Detection of external influence on Northern Hemispheric snow cover

Transcription

1 Detection of external influence on Northern Hemispheric snow cover Tianshu Ma 1, Xuebin Zhang 1,2 Helene Massam 1, Francis Zwiers 3 Georges Monette 1, David Robinson 4 1 Dept. of Mathematics and Statistics, York University, Toronto, Ont. 2 Environment Canada, Toronto, Ont. 3, University of Victoria, Victoria, BC 4 Department of Geography, Rutgers University, Piscataway, NJ

2 Outline Climate Research Division Introduction Methods Data Some results Photo: F. Zwiers

3 Introduction Objective is to determine whether human influence is detectable in snow cover extent Snow cover responds strongly to temperature, so approach will be to consider temperature and snow cover jointly Photo: F. Zwiers

4 Detection strategy D&A analysis on snow cover or temperature for individual continents snow cover and surface air temperature in a seemingly unrelated regression setting, for one continent (2 variables) and for two-continent (4 variables). Bayesian method used to assess posterior probability of detection / attribution (next slide)

5 Methods: seemingly unrelated regression Allows for multiple variable detection without requiring the same scaling factor Smaller variance in scaling factors compared with that from independent univariate regressions

6 Construction of the posterior on β Likelihood: Prior: Impose non-informative conjugate priors on hyperparameters K (Wishart) and µ (multivariate normal conditional on K) Multiply to obtain joint distribution on Integrate over µ, K, and apply Bayes Theorem to obtain

7 Posterior distribution on scaling factors Multi-variate normal: Detection probability for the joint multiple variables is computed within a defined detection cube Marginal probability gives detection or attribution probability for a particular variable

8 Snow cover observations Station observations Point observations from ~1950 on, large spatial gaps (esp Cdn Arctic), unrepresentative of surrounding terrain Satellite observations Available from 1966 onwards based on visible satellite imagery Subject to procedural changes, confusion of lake ice and snow cover during melt season, etc. Increasing resolution over time Passive microwave from 1978, calibration issues between SMMR ( ) and SSM/I, noisy for snow cover onset, useful information on snowmelt onset Merged/reconstructed snow cover data Longer period (some start from 1950s) Possibly more homogeneous Typically involves a snow cover model with temp and precip as inputs We use gridded NOAA satellite data record ( ) Extensive reanalysis of NOAA data to correct known problems (Robinson, et al, 1993 and updates)

9 Observed monthly snow cover variability (standard deviation in % of fraction of month snow covered) Jan July Courtesy Ross Brown, Environment Canada

10 Model simulated snow cover Very limited availability of 20 th century runs for which snow cover has been diagnosed GISS, model-e-h, 5 runs, 72x46, ALL GISS, model-e-r, 9 runs, 72x46, ALL MIROC3.2, medres, 3 runs, 128x64, ALL MRI, CGCM2, 3.2a, 5 runs, , ALL NCAR, CCSM3.0, 8 runs, 256x128, ALL CCCma, CGCM3.1, 5 runs, 96x48, ANT GISS, AOM, 2 runs, 90x60, ANT

11 Observed and simulated snow cover (10 8 km 2 ) Box plots of the snow cover extent climatology (monthly values averaged from , in 10 8 km 2 ) from model simulations for Eurasia (EA, upper panel) and North America (NA, lower panel). Observed snow cover climatology is marked by red circles Eurasia J F M A M J J A S O N D North America J F M A M J J A S O N D

12 Model data processing Limited availability is the reality Model data normalized for each run, to remove model bias Consider Eurasian and North American snow cover (north of 30N) Remove mean and divided by the standard deviation for the same period Pool all forced runs Non-overlapping 5-yr means Consider only half year once snow cover initially established (Jan-June)

13 Observed and simulated snow cover anomalies Eurasia North America Jun May Apr Mar Feb Jan Observed Observed Jun May Apr Mar Feb Jan Jun May Apr Mar Feb Jan Signal Signal Jun May Apr Mar Feb Jan

14 S T S T S T S T S T S T EA NA EA NA EA NA Scaling factors and confidence intervals for snow cover (S) and temperature (T) in Eurasia (EA) and North America (NA) in a 4- variable analysis that combines snow cover and temperature in two continents (4-var analysis), in 2-variable analyses that combine snow cover and temperature for individual continents (2-var analysis), and in one-variable (1-var analysis).

15 Detection probability Joint EA, marginal

16 Attribution probability (0.5, 1.5) Joint EA, marginal

17 Conclusions Multi-dimensional setting for d&a Strong evidence of detection for of external forcing on snow cover in EA and NA Some evidence for attribution Photo: F. Zwiers

18 Climate Research Division Photo: F. Zwiers Questions?