robert laplante christopher mello noaa/nws cleveland, oh gregory lang jia wang noaa/glerl ann arbor,...
TRANSCRIPT
Evaluation of Great Lakes Ice Model (GLIM) Real-time Ice Forecasts for
the 2012-2013 Ice seasonRobert LaPlante
Christopher MelloNOAA/NWS Cleveland, OH
Gregory LangJia Wang
NOAA/GLERL Ann Arbor, MI
9 April 2013
OutlineDescription of the GLIM Comparison of the GLIM to the National
Ice Center analysis for the 2012-2013 ice season
Addition of Vessel Icing parameterFuture plans for the GLIM
MODIS 4-9 March 2010
Description of the GLIMGLIM has been under development by GLERL since 2007 GLIM is a combination of the Princeton Ocean Model for
hydrodynamics and the Combined Ice Ocean Model tailored for the Great Lakes
GLIM which has two modules, runs twice a day for all five Great Lakes as part of the Great Lakes Coastal Forecast system (GLCFS) at NOAA/GLERL Hourly Nowcast runs - ingest surface meteorological and
daily NIC ice concentration observationsForecast runs from 00 to 12o hours – driven by the NDFD
The GLIM is run using the latest surface forecast meteorological grids from the NWS National Digital Forecast Database (NDFD) out to 5 days
Description of the GLIMGLIM Nowcasts of ice concentration are
“nudged” towards the daily NIC ice concentration field and form the initial conditions for the forecast module
Model output for the GLIM is posted to the web
NWS CLE retrieves via ftp the GLIM for AWIPS and GFE
GLIM Ice Concentration is nudged toward the NIC ice concentration field
which is updated daily
GLIMNIC
GLERL GLIM included in the GLCFSGreat Lakes Coastal Forecasting System
http://www.glerl.noaa.gov/res/glcfs/
Select “Ice”
Integration of the GLIM and NDFD results in the Forecast Generation of:Surface Water Temperature
Ice ConcentrationIce ThicknessIce Drift & Vessel Icing
http://www.glerl.noaa.gov/res/glcfs/glcfs-ice.php?lake=e&type=N&hr=00
Sources of Error in the GLIM Ice Concentration Forecast
Error in the daily NIC ice concentration analysisNIC analysis data sources are lagged: ~00 and ~12
UTC RADARSAT passes & previous day ~18 UTC MODIS imagery and current day ~18 UTC MODIS imagery are combined and posted about 00 UTC the next day Variable observation times add uncertainty due to
advective changes and growth and decay of ice cover Subjectivity of the ice analyst in specifying ice edge
and concentration
Error in the NDFD parameters (wind, temperature, or dewpoint) over the entire Great Lakes
Error within the GLIM: nudging procedure, vertical mixing, or wave mixing
Winter TemperaturesWhile still above normal, the 2012-13 winter was
colder than the unseasonably mild 2011-2012 winterResultant Great Lakes ice cover increased,
especially on shallow Lake ErieMaximum seasonal ice cover attained by lake
(GLERL) 2011-2012 2012-2013Lake Ontario 3% 15%Lake Superior 7% 38%Lake Erie 16% 85%Lake Michigan 17% 20%Lake Huron 20% 40%
Assessment of the NIC Ice AnalysisNIC ice analysis was notRealistic at times with ice Concentration decreasingWith air temperaturesBelow freezing
Temperature
Assessment of the NIC Ice AnalysisNIC Analysis exhibited Occasional spikes of increasedIce concentration which lastedBut a day or so and were notRealistic. In particular, weWill examine the second, or Day 62 spike on Lake SuperiorWhen the ice cover nearly doubledFor only a day
Lake Superior
Assessment of the NIC Ice AnalysisDay61
Day62
Day63
Ice Concentration
Note the abrupt increase and decrease in ice concentrationFrom Day61 through Day63 overWestern Lake Superior
This High Nowcast, due to the highNIC analysis, produced a high starting forecast
Which, with low air temperatures Leads to these high 1-day, 3-day, and 5-day forecasts
Lake Superior
Change in GLIM forecastIce Concentration vsChange in NIC ice analysis
NIC spikes are apparent in thesex/y plots of the GLIM days 1-3-5 forecasts
Too slow ice growth in GLIM
To fastIce growth
Too fastmelt
Too slowmelt
GLIM tends toGrow or meltIce too slowly inDay 3 and Day 5 forecasts
LakeErie
Summary of the GLERL GLIM2012-2013 ice season on Lake Erie
GLIM was successfully run over all the Great LakesGLIM nowcasts for 2012-2013 closely matched the
NIC ice concentration analysis with daily nudging across all the lakes
Since the NIC analysis is not a snap shot of the observed ice cover…some uncertainty is introduced into the GLIM“Spikes” in the daily NIC ice analysis appeared spurious
Compared to NIC analysis, the GLIM 3-day and 5-day forecasts tended to grow and melt ice too slowly on all the Lakes
• Difficult to separate error in NIC analysis and NDFD with error in GLIM
Sea Spray Icing AlgorithmUsing the work of Overland et al. (1986) &
(1990): Icing Predictor: PPR is defined by
GLERL modified the equation for fresh water and used output from the Great Lakes Ice Mode (GLIM) to produce icing classes for all five lakes
Icing ClassesIcing classes are determined with the
assumption that vessels are steaming into the wind and of 20 -75 meters in length
Potential Issues Shoreline affects are not accounted for in
icing potentialFor a given wind, waves and resultant freezing
spray will be less at the start of the fetchThe reduction in freezing spray due to lake
ice cover has not been determined, but at some high percentage of ice cover (>70%), freezing spray may be minimal
Vessel Icing NotesGLERL added the Vessel Icing parameter to
its web site http://www.glerl.noaa.gov/res/glcfs/
The quality of the freezing spray forecast is dependent on the inputs into the GLIM which include the NDFDSurface WindAir Temperature
So pay attention (American Forecasters) to the NDFD grids over the entire lake you forecast for
Future Plans for the GLIM• GLCFS POM (currently running the GLIM) will not be
modified• GLERL is in the process of developing a new model: an
operational FVCOM (Finite Volume Coastal Ocean Model) for Lake Erie• Variable grid and much higher resolution for harbors and bays• Heat fluxes are being calculated and testing of ice dynamics is
expected over the next year• Plan to test the Lake Erie FVCOM this fall at CSDL (Coastal
Survey Development Lab)• By late summer 2014, FVCOM may be delivered to NCEP• Remaining lakes will be modeled by early 2017