Transcript Caribbean Community Climate Change Centre - precis

The PRECIS regional climate
modelling system and an example
of its use
David Hein, Met Office Hadley Centre for Climate Change,
Exeter, UK
© Crown copyright Met Office
Local to regional – this
the scale at which
much of the climate
change related
information is most
needed
Continental – the scale of
much of the reliable
information coming from
Global Climate Models
(GCMs)
© Crown copyright Met Office
Why resolution is important (example)
© Crown copyright Met Office
Why resolution is important (example)
© Crown copyright Met Office
Winter precipitation over Great Britain
© Crown copyright Met Office
Higher resolution models are
needed in order to simulate
tropical storms / hurricanes
© Crown copyright Met Office
What is PRECIS?
• Providing REgional Climates for Impacts Studies
• Is a Regional climate modelling (RCM) system which can be run over
any region of the Earth
• Generates detailed climate change projections, and runs on a PC
© Crown copyright Met Office
Why PRECIS came to be
• The majority of climate models run on
supercomputers. Supercomputers are
expensive to purchase and to run
•Personal computers (PCs) are readily
available and have become much more
powerful in the past 10 years.
© Crown copyright Met Office
The Components of PRECIS
• PC version of the Hadley Centre’s HadRM3P Regional Climate
Model
• resolution 50km (25km for small areas)
• runs on the free Linux operating system
• Easy to use Graphical User Interface to set
up RCM experiments
• Data processing and display software
• Boundary conditions (input data)
• Training workshop and materials
• Technical Support
• The PRECIS web site and email address:
• http://precis.metoffice.com
• [email protected]
© Crown copyright Met Office
PRECIS: Workshops and Projects
 Over 200 trained users from over 60
countries
 Extensive regional networks in developing
countries (and some developed countries)
across the globe
 Local affiliated institutions:
 Cuban Institute of Meteorology (INSMET)
 Caribbean Community Climate Change Centre (CCCCC, Belize)
© Crown copyright Met Office
Daily weather events from an RCM
Daily
precipitation
and surface
pressure
over New
Zealand
© Crown copyright Met Office
PRECIS and extreme precipitation: a
case study
From “The Representation of Extreme Precipitation in the PRECIS
Regional Climate Model, Masters dissertation for David Hein (2008).
© Crown copyright Met Office
Experimental Purpose
• Extreme Precipitation can have a severe impact on
human life and livelihood
• Policy makers and planners have an interest in
determining how the frequency and intensity of extreme
precipitation could change in the coming century
• PRECIS can be used to generate detailed climate
projections which will include extreme precipitation
• It is thus important to establish whether PRECIS can
realistically simulate detailed extreme precipitation
• “Essentially, all models are wrong, but some are useful.”
-- George E. P. Box, Professor Emeritus, University of
Wisconsin-Madison
© Crown copyright Met Office
Experimental Set-up
• The PRECIS RCM (HadRM3P) was run over
four different areas of the world, each featuring
differing characteristics and influences on
climate. Output data from the RCM was then
compared to historical records of rainfall
amount (also called “observations”).
• The model was run between December 1958
and December 1999 over each region
• The input data was from the European Centre
for Medium Range Weather Forecasting “ERA40” quasi-observational data set (i.e. reanalysis
data)
© Crown copyright Met Office
Experimental Set-up (Domains)
Europe
Southern
Africa
© Crown copyright Met Office
USA
South
Asia
Experimental Set-up
• Precipitation varies in quantity (some days it
rains more than other days)
• Precipitation varies in time (it’s not always
raining)
• Precipitation varies in space (rain is a localised
weather phenomenon)
• Extreme precipitation is, by definition, a
relatively rare occurrence
• These factors must be taken into consideration
when comparing PRECIS output data against
historical observations of precipitation
© Crown copyright Met Office
Indices for Comparison
• Index 1: Multi-annual seasonal mean
precipitation
• Provides a “big picture” of how well PRECIS
simulates precipitation vs. historical
observations (i.e. what actually occurred)
• Seasons are abbreviated via the first day of the
month: DJF, MAM, JJA, SON.
© Crown copyright Met Office
Indices for Comparison
• Index 2: Wet day intensity
• Wet day intensity is defined as the multi-annual
seasonal mean of precipitation on “rainy days”.
A “rainy day” (also called “wet day”) is a day
which there is more than 0.1mm of rain.
• Allows for comparison of the total amount of
rain which PRECIS produces in comparison to
how much rain actually occurred (on “rainy
days”)
© Crown copyright Met Office
Indices for Comparison
• Index 3: Wet day frequency
• Wet day frequency is percentage of the total
days in which precipitation occurs
• Allows for comparison between how often
PRECIS produces precipitation vs. how often
precipitation occurred in historical records
© Crown copyright Met Office
Indices for Comparison
• Index 4: Extreme Precipitation
• Allows for comparison of the times when
PRECIS produces rainfall in the upper 5% vs.
the upper 5% of historical observations.
• Useful to gauge how well that PRECIS
simulates extreme precipitation
© Crown copyright Met Office
Indices for Comparison
• Index 5: Pooled Extreme Precipitation
• Because Extreme Precipitation is a rare occurrence,
the sample size is often insufficiently large to obtain
statistically significant results
• Spatial pooling considers values from neighbouring
grid boxes as sampling from the same precipitation
population due to being close together. This
increases the sample size and the “signal” of
extreme precipitation can be boosted.
© Crown copyright Met Office
Further information
• The Bias is the difference between the areaaveraged values of the PRECIS output data
and the observations.
• Pattern Correlation is a value which quantifies
how well that PRECIS matches the
observations spatially (i.e. is PRECIS producing
rain in the same areas that the observations
show)
© Crown copyright Met Office
Further information
• Descriptive terms are used to give a clear
assessment of how well the model is doing:
© Crown copyright Met Office
Europe
• For DJF/JJA/SON seasonal means, the model produces
values which are Good in comparison to observations in
quantity (Bias) and upper Fair to Good spatial correlation
(Patt Corr).
• MAM shows Fair bias and spatial correlation - PRECIS
produces 32% more precipitation than observations
© Crown copyright Met Office
Europe
• For JJA seasonal mean, PRECIS (left) simulates both the
amount and spatial distribution well in comparison to
observations (right).
© Crown copyright Met Office
Europe
• For MAM seasonal mean, PRECIS (left) produces too much
precipitation over mountainous areas in comparison to
observations (right). This is a common problem for models.
© Crown copyright Met Office
Europe
• For Wet Day Intensity, PRECIS is not producing enough
rainfall on “wet days” in comparison with observations, and
spatially matches Fair to Poor.
• For Wet Day Frequency, PRECIS is producing too many
days when precipitation occurs, although it captures the
spatial distribution well.
© Crown copyright Met Office
Europe
• PRECIS produces rainfall too often in
comparison with observations …
© Crown copyright Met Office
Europe
… But not enough when it does rain. There are
too many light rainfall days.
© Crown copyright Met Office
Europe (extremes)
• PRECIS simulates Extreme Precipitation for DJF and SON very well
(in amount and location)
• JJA shows a very low model bias, but Poor pattern correlation,
indicating that PRECIS is producing very close to the amount of
extreme rainfall for this season, but not in the correct locations
• Spatial pooling boosts pattern correlation at the expense (in some
cases) of an increase in model bias
© Crown copyright Met Office
Southern Africa
• The overall picture for Southern Africa is that PRECIS
produces too much precipitation (high biases)
• PRECIS does better when comparing wet day intensity, but
pattern correlation is Poor.
• For wet day frequency, pattern correlation is Good, but again
PRECIS is producing rain about twice as often as occurred
in the observations
© Crown copyright Met Office
Southern Africa
• This plot shows annual mean area averaged precipitation for
PRECIS and the observations. PRECIS gets the trends correct,
but is producing roughly twice as much precipitation.
© Crown copyright Met Office
Southern Africa (extremes)
• In general, pattern correlations are better than for the
other indices (WDI, WDF, etc), implying that PRECIS is
doing better at simulating the upper tail of the distribution
• Pattern correlation for MAM and SON is Poor, although
the biases are Fair or Good. For these seasons,
PRECIS is performing better in the amount of extreme
precip being produced, but not where it occurs.
© Crown copyright Met Office
Southern Africa (extremes)
• DJF pooled extreme precipitation shows PRECIS
reproducing the amount of spatial distribution (the
east-west difference in extreme precipitation)
© Crown copyright Met Office
Continental USA
• Overall, PRECIS performs well with Good or high Fair values
for both bias and correlation.
• PRECIS is slightly too wet in DJF and MAM and slightly too
dry in JJA and SON.
© Crown copyright Met Office
Continental USA
• PRECIS produces too much precipitation in the
mountainous west, but overall does well in
capturing the spatial patterns of wet day intensity
© Crown copyright Met Office
Continental USA
• PRECIS performs very well in capturing the spatial
distribution of wet day frequency for JJA. Example:
the observations show it rains almost everyday in
Florida, and rarely in California. PRECIS reproduces
this feature.
© Crown copyright Met Office
Continental USA
• Annual mean precipitation produced by PRECIS matches the trends
of the observations very well
© Crown copyright Met Office
Continental USA (extremes)
• Pattern correllation for pooled Extreme precipitation is
excellent, as well as biases for JJA and SON. In these
seasons, PRECIS simulates Extreme Precipitation
well, both in amount and spatially
© Crown copyright Met Office
Continental USA
• For DJF pooled extreme index, the pattern correlation
is extremely good -- PRECIS is producing extreme
precipitation in all the right places. However, it
produces too much extreme precipitation in the
southeast and mountain west
© Crown copyright Met Office
Conclusions
• Overall, PRECIS showed better performance in
time and location in the regions in which largescale (i.e. frontal) precipitation dominates (the
USA and Europe) than in the regions in which
convective rainfall dominates (South Africa and
India)
• Mountainous areas were problematic for
PRECIS in that it tended to overestimate rainfall
in these areas
• Important to note that a single index of model
performance (e.g. multi-annual seasonal mean)
can hide many problems and is therefore no
sufficient for a thorough model validation.
© Crown copyright Met Office
End
© Crown copyright Met Office