Causes and uncertainty of future summer drying over Europe

Abstract

Previous studies have shown that the continuing rise in anthropogenic emissions is likely to cause continental European summers to become substantially drier over the coming century. Since this predicted decline in rainfall and soil moisture (SM) would bring significant stress to society and ecosystems, it is essential that its reliability or otherwise is properly assessed. One approach is to gain a better understanding of the model’s mechanisms of regional climate change and integrate this with a knowledge of the model’s strengths and weaknesses. Here we propose a methodology that partitions some of the mechanisms of regional climate change, and apply it to the problem of summer drying over continental Europe. Earlier work suggests that a plausible partition of the mechanisms of future mid-latitude continental summer drying might be as follows: (a) an earlier and more rapid decline in SM during spring, leading to lower SM in summer, and hence less convective rainfall (‘Spring SM’); (b) a larger land–sea contrast in lower tropospheric summer warming, leading to reduced relative humidity in air advected onto the continent, and so reduced rainfall (‘Warming’); (c) other large-scale atmospheric changes, including remotely forced circulation changes (‘Large-Scale’); and (d) a positive feedback mechanism in summer, whereby the reduced rainfall dries the soil further, so reducing convective activity further (‘Summer SM Feedback’). We attempt to isolate these mechanisms by integrating a geographic subset of the high resolution global atmospheric model HadAM3P to assess their relative importance in generating the projected European summer drying. Each mechanism is approximately represented (and so isolated) using an appropriate mix of inputs to the model, with some matching a control integration and others matching a future scenario integration. These mixed inputs are: atmospheric composition (CO₂, aerosol and ozone), surface boundary data (SM and SSTs), and lateral boundary data (temperature, moisture, winds, and surface pressure). We describe this methodology and the experimental suite in some detail, as well as the constraints on our ability to fully separate these mechanisms. It is also shown that the separation of mechanisms is not compromised by interactions between them. For continental and southeastern Europe, it is found that both the ‘Warming’ and ‘Spring SM’ mechanisms are the primary drivers of the projected summer drying. ‘Summer SM Feedback’ plays an important secondary role, and ‘Large-Scale’ mechanisms, as represented here, have least influence. Since the two dominant mechanisms depend on processes in which we have reasonable confidence, this gives us high confidence in the sign of the projected summer drying over continental and southeastern Europe. Nevertheless, uncertainties in model formulation and future anthropogenic emissions mean that the magnitude of this future rainfall anomaly remains unclear. Over Great Britain and southern Scandinavia, our experiments show that the rainfall anomaly is dominated by opposing effects from the ‘Warming’ and ‘Large-Scale’ mechanisms, which in this area dictate increased and decreased rainfall respectively. Given this rivalry, and also that we have low confidence in the ‘Large-Scale’ mechanism, this suggests that even the sign of the projected drying here is uncertain.

Appendix A: Experiment details

1.1 A.1 Model description

Two minor details are noted here:

• 1. The only difference in the formulation of HadAM3P-Eur compared to its global parent model (HadAM3P), is that Fourier filtering can no longer be used to maintain stability at high latitudes, due to the restricted longitudinal domain. Instead the timestep between dynamics calculations is halved (but the timestep between physics calculations remains unchanged).

• 2. The application of lateral boundary data in HadAM3P-Eur is identical to that of Jones et al. (1995), except that an ‘inflow-outflow’ condition is applied to moisture and temperature variables in a similar approach to that of Giorgi et al. (1993). This minimises noise generated by inconsistencies at the outflow boundaries, which is otherwise found in the ‘Cntl+W’ and ‘Cntl+WS’ experiments, primarily in winter.

1.2 A.2 Isolating the impact of warming anomalies

Here we list a number of points of detail in the design of experiments that include the ‘Warming’ mechanism.

• 1. As noted in Sect. 2.3, aerosol concentrations at the lateral boundaries are taken from the global A2 integration. This provides mean aerosol flux anomalies across the boundaries that are consistent with the altered emissions. However, a slight weakness in experiments without the ‘Large-Scale’ mechanism is that daily synoptic variations in this aerosol flux are inconsistent with the daily circulation patterns, since the latter derive from the control integration. Nevertheless, further adjustment of the boundary aerosol data (to match their statistics with the A2 integration, and their daily anomaly patterns with the control integration) was judged to be unnecessary, given that the largest flow into the region is from the west where concentrations are low.

• 2. The annual cycle of 30-year mean temperature anomalies, used at the lateral boundaries, is computed at 6-hourly resolution for the 360 model days. This is then smoothed (over the 360 days) separately for each phase of the diurnal cycle with a low-pass Chebychev filter, with amplitude cutoff at 60 days, to remove most of the noise due to sampling errors.

• 3. Note that only the mean temperature anomaly is altered at the lateral boundaries. The more complex changes in the higher moment statistics (i.e. interannual and eddy components) are included in the ‘Large-Scale’ mechanism. This allows the processes included in each mechanism to have broadly consistent levels of uncertainty (low uncertainty for ‘Warming’, high uncertainty for ‘Large-Scale’; Sect. 5.1), aiding our assessment of the overall uncertainty of predicted summer drying. Nevertheless, the change in temperature variability is likely to be small at the maritime inflow (western) boundary, since the variability of SSTs is unaltered in these experiments (Rowell 2005).

• 4. The SM data used to force the model is taken from HadAM3P-Eur ‘Cntl’, not the global HadAM3P control, to provide greater consistency when the ‘Cntl+W’ and ‘Cntl’ experiments are compared.

• 5. From a technical point of view, the model is most easily forced with ‘Cntl’ SM by linearly interpolating monthly mean ‘Cntl’ SM data to daily values as the model integration proceeds. However, this approach introduces three additional changes to the model’s inputs that then contribute to our measurement of the ‘Warming’ mechanism for technical rather than scientific reasons, when it is measured by the difference between ‘Cntl+W’ and ‘Cntl’. First, daily SM anomalies (about the linear fit between monthly means) are removed. Second, the process of linear interpolation to daily values does not perfectly preserve the monthly mean SM, introducing a slight negative bias in the amplitude of the mean annual cycle and interannual variance (c.f. Sheng and Zwiers 1998). The mean absolute error in the 30-year seasonal mean SM from this source is nevertheless less than 10% (and usually less than 5%) of the SM difference between ‘Full-A2’ and ‘Cntl’. Third, it is conceivable that the removal of two-way coupling between SM and the atmosphere could itself lead to systematic differences. However, other assessments of this possibility in the context of SST forcing have found it to be negligible (S.J. Brown and J.R. Knight, personal communication, 2001). Thus, these three additional changes are likely to have only a minor impact on our measurement of the impact of ‘Warming’. Note that they do not contribute at all when ‘Warming’ is measured from the difference between ‘Cntl+WLS’ and ‘Cntl+LS’, since in this case both experiments use specified SM.

1.3 A.3 Isolating the impact of large-scale anomalies

Here we list some fine details in our experimental representation of the ‘Large-Scale’ mechanism, noting that their contribution to its role is likely to be rather small.

• 1. In the experiment ‘Cntl+LS’, synoptic variations of the aerosol flux across lateral boundaries are sourced from the global control integration, whereas the daily circulation patterns are sourced from the global A2 integration. A correction procedure for this low-impact inconsistency was however judged to be unnecessary, as discussed in 1st list entry in Appendix A.2.

• 2. We note that the ‘Large-Scale’ mechanism also includes the impact of changes to cloud moisture generated by the global model at the boundaries of HadAM3P-Eur.

• 3. We note that interannual anomalies of the lateral boundary data in experiments that include ‘Large-Scale’ mimic those of the global A2 integration, whereas in other experiments they mimic those of the control integration.

• 4. Interannual anomalies of SM are manipulated such that their pattern at each soil level becomes the same as that of ‘Full-A2’ for consistency with the circulation anomalies. In experiments without ‘Spring SM’ their 30-year mean for each month remains identical to that of ‘Cntl’, and their distribution (rough shape and extremes at each location and month) also remains close to that of ‘Cntl’. Their treatment in experiments that include ‘Spring SM’ is described in 2nd list entry in Appendix A.4.

• 5. We note that the ‘Large-Scale’ mechanism includes the impact of ‘Warming’ and SM changes within Europe on the large-scale circulation beyond Europe.

1.4 A.4 Isolating the impact of spring soil moisture anomalies

Two points of detail in the representation of the ‘Spring SM’ mechanism are as follows:

• 1. If a monthly SM value during June to October exceeds the range spanned by all 12 monthly values from both ‘Cntl’ and ‘Full-A2’ for that geographic location and soil layer, then the SM is constrained within this range. This prevents unreasonable values being used to force the model. The reduced anomaly is then persisted for the remainder of the season, or further reduced if necessary. In effect this allows some summer drainage of the component of SM arising from the May anomaly (usually corresponding to a reduction in actual drainage, since these anomalies are usually negative).

• 2. Interannual anomalies of SM are adjusted as follows. In wintertime, this is only necessary if ‘Large-Scale’ is not included, since this implies the circulation anomalies derive from ‘Cntl’ (whereas the SM derives from ‘Full-A2’). In this case the pattern of interannual SM anomalies is altered to follow that of the ‘Cntl’ experiment, but their distribution (mean, rough shape and extremes at each geographic location) remains close to that of the ‘Full-A2’ experiment. In summertime, their distribution (shape and extremes) is scaled to lie between (or close to) that of ‘Cntl’ and ‘Full-A2’, dependent on where the new 30-year mean SM lies relative to the control and A2 means. Their pattern in summer must be consistent with the circulation anomalies, so whether it mimics that of the ‘Full-A2’ or ‘Cntl’ SM depends on whether or not ‘Large-Scale’ is included.

1.5 A.5 Isolating the impact of summertime soil moisture feedback

Two further issues concerning the ‘Summer SM Feedback’ mechanism are as follows:

• 1. The minor effects noted in 5th list entry in Appendix A.2 of decoupling the SM model are reversed. Thus, daily SM anomalies are re-introduced, the small bias in the annual cycle and interannual variance of monthly mean SM is removed, and any systematic bias due to the lack of two-way interactions is removed.

• 2. We emphasise that this mechanism (as its name suggests) estimates the role of SM feedbacks only in summer. The full amplitude of future SM for the remainder of the year is included in the ‘Spring SM’ mechanism.