Transcript Climate data and impact assessment

Downscaling and its limitation on
climate change impact assessments
Sepo Hachigonta
University of Cape Town
South Africa
“Building Food Security
in the Face of Climate change”
4 the May 2010 , ICRAF, Nairobi
(1 − a)Sπr2 = 4πr2εσT4
GCMs
Primary source of information on climate change projections
 Incoming and outgoing radiation
 Wind, Temperature, humidity etc..
 Clouds formation
 Precipitation falls
 How ice sheets grow or shrink, etc.
 Feedback processes
Horizontal resolution of about
300km and 10 to 30 vertical
layers
GCMs
Downscaling
Process of generating higher resolution data or climate change information from
relatively coarse resolution GCMs relevant for adaptation and policy
Two main stream methodologies
Dynamic downscaling / Regional Climate Models (RCMs)
(e.g. RegCM, high resolution PRECIS)
Statistical /Empirical downscaling
• Weather typing (SOMD – University of Cape Town)
• Linear (and nonlinear) regression (SDSM – Rob Wilby)
• Artificial Neural Networks
• Weather generators
RCMs
• Essentially a model like a GCM but at
higher resolution and over a smaller finite
domain
• Uses a GCM to establish the boundary
fields of the RCM
• The RCM derives a dynamic solution at
higher resolution, which is physically
consistent with the larger scale circulation
of the forcing GCM
Image courtesy of the
UK Met. Office
(htp://www.metoffice.gov.uk).
RCMs
Pros:
• Accounts for sub-GCM grid scale forcing (e.g. topography)
• Information is derived from physically based models
• Better representation of some weather extremes as
compared to GCMs
Cons:
• Expensive to run RCMs as compared to statistical
downscaling over a large region
• Its dependence on GCM predictors
• It is a spatially smoothed product compared to station
scale
Statistical downscaling
• Involves the development of quantitative relationships between large
scale atmospheric variables (predictors) and local surface variables
(predictands)
Example: SOMD
• A Self Organising Map (SOM) is used to recognise commonly
occurring patterns within multi-dimensional data sets
• Identify modes of circulation over a particular region with each
circulation mode being associated with an observed precipitation
probability density function (PDF)
Calculate probability of rainfall for each
synoptic pattern
Statistical downscaling
Pros:
• Efficient and cheap computation requirements
• Its ability to provide point resolution climatic variables
from GCM outputs
• Its ability to directly incorporate observations
Cons:
• High dependence on the predictors
• Vulnerability to non-stationarity of the cross scale
relationships
Climate data and impact assessment
Downscaling does NOT seek to reproduce the real world
observation , but rather generate a realistic time evolution
that:
• At seasonal and inter-annual scales should match relative
magnitude of the temporal evolution of the forcing
• At daily time scales should match the statistics of the daily
events (e.g. frequency of events, etc)
Climate data and impact assessment
FACT : There will only be one time evolution into the
future, but many possible evolutions
Limitations include:
• Imperfect ability to model our knowledge into accurate
mathematical equations: e.g. physics, knowledge gaps
etc…
• Data formatting techniques (e.g. different spatial
resolution of systems)
Climate data and impact assessment
• Imperfect observation data
Data type
Provincial, Catchment, Station, Gridded
Station
Spatially
average
station yields over
each region and then
compare to observed
data.
Data estimation and uncertainty
Penman Montieth
FACT: Society cannot wait for perfect
models (GCMs and impact) and methods
• We need to make choices today based on the best current
scientific information
• We need to characterize baseline observational climate as
best as possible
• We need to use as many models as possible
• We need to downscale or upscale where possible
Thank you