LDAS Meeting Minutes
July 13th
2001
NWS/OHD
I. STATUS OF REALTIME N-LDAS FORCING AND LSM RUNS
Each of the four LDAS modeling groups reviewed the status of their realtime LDAS model executions at NCEP, using NCEP-produced hourly surface forcing on the common LDAS grid. The NOAH, VIC, and SAC LSMs continue to execute in realtime up to the current day, with NOAH and SAC having continuously cycled forward from 17 Apr 99 and VIC having continuously cycled forward from Oct 00. Owing to the concurrent effort by NASA/GSFC of preparing 3 years of retrospective forcing for retrospective LDAS LSM runs (see below) for the period of Sep 96 to Sep 99, the MOSAIC realtime runs have been suspended for a time, but will resume and catch-up to realtime in the near future.
Justin Sheffield of the VIC group of Princeton checked the common LDAS GRIB output (e.g. from the VIC LSM) to confirm the presence of sufficient precision (as specified in the common PDS table of the common LDAS GRIB output document) in the GRIB output to maintain water balance in the monthly water balance components. Justin found an acceptably small balance residual of less than 1 mm/mo, which with month-to-month residual sign-cancellation over a one-year period yielded an annual water balance residual of less than 0.3 mm month, which satisfies the criteria applied in the PILPS projects.
On 20 Jun 01, the domain size of the NESDIS operational GOES GEWEX product suite (including the solar insolation and skin temperature products used by the LDAS project) was expanded by about 10-20 percent to encompass the entire N-LDAS domain, based on work by Byron Raines of Dan Tarpley's Land Team in NESDIS/ORA, and simultaneously Dag Lohmann of NCEP changed the realtime LDAS forcing software to utilize this larger domain of GOES solar forcing.
Istvan Laszlo of NESDIS/ORA and Jesse Meng of NCEP/EMC (sponsored by Paul Houser's global LDAS project) presented results showing that both the retrospective and realtime GOES-based solar forcing retrievals suffer a substantial low bias over snow cover, mainly arising from the difficulty of distinguishing cloud cover from snow cover in the visible spectrum during the GOES retrieval algorithm. A proposal to average GOES and EDAS hourly solar forcing when the daily NESDIS IMS snow cover analysis shows snow cover at a point was turned down for practical reasons stemming from the fact that none of the LDAS LSMs ingest the NESDIS IMS snow cover. Rather, all four LDAS LSM executions simply cycle the physically modeled snow cover that physically evolves in each LSM in response to the common surface forcing (as done in the PILPS and GSWP tradition).
ACTION: NCEP will produce a document that describes the methodology and data sources that NCEP uses to produce the realtime hourly LDAS forcing used in common by all LDAS LSMs. This document will include a periodically updated log that marks dates of discovered forcing glitches and changes in the NCEP realtime LDAS forcing methods or data sources.
II. UPDATING APPROACH TO 3-YEAR RETROSPECTIVE N-LDAS RUNS
In order to overcome initial spin-up behavior in LDAS LSM runs (which can span many months) and to overcome the various occasional glitches (such as interruption of an input observation streams) that unavoidably plague realtime LDAS forcing, the LDAS group has decided to execute a clean 3-year retrospective LDAS for the period 30 Sep 96 to 27 Sep 99 (spanning three complete "water" years). NASA/GSFC (Brian Cosgrove in Paul Houser's LDAS group) has undertaken the large substantial effort of constructing the retrospective forcing fields, highly leveraging NCEP LDAS forcing software and GCIP product archives at NCAR (Eta/EDAS and Stage IV), NCEP (Higgins retrospective precipitation analyses), and U.Maryland (Pinker retrospective GOES GEWEX products).
At the present meeting, it was decided that each of the four LSM modeling groups in LDAS will execute their 3-year retrospective runs on their own local computer platforms, rather than on the realtime LDAS workstation platform at NCEP, which does not have the capacity to simultaneously host both the realtime and retrospective LDAS runs (one exception may be the SAC model, which requires a relatively little computational resource compared to the NOAH, VIC, and MOSAIC LSMs). All the hourly output from the four retrospective LDAS LSM runs for the cited 3-year retrospective period will be transferred by ftp or tape to an archive at NASA/GSFC.
After discussing various options for common initialization of the land states on the 30 Sep 96 initial date of the retrospective LDAS runs, it was decided to a) initialize zero snowpack (since the start date is around the date of annual snowpack minimum, b) initialize zero canopy water content (since this spins-up within about 1-day's time) and c) take initial soil temperature and soil moisture from the land states of the NCEP Global Reanalysis II (where the soil moisture will be provided in terms of a soil moisture availability fraction, in the range 0-1, with respect to soil saturation).
ACTION: For the 3-year retrospective LDAS runs, NCEP/EMC will derive common LDAS initial land states of soil temperature and soil moisture on the common LDAS grid from the NCEP Global Reanalysis II valid at 12Z on 30 Sep 96. The soil moisture state will be provided in terms of a soil moisture fraction relative to saturation.
ACTION: NASA/GSFC will have to establish remote login access for NCEP, Princeton U., and NWS/OHD to the NASA/GSFC LDAS computational platform.
III. CHOOSING TEST BASINS TO VALIDATE LDAS STREAMFLOW SIMULATIONS
John Schaake presented the approach for his choosing of test basins within the CONUS for LDAS streamflow simulation validation. In his first pass of choosing basins, he first choose basins that a) exceeded 2000 sq km in area (so as to span an area of at least 10 or so LDAS grid boxes) but did not exceed about 10000 sq km (so as to minimize the likelihood of human regulation that characterizes larger basins) and b) that have a sufficient density of precipitation measuring gage stations available to the LDAS data stream to provide decent precipitation forcing. These two criteria yielded 437 basins, of which 188 have realtime USGS streamflow observations available via ftp. From these 188 candidate test basins, John lastly selected 26 for initial LDAS streamflow validation exercises. For these 26 basins, John derived the streamflow connectivity files for our common 1/8-th degree LDAS grid and John emailed these connectivity files to all the LDAS project principals.
IV. VALIDATION OF LDAS SOLAR RADIATION AGAINS SURFRAD OBSERVATIONS
At the GAPP/GCIP annual PIs meetings in early May 01, Eric Wood and Justin Sheffield showed that realtime hourly EDAS solar insolation forcing provided as a backup to hourly GOES solar insolation forcing in LDAS forcing files had a 1-3 hour phase lag error. Jesse Meng showed results from his SURFRAD validation of various solar insolation products, that the LDAS phase lag error in EDAS solar insolation was a) greatly reduced following an Aug 99 fix by Curtis Marshall of a glitch in the realtime LDAS solar insolation software and b) rather negligible by the Dec 99 period of LDAS forcing, and completely absent by the May 00 period of LDAS forcing. Jesse also determined there is no solar phase-lag error in the retrospective forcing files constructed by Brian Cosgrove of NASA/GSFC, since Brian's forcing software leveraged that of NCEP's after the Aug 99 fix by Curtis Marshall of the main source of the phase lag error.
ACTION: Jesse Meng will expand his SURFRAD validation to early months in 2000 to better pinpoint the month between Dec 99 and May 00 when the EDAS solar phase-lag error disappears in LDAS forcing files.