anne stoner, katharine hayhoe texas tech university keith dixon, john lanzante, aparna...

18
Anne Stoner, Katharine Hayhoe Texas Tech University Keith Dixon, John Lanzante, Aparna Radhakrishnan GFDL COMPARING STATISTICAL DOWNSCALING METHODS: FROM SIMPLE TO COMPLEX

Upload: kaitlynn-whetten

Post on 31-Mar-2015

216 views

Category:

Documents


1 download

TRANSCRIPT

Page 1: Anne Stoner, Katharine Hayhoe Texas Tech University Keith Dixon, John Lanzante, Aparna RadhakrishnanGFDL COMPARING STATISTICAL DOWNSCALING METHODS: FROM

Anne Stoner, Katharine Hayhoe Texas Tech UniversityKeith Dixon, John Lanzante, Aparna Radhakrishnan GFDL

COMPARING STATISTICAL DOWNSCALING METHODS: FROM SIMPLE TO COMPLEX

Page 2: Anne Stoner, Katharine Hayhoe Texas Tech University Keith Dixon, John Lanzante, Aparna RadhakrishnanGFDL COMPARING STATISTICAL DOWNSCALING METHODS: FROM

Goal: Evaluate and compare multiple statistical downscaling methods using the same framework Monthly and daily versions of Delta, Quantile Mapping, and

Asynchronous Regional Regression ModelVariables –

Minimum, maximum daily 2m temperature Daily accumulative precipitation

Input: GFDL-HiRES experimental model as both model and observations OBS: 25km GFDL-HiRES (1979-2008) Model: 200km coarsened GFDL-HiRES (1979-2008, 2086-

2095)Output: Daily 25km downscaled Tmin, Tmax, Prcp

(2086-2095)

APPROACH

Page 3: Anne Stoner, Katharine Hayhoe Texas Tech University Keith Dixon, John Lanzante, Aparna RadhakrishnanGFDL COMPARING STATISTICAL DOWNSCALING METHODS: FROM

Calculates average diff erence between present and future GCM simulations, then adds that diff erence to the observed time series for the point of interest Here: individually for each high-resolution grid cell

METHOD 1: DELTA CHANGE

Assumptions – GCMs are more successful

at simulating changes in climate rather than actual local values

Stationarity in local climate variability

Page 4: Anne Stoner, Katharine Hayhoe Texas Tech University Keith Dixon, John Lanzante, Aparna RadhakrishnanGFDL COMPARING STATISTICAL DOWNSCALING METHODS: FROM

Projects PDFs for monthly or daily simulated GCM variables onto historical observations

METHOD 2: QUANTILE MAPPING (e.g. BCSD)

Changes the shape of the simulated PDF to appear more like the observed PDF, but allowing the mean and variance of the GCM to change in accordance with GCM future simulations

Page 5: Anne Stoner, Katharine Hayhoe Texas Tech University Keith Dixon, John Lanzante, Aparna RadhakrishnanGFDL COMPARING STATISTICAL DOWNSCALING METHODS: FROM

Asynchronous Regional Regression Model

METHOD 3: QUANTILE REGRESSION (e.g. ARRM)

Daily quantile regression using piecewise linear segments to improve fit for the training period

Individual monthly models allows for different distributions throughout the year

Page 6: Anne Stoner, Katharine Hayhoe Texas Tech University Keith Dixon, John Lanzante, Aparna RadhakrishnanGFDL COMPARING STATISTICAL DOWNSCALING METHODS: FROM

COMPARISON

The shape of the resulting downscaled distribution depends highly on the downscaling method used

Delta

Quantile Mapping

ARRM

Colorado National Monument, CO

Page 7: Anne Stoner, Katharine Hayhoe Texas Tech University Keith Dixon, John Lanzante, Aparna RadhakrishnanGFDL COMPARING STATISTICAL DOWNSCALING METHODS: FROM

MAXIMUM TEMPERATURE

Page 8: Anne Stoner, Katharine Hayhoe Texas Tech University Keith Dixon, John Lanzante, Aparna RadhakrishnanGFDL COMPARING STATISTICAL DOWNSCALING METHODS: FROM

MINIMUM TEMPERATURE

Page 9: Anne Stoner, Katharine Hayhoe Texas Tech University Keith Dixon, John Lanzante, Aparna RadhakrishnanGFDL COMPARING STATISTICAL DOWNSCALING METHODS: FROM

PRECIPITATION

Page 10: Anne Stoner, Katharine Hayhoe Texas Tech University Keith Dixon, John Lanzante, Aparna RadhakrishnanGFDL COMPARING STATISTICAL DOWNSCALING METHODS: FROM

PRECIPITATION

Page 11: Anne Stoner, Katharine Hayhoe Texas Tech University Keith Dixon, John Lanzante, Aparna RadhakrishnanGFDL COMPARING STATISTICAL DOWNSCALING METHODS: FROM

MAXIMUM TEMPERATURE

Page 12: Anne Stoner, Katharine Hayhoe Texas Tech University Keith Dixon, John Lanzante, Aparna RadhakrishnanGFDL COMPARING STATISTICAL DOWNSCALING METHODS: FROM

MINIMUM TEMPERATURE

Page 13: Anne Stoner, Katharine Hayhoe Texas Tech University Keith Dixon, John Lanzante, Aparna RadhakrishnanGFDL COMPARING STATISTICAL DOWNSCALING METHODS: FROM

PRECIPITATION

Page 14: Anne Stoner, Katharine Hayhoe Texas Tech University Keith Dixon, John Lanzante, Aparna RadhakrishnanGFDL COMPARING STATISTICAL DOWNSCALING METHODS: FROM

PRECIPITATION

Page 15: Anne Stoner, Katharine Hayhoe Texas Tech University Keith Dixon, John Lanzante, Aparna RadhakrishnanGFDL COMPARING STATISTICAL DOWNSCALING METHODS: FROM

DAILY DOWNSCALED TMAX

Page 16: Anne Stoner, Katharine Hayhoe Texas Tech University Keith Dixon, John Lanzante, Aparna RadhakrishnanGFDL COMPARING STATISTICAL DOWNSCALING METHODS: FROM

MONTHLY DOWNSCALED TMAX

Page 17: Anne Stoner, Katharine Hayhoe Texas Tech University Keith Dixon, John Lanzante, Aparna RadhakrishnanGFDL COMPARING STATISTICAL DOWNSCALING METHODS: FROM

Comparing multiple downscaling methods in a standardized framework gives us useful information

If someone has already used a certain downscaling method they can correctly interpret the biases

If someone is trying to decide which method to use, this can help their decision, because there’s no perfect method

Simple methods can be fine for studying monthly/annual means, daily output for low latitudes

More complex methods are required when studying climate extremes and high latitudes

CONCLUSIONS

Page 18: Anne Stoner, Katharine Hayhoe Texas Tech University Keith Dixon, John Lanzante, Aparna RadhakrishnanGFDL COMPARING STATISTICAL DOWNSCALING METHODS: FROM

Downscale relative humidityFigure out physical causes of the biases we’re

seeingExplore the influence of different predictorsIncorporate more downscaling techniques

NEXT STEPS