The NCAR WRF-Hydro Team

D. Gochis, A. Dugger, M. Barlage, R. Cabell, A. Fanfarillo, K. FitzGerald, A. Gaydos, J. Grim, L. Karsten, M. McAllister, A. McCluskey, J. McCreight, J. Mills, L. Pan, A. RafieeiNasab, R. Rasmussen, L. Read, K. Sampson, E. Towler, D. Yates, W. Yu, Y. Zhang

NOAA Office of Water Prediction

B. Cosgrove, E. Clark, F. Salas, T. Flowers, T. Graziano, Y. Liu, X. Feng, C. Pham, Z. Cui, F. Ogden, M. Glaudamens, N. Frazier, M. Smith

Description of the operational National Water Model

BiodiversityDroughtPollutionFlooding

An Array of Water Issues

“Hydrology of Now”

Spatial ScalesWatershed Global

Temporal ScalesDesign Climate Adaptation

• Operational forecast streamflow guidance for currently

underserved locations

• Spatially continuous estimates and forecasts of

hydrologic states for the nation through enhanced

physical accounting of major water cycle

components (snowpack, soil moisture, channel flow,

major reservoir inflows, flood inundation)

• Seamlessly interface real-time hydrologic products into

an advanced geospatial intelligence framework

• Implement an Earth system modeling architecture that

permits rapid model evolution of new data, science and

technology

Goals of the Initial Implementation of the National Water Model

http://water.noaa.gov/about/nwm

Development Team: NCAR/RAL, NOAA/OWP/NWC, USGS, CUAHSI,

Universities

Sponsor: NOAA Office of Water Prediction

System become fully operational beginning Aug. 16, 2016

• Real-time verification since June 2016 (Rwrfhydro)

• Multiple operational products created by NOAA, academia, private sector

The National Water Model

http://water.noaa.gov/about/nwm

The National Water Model Version 1.2: Technical Specs

Data Throughput:

• Input data per day: 4.45 Terabytes

• Output data per day: 3 Terabytes

• # of river channels: 2.7 million

• # of reservoirs: 1,260

• Total # of computational elements: ~360,000,000

Model Details:

• Number of lines of code: 74,740

• Computer usage: > 100,000 cpu-hours per day

National Streamflow Anomaly Map

Available online at: http://water.noaa.gov/tools/nwm-image-viewer

Development Team: NCAR/RAL, NOAA/OWP/NWC, USGS, CUAHSI, UniversitiesSponsor: NOAA Office of Water Prediction

Full NWM Ecosystem

NWM Forcing Engine

Column Land

Surface Models

Terrain Routing

Modules

Channel & Reservoir

Routing Modules

Geospatial

Pre-Processing

Streamflow Data

Assimilation

Model Evaluation &

Calibration

Web-Mapping

ServicesMO

DE

L P

HY

SIC

S

Multi-Scale

Spatial

Transformations

WRF-Hydro Options Current NWM Configuration

3 up-to-date column land

models: Noah, NoahMP (w/

built-in multi-physics options),

Sac-HTET

2 groundwater schemes: direct

aggregation storage-release: pass-through

or exponential model

2 subsurface routing

scheme: Boussinesq

shallow saturated flow,

2d aquifer model

NoahMP

Exponential model

Diffusive wave

3 surface routing schemes:

diffusive wave, kinematic

wave, direct basin aggregation

Column Land

Surface Model

Overland

Flow Module

Lateral

Subsurface Flow

Module

Conceptual

Baseflow

Parameterizations

Boussinesq shallow

saturated flow

Channel

Routing/

Hydraulics

5 channel flow schemes: diffusive wave,

kinematic wave, RAPID, custom-network

Muskingum or Muskingum-Cunge

Lake/Reservoir

Management

1 lake routing scheme: level-

pool management

Level-pool

management

Custom-network

(NHDPlus) Muskingum-

Cunge model

WRF-Hydro Physics Permutations

In Development: Upgrading to NWM V2.0 and Beyond

Foundation Established August 2016

Water Resource Model for 2.7 Million Stream

Reaches

First Upgrade May 2017

Increased cycling freq. and forecast length, initial calibration, improved

soil/snow physics

Second UpgradeEarly 2018

Extensive calibration, improved hydrofabric (terrain and stream

connections), improved data assimilation

Second UpgradeMay 2019

Expansion to Hawaii, medium range ensemble configuration, improved

physics, improved calibration, longer analysis cycle, enhanced code

modularity, 8

Versio

nStats

V1.0

+ 31,535 catchments

+ 257,802 km2

V1.1

+ 17,843 catchments

+ 64,450 km2

V1.2

+ 12,468 catchments

+ 94,824 km2

+596 gages

+233 lakes

TOTAL

2,677,135 catchments

8.03M km2

2,729,077 flowlines

7,542 gages

+5,000 lakes

NWM Geospatial Framework based on

EPA/USGS NHDPlusV2:

Version over Version Improvements

Simulated Flowline Gages used for DA

National Water Model V1.2/V2.0CONUS Analysis and Forecast Cycling Configurations

Cycling Forecast Forcing Outputs

Short-Range

Medium-Range

Long-Range

Hourly

4 x Day(7 members)

Daily Ensemble(16 members)

18 hours

Member 1: 10 daysMembers 2-7: 8.5 days

30 days

Downscaled HRRR/RAP

Blend

Downscaled 13kmGFS

Downscaled and Bias-

Corrected CFS

1km Land States,250m Sfc Routed Water,

NHDPlus Streamflow

All: NHDPlusStreamflow

Member 1: 1km Land States,250m Sfc Routed Water

1km Land States,NHDPlus

Streamflow

Analysis

Std: HourlyExt: Daily

Long-Range:4 x day

Std/Long: MRMSExt: RFC MPE

All: Downscaled HRRR/RAP

All: 1km Land States,NHDPlus Streamflow

Std and Ext: 250m Sfc Routed Water

Std: -3 HourExt: -28

HourLong-Range:

-12 hours

The National Water Model: Operational Cycling - CONUS

The NOAA National Water Model

Operational outputs:

1km Soil Moisture

250m Depth to soil saturation

1km Evapotranspiration

250m Surface overland flow depth

NHDPlus channel flow and velocity

Ensemble streamflow predictions

NWM v1.2 Medium Range Forecast Surface Overland Flow Water Depth (mm):Operational Analysis: 00UTC 15 Sep, 2018 Eastern N. Carolina, Hurricane Florence….Forecast guidance up to 6 days in advance

Operational outputs:

-Addition of Hawaii to NWM (3-hr Analysis and 60-hr Short-Range forecast, both forced by NAM-Nest NWP model)-Addition of Extended Analysis (daily 28-hour look-back using RFC-based MPE precipitation from Stage IV) -Addition of separate Long-Range Analysis configuration to initialize LR forecast-Addition of Medium Range ensemble forecast configuration (7 members 4 x day) (mem1=uses current GFS to 10 days, mem2-7=use time lagged GFS out to 8.5 days)

-Use of 13km GFS forcing (versus 0.25 degree in NWM V1.2)-Improved downscaling of GFS and CFS forcing via a Mountain Mapper-based approach

-Out-of-bank parameterization via compound channel and new empirically based channel parameters-Improved snow albedo formulation, new soil evaporation parameter and relaxation of ponded water threshold-Bug fix in the units in one of the groundwater bucket calculations and a fix in reservoir module.

-Improved calibration of parameters by using hourly streamflow data, expanding calibration from ~1100 to ~1400 calibration basins and improving parameter regionalization process. Also, utilized Mountain Mapper-downscaled NLDAS2 forcing in calibration so as to more closely match the forcing used in the new Extended Analysis cycle.

-Increased CONUS reservoirs from ~1500 to ~5500 (impact mostly on non-calibration basins)-Fixed 37 stream breaks-For Hawaii, added 58 USGS gauges for DA, 13,637 new flowlines, 10 reservoirs and 16,625 km^2 of basins -Inclusion of a new elevation base that is harmonized with the NHDPlus channel network

Mo

de

l

Co

nfi

gu

rati

on

Hyd

rofa

bri

cF

orc

ing

Ph

ys

ics

Ca

lib

rati

on

NWM Version 2.0 Enhancements

V1.0

NWM v2.0 Improvement: All USGS Gauges (Validation Retrospective)

V1.1 V1.2 V2.0

NWM v2.0 Streamflow Bias at USGS Gauges (WY 2014-2016) ● Streamflow bias improves at USGS gauged

basins with each version● Improvements more pronounced at directly

calibrated sites● Model now calibrated/validated against

hourly (previously daily) streamflow obs● Daily metrics also improve● Simulation is for WY2014-2016 (validation

period) and uses NLDAS-2 forcing data (with Mountain Mapper downscaling in v2.0)

● No assimilation of USGS obs

37% have bias < +/-20% 44% have bias < +/-20%31% have bias < +/-20% 34% have bias < +/-20%

A. Dugger, J. Mills, E. Towler, A. RafieeiNasab

V1.0

NWM v2.0 Improvement: All USGS Gauges (Validation Retrospective)

V1.1 V1.2 V2.0

● Streamflow correlation improves at USGS gauged basins with each version

● Improvements more pronounced at directly calibrated sites

● Model now calibrated/validated against hourly (previously daily) streamflow obs

● Daily metrics also improve● Simulation is for WY2014-2016 (validation

period) and uses NLDAS-2 forcing data (with Mountain Mapper downscaling in v2.0)

● No assimilation of USGS obs

NWM v2.0 Streamflow Hourly Correlation at USGS Gauges (WY 2014-2016)

18% have cor >= 0.8 21% have cor >= 0.8 27% have cor >= 0.8 33% have cor >= 0.8

A. Dugger, J. Mills, E. Towler, A. RafieeiNasab

NWM V2.0 Snowpack Analyses and Forecasts

Model vs. SNOTEL SWE

Scatterplots (all CONUS sites)

V1.2

V2.0

National Water Model geographic expansion(with NOAA OWP and National Water Center)

• Hawaii (with APRFC, Honolulu WFO)

– First NWS real-time streamflow guidance for the islands

– Operational with v2.0, scheduled to go live March 2019

– Challenges: Forcings, groundwater, diversions

• Great Lakes (with GLERL)

– US/Canada trans-boundary hydrography dataset (stream reaches, catchments, gage locations)

– Realtime Canadian gage DA to supplement USGS gages

– Scheduled for v2.1 (live March 2020)

– Challenges: Bilateral, complicated water management

• Puerto Rico & Virgin Islands (with SERFC)

– Scheduled for v2.1 (live March 2020)

– Challenges: Forcings, groundwater/karst, damaged infrastructure

• Alaska (with APRFC, LANL, Arctic Testbed, UAF)

– Research/early-stage development

– Challenges: Forcings, permafrost, glaciers

National Water Model Expansion Domains

Honolii Stream near Papaikou

Observed: Solid

NWM Forecast Cycles (10/8 12Z and 18Z): Dashed

NWM distributed soil moisture estimate for

Hawaiian Islands

• Environmental tracers for transport timing prediction• On-demand capability using existing operational NWM• Nearly instantaneous response• Amenable to stochastic perturbation of flows to generate probabilisitc guidance• Example of the Gold King Mine Spill

Model flowpath tracing:

Pearl Harbor - Oahu

NLCD 2011 Land Cover

Channel Properties & Hydraulics

WRF-Hydro Physics: Future Expansion Areas

Land Cover Change Impacts on Hydrology

A

Bw

αy

Twco

mp

h

• Improved methods to estimate channel parameters (geometry, roughness) across 2.7 million reaches **funded JTTI 2018**

• Improved methods to simulate overbank flow and channel-floodplain interactions

• Incorporate continuously updated, remote sensing based land cover datasets

• Simulate impacts of disturbance (e.g., fire, insect) on flood risk

Image courtesy of USGS

• Coupling WRF-Hydro with water quality tracer and sediment transport models

Image courtesy of USFS

Water Quality Prediction

Integrated Inland-Coastal Prediction

• Develop real-time coastal and riverine inundation maps

• Provide key support for hurricane landfall

• Dynamically couple streamflow module with tide and storm surge models **UCAR proposal submitted with MMM**

• Addition of tracer capabilities to WRF-Hydro (overland, subsurface, groundwater, and channel routing modules)**funded USGS Powell Center 2018**

The National Water Model: A community R2O-O2R effort

• NCAR Team:– Model and dataset

development, model implementation, evaluation

• NOAA OWP Team:– Model onboarding

– Model evaluation

– Forecast product development & water operations center

– R&D scoping

• NCEP NCO Team:– Operational cycling

and data flow

• Broader Community:– R&D, evaluation,

applications

WRF-Hydro® Software Ecosystem

• Ecosystem overview: https://github.com/NCAR/wrfHydro

• Model: https://github.com/NCAR/wrf_hydro_nwm_public

○ Public, community model, with version control system

○ Contributing guidelines, conventions, license, code of conduct

○ Python-based (pytest) testing framework (Python API)

• Python API: https://github.com/NCAR/wrf_hydro_py

• Docker containers: https://github.com/NCAR/wrf_hydro_docker

○ Standard portable environments for working with the model

• Continuous Integration with Travis on Github (Docker + Python)

• “Discontinuous integration” at scale (cheyenne)

○ Large jobs, compilers with licenses

• ARC GIS preprocessing toolbox:

https://github.com/NCAR/wrf_hydro_arcgis_preprocessor

• Analysis tool box: https://github.com/NCAR/rwrfhydro

• Training: https://github.com/NCAR/wrf_hydro_training

WRF-Hydro Website:

https://ral.ucar.edu/projects/wrf_hydro

GitHub:https://github.com/NCAR/wrf_hydro_nwm_public

Docker Hub:

https://hub.docker.com/u/wrfhydro/

National Water Model:

http://water.noaa.gov/about/nwm

Thank you!