Weather@home analysis

Experiment Design

We conducted thousands of simulations of possible weather under three different scenarios computed by participants around the world on their home computers. These scenarios are 1) 2015 actual conditions, 2) the world that would have been in 2015 without human influence (natural), and 3) conditions over the 1985-2013 period (representing typical weather over recent decades). These scenarios are described in the following paragraphs.

The model used is the MetOffice Hadley Centre regional atmospheric circulation model hadrm3p (Massey et. al. 2014), at 50km resolution over the European region, nested in the global model hadam3p. This model is forced by sea-surface temperatures (SSTs), sea-ice extent, and greenhouse gas concentrations. The initial conditions for the simulations are from previous simulations, and the large number of ensemble members are produced by applying perturbations to the potential temperature of the initial conditions.

The actual simulations were run from November 2015 to January 2016. The were driven with observed sea surface temperatures and sea ice conditions from the OSTIA dataset (Stark et. al. 2007) for November, with a smooth transition to using seasonal forecast SSTs and sea-ice from GloSEA5 (MacLachlan et. al. 2015) for December and January.

The natural simulations were run over the same period as the actual simulations. These use the same SSTs except with estimated patterns of anthropogenic warming subtracted (as per Schaller et. al. 2014). Preindustrial greenhouse gas concentrations were used in these simulations.

The climatology ensemble consists of single year simulations from the period 1985-2013 forced by OSTIA SSTs, and observed greenhouse gas concentrations. In contrast to the 2015 simulations initialised in November, the climatology runs are initialised in December, hence the first few days of December exhibit effects from model spin ups. This is to be rectified in an updated set of climatology simulations.

The regions analysed were for Central England: land points over region 51-54N, 3W – 0E (see King et. al. 2015) and Northern England: Land points over region 54-57N 6W – 2E (see van Oldenborgh et. al. 2015). Different numbers of simuations were used from each ensemble which were Actual: 3383, Natural: 5030, Climatology: 11891.

The weather@home model, hadrm3p is forced by SSTs which to a certain extent constrain the the behaviour of the model. However, the model does still have systematic biases due to limitations in representing all of the intricacies of the climate system, and due to finite spatial resolution. See Massey et. al. 2014 for more details of the model and its biases. In this study we apply a simple bias correction method to adjust the model climatological mean of each variable to match a reanalysis product representing observations, ERA-Interim (Dee et. al. 2011). This was done by calculating the the average of the climatology simulations and the average of the reanalysis for 1985-2013 period and subtracting the difference (bias) from each ensemble member. Biases are as follows for temperature: -0.13ºC in Central England and -1.02ºC in Northern England. For precipitation: -0.22 mm/day in Central England and 0.80 mm/day in Northern England.



For temperature the weather@home simulations robustly show the actual simulations being warmer than the natural and climatology simulations. This is also the case for Northern England (not shown).

The observed temperature in Central England of 9.7 ºC is greater than a 1 in 670 year event for the actual ensemble, greater than a 1 in 3900 year event in the climatology and not recorded in any natural simulations at all. This shows that even taking into account the particular sea-surface temperatures driving the climate in 2015, this event was extremely unusual and its large amplitude was mostly due to internal atmospheric variability.

Using the example of a 1 in 100 (82 -140) year event in the actual ensemble (8.8ºC), this becomes a 1 in 192 (143 – 277) year event in the natural ensemble and a 1 in 272 (191 – 543) year event in the climatology ensemble. So this is likely a double or greater risk due to both anthropogenic climate change and the particular SSTs occurring in 2015. Note that for such a high anomaly of 5ºC, natural fluctuations of the atmospheric circulation is the largest contributor. This can be thought of as the anthropogenic signal and SST patterns making an extremely unlikely event into a rare but possible event.

Figure 8. Return time plot of December 2015 Central England Temperature for weather@home simulations. The observed CET value is also shown.


For precipitation, the weather@home simulations show a difference between the Actual and Natural simulations in Northern England but not in Central England. In both regions, the precipitation is greater in the actual simulations than the climatology, showing that the particular SST patterns in 2015 favoured higher precipitation than normal.

For a 1 in 100 (81 – 147) year event in Northern England from the Actual ensemble, the event becomes a 1 in 158 (123 – 220) year event in the natural ensemble, and a 1 in 183 (153 – 230) year event in the climatology ensemble. This corresponds roughly to a 50% increase in likelihood due to anthropogenic climate change and 75% increase in likelihood due to the SST patterns in 2015 relative to 1985-2013.

This indicates that the particular SSTs during 2015, such as the North Atlantic SSTs or the El Niño SST pattern, increased the likelihood of a wet December, and this effect was equal to or greater than the anthropogenic signal.

Figure 9. Return time plot of Precipitation for December 2015 in: a) Central England and b) Northern England, for weather@home simulations.