Arctic Snow Cover Monitoring - World Meteorological ... Snow Cover Monitoring Chris Derksen and Ross...
Transcript of Arctic Snow Cover Monitoring - World Meteorological ... Snow Cover Monitoring Chris Derksen and Ross...
Arctic Snow Cover Monitoring
Chris Derksen and Ross Brown
Climate Research Division
Environment Canada
Thanks to our data providers:
Rutgers Global Snow Lab ● National Snow and Ice Data
Center ● World Climate Research Programme Working
Group on Coupled Modelling ● University of East Anglia –
Climatic Research Unit ● NASA Global Modeling and
Assimilation Office ● European Centre for Midrange
Weather Forecasting
Outline
• How do recent reductions in observed
spring snow cover extent (SCE)
compare to CMIP5 simulations? (historical + RCP8.5)
• Why are observed trends in the dates of
snow cover onset (relatively little
change) and snow cover
disappearance (widespread significant
trends to earlier loss of snow cover) so
different?
• What is the state of our observational
capability to assess CMIP5 simulations
of snow water equivalent (SWE)?
• A few suggestions for GCW…
• Binary snow extent (weekly) derived by analysts from primarily optical satellite imagery
since 1966.
• Maintained at Rutgers U by David Robinson http://climate.rutgers.edu/snowcover/].
• Subject to intensive validation, and is widely used in the climate community.
NA monthly
SCE anomalies
EUR monthly
SCE anomalies
Observed Arctic Snow Cover Extent –
NOAA CDR Dataset
Model Acronym Institution n lat n lon Resolution
(°)
CanESM2 CCCma (Canadian Centre for Climate Modelling and
Analysis, Canada)
64 128 2.81 x 2.81
CCSM4 NCAR (National Center for Atmospheric Research, USA) 192 288 0.93 x 1.25
CNRM-CM5 CNRM-CERFACS (Centre National de Recherches
Meteorologiques / Centre Europeen de Recherche et
Formation Avancees en Calcul Scientifique, France)
128 256 1.40 x 1.40
GISS-E2-R NASA-GISS (NASA-Goddard Institute for Space Studies,
USA)
90 144 2.00 x 2.50
INMCM4 Institute for Numerical Mathematics, Russian Academy of
Sciences, Russia
120 180 1.50 x 2.00
MIROC5 AORI (Atmosphere and Ocean Research Institute, The
University of Tokyo, Japan), NIES (National Institute
for Environmental Studies, Japan), and JAMSTEC
(Japan Agency for Marine-Earth Science and
Technology)
128 256 1.40 x 1.40
MPI-ESM-LR Max Planck Institute for Meteorology, Germany 96 192 1.88 x 1.88
MRI-CGCM3 Meteorological Research Institute, Tsukuba, Japan 160 320 1.13 x 1.13
CMIP5 Climate Model Simulations of
Arctic Snow Cover (historical + rcp8.5)
Historical + projected (8 CMIP5 models; rcp85 scenario) and observed (NOAA snow
chart CDR) snow cover extent for April, May and June for the Northern Hemisphere
Snow covered area is normalized by the maximum area simulated by each model.
• Observations are mostly within ±1 standard deviation of the multi-model ensemble in
April and May, but start to diverge from the model consensus in recent years.
• Marked reductions in June SCE observed since 2005 fall below the zone of model
consensus defined by +/-1 standard deviation from the multi-model ensemble mean.
Simulated vs. Observed Arctic Snow
Cover – CMIP5
Derksen, C Brown, R (2012) Geophys. Res. Letters
Changes in Snow vs. Sea Ice Extent
• Over the 1979 – 2012 time period, June snow extent is decreasing at a rate of -17.6%
per decade (relative to 1979-2000 mean).
• September sea ice extent is decreasing at -13.0% per decade over the same time
period.
Northern hemisphere June snow cover; June and September
Arctic sea ice extent, 1979-2012
Derksen, C Brown, R (2012) Geophys. Res. Letters
How reliable are the NOAA Snow Chart
CDR Trends?
• Similar anomaly trend results obtained
with three independent datasets. • Tendency for NOAA to consistently map
less spring snow (~0.5 to 1 million km2)
than the multi-dataset average since 2007.
• Accounting for this difference reduces the
June SCE trend to -15.0% per decade.
Standardized anomaly time series of
Northern Hemisphere SCE, 1981-2012,
from the NOAA snow chart CDR (blue),
MERRA (red) and ERAint (green)
Northern Hemisphere SCE time series,
1981-2012, for the NOAA snow chart CDR
(red) and average of
NOAA+MERRA+ERAint (blue)
NA monthly
SCE anomalies
EUR monthly
SCE anomalies
Observed Arctic Snow Cover Extent –
NOAA CDR Dataset
• Warming temperature trends in both spring and fall, so why the seasonal difference in
SCE anomalies?
• CMIP5 model output suggests seasonally symmetric reductions in SCE.
There is evidence of a
tendency in the NOAA
snow chart data record to
map relatively more snow
over Eurasia in the snow
onset period than other
datasets, which results in
an artificial trend (~+1.0
million km2 per decade)
October snow cover.
Explaining Seasonal Asymmetry in Arctic
Snow Cover Duration Trends
Difference between NOAA snow chart CDR and other
independent SCE datasets, 1982-2005
Static adjustment of -1.0 x 106km2 per
decade applied to the NOAA time series
An arbitrarily adjusted NOAA time
series:
1. agrees closely with multiple
evaluation datasets
2. is significantly correlated with
surface temperature
• Higher MODIS
cloud fraction in the
snow cover onset
period creates
challenges for
NOAA analysts
Explaining Seasonal Asymmetry in Arctic
Snow Cover Duration Trends
October May
Snow cover
onset date is
less strongly
coupled to air
temperature
than snow-off
date [weaker
feedbacks;
requires Tair <
0 AND precip >
0 to initiate
snow cover].
Date of freeze onset
vs. start of snow cover
Melt onset vs. end of
snow cover
Explaining Seasonal Asymmetry in Arctic
Snow Cover Duration Trends
MERRA
ERA-interim
reconstruction
Multi-Dataset SWE Time Series Dataset Method Domain Time Period Resolution Reference
GlobSnow Passive microwave + climate station Pan-Arctic 1980-2011 25 km Takala et al., 2011
ERA-int Snow accumulation model driven by
reanalysis temp and precip
Pan-Arctic 1980-2012 ~275 km Uppala et al., 2005
B2003 Surface analysis + snow model North America 1979-1997 ~35 km Brown et al, 2003
CMC Surface analysis + snow model Pan-Arctic 1998-2012 ~35 km Brasnett, 1999
MERRA SWE field from reanalysis Pan-Arctic 1980-2012 ~35 km Rienecker et al., 2011
Seasonal SWEmax time series (+/- standard error) from multi-datasets
• SWE datasets are reasonably consistent (+/- 10 to 25 mm) with no significant trends in the
multi-dataset average; SWEmax shows a step change over Eurasia in the late 1990s.
Mean annual maximum SWE (1998/99 – 2009/10) from CMC (left),
GlobSnow (middle), and merged dataset (right).
Hemispheric Snow Water Equivalent
Data for CMIP5 Evaluation
GlobSnow: combines satellite passive microwave measurements, forward snow
emission model simulations, and climate station snow depth observations in an
assimilation framework (lacks information in areas with complex relief).
CMC: snow depth observations combined with a first guess using a simple snow
accumulation and melt model (underestimates SWE over northern latitudes).
RMSE= 47 mm RMSE= 92 mm
GlobSnow Standard Algorithm
Takala M et al (2011) Remote Sens. Environ.
GlobSnow Algorithm
Evaluation
(Greater than 100 000 samples from across FSU)
CMC+GlobSnow
(mm)
10 model avg
(mm)
10 model avg
bias (mm)
CMIP5 Simulated vs. Observed
Arctic Snow Water Equivalent
• Models overestimate SWEmax over Arctic land areas
• The multi-model ensemble agree more closely with the observed data
than any individual model
10-model ensemble mean annual max monthly SWE (SWEmax) evaluation results
Primary Objective:
Quantify amount and variability of freshwater stored in seasonal snow packs,
and snow accumulation on glaciers
Scientific Impact:
To improve hydrological and climate modelling and Numerical Weather
Prediction by incorporation of direct observations of snow mass and snow
mass variability
Instrument:
SAR at Ku-(17.2 GHz) and X-Band (9.6 GHz)
Repeat Time:
3 and 15 days
Spatial Resolution:
50 x 50 m, Swath ≥100 km, Products at 500m
Cold Regions High
Resolution Hydrological
Observatory (CoReH20)
Conclusions
• The rate of June snow cover extent reductions
(-17.6% per decade since 1979) is greater
than the rate of summer ice loss (-13.0% per
decade over the same time period). Both are
declining at rates that exceed CMIP5 model
projections [Stroeve et al 2012, Derksen and
Brown 2012].
• The observed seasonal asymmetry in snow
cover extent response to warming Arctic
temperatures is due to: (1) internal bias in the
NOAA snow chart time series; (2) weak
coupling between snow cover and air
temperature at the start of the snow season.
• New SWE products (i.e. ‘GlobSnow’ merged
with the CMC analysis) are providing an
improved basis for model evaluation
Recommendations for
GCW Snow-Watch 1. Through dataset intercomparisons, develop high quality validation data sets
covering a range of spatial scales.
Anomalies of June Arctic SCE, 1982-2002
-3
-2
-1
0
1
2
3
1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002
SC
E s
td a
no
maly
NOAA
NCEP
ERA-40
ERA-40rec
CCRS
PMW
Avg
Various estimates of June 2002 Arctic snow cover extent (ranges from 2.3 to 5.4 million sq km)
Address:
1. regions and times when between-dataset
variance is the highest (i.e. snow cover onset
and melt periods over Arctic).
2. strong interannual variability: compare “good”
and “bad” years to identify issues
3. is the multi-dataset average a better estimate
of the “true” snow cover?
2. Provide near real time SCE and SWE anomaly tracking.
-requires continuous monitoring by credible, media savvy people (like
Dave Robinson at Rutgers)
2011 North American Snow Cover Extent (excl Greenland)
0
2
4
6
8
10
12
14
16
18
20
1 31 61 91 121 151 181 211 241 271 301 331 361
Julian Day
SC
E (
millio
n s
q k
m)
NOAA mly avg (1981-2010)
NOAA mly min (1966-2010)
NOAA mly max (1966-2010)
IMS-24 km
CMC > 0 cm
Earlier than
normal melt
Comparison of NA
SCE from CMC daily
snow depth analysis
and NOAA IMS-24
km daily snow cover
product
Below average
SCE during snow
cover onset period
Recommendations for
GCW Snow-Watch
3. Conduct an evaluation of snow reconstructions from detailed physical
snow process models (driven by reanalysis) in order to determine the
reliability and consistency of precipitation (and SWE) products.
Trend in simulated maximum winter SWE
(cm/decade) over 1979-2009 from Liston and
Hiemstra (2011) [Arctic snow process model
driven with MERRA reanalysis fields
downscaled to 10 km]
Recommendations for
GCW Snow-Watch
Interannual Variability:
Observations vs. Simulations
Standard deviation in monthly SCE from NOAA
observations and CMIP5 model simulations.
• CMIP5 simulations underestimate variability in the shoulder seasons.
• Peak spring variability in observations occurs in June; May for CMIP5.
Derksen, C Brown, R (2012) Geophys. Res. Letters
Linear trends (1967-2011): NOAA snow chart CDR
Arctic snow cover extent (km2 x 106 x decade-1);
CRUtem3v surface temperature over land areas
(ºC x decade-1).
Fall and spring Arctic snow cover
duration anomalies, 1967-2012
(NOAA Snow Chart CDR)
Derksen, C Brown, R (2012) BAMS
• Warming temperature trends in both spring and fall,
so why the seasonal difference in snow cover
duration anomalies?
• CMIP5 model output suggests seasonally
symmetric reductions in SCE.
Arctic Snow Cover Duration and
Temperature Trends SCE SAT
Month NA EUR NA EUR
September -0.02 -0.01 0.42 0.30
October 0.03 -0.02 0.38 0.53
November 0.07 0.01 0.50 0.53
April -0.18 -0.43 0.53 0.50
May -0.24 -0.58 0.30 0.46
June -0.35 -0.78 0.35 0.35
Trends in Passive Microwave Derived
Melt Onset
Trend P values (blue for P<0.05) Linear least squares trend (1979-
2011) for local significance (days/year)
+1
-1
0
Wang L et al (in press) Geophys. Res. Letters
Correlations of melt onset date and surface air temperature, 1979-2011.
(Correlations less than -0.3 are statistically significant at 95% confidence level)
March April
May June
Correlations Between Melt Onset Date
and Surface Air Temperature
Wang L et al (in press) Geophys. Res. Letters
Satellite brightness temperatures
(19 and 37 GHz)
Snow depth observations from synoptic weather
stations (ECMWF)
Observed field produced by kriging synoptic
weather station observations. Estimate of
kriging interpolation variance also obtained.
Weather station measurements of snow depth
used as input to forward snow emission model
simulations. Model fit to satellite measurements
by fluctuating the effective grain size.
Spatially continuous background field of
effective grain size (including a variance field)
produced by kriging.
Radiometer field produced through forward Tb
simulations using the kriged effective grain size.
Model fit to satellite measurements by
fluctuating SWE.
Assimilation procedure adaptively weighs the
observed and radiometer fields to estimate final SWE
and a measure of statistical uncertainty.
GlobSnow SWE Algorithm
2. Promote dataset intercomparisons to characterize the uncertainty in data
products
95% confidence interval in NA mean
SCE and between-dataset standard
deviation (% of mean SCE) for 6 SCE
datasets over 2001-2008 period
(MODIS, CMC, IMS-24, ERAint
reconstruction, PMW, NCEP proxy)
Stdev < 10% from Dec to April
Recommendations for
GCW Snow-Watch
3. Provide near real time information on hemispheric snow depth (or
SWE) anomalies.
-Highly visible product with many potential applications (e.g. water
resources, agriculture, forest fire potential)
Dec Feb April
Monthly snow depth anomaly (%) over 2011-2012 snow season from CMC daily
snow depth analysis (1999-2010 reference period) (Derksen and Brown, BAMS State
of Climate, 2012)
Recommendations for
GCW Snow-Watch