How does the CMIP6 ensemble change the picture for European climate projections?

In the summer the CMIP6 ensemble projects a warmer range of temperatures than the CMIP5 ensemble in all regions. The change in projected temperature for the CMIP6 ensemble (compared to CMIP5) for summer is statistically significant ($P \lt \textrm{or} =$ 0.05, see tables S4 and S5 for KS tests in supporting material and hatching on figure 1) in the central Europe and the Mediterranean regions by mid-century (see figure 1(a)) and this difference has increased by the end of century (figure 1(b)). Projected temperature differences between MIPs in the northern Europe region are smaller for summer than the other two regions and were not found to be statistically significant. End of century projections for the central Europe and the Mediterranean regions show an increased interquartile range for both ensembles compared to mid-century (figures 1(a) and (b)).

Zoom In Zoom Out Reset image size

In the winter the differences between the CMIP5 and CMIP6 projections at the regional scale are small for all European regions. Central Europe and the Mediterranean do see higher median temperatures projected in CMIP6 (figures 1(c) and (d)), but these changes are smaller than in the summer (with the Mediterranean showing the largest increase in mean of approximately 0.5^∘, compared to about 1.5^∘ in JJA) by end of century. The overall projected range for the Mediterranean and central Europe areas are almost unchanged. No statistically significant differences were found between the ensembles for winter temperature projections in for mid-century or end of century.

CMIP6 projected temperatures are consistently higher (shown here for both lower and upper percentiles) for all land regions of Europe, (figures 2(a) and (b)). CMIP6 land temperatures for central Europe, the Mediterranean and parts of northern Europe are consistently warmer and less uncertain as they project a narrow range of future temperature changes for all but parts of the most northern land regions of Europe (figure 2(d)). More than 80$\%$ of CMIP6 projected temperatures are warmer than median CMIP5 projected changes, in these regions, (figure 2). There is a hint that CMIP6 projects a wider range of changes of the North Atlantic SSTs, which may be important given their role in driving European circulation.

Zoom In Zoom Out Reset image size

Generally the temperature increases are much smaller for winter than for summer with parts of Northern Europe slightly cooler (figure S2). Spatial maps of the end of century temperature projections for CMIP6 and CMIP5 are included in the supplementary information (figures S1 and S2).

Overall the difference between the MIPs are clearer by the end of century due to a stronger climate signal and larger signal to noise. Where there are differences between the MIPs these are usually already apparent by mid-century however.

Precipitation changes are important for central Europe and the Mediterranean regions, where further drying may have significant impacts on agriculture. In previous studies higher temperature projections for central Europe and the Mediterranean are also linked to greater drying in these regions (Lionello et al 2012, IPCC 2013, Lionello and Scarascia 2018). We show here that there is a large reduction in the projected range of precipitation change in the summer CMIP6 ensemble for northern Europe and central Europe regions (figures 3(a) and (b)), which is projected for both mid-century and end of century. CMIP6 suggests a clearer shift to drier conditions in northern Europe in summer, however, these differences can not be said to be statistically different from the prior CMIP5 distribution.

Zoom In Zoom Out Reset image size

For winter the difference in all regions between CMIP5 and CMIP6 is small, for mid-century and whilst there is a suggestion that CMIP6 projects wetter conditions in northern and central Europe by the end of the century, neither can be excluded as being statistically different (see tables S4 and S5 for KS tests in the supporting material) from the earlier CMIP5 responses by the end of century (figures 3(c) and (d)).

The spatial comparison differences for the 25th and 75th percentiles (figures 4(a) and (b)), show that the stronger drying signal in central Europe for CMIP6 also extends to parts of northern Europe, with a drier pattern from the north of Spain extending to Moscow and in to a large part of Scandinavia. This is largely due to a decrease range in CMIP6 (d) which is from the upper quantile ((c) and (d)).

Zoom In Zoom Out Reset image size

The spatial changes for winter show a small increase in precipitation in most European land regions. Smaller regions, such as the southern UK, France and the eastern Mediterranean have a small projected increase in CMIP6 (figure S4). Spatial maps for end of century precipitation changes for CMIP5 and CMIP6 are also included in the supplementary information (see figures S3 and S4).

The difference between MIPs is only statistically significant by the end of century for the combined regions. The reduced range of predictions for northern Europe in the summer is not significant. While the difference for northern Europe is not found to be significant it does also indicate more of a trend in the northern Europe of a neutral or slight drying response to regional warming in the CMIP6 projections.

For summer temperature (left panels (a)–(c), figure 5) the regional temperature response is largely a function of the global warming change, in both ensembles. The small vertical spread, for a given global temperature change, is indicative of this. For summer temperature, the warmer shift towards higher regional responses in CMIP6 relative to CMIP5 appears to be driven largely by the large global warming responses. The similar slopes and widths of the relationships in figure 5 (left panels (a)–(c)) suggest that the relationship between regional and global responses remain similar between the two ensembles for northern Europe (NEU) and the Mediterranean (MED).

Whilst individual quantiles visually differ for temperature in figure 5(d), for most regions there is no consistent shift between the two ensembles, across the quantiles. Central Europe, however, shows consistently larger normalised temperature response across all the quantiles, highlighting a change in the RC consistent with the differences in slope identified in figure 5(b). These normalised differences between MIPs were statistically significant in central Europe, by mid-century and have increased by end of century (mid-century P = 0.05 and end of century P = 0.03). What is apparent (from all panels, figure 5) is that CMIP6 explores a number of larger global warming responses compared to CMIP5, particularly at end of century.

There is more scatter in the relationship in the right panels showing the regional precipitation response to global temperature (figures 5(e)–(g)), this illustrates the greater uncertainty in the regional response to global warming. Both ensembles show a similar overall response, with summer precipitation decreasing in response to increasing global mean temperatures. The difference in slope between the two ensembles is visibly greater than for temperature indicating that there is a larger difference between MIPs in their regional responses.

The summer precipitation change per °C global warming (figure 5(h)), has similar median values for CMIP5 and CMIP6 for most regions and timescales. CMIP6 projections are largely consistent with CMIP5 in terms of central estimates of the projected changes. The most evident difference between the two ensembles is the narrow spread of the projected range for northern and central Europe in CMIP6. This smaller range of projected changes is an interesting result, as it suggests a more confident picture of future precipitation change in both regions. The differences between the two normalised ensembles in nearly all cases are not significant.

To attempt to quantify the contribution of the RC, compared to that of the global temperature change, to the total regional projected summer temperature change, we applied a simple statistical model to the end of century projections. The linear model is described by equation (1):

$$\begin{equation} Y \approx XS, \end{equation} \tag{ 1 }$$

where Y is the ensemble mean regional change, X is the ensemble mean global change (GC) and S is the slope of the linear regression line (which is a measure of the RC to GC).

The total difference ($\delta Y = Y6-Y5$) in regional change between the two ensembles can be approximated by equation (2):

$$\begin{equation} \delta Y \approx X5~\times \delta S5 + S \times \delta X, \end{equation} \tag{ 2 }$$

where:

$$\begin{equation} \delta S = (S6-S5), \end{equation} \tag{ 3 }$$

$$\begin{equation} \delta X = (X6-X5). \end{equation} \tag{ 4 }$$

The numbers given relate to the CMIP5 ($X5$, $S5$) and the CMIP6 ($X6$, $S6$) ensembles respectively.

The contribution from the change in RC is taken from $X5~\times \delta S$ and the contribution from the GC is $S5~\times \delta X$.

The actual total value ($\delta Y$) of the mean regional change between CMIP5 and CMIP6 is given by equation (5) and it is assumed that the difference $\delta Y - (RC+GC)$ is a relatively small residual this has been recorded in table S3.

$$\begin{equation} \delta Y = (Y6-Y5). \end{equation} \tag{ 5 }$$

The change in the slope of the linear fit is taken to represent the contribution of the RC to the overall regional mean change. This simple statistical model is able to capture the differences in regional mean responses, within both MIPs, to the first order (see table S3 and text S3 in the supporting material where some further details of the method and full table of the calculations is included).

We can now assess the relative importance of either differences in the global response (represented by global mean temperature differences) or the RC (captured by the figure 5 regression slope) in explaining the CMIP6/CMIP5 differences in the mean regional response, by comparing the contribution to $\delta Y$ from $X5~\times \delta S$ (the contribution from the RC), to $S5~\times \delta X$, (the contribution to the change in the global annual mean temperature, GC). Further description and details of the calculation is given in the supporting material (text S3 and table S3).

The results for temperature confirm and help quantify what is seen visually in figure 5. For the Mediterranean the higher global sensitivity of the CMIP6 models accounts for nearly all of the increase in projected temperature for this region (93$\%$). There is some RC in northern Europe (21$\%$), but most of the change is due to the annual global temperature increase (79$\%$). In central Europe a significant percentage of the total mean regional temperature anomaly (42$\%$) is due to the RC (table 1).

Table 1. Summary table of linear model results. The results from the linear fit from the model ensemble is given in the first two columns RC (contribution from regional sensitivity), GC (contribution from annual global temperature change). The median, 25th and 75th percentile of contribution of the regional sensitivity (RC) from the bootstrap resampling is also summarised.

Region	$\%$RC	$\%$GC	Median $\%$RC	25th $\%$RC	75th $\%$RC
	Temperature (tas)
NEU	21	79	21	7	32
CEU	42	58	42	35	49
MED	7	93	7	−3	18
	Precipitation (Pr)
NEU	102	−2	102	99	104
CEU	70	30	71	56	80
MED	200	−100	190	156	252

The level of scatter around the linear regression fit in figure 5 indicates that it is unclear how sensitive this result is to any sub-setting of the model ensemble. To address this a bootstrap resampling without replacement of 10 000 resamples was applied, drawing a sub-sample of 20 models from each ensemble per sample. The results are summarised in table 1. The interquartile range from the bootstrap resampling does indicate some sensitivity in the result with sub-setting (with a range of about 15$\%$–20$\%$ for the Mediterranean and central Europe). Northern Europe has a larger uncertainty range with the 75th percentile suggesting RC changes may be important for some CMIP6 models. It is noted however that the only region where the interquartile range of the contribution of the RC reaches over 40$\%$ to nearly 50$\%$ for the 75th percentile, is central Europe. This indicates that the finding of a greater contribution from the RC in central Europe is not due only to a couple of outlying models in the ensemble but is a result that appears to be consistent across the CMIP6 ensemble.

The same model was applied to the precipitation results where equation (1) was found to be a good model of mean regional precipitation response in both ensembles (see supporting material table S3). The greater scatter for precipitation illustrates the greater uncertainty in the regional response to global warming. Further this scatter indicates that changes in the mean precipitation are less informative of the changes in the broader distribution of the precipitation response. The results of the linear model can still provide some indicative insights in the relative roles of contribution of global warming and RC in the mean regional response. In this case the results showed a considerably larger contribution from RC than to the global mean temperature change. The contribution of the RC was found to be 102$\%$ for northern Europe, 71$\%$ for central Europe and 200$\%$ for the Mediterranean (with the GC acting in the opposite direction to the regional in northern Europe and the Mediterranean). It is interesting in the Mediterranean that the RC in CMIP6 results in slightly less drying than in CMIP5 (shown by a slightly less negative gradient in the slope) despite the increased global sensitivity in CMIP6 (the large percentage changes are due to the small absolute values in this region). It is the opposite case in the other two regions (where RC in CMIP6 results in further drying), although the difference in regional change between MIPs is small (see supporting information text S3 and table S3). The regional precipitation sensitivity to global temperature change has a larger degree of uncertainty than regional temperature as can be seen in the scatter and the projected ranges in in figure 5. Bootstrap resampling of this result is summarised in table 1.