Statistical Downscaling of General Circulation Models to Produce Climate Change Scenarios for Halifax, NS

1
Statistical Downscaling of General Circulation
Models to Produce Climate Change Scenarios for
Halifax, NS

• Lee Titus BSc (physics), Dmet, MSc Candidate
  (physics)
• Environment Canada
• Meteorological Service of Canada
• Climate Change Section
• In collaboration with Dr. Richard Greatbatch,
  Dr. Jinyu Sheng and Dr. Ian Folkins in the
  Department of Atmospheric Science at Dalhousie
  University

1
2
(No Transcript)
3
Typical grid spacing in a GCM
4
4
From IPCC 2001
5
METHOD

• Remove seasonal cycle from predictand/predictors.
• Focus on Winter (DJF)
• Predictor Selection
• Principal component analysis (PCA).
• Multiple linear regression on Tmax and NCEP PCs.
• Validation
• Use CGCM3 predictors to hindcast historical
  distribution.
• Use future CGCM3 predictors to make projections.

6
Seasonal cycle removal
7
Predictor Selection

• mean sea level Pressure
  Redundancy inspired the removal of
• 500hpa geopotential height
• 500hpa zonal windspeed.
• 500hpa meridional windspeed
  total windspeed
• 500hpa vorticity
• 850hpa geopotential height
  
  
• 850hpa zonal windspeed
  
• 850hpa meridional windspeed
  wind direction
• 850hpa vorticity
• surface zonal windspeed
  
  
• surface meridional windspeed
• surface vorticity
  divergence
• 500hpa specific humidity
• 850hpa specific humidity
• surface specific humidity
• surface mean temperature

8
PCA/MLR BASICS
9
PCA

• mean sea level pressure -0.17
• 500hpa geopotential height 0.14
• 500hpa zonal windspeed -0.05
• 500hpa meridional windspeed 0.48
• 500hpa vorticity -0.14
  CORRELATION 0.62
• 850hpa geopotential height -0.01
  40 of the variation in TMAX is
• 850hpa zonal windspeed -0.08
  explained by temperature advection
• 850hpa meridional windspeed 0.58
• 850hpa vorticity 0.10
• surface zonal windspeed -0.13
• surface meridional windspeed 0.52
• surface vorticity
  0.24

10
PCA
850GPH
500GPH
????
MSLP
11
REGRESSION INFO

• Explained variance (percent) 79
• Regression error variance 0.07
• Number of predictors (PC's) 10
• Gamma squared 0.21

12
VALIDATION
13
NCEP PREDICTION
14
NEXT

• Once the best historical model from NCEP has been
  madehindcast using CGCM3 predictors.
• Project the CGCM3 predictors onto the
  eigenvectors created from NCEP.

15
CGCM3 PREDICTION
16
PROJECTIONS
1

17
SUMMARY

• Predictor Selection/PCA improves regression
• Reduces autocorrelation of regression errors
  (removal of seasonal cycle).
• Identify governing physics by season (gives
  confidence in the future).
• Normal distributions are a much better assumption
  using seasons compared to annual.

18
My Next Steps

• FDEOFR to get regression coefficients as a
  function of frequency.
• FDEOFR gives candidate downscaling frequencies
  and helps sort out physics (climate vs. weather)

19
CLIMATE QUOTES

• The reasonable man tries to adapt to the world
  around him, while the unreasonable tries to
  adapt the world to himself. Therefore all
  progress depends on the unreasonable man
• --George Bernard Shaw
• The real problem is not global warming. It is
  in fact the majoritys level of awareness. As if
  we are somehow separate from nature
• --Lee Titus
• The laws of Congress and the laws of physics
  have grown increasingly divergent, and the laws
  of physics are not likely to change."--Bill
  McKibben