Assessment of regional climate change & effectiveness of pattern scaling approaches

Deliverable Report 2.4 by Craig Wallace and Tim Osborn of the Climatic Research Unit at the University of East Anglia and Jason Lowe and Ban Bernie of the UK Met Office Hadley Centre.

Pattern scaling is a widely-used method for creating multiple climate change scenarios with
which to investigate climate change impacts. Although generally sufficient to capture the
climate-change response of General Circulation Models (GCMs) over land, the technique has
known limitations and its apparent accuracy is also sensitive to the validation data set to
which it is compared (especially the ensemble size).

For this deliverable we examine and quantify the effects upon performance of the ‘ClimGen’
pattern scaler in emulating a specific GCM (HadgEM2-ES under RCP8.5 forcing) arising from
both the validation data ensemble size and potential non-linearities in the climate change
response of the GCM. We especially investigate the emission-dependence of the patternscaler
– that is investigating whether patterns diagnosed from GCMs under some emission
scenarios are acceptable for use in pattern scaling to emulate other (namely high-end)

We show here that sensitivity of performance to the ensemble size of the validation data set
is large, as use of multiple ensemble members (limited to 4 in this work) strengthens the
true climate change signal which the pattern scaler is attempting to reproduce. Even
accounting for this fact, however, we identify degradation in the pattern-scaling
performance when mean global warming approaches 3.5C. This can be attributed to a
tendency for patterns to underestimate the GCM warming over many land areas. Of note to
the HELIX community is the relative superior performance of the pattern diagnosed from
the high-end (RCP8.5) GCM simulations for generating pattern-scaled projections for global
warming of more than 3.5C in comparison to other patterns including ClimGen’s default
pattern which would typically be used in generating scenario data.

Re-calculating performance metrics whilst spatially limiting the pattern-scaled and
validation data to exclude regions where HadGEM2-ES is known to exhibit non-linear
climate responses shows little affect upon apparent performance. This is unexpected,
especially for global warming levels of 6C. The seemingly small influence may be a
consequence of our limited data pool for high-end warming (just one ensemble member for
validation) whereas lower warming thresholds lie within the scope of the full ensemble of
four members.

Download Deliverable Report 2.4 Effectiveness of pattern scaling