Future Science Developments in NWSRFS

1
Future Science Developments in NWSRFS

• Mike Smith
• NWS/OHD/Hydrology Lab
• October 21, 2003

michael.smith_at_noaa.gov
2
Goals of New Science

• Improve accuracy
• Quantify uncertainty

3
New ScienceIncrease Accuracy Quantify
Uncertainty

• Increase accuracy
• Distributed modeling
• Improved snow frozen ground modeling
• Improved estimates of precipitation, temperature,
  evapotranspiration
• Advanced hydraulic routing procedures
• Advanced parameter estimation/calibration
  procedures
• VAR data assimilation

4
New ScienceIncrease Accuracy Quantify
Uncertainty

• Quantify Uncertainty
• Short term ensemble forecasting
• Use of longer term forecast forcings
• Major science questions

5
What is a distributed hydrologic model?

• -a model which accounts for the spatial
  variability of factors affecting runoff
  generation
• precipitation
• terrain
• soils
• vegetation
• land use
• channel shape

6
Hydrologic Modeling Approaches
Distributed
Lumped

• Rainfall, properties averaged over basin
• One rainfall/runoff model
• Prediction at only one point

• Rainfall, properties in each grid
• Rainfall/runoff model in each grid
• Prediction at many points

7
Benefits of Distributed Modeling

• 1. Finer Scale Modeling Can Lead to Better
  Results
• Ability to Model/Predict Processes in Basin
  Interior Flash Flood
• Land Use Change Analysis
• Parameterization Simplified using GIS Data

8
Which distributed model to use?

• Strategy involves
• HL Distributed Model
• Distributed Model Intercomparison Project (DMIP)

9
Hydrology LabDistributed Model(HL-Research
Modeling System HL-RMS) Project Leader Victor
Koren

• Modular, flexible modeling system
• Gridded (or small basin) structure
• Independent rainfall-runoff calculations for each
  grid cell (SAC-SMA)
• Grid to grid routing of runoff (kinematic)
• Channel routing (kinematic)

10
Sacramento Soil Moisture Accounting Model
11
Treatment of SAC-SMA runoff components

• Fast runoff components
• Surface
• Direct
• Impervious

• Slow runoff components
• Interflow
• Supplemental baseflow
• Primary baseflow

Hillslope routing
Channel routing
12
Conceptual Channels
Drainage Density Illustrated 1.07
0.007
0.0019
0.009
0.08
0.01
Actual Channel Network
Conceptual Channel Network
13
Example A-Priori Soil Parameter Grid
UZTWM
14
Application of HL Distributed Model
15
Hydrologic Response at Different Points in the
Blue River Basin
200
160
120
Flow (CMS)
80
40
0
4/3/99 000
4/3/99 1200
4/4/99 000
4/4/99 1200
4/5/99 000
4/5/99 1200
4/6/99 000
Hydrograph at Location A
200
Distributed Lumped Observed
160
A
120
Flow (CMS)
80
40
B
0
4/3/99 000
4/3/99 1200
4/4/99 000
4/4/99 1200
4/5/99 000
4/5/99 1200
4/6/99 000
Hydrograph at Location B
200
160
Flow (CMS)
120
80
40
0
4/3/99 000
4/3/99 1200
4/4/99 000
4/4/99 1200
4/5/99 000
4/5/99 1200
4/6/99 000
Hydrographs at Basin Outlet
16
Large Area Application of HL Distributed Model
ABRFC
25,000 computational elements
outlet
17
Distribution of Upper Zone Free Water
Feb. 12, 1997 600 (mm)
18
Distributed Model Intercomparison Project-
DMIP(First) Broad Comparison of Lumped and
Distributed Models

• Sponsorship NWS OHD/HL and GCIP/GAPP
• Results used to guide NWS future research into
  finer scale modeling
• Participation from universities, other Federal
  agencies and internationally

19
Why is DMIP Needed?

• Not a clear path to distributed model for NWS
• Hypothesis high resolution data will lead to
  better results
• NWS needs scientific community to help guide its
  research
• To understand the value of NEXRAD and spatial
  data to improve river forecasting
• To provide a venue to compare complex models to
  those used in operational forecasting

20
DMIP Elements

• HL provided data sets for test basins on DMIP web
  site
• Participants accessed data, generated simulations
• HL compared simulations to NWS lumped and
  distributed model runs
• HL hosted summary workshop

21
DMIP Study Basins
6
5
4
3
1
2
Elk R.
Arkansas R.

• Ill. R. at Tahlequah, OK
• Baron Fork at Eldon, OK
• Peacheater Creek, OK
• Flint Creek at Kansas, OK
• Ill. R at Watts, OK
• Ill. R at Savoy, OK

Red R.
Blue R.
22
DMIP Data Sets

• NEXRAD hourly 4km. estimates (93-00)
• DEM data
• Cross sections
• Meteorological data
• Observed hourly flow data
• Soils information
• Vegetation

23
DMIP Participants

• Blacklands Research Center
• Utah State U. and NIWA, New Zealand
• Wuhan U., China
• U. of Waterloo, Canada
• UC Berkeley

• MIT
• NCEP/EMC
• HL
• U. of Ariz.
• HRC
• DHI
• U. of Ok.

24
DMIP Schedule

• March 2001 Data sets were available
• March 31, 2002 All simulations were due to HL
• August 21-23, 2002 - Workshop discussed of
  results, publications, and future DMIP phases
• Current Publication of results in Special Issue
  of Journal of Hydrology

25
DMIP Results
basin outlets
interior points
26
DMIP Results
interior points
basin outlets
27
DMIP Overall Results

• Proved that for certain events and basins,
  distributed models provided benefit compared to
  lumped.
• Demonstrated that reasonable hydrographs can be
  predicted at interior points.
• Exposed research models to operational data and
  concerns.
• Acknowledged that DMIP is a powerful method to
  get science from research community to
  operations.
• Showed need to understand data errors and
  streamflow.
• Established that OHD/HL distributed model
  research is very competitive.

28
Challenges of Distributed Modeling

• Space/Time
  Variability
• Does accounting for the space/time variability of
  input data and parameters guarantee better
  results?

Effect of noisy rainfall data on the peak volume
at different simulation scales.
75
50
Error
25
0
Increasing model resolution
Koren et al., 1999, Scale dependencies of
hydrologic models to spatial variability of
precipitation , Journal of Hydrology
29
Future Phases of DMIP

• Basins in other climatic regimes
• Snow
• Orographics
• Other data for verification
• Soil moisture
• Nested basins
• Synthetic simulation analyses

30
Snow Modeling SNOW-17, Energy Based, and
Distributed Snow Modeling
31
Motivation

• Improve performance of SNOW-17
• Make possible to run it in a distributed mode
• Investigate performance of energy-budget snow
  models in quasi-operational setting

32
SnowMIP results for Sleeper River (USA) site 23
energy based models and SNOW-17 (Uncalibrated
model results).
33

• Intermediate Snow-17 Modifications
• Addition of snow depth simulations
• Dynamic wind function
• Liquid water refreezing formulation

34

• OHD PLANS
• Distributed snow modeling
• Include SNOW-17 into HL-RMS to make it
  available for RFC
• tests
• Work on a grid type SNOW-17 parameterization
• Energy based snow modeling
• Collect and analyze new data sources to run an
  energy based snow model over US
• Analyze dependency of simulation results from
  energy and
• SNOW-17 models
• Define the most reliable data sources to run
  an energy snow model in the HL-RMS setting
• We expect to hire a person to perform data and
  comparison
• analyses

35
Frozen Ground Modeling Efforts

• Existing parameterization
• Need in more physical approach to describe heat
  fluxes
• Large number of parameters
• Requirements for new method
• Simple enough procedure to run with limited noisy
  operational data
• Procedure should be compatible with SAC-SMA
  complexity

36
Existing Frozen Ground Model

• SAC-SMA frozen ground component has two parts
• Frost index calculation component
• Modification of the water balance using frost
  index
• Frost index component
• Mimics Stefans solution for freezing depth
• Accounts conceptually for snow depth
• Has four empirical parameters
• Cg bare ground frost coefficient
• Cs reduction in Cg due to snow pack
• Hc daily thaw rate from ground heat
• Ct thaw coefficient for water entering the soil

37
Existing Frozen Ground Parameterization

• Modification of the water balance (percolation
  and interflow reduction) using frost index
• Mimics empirical relationship between losses
  reduction and soil saturation and freezing depth
• Uses threshold type dependency on the frost index
• Has three empirical parameters
• FIL frost index threshold
• Cr water withdrawal reduction under saturated
  conditions
• x exponent of lower zone soil moisture
  deficiency ratio

38
Linking of soil moisture (SAC-SMA) and heat states
39
Frost Index Replacement

• N-layers soil column
• The layer-integrated form of diffusion
  equation
• Soil moisture and heat fluxes are simulated
  separately at each time step
• Surface temperature is equal to air
  temperature
• Lower boundary is set at the climate annual
  air temperature
• Unfrozen water content is estimated as a
  function of soil temperature,
• saturation rate, and ice content

40
River Mechanics

• Routing of large rivers
• Tidal effects
• Flood inundation mapping
• Dam failure analysis
• Sediment transport

41
Recent Enhancements to FLDWAV Model
Stand-Alone model - Ability to model channel
networks including bifurcations - Ability to
handle floodplain compartments (network levee
options) - Ability to model mudflows -
Ability to export information for FLDVIEW and
FLDAT applications - Muskingum-Cunge Routing
method Operational model - Added ability to
blend observed and NOS tide time series -
Added ability to blend observed adjust computed
stage time series - Tested several internal
boundary options including rating curves
bridges - Ability to model channel networks
including bifurcations - Ability to handle
floodplain compartments (network levee
options) Ability to export information for
FLDVIEW and FLDAT applications
42
Flood mapping
43
Future Enhancements to FLDWAV Stand-Alone Model
Add new capabilities Sediment transport
Pollutant transport Routing flow thru
culverts - Channel losses - Practical
Updating Techniques - Ice Jams Add
DWOPER/DAMBRK capabilities not currently in
FLDWAV - Landslide generated waves -
Multiple movable gates - Improved method of
modeling wind effects
44
MOPEX(Model Parameter Estimation
Experiment)GOALS

• Develop improved methods for a priori estimation
  of model parameters for hydrological and
  hydrometeorological models operating at space
  (100-100,000 km2) and time scales (10 min to 1
  day) appropriate for a range of applications
  including river forecasting and water resources
  management, weather forecasting, climate
  prediction, and climate change study.

45
MOPEX Participating Models

• Simple Water Balance - NWS
• Sacramento - NWS
• ISBA
• SWAP Gusev/Nasonova
• VIC Princeton/Washington
• USGS - PRMS
• Sacramento Thian Yew Gan, U. Alberta Canada
• NOAH NWS
• Kuni Takeuchi - Japan

46
Data Assimilation

• Bring in all available data (observations,
  estimates, model output, etc.) into the
  hydrologic model while maintaining dynamic
  consistency of the physical processes being
  modeled
• Improves forecast by reducing errors in the
  initial and boundary conditions
• Gives a better handle on uncertainty, scale

47
Assimilation Background

• Two biggest hurdles to objective operational
  hydrology
• Model calibration
• Run-time modification (MOD)
• (Semi-) Automatic run-time modification proven
  difficult

48
Assimilation Background (cont.)

• Significant issues with SS-SAC recognized
• Variational assimilation (VAR) approach proposed
  (May 2000)
• 1st-yr summary presented (Sep 2001)
• Directed to implement the prototype at a forecast
  office
• Collaboration with WGRFC started (Dec 2001)

49
VAR Current Status (cont.)

• WGRFC started running the VAR operation for 6
  headwater basins (Jan 2003)
• 1-hour time step
• RFC (6-hour) SAC parameters
• 1-hour empirical UH

50
VAR Near-term plan

• Refine the empirical UH estimation technique
• Purely empirical
• (Using geomorphological UH as a priori)
• The computer code is to be delivered to WGRFC by
  May 1, 2003

51
VAR Near-term plan (cont.)

• Develop a technique and a computer code for
  automatic adjustment/refined of SAC parameters
• Using the RFC operational as a priori
• Using the soils-based as a priori
• SCE-based
• Explicit gradient (adjoint)-based
• (Hybrid approach)

52
Probabilistic/Ensemble Hydrologic Forecasting

• A new paradigm in hydrologic forecasting
• To account for uncertainties in
• boundary conditions (e.g., observed and forecast
  precipitation)
• initial conditions (e.g., soil moisture)
• model errors (including parameter uncertainty)
• Quasi-analytical vs. numerical

53
Ensemble Streamflow Prediction (ESP)
54
(No Transcript)
55
Existing Capability to Produce Ensemble
Precipitation Forecasts

• NWSRFS includes techniques to create ensemble
  products from existing NWS forecasts
• All time scales from the next few hours up to
  seasonal to inter-annual

56
Emerging Capabilities - Weather and Climate
Ensemble Forecast Applications

• Simplified short term ensembles using existing
  deterministic QPF (1-5 days)
• Medium range global ensembles (lt2wks) (NCEP and
  ECMWF)
• Short range regional ensembles (lt2days)
• Long range coupled ocean land atmosphere
  ensembles (gt2wks)

57
Short Term Ensemble Precipitation Forecasts from
Deterministic Forecasts

• Analyze joint distribution of historical
  forecasts and corresponding observations
• Estimate joint distribution parameters
• Derive conditional distribution given a
  deterministic forecast varies spatially
• Create synthetic ensemble forecasts consistent
  with conditional distribution
• Evaluate results

58
Questions

• Utilization of ENSO signals
• Re-/down-scaling of boundary conditions
• Probabilistic characterization of errors in
  initial conditions and model parameters
• Sample size
• Extreme values
• Validation of probabilistic/ensemble prediction

59
Thank you!
For more information - http//www.nws.noaa.gov/oh/
hrl