Report on Statistical Downscaling
Scenario generation task of SIS06:
The Threat of Dengue Fever - Assessment
of Impacts and Adaptation to Climate
Change in Human Health in the
Student: Lawrence Brown
Supervisors: Anthony Chen, Albert Owino
The purpose of climate scenarios
To provide data for impact/adaptation/assessment
To act as an awareness-raising device
To aid strategic planning and or policy formation
To scope the range of plausible futures
To structure our knowledge (or ignorance) of the
To explore the implications of decisions
To function as learning machines, bridging analyses
and encouraging participation
Basically, downscaling is any process where large (coarse)
scale output of models is reduced or made finer.
The first attempt to DownScale the global models was by
the production of regional models. These regional models
resolutions of 70 to 50 km square.
However there is still a need to go even further by
This is what is require by impact researchers.
What we require from downscalin
General Circulate Models supply...
Impact models require ...
Regional models supply
Objectives of SIS06 Downscaling:
To develop future climate scenarios so
that the climate health linkages being
developed and the impacts being studied
can be used to develop adaptaton
To contribute to the National
Communications under the UNFCCC
Stochastic DownScaling - Probabalistic
Dynamical DownScaling - running a higher resolution
RCM within a coarser resolution GCM.
Weather Typing approaches - grouping of countries with
Regression-based DownScaling - empirical relationships
between local scale predictand and regional scale
Statistical DownScaling Model (SDSM) developed
by Dr. Rob Wilby, Dr. Christian Dawson and Dr. Elaine Barrow
Downscales GCM output from course grid to island or
SDSM is a hybrid of the multiple Regression and
Stochastic DownScaling models.
The model assumes Stationarity - statistical properties of
the variable being downscaled will not change over time.
Advantages of SDSM
Cheap - not computational demanding and is
Flexible - can be tailored to target a variety of
variable e.g. storm surge.
Generates ensemble of climate scenarios - allows
risk (uncertainty) analysis.
Rapid application to multiple GCM’s.
Complementary to regional modelling
Fairly simple to use and thus accessible beyond
Disadvantages of SDSM
Dependent on the realism of the GCM boundary forcing.
DownScaling in general propagates the GCM error.
Require high quality (daily, in our case) data for
calibration of the model.
Relationship between predictor and predictand are often
Predictor variables used explain only a portion of the
Low frequency climatic variables are problematic to
Steps in downscaling
The process involves basically four steps
Gathering of the predictor (model output) and predictand
Screening: to decide which GCM to use, which variables
are most relevant, the best predictor predictand
relationships, the various locations in the different
countries to downscale, the best transfer scheme to use.
Calibrating the model and Developing results for the
different time slices.
Interpretation and Presentation of findings.
Overview of the SDSM:
Key functions of SDSM - Menu driven
‘Quality Control’ and ‘Transform’ data
‘Screen Variables’ (Selection of predictors)
‘Calibrate Model’ (Regression equations)
‘Weather Generator’ (Synthetic daily weather);
‘Analyse Data & Model Output’ (Statistics)
‘Compare Results’ (Graphs)
‘Generate Scenario’ (Synthetic daily weather
using model predictors) 11
Domain of interest - map
Domain of interest
The coordinates for the area we will be looking at is latitude
10 to 22 degrees north and longitude 55 to 90 degrees
This area is made up of 27 countries and also sections of
five other countries.
At first we will be looking at a subgroup which include
Trinidad & Tobago, Jamaica, St Kitts and Barbados
The land area represent nearly 10% of the area of
GCM outputs used and to be used:
Preliminary _ 1st generation Couple General
Circulation Model developed by the
Canadians (CGCM1)- readily available,
output formatted for use in SDSM
Primary model - HadCM3, simulates
Caribbean climate best (Santer, 2001) -
predictors being prepared in an SDSM
Emission Scenario being used - IS92a
IS92 readily available, later SRES not readily available at start of
IS92 was proposed in the 1992 Supplement (IPCC,1992) to the
IPCC First Assessment Report of 1990. Basically the IS92
scenarios ranged from a-f (IS92a-f) and considered various
factors that would affect emissions up to the year 2100.
IS92a - a middle range scenario: population reaches 11.3 billion,
convention and renewable energy sources are used. Only
emission scenarios internationally agreed on and national
policies enacted into law are included, e.g., London Amendments
and the Montreal Protocol.
SRES emission scenarios to be used
SRES emission developed after IS92 was evaluated in
1995: New knowledge relating to: e.g. carbon intensity of
energy supply, income gap between developed and
developing countries and sulphur emissions.
Special Report on Emission Scenarios (SRES) has 4
groups and 6 scenarios A1 (which has 3 sections A1FI,
A1T and A1B) and A2, B1, B2. - Only A2 and B2 readily
available for use in SDSM, others will be considered.
A2 SRES scenario:
A very heterogeneous world with an underlying theme of self
reliance and self preservation of local identities.
Fertility across regions converges very slowly and this results in
continuous increase in population.
The A2 world “consolidates” into a series of roughly continental
economic regions, emphasizing local culture and roots, e.g.,
some social and political structure moving towards stronger
welfare systems and reduced income inequity while others move
towards lean government.
Environmental concerns relatively weak. Economic growth and
technological change are more fragmented and slower than
other storylines. 18
B2 SRES scenario:
A world in which emphasis is placed on local solutions to
economic, social and environmental sustainability.
Education and welfare are widely pursued, reductions in mortality
and to a lesser extent fertility, population reaching about 10 billion
people by 2100.
Income per capita grows at an intermediary rate to reach about
US$12,000 by 2050.
This generally high educational level promotes both development
and environmental protection.
Technological are pushed less than in A1 and B1 but higher than
in the A2 scenario. 19
Work done – training and learning
Training under auspices of Caribbean/Canadian CC
Contact made with various agencies and organisations in
an effort to fill the gaps in the CSGM database.
Predictors for the HadCM3 (monthly) and CGCM1
The transformation of predictors and predictands to use
The generation of test/preliminary scenarios for Piarco
Airport in TNT, Grantley Adam Airport Barbados,
Agronomy St. Kitts and Guantanimo Bay Cuba.
Analysis to date –Piarco airport TNT
Predictants- Rainfall and Temperature
Predictors (and predictor code) used
Surface vorticity - ncepp__zna.dat
850 hpa vorticity - ncepp8_zna.dat
Specific humidity at 500 hpa - nceps500na.dat
Specific humidity at 850 hpa - nceps850na.dat
Mean temperature at 2 metres - nceptempna.dat
Analysis to date –Piarco airport (TNT) temp
Predictors used and their code
Mean seas level pressure ncepmslpna
Surface air flow strength ncepp__fna
Surface zonal velocity ncepp__una
Surface meridonal velocity ncepp__vna
Surface vorticity ncepp__zna
Surface divergence ncepp_zhna
500 hpa air flow strength ncepp5_fna
500 hpa vorticity ncepp5_zna
500 hpa geopotential height ncepp500na
850 hpa zonal velocity ncepp8_una
850 hpa specific humidity nceps850na
Near surface specific humidity spncepsphuna 22
Mean temperature at 2 metre nceptempna
SDSM Piarco rainfall validation
Sample result - Rainfall
No significant difference in the annual
mean over 90 years
SDSM Piarco temp validation
Jan 21.3096 20.4857
Feb 21.2146 20.6114
Mar 21.6342 21.1696
Apr 22.7023 22.2768
May 23.5374 23.4493
Jun 23.6557 23.4903
Jul 23.2332 23.1077
Aug 23.1371 23.2561
Sep 23.1903 22.9814
Oct 23.1481 22.9169
Nov 22.8687 22.5199
Dec 22.1423 21.5914
Sample result - temp
Jan 21.3096 20.7753
Feb 21.2146 20.7504
Mar 21.6342 21.3123
Apr 22.7023 22.2377
May 23.5374 23.1525
Jun 23.6557 23.3185
Jul 23.2332 22.9726
Aug 23.1371 22.8925
Sep 23.1903 22.8695
Oct 23.1481 22.8291
Nov 22.8687 22.475
Dec 22.1423 21.573
Assessment of Results to date
These exercises are a learning experience
No importance is placed on these results
More detailed results required to know if the
length of dry or wet spells will increase,
time of year that increases occur, etc.
Continue collection of data
Look at other GCM outputs – consider
Look at SST as possible predictor – need
AOGCM SST output
Compare SDSM results with RCM results
AREAS OF CONCERN
The SDSM uses daily climate data - Do we have
enough, high quality daily data (especially
Will we have station data at locations to
develop correlations with areas of Dengue risk?
SST plays important role in Caribbean weather
but is not on the list of predictors for HadCM3.
Only 10% of area of domain is land - GCM see
many area as ocean, may lead to damping of 30
Steps being taken to reduce problem
Agencies and met offices contacted for data by both
the climate and Epidemiology group data
In the case of Jamaica where there was a fire some
years ago, steps taken to access data on tape and
interaction with the met office has resulted in
significant data being gathered.
We are looking at using monthly time scale in case
the daily time scale is not feasible as the seem to be a
plethora of monthly data around.
Checks will be made to ascertain the level of
dampening that might occur.