/Por/Por.html#niche; Porter & Mitchell 2006). This software includes a microclimate model that translates weather-station records into near-surface air temperature and wind speed profiles and a soil surface energy balance (Porter et al. 1973). The model includes a first-principles model of solar radiation (McCullough & Porter 1971) and uses cloud cover records to predict long- and short-wave radiation loads. Individuals were assumed to be 3 cm above the ground in grass tussocks (75% shade), based on our observations of larval behaviour. Microclimate model predictions compared well with field microclimate measurements (figure S1, electronic supplementary material).

Animals were assumed to be at the 3 cm air temperature. Predicted daily cycles in animal body temperature were converted to constant temperature equivalents (CTEs) (as implemented in Mitchell et al. 2008), which were then used as independent variables in the fitted equations for thermal dependence of development rate. There was a good agreement between modelled CTE-based development times and the development times measured under natural, fluctuating thermal conditions in grass tussocks in Melbourne over the winter of 2007 (table S1, electronic supplementary material). We encoded the equations for development rate and CTE calculations into the Ectotherm model of Niche Mapper. The Ectotherm program tracked developmental stage on an hourly basis given a specified laying date (11 April—females lay from around March to April in Melbourne; Pearse 1978) and outputted the time of completion of development based on observed meteorological data from 1944 to 2005. We calculated separate emergence dates for each year and then took the average value for each 10-year window as the final emergence date. Note that we did not solve an energy balance specifically for the caterpillars with the Ectotherm model but rather assumed that they were at the shaded 3 cm air temperature.

The thermal response of development rate varies considerably across the life stages (no fifth instars survived at constant temperatures at or above 25°C) (figure 1c). Combining the physiological functions of thermal sensitivity with the microclimate model and historical monthly climate data, we predicted a shift in emergence date of −1.5 days per decade (R² = 0.866, p = 0.024), close and statistically indistinguishable from the observed rate of −1.4 days per decade (heterogeneity of slopes test, p > 0.944; figure 1a). Importantly, the physiological model provided a better prediction of emergence date (R² = 0.652, p = 0.028) than did air temperature (R² = 0.426, p = 0.112). While the modelled H. merope temperatures are mainly a function of radiation, wind speed and air temperature, sensitivity analyses showed that most (80.6%) of the predicted shift in emergence date was driven by changes in air temperature. Very little (less than 1%) was independently attributable to wind speed and cloud cover, but the predicted effects of the latter two variables were inversely correlated (r = −0.65, p = 0.01).

In our model, egg, caterpillar and pupal temperatures were a function of air temperature and also of radiation (via cloud cover) and wind speed. Cloud cover and wind speed variation did explain part of the predicted shift in emergence date when considered in isolation but had no net effect because they acted in opposite directions. As a consequence, air temperature shifts were seen to dominate the predicted phenological response in this particular case, but it should be clear that this need not necessarily be so. In other cases, concurrent changes in other environmental variables may suppress or exacerbate the effects of air temperature and such complexities can only be considered through a mechanistic analysis of microclimatic impacts on phenology.

The observed shift in air temperature of 0.16°C per decade in the vicinity of Melbourne can very likely be explained through the effects of greenhouse gases emitted by humans (figure 1d). Our analysis thus provides direct causal linkages between the emission of greenhouse gases by humans, a shift in local air temperature, and the physiological response of a butterfly resulting in earlier spring emergence.