Transcript Climate Change Experiment - International Meetings on Statistical

Detecting change in UK extreme
precipitation using results from the
climateprediction.net BBC Climate
Change Experiment
11th International Meeting on Statistical Climatology
Edinburgh University, UK
15th July 2010
Hayley Fowler (Newcastle University)
Dan Cooley (Colorado State University), Steve Sain
(NCAR), Milo Thurston (Oxford University
Aims
 To estimate how seasonal extreme
precipitation will change for regions across the
globe using results from the BBC
climateprediction.net experiment
 To explore when changes in extreme
precipitation due to climate change will be
detectable
 To explore which parameters of climate models
affect extreme rainfall simulation and
detectability
Defining “detection”
 Detection is the process of demonstrating that
climate has changed in some defined statistical
sense without providing reasons for the change
 Change is detected in observations when the likelihood of
an observation (e.g. an extreme temperature) lies outside
the bounds of what might be expected to occur by chance
 Changes may not be found if the underlying trend is weak
compared with the “noise” of natural climate variability;
conversely there is always a small chance of spurious
detection (e.g. outlier at end of the observational record)
The ‘BBC’ Climate Change
Experiment (climateprediction.net)
1.
2.
The HadCM3L coupled
atmosphere-ocean global
climate model was run in
‘grand ensemble’ mode for a
control and transient
integration (observed
forcings applied from 1921
and A1B emissions scenario
from 2000) for 1921 to 2080,
varying 34 parameter values
and initial conditions
There are now more than
5000 complete matching
control/transient runs
available
Data Available

EA Rainfall and Weather
Impacts Generator
 Developed for EA as a catchment
scale Decision Support Tool
 Generates series of daily rainfall,
T, RH, wind, sunshine and PET
on 5km UK grid
 Based on UKCIP02 scenarios
 Combines NSRP rainfall model
with CRU Weather Generator
 Increasingly used throughout UK
 Being used as part of UKCIP08
scenarios
UK Case Study
 8 UK Grid Cells
 Extract 1 day monthly maximum
precipitation for 1921-2080 for 304
model pairs (complete models from
524 model set used at Oxford for
mean temperature change analysis)
 Calculate time series of seasonal
maxima for each model pair and
grid cell
hl
ni ne
ei wm la
cw kt
Structure of statistical model
To establish the structure of the statistical model needed to
fit the data several combinations of the GEV tested:
1. Control has no forcings applied:
 time-invariant GEV model fitted
2. Scenario 1920-2000 has observed forcings and 20002080 has SRES A1B forcings applied:
 We assume GEV location parameter, μ, most likely to change – but
use a model selection exercise that increasingly adds complexity (i.e.
no change, level shift, linear trend in μ from 1920, change in linear
trend in μ from 2000) – and use AICc criterion to determine best fit
 We also test whether allowing σ and ξ to vary improves model fit
Results of model selection
 Different behavior between seasons:
 winter and spring tend to choose hinge-type models (change in
linear trend in μ from 2000)
 summer and autumn - common model for the Control and
Scenario runs more likely to be selected (less evidence of change
in extreme precipitation)
 Marginal improvement to model from inclusion trend in
σ and ξ – therefore restricted to μ
 Best GEV model used in each case to provide estimate
of 20y RP for each year of the time series
Projected changes in extreme
precipitation

percentage
changes to 20y
return level of 1d
winter extreme
precipitation by
2020, 2050 and
2080 from 1961-90
projected by CPDN
BBC CCE
Testing for detectibility
 Detectible increase in extreme precipitation defined as
the year at which we would reject (at the α = 0.05 level)
the null hypothesis that the 20-year return levels from
the two runs are equal in favor of the alternative
hypothesis that the 20-year return level from the
Scenario run is greater than that from the Control run
(1) Hinge-type model:
Detection time = 84 years
(2) Linear model:
Detection time = 71 years
Results for detectability
 For the winter season, more than 50% of the
climate model pairs found a detectable difference
by 2010
 For the spring season, more than 50% of the
climate model pairs found a detectable difference
by 2030
 However, for the summer and fall seasons
respectively, only ∼30% and 40% of the data sets
showed a detectable difference by 2080
Climate model parameter effects
GEV shape parameter (ξ)
 Not well understood which climate model parameters could
possibly affect ξ (affects tail of extreme precipitation
distribution)
 Ran sequence of simple one-way analysis of variance
(ANOVA) tests for each of the 34 different model
parameters for Control runs only - ξ serves as our variable
and the different parameter settings serve as treatment –
looked at p-values and also “importance”
 Two climate model parameters, the entrainment coefficient
(entcoef) and the icefall speed (vf1), have an important
effect on the tail weight in summer, explaining respectively
35% and 9%of the total variability found in the estimates of ξ
Climate model parameter effects
GEV shape parameter (ξ)
 Previously found to affect climate sensitivity
 Entrainment coefficient affects how air is diluted in rising
cumulus cloud columns, thus partially controls the amount of
convective activity – suggests precipitation efficiency must
play a critical role in the occurrence of heavy precipitation (as
Wilson and Toumi, 2005 suggest for observations)
 Ice fall speed has a major impact on cloud cover and cloud
optical properties - reducing this parameter results in
increased long-wave clear sky and increased low-level layer
clouds, allowing air to remain moister - causes simulation of
increases in extreme precipitation in the climate model output
Climate model parameter effects
Detectability
 Contingency table analysis used to determine effect on
detectability - run only for winter season
 Two climate model parameters seem to affect the time of
detection: “ct” – the accretion constant, and: “anthsca”, which
describes the scaling factor for emissions from anthropogenic
sulfur aerosols
Conclusions
 Climateprediction.net GCMs suggest that most UK
regions will experience increases in extreme 1 day
precipitation, particularly in winter
 Models suggest that changes are likely to be detectable
in the near future (by 2050)
 changes are more likely to be detectable in winter than in
other seasons
 Two climate model parameters have an important effect
on the tail weight in summer, and two others seem to
affect the time of detection in winter
 Climate model simulated extreme precipitation has a
fundamentally different behavior to observations,
perhaps due to the negative estimate of ξ
Fowler, H.J, Cooley, D, Sain, S.R and Thurston, M. 2010. Detecting
change in UK extreme precipitation using results from the
climateprediction.net BBC Climate Change Experiment. Extremes, 13(2),
241-267, doi:10.1007/s10687-010-0101-y.
Fowler, H.J. and Wilby, R.L. 2010. Detecting changes in seasonal
precipitation extremes using regional climate model projections:
Implications for managing fluvial flood risk. Water Resources Research, 46,
W03525, doi:10.1029/2008WR007636.