15th AMS Conference on Hydrology
9-14 Jan 2000, Long Beach, CA
1.1 THE COLLABORATIVE GCIP LAND DATA ASSIMILATION (LDAS) PROJECT
AND SUPPORTIVE NCEP UNCOUPLED LAND-SURFACE MODELING INITIATIVES
K. Mitchell, C. Marshall, D. Lohmann, M. Ek, Y. Lin, P. Grunmann, P. Houser1, E. Wood2, J. Schaake3,
D. Lettenmaier4, D. Tarpley5, W. Higgins6, R. Pinker7, A. Robock8, B. Cosgrove1, J. Entin1, Q. Duan3
Environmental Modeling Center (EMC)
National Centers for Environmental Prediction (NCEP)
1NASA Goddard Space Flight Center (GSFC)
3National Weather Service (NWS) Office of Hydrology (OH)
4University of Washington
5NESDIS Office of Research and Applications (ORA)
6NCEP Climate Prediction Center (CPC)
7University of Maryland
The soil moisture and runoff from the 4-D data assimilation systems (4DDA) of coupled land-atmosphere models often suffer substantial errors and drift owing to biases in the precipitation, surface air temperature, and surface radiation in the land-surface forcing of the coupled system. To constrain such errors and drift, some developers apply soil moisture nudging techniques in coupled 4DDA; but such nudging can introduce other undesirable behavior, such as over-amplified annual cycles of soil moisture or lack of water conservation.
As an appealing alternative to coupled land-surface 4DDA, a consortium of GCIP-supported groups has undertaken the development and execution of an uncoupled Land Data Assimilation System (LDAS). The LDAS execution is hosted on an NCEP developmental computing platform, and supported by NCEP in collaboration with NASA/GSFC, NWS Office of Hydrology, NESDIS /ORA, Princeton and Rutgers Universities, and Universities of Washington and Maryland.
Specifically, these close partners are developing, executing, and validating a prototype, national,
realtime, hourly, 1/8-th deg, distributed, uncoupled, land surface simulation system. This system consists of several leading, physically-based, land-surface models (LSMs) running in tandem on a common grid and driven by common surface forcing fields. The hallmarks of the forcing fields are observed, hourly, gage/radar precipitation and observed GOES-based satellite-derived surface solar insolation. Additionally, a common streamflow routing model is being applied to each LSM's gridded runoff on the shared common grid. Finally, the "DA" for "data assimilation" in LDAS denotes a later Phase II thrust that will include the assimilation of satellite-derived land-surface fields, such as skin temperature, soil moisture, snowpack, and vegetation density and greenness. Another future phase will be a forecast component, where LDAS will be integrated days, weeks, and months into the future using ensemble surface forcing (including ensemble QPF) from the weather and seasonal prediction models of NCEP.
The highly successful GCIP-sponsored regional PILPS-2(c) project (Wood et al., 1998) and the ISLSCP-sponsored Global Soil Wetness Project (GSWP, Dirmeyer et al., 1999) demonstrated the viability of executing physically-based, distributed, uncoupled, land-surface models over very large spatial domains, provided suitably reliable and reasonably dense observed precipitation forcing was available. Having participated in these retrospective LDAS projects, the aforementioned institutions have joined together to construct and demonstrate a realtime LDAS.
Key thrusts of early phases of the LDAS project are a) LSM model validation and intercomparison (e.g. using streamflow, flux station measurements, measured soil moisture and temperature, and satellite-derived surface skin temperature) and b) identifying, understanding, and sharing successful physical process and parameter estimation algorithms and methods.
2. LAND FORCING ERROR IN COUPLED 4DDA
Substantial surface forcing errors (especially in precipitation and solar insolation) in coupled land-atmosphere 4DDA systems are a key reason for pursuing the alternative LDAS approach. Here we provide examples of surface forcing errors from the NCEP mesoscale Eta model and its associated Eta-
based 4-D data assimilation system (EDAS, Rogers et al., 1996). The EDAS began fully continuous self cycling of all atmospheric and land state variables on 03 June1998, which has yielded 16 months of full cycling at the time of this writing.
The summer months are traditionally the months wherein the EDAS exhibits its largest precipitation and radiation biases because of the greater difficulty of predicting precipitation and cloudiness during the convective season. Fig. 1 provides the July 1999 monthly total of U.S. precipitation from the EDAS (left) and gage observations (right). Notable in this figure is the large positive EDAS precipitation bias in the southwest (monsoon region), over the Gulf sates, and the eastern seaboard. Of particular interest is the northern half of the eastern seaboard, which was in the grips of drought during July 1999. Fig. 1 shows the EDAS generated substantially higher precipitation amounts over the drought region (e.g. eastern Pennsylvania, eastern New York, most of New England). Additionally, the EDAS misplaced the Wisconsin precipitation maximum to the southwest over Iowa and eastern Nebraska.
Fig. 2 is a snapshot example of solar insolation biases in the EDAS (and Eta), which arise from errors in both a) forecast cloudiness and b) insufficient atmospheric absorption of solar radiation in clear skies. Over the west and northeast in Fig. 2a, the EDAS cloud cover (not shown) was too low for this chosen case. The Eta positive bias in surface solar insolation is of order 10-12 percent and is persistent across the U.S. domain and seasons, as shown by a number of Eta model validation studies (Betts et al., 1997; Marshall et al., 1999; Berbery et al., 1999; Hinkelman et al., 1999). Most recently, NESDIS/ORA and NCEP/EMC have collaborated to provide monthly and regional scatter plots of point-to-point match-ups of EDAS modeled versus GOES-8 retrieval of surface solar insolation. These scatter plots may be viewed in realtime at http://orbit-net.nesdis.noaa.gov/goes/gcip/html/scatter.html. Eta and GOES-8 insolation maps like Fig. 2 may be viewed via the last hyperlink at above site. Comparison plots like Fig. 1 of EDAS and observed precipitation may be viewed at http://www.emc.ncep.noaa.gov:/mmb/gcip.html.
Fig. 1 The accumulated monthly precipitation (mm) for July 1999 from A) the NCEP operational EDAS
and B) gage observations (as analyzed to a 0.25-degree grid).
Fig. 2 Instantaneous downward surface solar insolation at 18 GMT on 30 July 1999 from A) the NCEP
operational EDAS and B) the NESDIS GOES-8 satellite retrieval.
3. LDAS CONFIGURATION
The LDAS grid shared by all LDAS LSMs is a 464 X 224 1/8th-degree grid bounded by 25 N,
53N, 67W, and 125W. On this grid, NASA/GSFC used high resolution source datasets (typically 1-km) to derive common 1) land/sea mask, 2) terrain heights, 3) dominant and subdominant vegetation classes, and 4) vegetation parameters for use by all LDAS LSMs. Similarly on this grid, NWS/OH used high resolution soils databases to derive soil characteristics (such as texture) and companion soil physical parameters. The above fields (plus the common LDAS forcing and LDAS LSM outputs) may be viewed at the LDAS web site, hosted and supported by NASA/ GSFC at http://ldas.gsfc.nasa.gov. A fundamental feature of this LDAS project is the simultaneous, realtime, parallel execution of several modern-era LSMs.
At present these designated parallel LSMs are 1) the NASA/GSFC MOSAIC LSM (Koster and Suarez, 1996), 2) the Princeton/Washington VIC LSM (Liang et al., 1996) and 3) the NCEP Eta LSM (Chen et al., 1997) -- now called the Community "NOAH" LSM (see history, physical description, and availability of the NOAH LSM in Paper P1.22 this volume). In the future, NWS/OH will add the Sacramento LSM as a traditional "lumped" and calibrated LSM for comparison. We expect (as resources and support allow) to add other LSMs and partners to our realtime setting (e.g.TOPLATS and CLM-Common Land Model). The LSMs are executing with physical time steps of order 15 minutes with around 3-6 soil layers, and providing hourly output of state variables, water and energy fluxes, and diagnostic quantities. All LDAS surface forcing and LSM model outputs are being archived for wider use in a common format (the WMO GRIB standard). Each LDAS partner is pursuing retrospective LSM runs covering 3-50 years .
NCEP/EMC is providing the realtime LDAS computing platform (a 4-processor SGI Origin 2000) and the supporting infrastructure therein (hardware maintenance, 24/7 power backup, data migration and mass storage, compilers, and software libraries, including GRADS / GRIB).
4. COMMON LDAS FORCING
Following PILPS (Wood et al., 1998) and ISLSCP (Sellers et al., 1996), a common unified LDAS forcing data set is produced and applied to all the LDAS LSMs. The development, assembly, QC, production, and archive of these forcing data sets in realtime is a substantial undertaking by NCEP/EMC. These surface forcing sets, which are hourly and rendered on the common land mask of the LDAS grid, include downward solar, PAR, and longwave radiation, precipitation, 2-m temperature and humidity, 10-m wind speed, and surface pressure. Aside from precipitation and downward insolation/PAR, the remaining forcing fields are taken from the 3-hourly, 40-km, EDAS analyses, which we temporally and spatially interpolate to hourly on the LDAS grid. A notable final step, using a standard lapse rate, is to apply a terrain-height correction to 2-m temperature and surface pressure, to account for the finer terrain resolution of the LDAS grid versus the EDAS grid. Keeping relative humidity constant, the specific humidity and downward longwave are also adjusted to reflect the temperature adjustment. The final and crucial surface forcing fields (model independent) are 1) the GOES satellite retrieval of hourly 0.50 degree surface insolation (e.g Fig. 2b) and PAR using the retrieval algorithm of Pinker et al., (1999), 2) the daily 0.25-degree gage-only rainfall analysis of Higgins (1999, private communication) typically using around 5-6 thousand daily gage reports of precipitation (e.g. Fig. 1b), and 3) the hourly, 4-km, national "Stage IV" radar-dominated rainfall analysis of Baldwin and Mitchell (1997). The insolation and daily rainfall are interpolated to the LDAS grid. Finally, the hourly Stage IV precipitation analysis is used only to derive hourly temporal weights on the LDAS grid. These weights are used soley to partition the daily gage-only precipitation into hourly amounts. The NCEP LDAS forcing fields may be viewed at http:// ldas.gsfc.nasa.gov or ftp'd from ftp.ncep.noaa.gov at /pub/gcp/ldas/noahoutput.
5. PILOT LDAS RESULTS OF NOAH LSM
From the start of the LDAS collaboration in early summer 1998, it took about one year to spin up the computing resources, databases, forcing fields, realtime data links, and surface characteristics. Unbroken real-time LDAS forcing and archive began on 16 April 1999. The terrain-height adjustments in the forcing fields began in mid July 1999. At the time of this writing, only the NOAH LSM is executing daily and cycling forward using LDAS forcing. A pilot realtime MOSAIC LDAS has been executing since summer of 1998 from EDAS-only forcing (see Paper 1.3 this volume). MOSAIC LSM cycling using the new NCEP LDAS forcing described here began just at the time of submission of this paper. Realtime VIC LDAS cycling is imminent (see Paper 1.4 this volume). Results of NOAH, MOSAIC, and VIC LDAS cycling will be presented at the conference.
For a first LDAS result, we again focus on the end-of-July 1999 eastern seaboard drought episode. Fig. 3b shows the top 1-m soil moisture fraction of the NOAH LSM LDAS at that time, after 107 days of cycling, which includes the observed 30-day rainfall of Fig. 1b. Fig. 3a shows the corresponding soil moisture from the cycled coupled land/atmosphere EDAS, which includes the 30-day model dependent EDAS rainfall of Fig. 1a. A northeast U.S. drought signature is clear and widespread in the LDAS, but virtually absent in the EDAS.
Fig. 3 The percent soil moisture (wrt wilting point and saturation) of the top 1-meter soil column at 12 GMT on 31 Juy 1999 for A) the NCEP EDAS and B) the NOAH LSM LDAS.
All LDAS LSMs produce gridded runoff on the LDAS grid. For this, NCEP has developed a common streamflow routing model (see Paper 1.2 this volume) to be applied uniformly to all the LDAS LSMs for intercomparison and validation of simulated streamflow. Fig. 4 (left) shows the basins where the routing model has been constructed to date. Additional basins will be added in the near future. Applying the routing model to NOAH LDAS output over the Arkansas-Red River basin, and a smaller sub-basin therein, yields the streamflow time series in Fig. 4 (right). At conference, streamflow from other LDAS LSMs and observations will be added.
Fig. 4 Left: Plot of major river basins for which the common LDAS routing model has been developed
to date. Right: NOAH LSM streamflow times series for the Arkansas-Red River basin (top) and a smaller
interior basin toward the southeast therein(bottom) from 16 April to 04 September 1999.
Baldwin and Mitchell,1997: The NCEP hourly multi-sensor U.S. precipitation analysis for operations and research. Preprnt,13th AMS Conference on Hydrology,54-55.
Berbery, E. H.,1999: Assessment of land-surface energy budgets from regional and global models. J. Geophys. Res., 101, 329-19,348.
Betts, A., et al., 1997: Assessment of the land surface and boundary layer models in two operational versions of the NCEP Eta model using FIFE data. Mon. Wea. Rev., 125, 2896-2916.
Chen, F, et al.,1997: Impact of atmospheric surface-layer parameterizations in the new land-surface scheme of the NCEP mesoscale Eta model. Boundary-Layer Meteo, 85,391-421.
Dirmeyer, P., J. Dolman, and N. Sato, 1999: The pilot phase of the Global Soil Wetness Project. Bull. Amer. Meteor. Soc., 80, 851-878.
Hinkelman, L. M. et al., 1999: An evaluation of NCEP Eta model predictions of surface energy budget and cloud properties by comparison with measured ARM data. J. Geophys. Res., 104, 19,535-19,549.
Koster, R., and M. Suarez, 1996: Energy and Water Balance Calculations in the MOSAIC LSM. NASA Tech Memo 104606, Vol 9.
Liang, X., E. Wood, and D. Lettenmaier, 1996: Surface and soil moisture parameterization of the VIC-2L model: Evaluation and modifications. Global Planet. Change, 13, 195-206.
Marshall,C,et al,1999: Evaluation of the new land-surface and planetary boundary layer parameterization schemes in the NCEP mesoscale Eta model using Oklahoma Mesonet observations. Preprint, 14 AMS Conference on Hydrology, 265-268.
Pinker, R., D. et al., 1999: Surface radiation budgets in support of the GEWEX Continental Scale International Project (GCIP). Submitted to J. Geophys. Res.
Rogers, E., et al., 1996: Changes to the operational “early” Eta analysis/forecast system at the National Centers for Environmental Prediction. Wea. Forecasting, 11, 391-413.
Sellers, P. J. et al., 1996: The ISLSCP initiative I global datasets: surface boundary conditions and atmospheric forcings for land-atmosphere studies. Bull. Amer. Meteor. Soc., 77, 1987-2005.
Wood, E. F. et al., 1998: The Project for Intercomparison of Land-surface Parameterization Schemes (PILPS) phase 2(c) Red-Arkansas River basin experiment: 1. experiment description and summary intercomparisons. Global and Planetary Change, 19, 115-135.