1

Mike Smith

OHD/HLHydrologic Science and Modeling

Branch

Introduction

Lecture 1 DHM/HL-RDHM Workshop

ABRFCJune 7, 2007

2

Overview

• Introductions

• Acknowledgements

• Review of Goals

• Expectations

• Strategy

3

Attendees• Norm Bingham NERFC• Paula Cognitore MARFC• Tom Adams OHRFC• Jonathon Atwell SERFC• Jeff Dobur SERFC• Eric Jones LMRFC• Katelyn Schnieda LMRFC• Paul McKee WGRFC• Mike Shultz WGRFC• Eugene Derner MBRFC• Ed Clark CBRFC• Craig Peterson CBRFC

• Pete Fickenscher CNRFC• Kevin Berghoff NWRFC• Kevin Werner WR• Kris Lander CR• Diane Cooper SR• Randy Rieman HSD• JJ Gourley NSSL• Suzanne Van Cooten NSSL• Prafulla Pokhrel U. Arizona• Michael Thiemann RTi• Mike Pierce ABRFC• John Schmidt ABRFC• Bill Lawrence ABRFC• ABRFC others

4

Workshop Objectives• To train RFC personnel how to set-up and run

the new AWIPS-NWSRFS DHM operation.• To train RFC personnel how to use the HL-

RDHM and related tools to parameterize and calibrate the DHM.

• To provide an overview of the vision and plan to use distributed models for RFC and WFO river and water resources forecasting operations.

• To provide an overview of the science and systems R&D for NWS distributed modeling, obtain feedback, and promote collaborative development.

“If you aim at nothing, you are sure to hit it!”

5

Goals and Expectations

• Potential– History

• Lumped modeling took years and is a good example• We’re second to do operational forecasting with dist. models

– Expectations• ‘As good or better than lumped’• Limited, but growing experience with calibration• May not yet show (statistical) improvement in all cases due to errors

and insufficient spatial variability of precipitation and basin features… but is proper future direction!

– New capabilities• Gridded water balance values and variables e.g., soil moisture• Flash Flood e.g., DHM-TF• Land Use- Land cover changes

6

(a) Lumped Basin

(c) Basin disaggregatedInto 16 cells

(d) Basin disaggregated into 100 cells

(b) Basin disaggregatedinto 4 cells

“Truth Scale” and“Truth Simulation”

Expectations: Effect of Data Errors and Modeling Scale

7

Expectations: Effect of Data Errors and Modeling Scale

Relative Sub-basin Scale A/Ak

1 10 100

10

15

20

25

30

0

5Re

lativ

e e

rro

r, E

k , %

(lumped) (distributed)

Noise 0% 25% 50% 75%

Data errors (noise) may mask the benefits of fine scale modeling. In some cases, they may make the results worse than lumped simulations.

Sim

ulat

ion

erro

r c

ompa

red

to fu

lly d

istr

ibut

ed

‘Truth’ is simulation from 100 sub-basin model

clean data

8

Rationale for Distributed Modeling

• Scientific motivation– Finer scales > better results– Data availability

• Field requests

• NOAA Water Resources Program

• NIDIS

• Flash flood improvements

Goals and Expectations

9

Applicability

• Distributed models applicable everywhere

• Issues– Data availability and quality needed to realize

benefits– Parameterization– Calibration– Run-time mods/assimilation

Goals and Expectations

10

R&D Implementation

Use

Distributed Modeling Strategy

11

Strategy for R&D

OHDParameterization: SAC-SMA, Snow-17, routingCalibration: manual, auto, spatially variableAssimilation: streamflow, soil moistureNew process modelsDMIP 1, 2 Data analysisLink to dynamic routing

RFCs WFOs

PartnersDMIP 1, 2MOPEXCollaborative Univ. Research PartnersNOHRSCRTi

Prototype testing of models, calibration, new science, etc

Components

12

XDMSand other

applications

AWIPS Oper. Baseline OB7.2/OB8.1

HL-RDHM and tools

Display xmrg grids,calibration

Calibration of baseline DHM;Generate gridded FFG;Prototype new capabilities

OB7.2 → Feb, 2007OB8.1 → July 2007

DHM

IFP

D-2D

Grids display

Display time series

OFSRuns DHM

ArcViewextensions

Calibration

(CAP)

Strategy for Implementation (1)

Distributed Model• SAC-runoff• Kinematic routing• No snow

DHM Approach for OB7.2/ 8.1

13

DHM Mods

DHM

IFP

D-2D

DHM Grid Editor

ASM-maintained application, enhanced

Field-developed application, enhanced

AWIPS Operational Baseline OB8.2

Grids display

HSMB prototype, enhanced

Display distributed time series

Distributed Model• SAC-runoff• Kinematic routing• No snow

OB8.2 → Jan 2008

(CAP)

DHM Approach for OB8.2

Strategy for Implementation (2)

XDMSand other

applications

HL-RDHM and tools

Display xmrg grids,calibration

Calibration of baseline DHM;Generate gridded FFG;Prototype new capabilities

ArcViewextensions

Calibration

14

Retire OHD-developed application

ASM-maintained application, enhanced

Field-developed application, enhanced

HSMB prototype, enhanced

OB8.3 → June 2008OB9 → June 2009

(CAP)

DHM Approach for OB8.3, OB9Strategy for Implementation (3)

DHM Mods

DHM

IFP

GFE

AWIPS Operational Baseline OB8.3

Basic Grid Editor and display

(replaces D-2D and DHM-Grid Editor)

DHM Grid Editor

Distributed Model• SAC-runoff• Kinematic routing• No snow XDMS

and other applications

HL-RDHM and tools

Display xmrg grids,calibration

Calibration of baseline DHM;Generate gridded FFG;Prototype new capabilities

ArcViewextensions

Calibration

15

Strategy: Use

• Use with, not instead of, lumped model at same time step (Example BLUO2)

• Part of natural progression to finer scalesLumped 6-hr Lumped 1-hour Distributed 1-hour

• Calibration is good training process for forecasting

• Current:– DHM: operation in NWS for headwaters, locals– HL-RDHM: Large area, soil moisture, FFG, etc

• Feedback to OHD

16

DHM/HL-RDHM Workshop

A. DHM and HL-RDHM Overview

B. Capabilities1. SAC-SMA and SAC-HT

2. Snow-17

3. Hillslope and channel routing

4. Manual and auto calibration

Overview of Capabilities

17

HL-RDHM

SAC-SMA, SAC-HT

Channel routing

SNOW -17

P, T & ET

surface runoff

rain + melt

Flows and state variables

base flowHillslope routing

SAC-SMA

Channel routing

P& ET

surface runoff

rain

Flows and state variables

base flowHillslope routing

DHM

Mods

AutoCalibration

DHM-TF

ForecastingCalibration

18

INTERFLOWSURFACERUNOFF

INFILTRATIONTENSION

TENSION TENSION

PERCOLATION

LOWERZONE

UPPERZONE

PRIMARYFREE

SUPPLE-MENTAL

FREE

RESERVED RESERVED

FREE

EVAPOTRANSPIRATION

BASEFLOW

SUBSURFACEOUTFLOW

DIRECTRUNOFF

Precipitation 1. Sacramento Soil MoistureAccounting Model

Source: U. Arizona

19

UZTWC UZFWCL

ZT

WC

LZ

FS

C

LZ

FP

C

UZTWC UZFWC

LZ

TW

C

LZ

FS

C

LZ

FP

C

SMC1

SMC3

SMC4

SMC5

SMC2

Sacramento Model Storages

Sacramento Model Storages

Physically-basedSoil Layers andSoil Moisture

1. Modified Sacramento Soil Moisture Accounting Model

In each grid and in each time step, transform conceptual soil water content to physically-based water content

SAC-HT

Soil moistureSoil temperature

20

1. SAC-HT Background

• Originally developed for the NOAH Land Surface Model– Documented improvements

• Koren, V., others, 1999. A parameterization of snowpack and frozen ground intended for NCEP weather and climate models. J Geo. Research, Vol. 104,

• Designed as replacement of the existing conceptual SAC-SMA frozen ground option.– Does not need calibration– Generates soil moisture and temperature versus depth– Can be used with local soil properties to adjust soil moisture to

local conditions.

21

Soil temperature

Soil moisture

Computed and observed soilMoisture and temperature: Valdai, Russia, 1972-1978

Validation of Modified Sacramento Model1. SAC-HT

22

Validation of Modified Sacramento Model

Comparison of observed, non-frozen ground, and frozen ground simulations: Root River, MN

Observed

Frozen ground

Non frozen ground

2. SAC-HT

23

NOAA Water Resources Program:Prototype Products

Soil moisture (m3/m3)

HL-RDHM soil moisture for April 5m 2002 12z

2. SAC-HT

24

Soil moisture RMS at layer 0-25 cm

0

0.05

0.1

0.15

0.2

0 0.05 0.1 0.15 0.2

Simulated using local properties

Sim

ula

ted

w/o

use

of

loca

l pro

per

ties

Simulated using local soil propertiesSim

ula

ted

w/o

lo

cal

soil

pro

per

ties

Tailor Soil Moisture Simulations for Local Soil types

Technique used in NOAAWater Resources EconomicsStudy

2. SAC-HT

25

HL-RDHMMOSAIC

Source: Moreda et al., 2005.

0

0.5

1

0 2000 4000 6000

Area km2

Cor

r.

0

0.5

1

0 2000 4000 6000

Area km2

Corr

.

Lower 30cmUpper 10cm

Comparison of Soil Moisture Estimates

HL-RDHM:HigherCorrelation

2. SAC-HT

26

‘We are also interested in the modified SAC model, particularly since we are somewhat “on the hook” to try to develop a soil moisture product (graphic) which conveys the current model states. This has been a recurring request (several years) which we have delayed, but was recently placed on a list in Central Region which specifies that we begin attempts to address this.”-John Halquist

Use of SAC-HT for Soil Moisture to Meet RFC Needs

27

SAC-SMA Parameters

1. Based on STATSGO + constant CN– Assumed “pasture or range land use” under “fair”

hydrologic conditions – National coverage – Available now via CAP

2. Based on STATSGO + variable CN– National coverage– Being evaluated

3. Based on SSURGO + variable CN– Parameters for 25 states so far – Being evaluated

Objective estimation procedure: produce spatially consistent and physically realistic parameter values

28

Demonstration of scale difference between polygons in STATSGO and SSURGO

SSURGO

STATSGO

Soils Data for SAC ParametersSoils Data for SAC Parameters

29

Status of SSURGO – Based SAC-SMA Parameter Derivation

30

SSURGO and STATSGO SAC-SMA Parameters

UZTWM- SSURGO

UZTWM- STATSGO

UZFWM-SSURGO

UZFWM-STATSGO

31

STATSGO and STATSGO Variable CN SAC-SMA Parameters

STATSGOSTATSGO Varable CN

DIfference

32

STATSGO vs SSURGOSTATSGO vs SSURGO ResultsResults

Hydrograph Comparison

__ Observed flow

__ SSURGO-based

__ STATSGO-based

TALO2TALO2

33

Hydrograph Comparison

__ Observed flow

__ SSURGO-based

__ STATSGO-based

CAVESPCAVESP

STATSGO vs SSURGOSTATSGO vs SSURGO ResultsResults

34

2. Distributed SNOW-17 Model

• SNOW-17 model is run at each pixel (hourly ok)• Gridded precipitation from multiple sensor products are

provided at each pixel• Gridded temperature inputs are provided by using DEM

and regional temperature lapse rate • The areal depletion curve is removed because of

distributed approach• Other parameters are either replaced by physical

properties or related to physical properties• Melt factors can be related to topographic properties:

slope & aspect• Parameters to be available through CAP

… Distributed Snow-17

35

Case Study 1: Juniata River

Outlet, Juniata at Newport

Saxton, Interior point

Williamsburg, Interior point

• Model resolution 4km x 4km

• Total number of pixels =497

• Watershed area = 8687 km2

• Model parameters = a priori

• Channel parameters are derived from USGS measurements at Newport.

WLBWLB

SPKSPK

SXTSXT

HUNHUN

PORPOR

REEREE

RTBRTB

LWSLWS NPTNPT

SLYSLY

MPLMPL

… Distributed Snow-17

36

Flow simulation during snow periods

0

200

400

600

1101200200 1201200220 0101200316 0201200312 0304200308 0404200304

Flo

w m

3/s

0

20

40

60

80

100

120

140

Sn

ow

Wat

er E

qu

ival

ent

(mm

)

Hyd_obs Hyd_simul swe

0

50

100

1101200200 1201200220 0101200316 0201200312 0304200308 0404200304

Sn

ow

co

ver

%

Simulated and observed hydrographs generally show good agreement, with the exception of some events where flows are extremely low/high compared to observed . This may be due to quality of temperature data

37

DEMAspect Slope Vegetation Type

Vegetation Percent Land Use Map

MFMIN MFMAX

Forest Cover MFMAX MFMIN

Coniferous forest /persistent cloud cover

0.5 -0.7 0.2 - 0.4

Mixed forest Coniferous plus open and/or deciduous

0.8 – 1.2 0.1-0.3

Predominantly Deciduous 1.0-1.4 0.2- 0.6

Open Areas flat terrain 1.5-2.2 0.2-0.6

Mountainous terrain 0.9-1.3 0.1-0.3

Computation of MFMAX and MFMIN

Eric AndersonRec’s.

38

2. Distributed modeling and snow

Parameterization of Distributed Snow-17

Min Melt Factor

Max Melt Factor

Derived from:1. Aspect2. Forest Type3. Forest Cover, %4. Anderson’s rec’s.

39

2. Distributed SNOW-17• Completed Activities

– Implementation of distributed SNOW-17 for the entire CONUS, proof of concept for computation of snow water equivalent and snow water covers

– Use/test of CONUS wide forcings such as archived STAGE II and Stage IV data for 2002 cold season

– Use and test of CONUS wide temperature from RUC model– Implement method of deriving gridded temperature for local

application on river basin scales. Two methods are used:• Disaggregation of MAT to grids by using DEM and basin

wide lapse rate.• Generating grids from gages within and near the basin

– Implemented concepts of removing areal depletion curve and substituting by simple linear curve for a pixel level simulation

– Generated a priori estimate of two major parameters MFMAX and MFMIN using properties of watershed

40

Overland flow routed independently for each

hillslope

(adapted from Chow et al., 1988)

HRAP Cell (~ 4 km x 4 km) Uniform, conceptual hillslopes within a modeling unit are assumed

• Drainage density illustrated is ~1.1 km/km2• Number of hillslopes depends on drainage density

Conceptual channel provides cell-

to-cell link

Overland flow routed independently for each

hillslope

(adapted from Chow et al., 1988)

HRAP Cell (~ 4 km x 4 km) Uniform, conceptual hillslopes within a modeling unit are assumed

• Drainage density illustrated is ~1.1 km/km2• Number of hillslopes depends on drainage density

Conceptual channel provides cell-

to-cell link

Real HRAP Cell

Hillslope model

Cell-to-cell channel routing

3. Routing Model

41

ABRFC ~33,000 cells

MARFC ~14,000 cells

• OHD delivers baseline HRAP resolution connectivity, channel slope, and hillslope slope grids for each CONUS RFC

• HRAP cell-to-cell connectivity and slope grids are derived from higher resolution DEM data.

HRAP Cell-to-cell Connectivity Examples

42

3. Channel Routing Model• Uses implicit finite difference solution technique• Solution requires a unique, single-valued

relationship between cross-sectional area (A) and flow (Q) in each grid cell (Q=q0Aqm)

• Distributed values for the parameters q0 and qm in this relationship are derived by using – USGS flow measurement data at selected points– Connectivity/slope data– Geomorphologic relationships

• Training on techniques to derive spatially distributed parameter grids is provided in this workshop

43

4. Manual and Auto Calibration• Adjustment of parameter scalar multipliers• Use manual and auto adjustment as a strategy• Parameters optimized:

– SAC-SMA– Hillslope and channel routing

• Search algorithms– Simple local search– Next: Rosenbrock, others

• Objective function: Multi-scale

44

48

5632 62

4230 44

40

4442

32

36 42

40

24

2816 31

2115 22

20

2221

16

18 21

20

Multiply each grid value by the same scalarfactor.

x 2 =

Calibrate distributed model by uniformly adjusting all grid valuesof each model parameter (i.e., multiply each parameter grid value by the same factor)

1. Manual: manually adjust the scalar factors to get desired hydrograph fit. 2. Auto: use auto-optimization techniques to adjust scalar factors.

Example: Ith parameter out of N total model parameters

Calibration Approach

Preserve Spatial Pattern of Parameters

45

HL-RDHM

SAC-SMA, SAC-HT

Channel routing

SNOW -17

P, T & ET

surface runoff

rain + melt

Flows and state variables

base flowHillslope routing

AutoCalibration

Execute these components in a loop to find the set of scalar multipliers thatminimize the objective function

46

km

iiksiko

k k

XqqJ1

2

,,,,1

2

1n

Multi-Scale Objective Function (MSOF)

• Minimize errors over hourly, daily, weekly, monthly intervals (k=1,2,3,4…n…user defined)

• q = flow averaged over time interval k• n = number of flow intervals for averaging

• mk = number of ordinates for each interval

• X = parameter set

k1

Weight: -Assumes uncertainty in simulated streamflow is proportionalto the variability of the observed flow-Inversely proportional to the errors at the respective scales. Assume errors approximated by std.

=

Emulates multi-scale nature of manual calibration

47

Average monthly flow

Average weekly flow

Average daily flow

Hourly flow

Calibration: MSOF Time Scales

Multi-scale objective function represents different frequencies of streamflow and its use partially imitates manual calibration strategy

48

Beforeautocalibrationof a prioriparameters

After autocalibration

Observed

Example of HL-RDHM Auto Calibration: ELDO2 for DMIP 2 Arithmetic Scale

Auto Calibration: Case 1

49

Example of HL-RDHM Auto Calibration: ELDO2 for DMIP 2 Semi-Log Scale

Auto Calibration: Case 1

Beforeautocalibrationof a prioriparameters

After autocalibration

Observed

50

Beforeautocalibrationof a prioriparameters

After autocalibration

Observed

Auto Calibration: Case 2Example of HL-RDHM Auto Calibration: ELDO2 for DMIP 2

Arithmetic Scale