Multi-scale water cycle predictions using the community WRF-Hydro

modeling system

May 2, 2017

D. Gochis, W. Yu, A. Dugger, J. McCreight, K. Sampson, D. Yates,A. RafieeiNasab, L. Karsten, L. Read, L. Pan, Y. Zhang

National Center for Atmospheric Research

Purpose

Purpose: Provide a update of multi-scale water cycle modeling capabilities using the community WRF-Hydro system and description of recent prediction applications

Scientific Imperative for WRF-Hydro:

Addressing water cycle prediction questions:1. How do horizontal routing processes impact the partitioning of water and energy

at the land-atmosphere interface?

2. How does organization of fine-scale heterogeneity impact boundary layer exchange and atmospheric circulation features?

3. How will eco-hydrologic processes evolve under various disturbance mechanisms such as landscape and climatic change?

4. What is the ‘coupled-system’ predictability of extreme hydrological events and how does physical process representation impact the lead-time dependence of forecast skill?

5. NOAA-National Water Model: Providing a framework for ‘total water prediction’

Water Cycle Modeling and Prediction within the WRF-Hydro System:

Great Colorado Flood of 11-15 Sept. 2013

Accumulated Precipitation (shaded colors)100m gridded streamflow (points)

WRF-Hydro within an Operational Forecasting Workflow:

1. Forecasts of water flow everywhere all the time

Moving Beyond Point Flow Forecasts

Current efforts are demonstrating the feasibility of Operational Quantitative StreamflowForecasting (QSF):

– NSSL-FLASH, WRF-Hydro, LISFLOOD (UK), RAPID

– Spatial resolutions > 100m better

– Allows cycling from QPE and forecasting from QPN/QPF

– Emphasis on 0-6 hr gap

1. Forecasts of water flow everywhere all the time: The NOAA National Water Model

Snow Water Equivalent (SNEQV): March 1, 2015 Total Column % Saturation (“SOILSAT”): Sept. 13, 2013

1. Forecasts of water flow everywhere all the time: The NOAA National Water Model

2. Coupled system flux predictions:

• Variability in surface fluxes are strongly coupled to convective initiation and cloud formation. Complex, non-linear feedback require coupled system resporesentation

FRNG_1km_cloudwater_tskin_NARR_7_18_2004_1800z FRNG_1km_cloudwater_tskin_WRF-Hydro_rtg_7_18_2004_1800z

3. Moving beyond ‘natural flows’ towards explicit accounting of infrastructure

Design storm streamflow capture by Barker Reservoir and Gross Reservoirs. Colorado Front Range

Including the control effects of, and impacts on infrastructure:

– Dams and reservoirs (passive and actively managed)

– Overbank storage and attenuation

– Diversion structures, headgates

– Levees, dikes

– Failures of infrastructure (exceeding design capacity)

* Needs Infrastructure & Operations

Data Standards

4. Probabilistic Framework for Meaningful Risk Forecasting

HMT SE Flooding July 28, 2013

Quantify analysis and forecast uncertainty to provide meaningful risk guidance

Provide forecasters and decision makers with probabilities of:

• Locations and time of rapid river stage increase

• Duration of high waters and inundation

Requires maximizing the utility of High Performance Computing (HPC)

5. Hydro-system Dynamics

Improving representation of landscape dynamics essential to flood risks:• Geomorphological:

– Bank stability– Sediment transport/deposition– Debris flows

• Land cover change due fire, urbanization, ag/silvaculture

* Needs improved channel, soils andland cover geospatial data

Overarching WRF-Hydro System Objectives

A community-based, supported coupling architecture designed to provide:

1. An extensible multi-scale & multi-physics land-atmosphere modeling capability for conservative, coupled and uncoupled assimilation & prediction of major water cycle components such as precipitation, soil moisture, snowpack, groundwater, streamflow, inundation

2. ‘Accurate’ and ‘reliable’ streamflow prediction across scales (from 0-order headwater catchments to continental river basins & minutes to seasons)

3. A robust framework for land-atmosphere coupling studies

WRF-Hydro Operates in 2 Major Modes: Coupled or Uncoupled to an Atmospheric Model

• Uncoupled mode critical for spinup, data assimilation and model calibration

• Coupled mode critical for land-atmosphere coupling research and long-term predictions

• Model forcing and feedback components mediated by WRF-Hydro:

• Forcings: T, Press, Precip., wind, radiation, humidity, BGC-scalars

• Feedbacks: Sensible, latent, momentum, radiation, BGC-scalars

One-way (‘uncoupled’) →

Two-way (‘coupled’) ↔

Version 4.0 physics components:

• physics-based runoff processes

Overland Flow -Diffusive waveKinematic*Catchment aggregation*

Groundwater Flow –Boussinesq flowCatchment aggregation*

Channel Flow –Diffusive waveKinematic*Reach-based Muskingam*

WRF-Hydro v4.0 Physics Components:

• Optional conceptual ‘catchment’ modeling support:– Benchmarking simple versus complex model structures

– Enable very rapid ‘first-guess’ forecasts with reduced runtime/computational demand

– Bucket discharge gets distributed to channel network channel routing (e.g. NWM & RAPID coupling)

DA with WRF Hydro

Current capabilities• Ensemble DA:

• Offline WRF Hydro + DART =“HydroDART”

• Ensemble generation: • Initial state & parameter perturbation,

ensemble runs

Future capabilities• Variational DA and/or nudging:

• Faster & computationally cheaper for large-scale applications.

• Variational DA not rank-deficient• Other kinds of DA (hybrid, MLEF, …)• Bias-aware filtering / Two-stage bias estimation

(Friedland, 1969; Dee and de Silva, 1998; De Lannoy et al., 2007)

Open loop

DA

0.0

2.5

5.0

7.5

0.0

0.5

1.0

May 15 Jun 01 Jun 15 Jul 01 Jul 15 Aug 012012

Str

ea

mflow

(cub

ic m

ete

rs p

er

se

co

nd

)

Obs 95% uncertOpen loopOpen loop meanEns. meanEns. spread

HydroDART Overview

‘WRF-Hydro’ Process Permutations and System Features:

• ~180 possible ‘physics’ component configurations for streamflow prediction: – 3 up-to-date column physics land models (Noah,

NoahMP, CLM)– 3 overland flow schemes (Diffusive Wave,

Kinematic Wave, Direct basin aggregation)– 4 lateral/baseflow groundwater schemes

(Boussinesq shallow-saturated flow, 2d aquifer model, Direct Aggregation Storage-Release: pass-through or exponential model

– 5 channel flow schemes: Diffusive wave, Kinematic Wave, RAPID-Muskingam, Custom Network Muskingam/Muskingam Cunge

• Simple level-pool reservoir with management• Data Assimilation:

– DART, filter-based hydrologic data assimilation– Nudging-based streamflow

Ensemble Flood Forecasting in the Southeast U.S. with WRF-Hydro2014 WRF User’s Workshop, K. Mahoney (NOAA-ESRL)

WRF-Hydro System-Level Coupling Capabilities

Completed:• Stand-alone, ‘Un-coupled’ (1-d Noah & NoahMP land model driver)• Coupled with the Weather Research and Forecasting Model WRF-ARW)• Coupled with LIS (WRF-Hydro v1.0, LISv6.1)• Coupled into DART…

In Progress:• NOAA/NEMS (NOAA Environmental Modeling System-Cecilia DeLuca) • Update of LIS coupling to LIS v7/WRF-Hydro v2.1• Coupled with CLM under CESM coupler (working on recent release of CLM

in WRF)

‘WRF-Hydro’ Software Features:

• Modularized F90/95 (and later) • Coupling options are specified at compilation and WRF-Hydro is

compiled as a new library in WRF when run in coupled mode• Physics options are switch-activated though a

namelist/configuration file• Options to output sub-grid state and flux fields to standards-based

netcdf point and grid files• Fully-parallelized to HPC systems (e.g. NCAR supercomputer) and

‘good’ scaling performance• Ported to Intel, IBM and MacOS systems and a variety of compilers

(pg, gfort, ifort)

Wei Yu (RAL) – lead engineer

The WRF-Hydro Workflow

WRF-Hydro Implementation Workflow:

Collect geospatial terrain and

hydrographic data

Prepare:Land model grids (WPS)Routing Grids/Networks

(ArcGIS)

Conduct uncoupled model runs-physics selection

-calibration-assimilation &/or spinup

Execute uncoupled forecast cycles:

Nowcasts, NWP QPF

Execute coupled-model forecast cycles

Create output forecast & evaluation products

Collect & Prepare Meteorological

Forcings:(uncoupled runs)

Prepare Atmospheric Model:

(coupled runs)

Model System and Components:

• GIS Pre-processor – Physiographic data processing

• Meteorological Forcing Engine (MFE) – Met. Pre-processing

• Core WRF-Hydro Model – Model physics

• Hydro-DART – Data assimilation

• Rwrfhydro – Analysis, verification, visualization

WRF-Hydro Setup and Parameterization:Python Pre-Processing Toolkit: K. Sampson - developer

• Python-based scripts

• ESRI ArcGIS geospatial processing functions

– Support of multiple terrain datasets

• NHDPlus, Hydrosheds, EuroDEM

Outputs: topography, flowdirection, watersheds, gridded channels, river reaches, lakes, various parameters

Meteorological Forcing Engine:• NLDAS, NARR analyses• QPE products: MPE, StgIV,

NCDC-served, dual-pol, Q3/MRMS, gauge analyses, CMOPRH, TRMM, GPM

• NOAA QPF products: GFS, NAM, RAP, HRRR, ExREF

• Nowcast (NCAR Trident/TITAN)

• NOHRSC SNODAS

• ESMF/ncl regridding tools

Boulder

Ft. CarsonAurora

Long’s Peak (~14,200’)

Pikes Peak (~14,300’)

Regridded MPE precipitation during the 2013 Colorado FloodsUnidata IDV display

NWM: Meteorological Forcing Engine (MFE): Examples

Blended MRMS-HRRR Precipitation

Seasonally-varying MRMS RQI

Visual forecast products…Web map service interfaces: GoogleMaps/Earth , ESRI ArcGIS, OpenLayers

GoogleEarth, GoogleMaps. ArcGISWMS display

Rwrfhydro Evaluation, Verification and Visualization Tools

• Set of R tools to support WRF-Hydro pre- and post-processing

• Open source, community tool

• Full documentation and training vignettes

• Major Features:

– Domain visualization

– Remote sensing & geospatial data prep

– Output post-processing

– Observation data acquisition and processing

– Model output evaluation and visualization

– Generally model agnostic

• Developed in parallel with NWM v1.0

https://github.com/mccreigh/rwrfhydro

Rwrfhydro: R package for Hydrological Model Evaluation

Rwrfhydro: R package for WRF-Hydro Model Evaluationhttps://github.com/mccreigh/rwrfhydro

Dataset Data type/format rwrfhydro functionality

Climate

GHCN point obs download, format/process, statistics

SNOTEL point obs download, format/process, statistics

Snow

SNOTEL point obs download, format/process, statistics

SNODAS raster download, regrid, statistics

MODIS SCA raster download, mosaic/resample, statistics

Soil Moisture

SCAN point obs download, format/process

NASMD point obs download, format/process (in process)

Energy

Ameriflux point obs download, format/process, statistics

ARM point obs download, format/process (in process)

MODIS ET, TRAD, ALBEDO raster download, mosaic/resample, statistics

Streamflow

USGS point obs download, format/process, DA prep, statistics

CO DWR point obs download, format/process, statistics

Rwrfhydro Model Calibration Toolkit

Developed by A. Dugger with contributions from Logan Karsten and A. RafieeiNasab

• Domain subsetting tools

• Parameter Sensitivity

Analysis

• Calibration:

• ‘Supervised’ Dynamically Dimension Search

(DDS) method (Tolson, B. A., and C. A.

Shoemaker: 2007)…..Implementation of

Shuffled Complex Evolution (in progress)

• Split sample calibration/validation

• Multiple criteria monitoring (NSE, RMSE, %

bias, correlation, KGE)

• Automated Rwrfhydro-NWM workflow

• Re-validation of full domain results

Rwrfhydro Model Calibration Toolkit

NWM v1.0 (5-yr NLDAS Percent Bias)

Developed by A. Dugger with contributions from Logan Karsten and A. RafieeiNasab

WRF-Hydro Support Services

• Web Page:– Code distribution (GIT

repository)– Documentation (v2, 120 pages)– Test cases (coupled and

uncoupled)– Script Library (file prep,

reformatting, viz)– ArcGIS preparation tools– Email help support (staff limited)

http://www.ral.ucar.edu/projects/wrf_hydro/

WRF-Hydro Support Services

• Training classes:– Semi-annual WRF tutorial

training sessions (short 1-hr system overviews)

– University hosted visits (~1-2/yron the order of 1-3 days)

– International training seminars and colloquia (~1-2/yr, on the order of 1-3 days)

http://www.ral.ucar.edu/projects/wrf_hydro/

1st European Fully Coupled Atmospheric-Hydrological Modeling and WRF-Hydro Users workshop, U. of Calabria, Italy, June 2014

Current WRF-Hydro Applications around the world:1. Operational Streamflow Forecasting:

– U.S. National Weather Service, National Water Center– Israeli Hydrological Service– State of Colorado-Upper Rio Grande River Basin (CWCB, NSSL)– NCAR-STEP Hydrometeorological Prediction Group– U. of Calabria reservoir inflow forecasting

2. Streamflow prediction research (U. Ankara, Arizona State U., Karlsruhe Inst. Tech.)3. Diagnosing climate change impacts on water resources

– Himalayan Mountain Front (Bierknes Inst.)– Colorado Headwaters (U. Colorado)– Bureau of Reclamation Dam Safety Group (USBR,NOAA/CIRES)

4. Diagnosing land-atmosphere coupling behavior in mountain-front regions of the U.S. and Mexico (Arizona State U., U. Arizona)

5. Diagnosing the impacts of disturbed landscapes on coupled hydrometeorlogical predictions– Western U.S. Fires (USGS)– West African Monsoon (Karlsruhe Inst. Tech)– S. America Paraná river (U. Arizona)– Texas Dust Emissions (Texas A&M U.)– Landslide Hazard Modeling (USGS)

6. Hydrologic Data Assimilation, WRF-Hydro/DART coupling

Acknowledgements

External Contributors• K. Mahoney (CU-CIRES)

• Brian Cosgrove (NOAA/OHD)

• B. Fersch, T. Rummler (KIT-Germany)

• Alfonso Senatore (U. Calabria-Italy)

• A. Parodi and E. Fiori (CIMA-Italy)

• Amir Givati and Erik Fredj (Israeli Hydr. Service)

• Lu Li (Bierknes Inst.)

• Col. State Univ. CHILL-team

• Logan Karsten (NOHRSC)

• Sujay Kumar, Christa Peters-Lidard (NASA-Goddard)

• Peirong Lin, Z.-Liang Yang (U. Texas-Austin)

• I. Yucel, (U. Ankara-Turkey)

Support provided by:• NSF- NCAR-STEP program, EarthCube, ETBC, WSC• NOAA-OHD• NASA-IDS• CUAHSI• DOE-ESM• USBR WaterSmart & Dam Safety Programs• Colorado Water Conservation Board• Texas Dept. of Environmental Quality & Texas A&M U.

NCAR Development, Evaluation and Advising Team: Wei Yu, David Yates, Kevin Sampson, Aubrey Dugger, James McCreight, Mike Barlage, Yongxin Zhang, Mukul Tewari, Roy Rasmussen, Andy Wood, Fei Chen, Martyn Clark, Matthias Steiner

End

WRF-Hydro: http://www.ral.ucar.edu/projects/wrf_hydro/