HRL Distributed Modeling
strategy and initial results
HL-RMS Documentation future
Papers at AGU Conference 2001
The main goal of this research is to develop approaches for
exploitingthe spatial and temporal information in the NEXRAD Stage
III precipitationestimates to improve forecast accuracy at basin
outlets for NWS RiverForecast Centers. These approaches may
include re-calibrated and/ormodified lumped models, semi-distributed
applications of existingNWS models, gridded hydrologic models,
linear translation models,and others. The criteria for evaluating
new approaches primarilyfocuses on the ability to consistently out-perform
current continuous NWSlumped modeling approaches. Other evaluation
criteria include ease of bothparameter estimation and calibration.
Ease of operational applicationis also important and computational
time and storage requirements mustalso be considered.
It is hoped that this research will alsolead to improved methods of
flash flood forecasting.
Strategy and Initial Results
Our strategy was to first evaluate the use of
existingNWS hydrologic models being forced by the NEXRAD Stage
III precipitationestimates. This initial strategy is outlined
Phase 1 Results.
This is to be followed by Phase II, in which we will examine other
modelingapproaches including gridded distributed parameter models.
Primarily, work under Phase I focused
onapplying the Sacramento Soil Moisture Accounting Model
(SAC-SMA) in a lumped and semi-distributed manner. Before
any modeling wasperformed, several issues were investigated.
First, we firstcarried out several analyses of the mean areal precipitation
values derivedfrom traditional rain gage networks and Stage III
data. We foundthat seasonal and annual biases on the order
of 5-10% often exist betweenthe two products, indicating that the
SAC-SMA calibrated at a 6-hour timestep with rain gage data must
be recalibrated for use with hourly meanareal values from NEXRAD
Stage III. These tests have been expandedto include several
more years of radar data as well as data from the calibrationrain
gage network (Wang, et. al, 2000) In addition, extensive
tests were performed to analyze the sensitivity of the SAC-SMA and
threeother models to spatial scale. These tests revealed that
the fastresponding runoff components from the SAC-SMA
are very sensitiveto changes in the scale of precipitation forcing.
Also, it was determinedthat the Reformulated SAC-SMA was the least
sensitive to changes in spatial scale. Multi-year continuous
lumped simulations drivenby Stage III data and traditional rain
gage data were derived forseveral basins in the Illinois River
watershed in Oklahoma. In thesetests, it was found that the
radar forced simulations were better thanthe gage forced simulations.
Lumped parameter simulations of these basinsdriven by Stage III
data compared well to the observed streamflowrecords.
Thus, for these basins, lumped hourly applications of theSAC-SMA
proved to be an effective way to use the NEXRAD Stage III data.
Simulations from semi-distributed applicationsof
the SAC-SMA to the same basins were also generated.
Theintent of this work was to capture the spatial variability
of precipitation while keeping the Sacramento model parameters spatiallyuniform.
Multi-year simulations of lumped and semi-distributed applicationsof
the SAC-SMA were evaluated. These tests are among the first
comparisonsof lumped vs. distributed continuous modeling
approaches usinghigh resolution radar estimates of precipitation.
Surprisingly, itwas found that semi-distributed modeling approach
did not provide significantimprovement over the hourly lumped simualations,
even in cases with identifiedspatial variability of precipitation.
Problems related to sub-basinSAC-SMA parameter identification and
calibration are identified and discussedin the Phase 1 report.
The results from Phase I investigations have led to
theitems under Current Projects
Projects with Academic Institutions and other Organizations
- Distributed Modeling with
HL has begun development of the Research Modeling System (RMS).
Click here to see a brief overview and actual
- Development and Testing
of the Reformulated SAC-SMA
This work is a short term practical effort to modify a lumpedmodel
to exploit the spatial variability of precipitation as measured
Reformulationof the SAC-SMA Model to Account for the Spatial
Variability of Rainfall) This allows the
lumped SAC-SMA to generate fast response surface runoffsomewhere
in the basin even though the mean areal precipition amountis
not enough to generate surface runoff from the entire basin. The
upperzone tension water (UZTWM) and upper zone free water
(UZFWM) componentsof the SAC-SMA have been modified to accept a
distribution function ofrainfall derived from hourly Stage
III data. Analysis of the spatialvariability of hourly precipitation
indicates that a gamma distributioncan be used as and approximation
of the empirical distriubution. The parametersof the distribution
are computed for the rainy area and are based on meanareal precipitation
and coefficient of variation. Preliminary testingof the model
has shown that it is less sensitive to changes in spatialscale (
DistributedModeling: Phase 1 Results, ScaleDependencies
of Hydrologic Models to Spatial Variability of Precipitation).
Initial testing is currently underway with the Elk River above the
USGSgage at Tiff City, MO. Plans are to test this approach
further withBaron Fork above Eldon, OK, Peacheater Ck. above Christie,
OK, and the Blue River above Blue, OK.
- Development and Testing
of the LinearDistributed Routing Model (LDRM)
The LDRM has a grid point structure. Connectivity of the grid pointsis
determined based on both stream network definition and elevation
datausing an eight direction rule. It is assumed that each grid
cell consistsof a number of the hillslopes which are the source
of water for the ‘main’grid channel. All hillslopes have the same
characteristics inside eachgrid cell but they may be different from
pixel to pixel. It is also assumedthat all surface and sub-surface
runoff of a pixel reaches the ‘main’ channelof that pixel. The ‘main’
channel is the only source of water exchangebetween neighboring
pixels. A linearized kinematic wave model integratedover each pixel
is used to route runoff within the cell to the stream (hillsloperouting),
and to move water from one grid cell to another (channel routing).The
grid cell size can be a few hundreds of meters to tens of kilometers.The
lower limit of grid cell size is determined by the resolution of
inputdata and geographical information. The upper limit is determined
by thedegree of heterogeneity and stream channel network structure
of the basin.Two parameters should be estimated for each grid cell:
hillslopeand channel lag time parameters, although constant values
throughout allpixels can provide reasonable practical results in
some cases. Lag timeparameters can be estimated based on an available
unit hydrograph at abasin outlet (case of constant parameters for
all pixels), or by usingempirical relationships between GIS data
and water velocity. A preliminaryversion of the LDRM is available
with the HRAP grid resolution. Distributed (with HRAP grid resolution)
or lumped over entire basin or a number ofsub-basins total runoff
can be used as an input to LDRM. DistributedHRAP grid rainfall
can be used as a pattern to distribute runoff if lumpedinput version
- Semi-Distributed Modeling
of the BlueRiver, OK.
We are continuing our work with SAC-SMA semi-distributed modeling
withan investigation of the Blue River above the U.S.G.S. gage at
Blue, OK. Preliminary lumped modeling of this basin with Stage
III data indicatedpotential improvements to be gained by distributed
modeling approaches.This elongated basin has been disaggregated
into 8 sub-basins, each withits own synthetic unit hydrograph
and SAC-SMA parameters. TheSAC-SMA parameters were derived
using a procedure developed by VictorKoren (
Useof Soil Property Data in the Derivation of Conceptual Rainfall-Runoff
ModelParameters) Initial testing of this semi-distrbutedapproach
has shown clear improvement in basin outlet simulation accuracycompared
to lumped modeling. A Muskingum-Cunge routing scheme isbeing
tested for improved channel routing of runoff. Channel cross sectiondata
has been obtained from local agencies as well as a field visit.
A scheme for calibrating a semi-distributed model is also being
- Derivation of A Priori
Estimates ofSAC-SMA Model Parameters from GIS-Soils Data
This study is focused on developing a procedure to derive the SAC-SMAmodel
parameters based on soil texture data. It is hoped that thismethod
can be used to derive initial estimates of SAC-SMA parameters forboth
lumped areas and for sub-basins in a semi-distributed applicationof
the SAC-SMA. To quantify relationships of model parameters
withsoil properties, the assumption was made that the SAC-SMA tension
waterstorages relate to an available soil water, and that free water
storagesrelate to gravitational soil water. Porosity, field
capacity, andwilting point derived from STASGO dominant soil texture
for eleven standardlayers were used in estimating available and
gravitational water storages. SCS runoff curve numbers and saturated
hydraulic conductivity of differentsoils were also used. Analytical
relationships were derived for 11SAC-SMA model parameters.
Preliminary tests on a few basins in differentregions suggest that
most parameters derived from soil properties agreedreasonably well
with calibrated parameters for those basins. Accuracystatistics
of hydrographs simulated using calibrated and derived parameterswere
also close. Although calibrated parameter simulations usuallygive
higher accuracy, the gain is not significant. It means thatparameters
derived from soils data are very reasonable, and can beimproved
by using calibration if observed historical data are available.
The procedure has been tested on several basins so far with promising
and Programs Developed to Assist with Flash-Flood Guidance (derivingthreshold
runoff values) and Distributed Modeling
Overview: The Hydrologic Research Lab is activelyengaged
in cooperative research with a number of organizations. Throughthese
efforts and HRL internal research, it is hoped that we can achieveoptimal
use of high resolution NEXRAD precipitation products to improvethe
ability of the NWS to forecast river flows.
- MIT-HRL Cooperative Research
The purpose of this collaborative research is to compare simulationsfrom
a fully distributed, physically-based hydrologic model withthose
from a lumped application of the Sacramento Soil Moisture AccountingModel.
Comparisons will be performed on several basins representing differentclimatic
regimes across the country. Initially, tests will be conductedon
Baron Fork at Eldon, OK (307 sq. mi), Peacheater Creek at Christie,OK,
(25 sq. mi. sub-basin of Baron Fork, and the Blue River at Blue,
OK( 476 sq. mi.). Following these, tests
will be conductedon the Cheat River in W.VA.,
the Marais Des Cygnes Riverin Kansas and a small river basin in
Massachusetts. The MITmodel uses a TIN representation of the
landscape and computes runoff usingboth the Hortonian Infiltration
Excess and Saturation Excess mechanisms.
- HRC-HRL Cooperative Research
HRLis continuing its research with theHydrologic
Research Center. This project focuses on Monte-Carlo randomsampling
of parameters and radar rainfall input perturbations in the generationof
an ensemble of flow traces from a distributed model. Several basinsaround
the U.S. will be examined. Questions such as "Is spatiallydistributed
model event response different from the spatially-lumped modelresponse
in view of parametric and rainfall input uncertainty"? and "Howdependent
are the answers on model discretization scale?" will be addressed.
We hope to identify characteristics that make basins good
candidatesfor distributed modeling.
- University of Arizona Cooperative
HRL has had a long standing cooperative research program with the
Universityof Arizona regarding automated methods for calibrating
lumped hydrologicmodels. Lately, this work has been expanded
to include approachesfor calibrating semi-distributed versions of
(links to .pdf files)
HRL is planning to host an "Intercomparison
of distributed, semi-distributed, and lumped hydrologic models for
operational forecasting." We plan to announce this intent
during the AMS Hydrology Committee meeting Wednesday January 12,
2000. We will invite others (other Federal agencies, universities,
the private sector) to help us plan this activity. We will
have final details in place for announcement during the Spring 2000
AGU meeting (May 30-June 3 in D.C.) at the NWS/OH organized session
on distributed modeling.
In general, we plan to provide Intercomparison objectives and test
procedures,as well as test data sets via the web. We anticipate
Intercomparisonparticipants to download data, conducts tests at
their facilities, andpost results on the Intercomparison web site.
We will host a workshopafter tests have been completed for Intercomparison
participants to discussmodel pros and cons, and to record strengths
and weaknesses in an operationalsetting.
In the course of research, we have developed the followingtools
for use in HRL and in the RFCs:
Publications (see also list of all HRL publications click here)
- GIS tools
- MAPX preprocessor: This program is used to derive multi-year
timeseries of mean areal values of precipitation derived from
the Stage IIIxmrg files. These time series are called
MAPX time series inthe NWS, and are used in calibration and simulation
experiments. . A basinboundary in Lat/Lon coordinates is required.
A beta version of thisprogram has been released to the RFCs, and
can be used to derive MAPX timeseries for lumped basins as well
as semi-distributed sub-basins. A formal version of this program
is planned for this year.
- Plotting Routines for Manual Calibration: The Interactive
CalibrationProgram (ICP) has been enhanced with the addition of
the PLOT-TS function. This graphical routine allows the user to
display up to six panes, eachcontaining a number of plotted time
series. Time series with differenttime steps can be plotted.
- Hourly Statistics program: This code is used for evaluating
resultsfrom hourly time step modeling experiments, both lumped
and semi-distributed.It is planned that a final version of this
code will be incorporated intothe ICP.
- NOAA Technical Report NWS 51: NOAA Distributed Model Research and Development (2004) 2.7mb
- NOAA Technical Report NWS 50: Distributed
Modeling: Phase 1 Results (1999) 221 pages (ordering info: contact Mike Smith)
- B.D. Finnerty, M.B. Smith, V. Koren, D.J. Seo, G. Moglen, Space-Time Scale Sensitivity of the Sacramento Model to Radar-Gage Precipitaton Inputs, Journal of Hydrology, 203 (1997) 21-38.
- Distributed Parameter Hydrologic Modeling and NEXRAD for River Forecasting: Scale Issues Facing the National Weather Service
- D. Johnson, M.B. Smith, V. Koren, and B. Finnerty, Comparing Mean Areal Precipitation Estimates from NEXRAD and Rain Gauge Networks, Journal of Hydrologic Engineering, Vol. 4, No. 2 April, 1999, 117-124.
- D. Wang, M.B. Smith, Z. Zhang, S. Reed, V. I. Koren, Statistical Comparison of Mean Areal Precipitation Estimates from WSR-88D, Operational,and Historical Gage Networks, 80th Meeting of the AMS, Long Beach, Ca., January 10-14, 2000.
- M.B. Smith, V. Koren, Z. Zhang, D. Wang, Semi-Distributedand Lumped Modeling Approaches: Case Study of NEXRAD Data Application toLarge Headwater Basins in the Arkansas River Basin, 1999, Spring Meeting of the AGU, Boston.
- V. I. Koren, B.D. Finnerty, J. C. Schaake, M. B. Smith, D. J. Seo, Q. Y. Duan, Scale Dependenciesof Hydrologic Models to Spatial Variability of Precipitation, Journal of Hydrology 217(1999) 285-302.
- ThePotential for Improving Lumped Parameter Models using Remotely Sensed Data, V. Koren Paper J1.10, 13th Conference on Weather Analysis and Forecasting, August 2-6, 1993, Vienna, Virginia, 397-400.
- V. Koren, M.B. Smith, D.J. Seo, B.D. Finnerty, Reformulation of the SAC-SMA Model to Account for the Spatial Variability of Rainfall. (HRL internal publication. Contact Mike Smith for a copy)
- V. I. Koren, M. Smith, D. Wang, Z. Zhang, Use of Soil Property Data in the Derivation of Conceptual Rainfall-Runoff Model Parameters, 80th Annual Meeting of the AMS, Long Beach, Ca. January
- S. Lindsey, Strategy for Utilizing Radar-Based Precipitation Estimates for River Forecasting, ASCE International Symposium on Engineering Hydrology, San Francisco, Ca., July 25-30, 1993, 940-945
- Wang, D., M. Smith, Z. Zhang, S. Reed, V. Koren, Statistical Comparison of Areal Precipitation Estimates from WSR-88D, Operationaland Historical Gage Networks, AMS 15th Conferencein Hydrology, Jan. 10-14, 2000, Long Beach, CA
1) Mathematical models and their application to flash and
spring flood predictions
2) Numerical solution of unsteady flow equations and numerical experiments
- Koren, V., J. Schaake, K. Mitchell, Q.-Y. Duan, F. Chen, J.
Baker. A parameterizationof snowpack and frozen ground intended
for NCEP weather and climate models,JGR, Vol. 104, No. D16, 19.569-19585,
- Koren, V., Q.-Y. Duan, J. Schaake, and K. Mitchell. Validation
ofa snow-frozen ground parameterization of the Eta model.
14thConference on Hydrology, 10-15 January 1999, Dallas,
TX, AMS, J1.3,410-413, 1999.
- Koren, V., F. Kogan, C. Barrett. Parameterization of
hydrologicalmodel using NOAA/AVHRR data. Adv. Space Res., Vol.
19, No. 3, 507-510,Pergamon, 1997.
- Schaake, J., V. Koren, Q.-Y. Duan, K. Mitchell, and F. Chen.
Simplewater balance model for estimating runoff at
different spatialand temporal scales, JGR, vol. 101, No.
D3, 7461-7475, 1996.
- Koren, V., J. Schaake, Q.-Y. Duan, b. Koch, F. Kogan.
Parameter estimationof large- scale hydrological models using
land surface characteristics.Extended abstract on “Second
International Scientific Conferenceon the Global Energy and Water
Cycle”, 17-21 June, 1996, Washington, DC,USA, 1996.
- Koren, V., C. Barrett. Satellite Based, Distributed
Monitoring,Forecast, and Simulation (MFS) System for the
Nile River, in Book‘Applications of Remote Sensing in Hydrology’,
Canada, 187-200, 1995.
Use of Digital Soil Maps in a Rainfall Runoff Model
- Reed,S.M., and D.R. Maidment Use of Digital
SoilMaps in a Rainfall-Runoff Model, CRWR Online Report 98-8,
265 pp., December1998, http://www.ce.utexas.edu/org/crwr/reports/online.html.
- Reed,S.M., D.R. Maidment, and J. Patoux, Spatial Water Balance
of Texas,CRWROnline Report 97-1, 126 pp., February 1997, http://www.ce.utexas.edu/org/crwr/reports/online.html(Also
published as Texas Water Resources Institute Technical Report
- Reed, S.M., and D.R. Maidment A GIS Procedure for Merging NEXRAD
PrecipitationData and Digital Elevation Models to Determine Rainfall-Runoff
ModelingParameters, CRWR Online Report 95-3, 119 pp., September
and Maidment, D.R., 'Coordinate Transformations for Using NEXRADData
in GIS-Based Hydrologic Modeling', Journal of Hydrologic Engineering,April,
PhD Dissertation: A GIS-BasedDistributed
Parameter Hydrologic Model for Urban Stormwater Protection
- Smith, M.B. and Vidmar, A., 'Data Set Derivation for GIS-Based
Urban HydrologicalModeling', Photogrammetric Engineering and
Remote Sensing, Vol.60, No. 1, January 1994, 67-76.
- Smith, M.B., 'A GIS-Based Distributed Parameter Hydrologic
Model for UrbanAreas', Hydrological Processes, Vo. 7, 1993,
- Smith, M.B., and Brilly, M., 'Automated Grid Element Ordering
for GIS-BasedOverland Flow Modeling', Photogrammetric Engineering
and Remote Sensing,Vol. 58, No. 5, May 1992,
- Smith, M.B., Koren, V. Zhang, Z., and Wang, D. 'Lumped and
Semi-distributedModeling Using NEXRAD Stage-III Data: Results
From Continuous Multi-yearSimulations', Proc. of the 1999 Georgia
Water Resources Conference, pp.359-362, Kathryn J. Hatcher (Ed.),
March 30-31, 1999.
- Berich, R.H. and M.B. Smith, 1985, ‘LandSAT and Micro-GIS for
WatershedModeling’, Proceedings of the ASCE Specialty Conference
- Hydraulics andHydrology in the Small Computer Age, Lake Buena
Vista, Florida, August12-17, pages 668-673.
Development of a Spatially Distributed Model of Arctic Thermal and
- Ziya Zhang, Douglas L. Kane and Larry D. Hinzman. Development
andApplication of a Spatially Distributed Arctic Hydrologic and
Thermal ProcessModel (ARHYTHM), An International Journal of Hydrological
Processes, inpress, 1999.
- D. J. Morton, Ziya Zhang, L. D. Hinzman, and S. O'Connor.
The Parallelizationof a Physically Based, Spatially Distributed
Hydrologic Code for ArcticRegions. In proceedings of the
1998 ACM Symposium on Applied Computing,pp. 684-689, Atlanta,
- Ziya Zhang, Douglas L. Kane and Larry D. Hinzman.
Developmentof a Physically Based, Spatially Distributed
Hydrologic Model forArctic Regions. Abstract and poster presented
at 1996 Fall meeting of theAmerican Geophysical Union.
- Ziya Zhang, Douglas L. Kane, Larry D. Hinzman and Douglas
J. Goering.Ground Dominates Arctic Hydrologic Response. Proceedings
of the Workshopon Environmental Technology and Resource Development
in Alaska, M. A. Tumeoand C. R. Woolard (Eds.), Fairbanks, Alaska,
May 22-24, 1996.
- Ziya Zhang, Doug Kane and Larry Hinzman. Simulations
of Hydrologicand Thermal Processes in Arctic Regions. Landscapes,
American Associationfor the Advancement of Science, The 46th Arctic
Division Science Conference,Abstract and presentation, 19-22 September
- Larry D. Hinzman, Douglas L. Kane and Ziya Zhang. A Spatially
DistributedModel for Arctic Regions. International GEWEX
Workshop on Cold-Season/RegionHydrometeorology, Summary Report
and Proceedings, 22-26 May 1995, Banff,Alberta, Canada, Int. GEWEX
Project Office Publication, Series, No. 15,Washington, D.C., 236-239.
- Douglas J. Goering, Ziya Zhang, Larry D. Hinzman
and DouglasL. Kane. A Spatially Distributed Hydrologic Model Applied
to an ArcticWatershed. Wadati Conference on Global Change and
the Polar Climate, 7-10 November, 1995, Tsukuba Science
City, Japan. GeophysicalInstitute, University of Alaska
and Wadati Conference Local OrganizingCommittee, Japan.
Simulations of Hydrologically Significant Winter Storms over a MountainousWatershed