The Available NCEP Reanalyses Wesley Ebisuzaki Climate Prediction Center National Centers for Environmental Prediction NWS/NOAA Maryland, USA wesley.ebisuzaki@noaa.gov
Topics Introduction R1, R2, NARR, CFSR Grib 1 and 2 and various utilities Getting the data by Nomads Will try to cover these topics from a user perspective. Some overlap with NOMADS and GrADS presentations.
Download links http://www. cpc. noaa. gov/products/ wesley/ams2010_sc Download links http://www.cpc.noaa.gov/products/ wesley/ams2010_sc.html Documentation package includes a pdf version. The pdf has more details and covers more material.
Are Reanalyses Truth? Convenient Run through quality control programs Consistent (ex. winds, temperature, heights) Almost every variable that you would ever want Better than interpolation of observations
But Need observations Some fields depend on the model physics Some regions are poorly observed Some levels are poorly observed Some fields depend on the model physics Clear sky radiation is pretty good Clouds and other moist process are bad BL is somewhere in between
Reanalysis Errors Best for winds, temperature and heights Humidity Observations, consistency relationships Near surface more dependent on BL physics Humidity Fewer observations, model physics important Other fields More model physics dependent
Not all observations are equal Sondes and aircraft data are good quality Surface pressure is good quality Surface winds, temperatures, humidities Hard to assimilate, elevation, representativeness Satellite data Retrievals=old, Radiances=new Hard to get a consistent record with historical data A concern for trend analyses
Analysis Uncertainties Factors to Consider Observation density Sensitivity to model parameterizations Dependent on lesser quality observations Model and data assimilation system
Are Observations Truth? Data assimilation could fit the observations exactly but produce a worse forecast! Error by equipment manufacturer Representativeness error Observation in an air parcel is not the same as average value in the grid cell
Finally a plot spread from an ensemble of opn analyses and reanalyses simple way to get a error estimate
Averaged over a year
Monthly Means
Analysis Uncertainties Not simple Uncertainty for a day: synoptic and data Averaged over a year: data Monthly/seasonal means have smaller uncertainties than daily field, biases
NCEP Reanalyses NCEP/NCAR Reanalysis (R1, CDAS) Mid 1990s, 1947-present, 2.5 degree grid, global NCEP/DOE Reanalyses (R2) Late 1990s, 1979-present, 2.5 degree grid, global North American Regional Reanalysis (NARR) Early 2000's, 1979-present, 32 km grid Climate Forecast System Reanalysis (CFSR) 2010, 1979-present, 0.5 degree grid, global
Others Reanalyses ECMWF: ERA-15, ERA-40, ERA-interim JMA, CRIEPI: JRA-25/JCDAS NASA/GSFC: MERRA ERSL (different approach) Check for robust signal
Support of NCEP Reanalyses 1) Journal articles (BAMS) 2) Web: NCEP, NCAR, ERSL, NCDC 3) Questions to provider of data 4) NCDC, NCAR, ERSL will forward tough questions to NCEP 5) At NCEP: R1, R2, NARR: wesley.ebisuzaki@noaa.gov CFSR: to be determined
Nuts and Bolts: data formats grib1, grib2 grib is a WMO standard and national meteorological centers use WMO standards for day-to-day operations Reanalyses run at highest resolution possible Large portion of supercomputer, massive tape storage grib files are smaller than netcdf Netcdf NCAR and ERSL often translate into netcdf and redistribute the NCEP reanalyses
My grib1 toolbox wgrib: inventory, get values, database tool GrADS: plots, some computations copygb: convert to different grid Calculations often easier on lat-lon grid Save space ggrib and lcgrib: subset of lat-lon, lambert- conformal grids, faster than copygb C/fortran programs: ieee -> grib
My grib2 tool box wgrib2: inventory, contents, database, encode GrADS: plots and some computations copygb2: convert to a different grid, computations are often easier on lat-lon grid cnvgrib: convert between grib1 and grib2 ggrib and lcgrib functionality in wgrib2 wgrib2: ieee -> grib2
grib2 to grib1: cnvgrib grib2 is new, many people use cnvgrib to convert from grib2 to grib1 Long term solution? NO! NCO dropped support for cnvgrib (5 years) New variables in grib2 are not in grib1 grib2 files are compressed, easier to use New features in grib1 utilities?
Short Grib1 Inventories -sh-3.00$ wgrib -s narr.t09z.awip32.merged 1:0:d=09102809:MSLET:MSL:anl:NAve=0 2:166602:d=09102809:PRMSL:MSL:anl:NAve=0 3:333204:d=09102809:PRES:hybrid lev 1:anl:NAve=0 .. -s is the short inventory column 1 = message (record) number column 2 = byte location starting from 0 column 3 = analysis time or initial time of the forecast column 4 = variable name column 5 = level/layer column 6 = timing information, anl=analysis, acc=accumulation, ave=average column 7 = number of fields used to make an ave/acc
Short Grib2 Inventories --sh-3.00$ wgrib2 pgblnl.gdas.2007010100.grb2 -s 1:4:d=2007010100:HGT:1 mb:anl: 2:16552:d=2007010100:TMP:1 mb:anl: 3:22064:d=2007010100:RH:1 mb:anl: .. column 1: message or message.submessage number column 2: the byte location of the grib message column 3: the analysis or start of forecast time, use -T to see the minutes and seconds column 4: variable name column 5: level column 6 = timing information, anl=analysis, acc=accumulation, ave=average, fcst = forecast
Names HGT = geopotential height (m) TMP = temperature (K) UGRD = zonal wind (m/s) VGRD = meridional wind (m/s) see NCEP tables on web (see pdf file) or use -v option in wgrib/wgrib2
Grid information: grib grib1/grib2 support many different grids internally grib stores data in different orders, 8 in grib1, 16 in grib2 up to software to figure it out wgrib -V wgrib2 -grid
Values at specified locations Grib1: understand the grid R1, R2: WE:NS storage NARR: WE:SN storage Global – easy to figure out the lat-lon of points NARR – need file with the lat-lon of the points rr-fixed.grb is with other course files On your own
Values at specified locations Grib2: harder to understand and easier to use 16 storage orders (3 in common use) wgrib2 convert data to WE:SN order by default Can use old way to pick up data (i.e. get n-th point) Can use wgrib2 to get the data (-lon option) See documentation for wgrib2 examples
Winds and other vectors Winds have two orientations North can point to the north pole (earth relative) North can point to the north grid point (grid relative) Non-staggered grid: (ix,iy) -> (ix,iy+1)
NARR Output Grid
Winds and other vectors NCEP convention is grid relative Lat-lon, Gaussian, Mercator is not an issue Lambert-conformal, polar stereographic an issue NARR is earth relative (anti-NCEP convention) Good for new users Bad for users of other NCEP regional products
GrADS and grib Widely used, open source, visualization + more Data model GrADS x,y,z,t and ensemble Grib time is more complicated forecast verification time Start of forecast time (or analysis time) average from 6-12 hours into the forecast Monthly average of one of the above Need to map grib times into GrADS time
Making plots with GrADS: grib1 Analyses grib2ctl.pl grib_file >ctl_file gribmap -0 -i ctl_file (run grads) Forecasts grib2ctl.pl -verf grib_file >ctl_file gribmap -i ctl_file
Making plots with GrADS: grib2 Analyses g2ctl.pl -0 grib_file >ctl_file gribmap -0 -i ctl_file (run grads) Forecasts g2ctl.pl grib_file >ctl_file gribmap -i ctl_file Unified options, -b option is working
See Jennifer's presentations More GrADS See Jennifer's presentations
Getting data the Nomads way On-line is easiest way – if downloading time ok Big jobs: spend more time optimizing the transfer and reduce the amount of data transferred Small jobs: ease of use is important, ex. plots, OpeNDAP (lat-lon grid, text output).
Downloading Methods Partial http downloading: download the fields that you want select by field/time/level data is compressed (grib2) or packed (grib1) efficient for the server (support many clients) easy to script (not point and click) example in the documentation package good for large downloads
Downloading Methods g2subset (grib-filter): download the fields that you want select by field/time/level select an optional regional subset grib2 only point and click to learn or a few files moderately easy to script example in the documentation package more server resources, less data transferred
Downloading Methods OPeNDAP: standard protocol (text based) select field/time/level/region data is read by the server interpolated to a lat-lon grid if needed sent to the client by a standard protocol advantages: supported by software easy to use .. can even use a browser disadvantages: server overhead data may not be compressed
Downloading Methods Plots: Part of the development nomads at NCEP Point and click Designed for casual use of the data Not designed to do everything Research: download and do own plots For some people, this all that they need
Downloading Methods Full File: Can use browser, etc to download full file from http server, get the directory listing and right click. Easy Good to get a sample file to help plan the download.
Typical Nomads front page
Summary Reanalyses are not truth Observations are not truth Tools for grib1 and grib2 Introduction to Nomads Save time in downloading data Select fields that you will need Can alway get other fields later Select access method