Analysis of WRF-Hydro Simulations of the 2008 Iowa Floods: Effects and Sensitivity of

Model Structure and Precipitation Forcing

Mohamed ElSaadani1, Tim Lahmers2

1 Iowa Flood Center, University of Iowa, Iowa City, Iowa 2 University of Arizona, Tucson, Arizona

Abstract:

WRF-Hydro is run From 1 April 2008 to 30 September 2008 with NLDAS atmospheric forcing.

The model is run in the following two configurations: 1) using the Noah MP LSM, channel

routing, sub surface routing, overland flow, and baseflow and 2) using only the Noah MP LSM

to force RAPID (i.e. NFIE Hydro). Using these two configurations, WRF-Hydro is forced with

both NLDAS precipitation and NCEP-Stage-IV precipitation. The skill of the model using these

configurations and forcing is examined at USGS stream gauges throughout the Iowa study area.

---------------------------------------------------------------------------------------------------------------------

Sensitivity analysis and calibration of NFIE-Hydro in Texas

Tim Ivancic1, Taereem Kim2, Manab Saharia3

1 SUNY College of Environmental Science and Forestry, Syracuse, New York 2 Yonsei University, Seoul, South Korea 3University of Oklahoma, Norman, Oklahoma

Abstract:

Uncertainty in precipitation input can take on a number of forms. In addition to error in

measured precipitation, there can be uncertainty in intensity and spatial and temporal distribution

of forecast data. When propagated through a hydrologic model, these errors can have a

measurable impact on streamflow predictions and impact the ability to properly assess flood

extent in events such as the 2015 Memorial Day Flood in Austin and Houston, Texas. In this

project, we force the WRF-Hydro (NoahMP) using various synthetic and measured precipitation

products. The runoff from the WRF-Hydro model is then routed using RAPID. The first input

ensemble measures the impact of a spatially independent gaussian precipitation error on the

streamflow output. The second consists of uniform pulses of precipitation with varying intensity

and duration and is intended to estimate the sensitivity of streamflow to forecasted precipitation

intensity. We also force the model with uniform pulses corresponding to several return periods

with values extracted from a point distribution. The third uses a spatially and temporally

redistributed precipitation signal to estimate the sensitivity of streamflow to forecasted

precipitation distribution. The model results are validated by comparing with the observed USGS

station streamflow.

---------------------------------------------------------------------------------------------------------------------

Spatio-temporal validations of the WRF-Hydro/RAPID simulated water fluxes over Texas:

interplay between uncertainties in modeled evapotranspiration and surface/subsurface

runoff

Peirong Lin1, Mohammad Adnan Rajib2, Marcelo A. Somos-Valenzuela3, Yan Wang4

1Jackson School of Geosciences, University of Texas at Austin, Austin, Texas 2Lyles School of Civil Engineering, Purdue University, West Lafayette, Indiana 3Center for Research in Water Resources, University of Texas at Austin, Austin, Texas 4College of Hydrology and Water Resources, Hohai University, Nanjing, China

Abstract:

Appropriate model diagnostics is important for successful implementation of the National Flood

Interoperability Experiment (NFIE) toward national-scale flood monitoring and forecasting. In

this project, the WRF-Hydro/RAPID modeling framework is evaluated by focusing on the spatial

and temporal uncertainties in three hydrological variables, namely evapotranspiration (ET),

surface runoff and subsurface runoff. The model outputs for a wet year (2008) and an extreme

drought year (2011) are compared against the MODIS ET products and USGS gauge

observations covering the entire state extent of Texas. A novel python-based algorithm is

employed to dynamically extract and populate ~5 km model grids with 8-day MODIS ET

estimates. To separate the surface and subsurface runoff components, a recursive digital filter is

run with the daily streamflow data from 486 USGS gauging sites. Then, the inverse distance

weighted (IDW) interpolation technique is employed to generate the surface and subsurface

runoff fields at ~5 km spatial resolution, appropriate to be used as observational datasets. It is

found that ET is consistently under-predicted over the wet regions, and over-predicted over the

arid to semi-arid regions of Texas, with possible significant differences in absolute values

beween model simulation and corresponding MODIS ET estimates. Accordingly, this distinct

spatial pattern in the ET biases is related to the biases in surface and subsurface runoff fields to

understand their mutual cause-effect relationship and interplay.

---------------------------------------------------------------------------------------------------------------------

Evaluation of Forecast values using WRF-Hydro and Rapid through error estimation

Gonzalo Espinoza1, Kayla Cotterman2

1 Center for Research in Water Resources, University of Texas at Austin, Austin, Texas 2 Michigan State University, East Lansing, Michigan

Abstract:

Floods are one of the most devastating natural disasters, therefore it is imperative to have flood

forecasts that are early and accurate to help prevent life and property loss. The National Flood

Interoperability Experiment (NFIE) forecasts stream discharge for all river reaches in the

National Hydrographic Dataset (NHD). An assessment of the quality of the predicted values is

required for validating the model. This research uses the forecast discharge produced by two

models: WRF-Hydro and the Routing Application for Parallel Computation of Discharge

(RAPID). WRF-Hydro is used for atmospheric coupling and hydrologic modeling (Gochis, Yu,

& Yates, 2013), and RAPID calculates stream discharge over the NHD network (David, Yang, &

Hong, 2013). The present research evaluates the error in the predictions from the WRF-Hydro

and RAPID models by comparing to the discharge values taken from the USGS stream gauges

network for a week of data from 06/10/2015 to 06/16/2015. The present research proves that the

forecast values are consistent due to the error decreasing as the time for the predicted values

approaches the predicted forecast time. In addition, the error is computed and compared for the

USGS II network, which is a subset of the gauges not affected by human activity. The research

estimates the standard error made in the forecast. It also identifies the hydrologic regions where

the model is performing well and areas where a more detailed model is needed.

---------------------------------------------------------------------------------------------------------------------

Evaluation of model skill for the newly developed NFIE streamflow forecast workflows

Scott Christensen1, Shawn Carter2, Chad Drake3

1 Brigham Young University, Provo, Utah 2 University of Alabama, Tuscaloosa, Alabama 3 University of Iowa, Iowa City, Iowa

Abstract:

The National Weather Service provides hydrologic forecasts at nearly 4,000 locations across the

United States. These forecasts, developed by 13 River Forecast Centers as part of the Advanced

Hydrologic Prediction Service (AHPS), are based on the lump-parameter models Sacramento

Soil Moisture Accounting model (SAC-SMA) and Snow17. However, recent advances in

hydrologic modeling have made it possible to provide hydrologic forecasts using physics-based

models for each of the 2.67 million reaches of the NHDPlus dataset. While these advances have

the potential to transform hydrologic forecasting, the new models are still under development

and require further evaluation and testing; for example, currently these new models do not

account for human influences on streamflow. To accurately measure the performance of these

models while eliminating the need to account for anthropogenic impact, we selected locations

that are part of the Hydro-Climate Data Network. Several measures of model performance of

simulated streamflow were automated into the Streamflow Prediction Tool in Tethys. For a given

forecast, major hydrograph components including peak flow, time of peak flow, and total runoff

volume are summarized for each reach. At USGS gaging locations, model performance is

summarized using several common goodness of fit statistical measures, including Nash-Sutcliffe

efficiency (NSE), coefficient of determination (R2), and percent bias, among others. This work

provides the foundation for evaluation of model skill as a function of lead time, generating

summary statistics for a given basin or region based on individual reach performance, and

extending this analysis to the broader national scale.

---------------------------------------------------------------------------------------------------------------------

A GENERAL ADAPTER FOR INTEGRATING A PHYSICS-BASED SNOWMELT

MODEL INTO OPERATIONAL STREAMFLOW FORECASTS

John Koudelka

Utah State University, Logan, Utah

Abstract:

In the Spring of 2010, flooding occurred on several rivers in the Bear and Weber basins of the

Upper Colorado River Basin during the spring runoff. Streamflow observations surpassed the

10% exceedence probabilities that were forecasted by the Community Hydrologic Prediction

System (CHPS) at the Colorado Basin River Forecast Center (CBRFC).

The CHPS system produces streamflow forecasts using the Sacramento Soil Moisture

Accounting Model (SAC-SMA). This system relies on the SNOW17 conceptual snowmelt model

to provide additional water from melting snow into SAC-SMA. This work attempted to

investigate the ability of a physics-based snowmelt model to improve streamflow forecasts

during this event. The year was characterized with lower than average snowpack and one that

persisted beyond the 30 year median and average. Preliminary investigations into the forecasting

of the event suggest that the current SNOW17 model did not model the rapid snowmelt that

occurred.

The Deltares Flood Early Warning System (FEWS), which provides the backbone of the CHPS

system, is an open data handling framework. Model integration is performed through

development of a General Adapter. This work describes the General Adapter that was designed

to integrate the physics-based Utah Energy Balance (UEB) snowmelt model into the CHPS

system for evaluating the performance of this model to produce better forecasts for the Spring

2010 runoff event.

---------------------------------------------------------------------------------------------------------------------

An Arc-Map Add-in for LISFLOOD-FP Model

Zhu Liu1, Hojun You2, Tong Wan3

1 Purdue University, West Lafayette, Indiana 2 Dankook University, Yongin, South Korea 3 University of Alabama, Tusacloosa, Alabama

Abstract:

LISFLOOD-FP is a well performed hydraulic model that is used by European Flood Alarm Syste

m(EFAS) to simulate and predict flood. Currently, however, there exists few GUI for LISFLOO

D-FP model and therefore it is difficult for beginners to use it. In our project, a LISFLOOD-FP

Arc-Map add-in that include functions to run the diffusive solver and subgrid solver is developed

under the environment of C# and ArcObject. With this tool, LISFLOOD-FP users can get rid of t

he difficulties for preparing all the input files in GIS manually. Instead, a friendly LISFLOOD-F

P GUI is provided that can generate cross-sections based on certain spacing, creating profiles for

each cross-section, making the various input files needed by the solvers, running the model and v

iew the simulation results. All these processes are automated by using DEM and Stream line and

users just need to follow the steps in the tool and set proper parameters/configurations. Also, we t

ested our tools with several reaches in United Stated based on 100-year flood and it turns out that

the simulations results are pretty good compared with FEMA 100-year map by checking F statist

ics and running speed is rapid.

---------------------------------------------------------------------------------------------------------------------

Assessment of Reservoir Management During Flood Events

Herman Dolder1, Tingting Zhao2, Lian Zhu3

1 Brigham Young University, Provo, Utah 2 University of Illinois at Urbana-Champaign, Urbana, Illinois 3 University of Alabama, Tuscaloosa, Alabama

Abstract:

The purpose of this project is to investigate the impacts of the reservoir management on the flood

inundation. There are three main parts in the project: 1) use data-driven approach to predict

reservoir inflows; 2) define a standard format to represent reservoir characteristics and operation

rules, and thus allow the development of reservoir modules into RAPID and other routing

models, and evaluate how the reservoir operation affect the river streamflow; 3) evaluate the

economic loss based on the river streamflow from part 2). Regression methods are applied to

predict reservoir inflows using soil moisture and precipitation data. Hazus-MH 2.2, WinEPIC V6

and an Input-Output based indirect economic consequence model were integrated to achieve a

comprehensive economic impact of flood. Details of economic loss with and without reservoir

operation are compared. The economic analysis part provides decision support for the reservoir

operation system in this project. The case study area in this project is the lower Colorado river

basin.

---------------------------------------------------------------------------------------------------------------------

Cross Section Information Integration for Continental Scale Hydraulic Modelling

Chi Chi Choi1, Peng Shang2, Kyungmin Sung3, Solmon Vimal4, Cheng-Wei Yu5, Xing Zheng5

1 Iowa Flood Center, University of Iowa, Iowa City, Iowa 2 University of Alabama, Tuscaloosa, Alabama 3 Purdue University, West Lafayette, Indiana 4 UNESCO-IHE, Delft, Netherlands 5 University of Texas at Austin, Austin, Texas

Abstract:

River geometry information is essential to compute stage and flow using 1-D hydraulic models.

In hydraulic modelling, cross section is the most typical unit to organize river channel data. In

this project, methods for extracting cross section data from different sources have been

developed. The data sources were Flood Risk Information System (FRIS) HEC-RAS models and

high resolution LiDAR (1m) and detailed bathymetry composite DEM. Workflows for using

tools such as AutoRoute (USACE) and cross section automatic generation tool (choi et al., 2015)

were used and the workflows have been identified and demonstrated for the nation wide

contiguous river network - National Hydrography Dataset (NHD). The cross-sections extracted

and assembled for each NHD reach have been integrated into a unique database and linear

referenced onto National Hydrography Dataset (NHD) river network. Using cross section

information stored in the information system, relationship between hydraulic parameters (wet

area, wetted perimeter and cross section top width) and water depth have been developed. In

order to run a Very Large Scale Integrated (VLSI) network hydraulic model, namely 'Simulation

Program for River Networks (SPRNT)', approaches for representing cross-section based

hydraulic parameter functions and river channel geometry information at a river reach have been

tested. Taking flow data from North American Land Data Assimilation Systems (NLDAS) and

United States Geological Survey (USGS) gage observations and reach-scale river geometry data,

SPRNT model will be run for the Upper Alabama River Region for the period the region was

affected by Hurricane Ivan in 2004. The simulation results will be compared with Federal

Emergency Management Agency (FEMA) flood inundation map and floodplain generated from

AutoRAPID.

---------------------------------------------------------------------------------------------------------------------

Identify effective relationships using multiple techniques and resources for data

assimilation flood forecasting

Tim Petty1, Allison Tarbox2

1 University of Alaska at Fairbanks, Fairbanks, Alaska 2 University of Alabama, Tuscaloosa, Alabama

Abstract:

A suite of tiered application tool-sets (provided by NFIE RAPID/ECMWF/Tethys

CloudApp/ESRI ArcGIS/Microsoft Azure/AutoRoute) were applied to the Boise River Basin in

Idaho to assess input & output methodology for optimizing flood inundation mapping. Flood

boundaries were created for the 50, 100, 200, and 500 year floods through the use of AutoRoute,

USGS flow regressions, and DS412 data for Idaho, focusing on the Boise River Basin. Large

historical flows found using RAPID/ECMWF computations will be run in AutoRoute and

compared to the flood boundaries for the flow regressions. Research attempts to run time

correlation and multiple regression analysis on in-situ USGS streamgage data stations within a

designated watershed and compare & contrast river/stream watershed data using Microsoft

Azure Machine Learning tool sets to forecasting model sets. Identifing multiple technigues using

both AutoRoute watershed relationships that build on bothTethys and ArcGIS toolsets that will

utlize forecasting ECMWF precipiation forecasting for Azure streamgage prediction models.

---------------------------------------------------------------------------------------------------------------------

Probabilistic Flood Inundation Delineation Using River Cross-Section Rating Curve

Library Approach

Caleb Buahin1, Cassandra Fagan2, Nikhil Sangwan3, Curtis Rae4

1 Utah State University, Logan Utah 2 Center for Research in Water Resources, University of Texas at Austin 3 Purdue University, West Lafayette, Indiana 4 Brigham Young University, Provo, Utah

Abstract:

Many researchers have recommended explicit incorporation of uncertainty in flood inundation

mapping. A typical approach to accomplish this goal is to use an ensemble of models with

different conceptualizations, initial conditions, and parameterizations that capture the uncertainty

in the model parameter space and model structure. However, the computational expense

commonly associated with hydraulic models, prohibit their use for such evaluations, especially

when dealing with large spatial scales and for real time forecasting scenarios. To overcome this

challenge, we use a rating curve library approach for floodplain delineation. This approach

involves pre-computing a rating curve at various cross-sections along a river by running a

hydraulic model over a wide range of flows. River stage at each cross-section for a given flow

forecast can then be readily derived from the rating curve library, thereby providing a

computationally economical way to delineate local inundation areas. We apply this approach

using a forecast automation software tool we have developed to streamline the workflow. This

software extracts cross sections and ratings curves from an existing HEC-RAS model;

downloads real-time ensemble flow forecasts from the European Center for Medium range

Weather Forecasting (ECMWF) model routed through the RAPID model; and dynamically maps

the inundated areas and their corresponding probability of flooding in immediate future. The

automation tool additionally exports the set of inundation depths and flooding probabilities

rasters to a Tethys web mapping application developed for emergency responders and local

residents.

---------------------------------------------------------------------------------------------------------------------

Integrating the social dimension into flood management: A framework to understand

human response to probabilistic flood warnings and the value of volunteered geographic

information

Erhu Du1, Marc Girons Lopez2, Samuel Rivera1

1 University of Illinois at Urbana-Champaign, Illinois 2 Uppsala university, Sweden

Abstract:

Flood forecasting and warnings provide critical information needed to prevent the loss of human

lives and mitigate economic impacts during flood events. These actions are often approached

from technical perspectives while disregarding the social dimension. To improve current flood

management and emergency response a better understanding and integration of human behavior

and data is needed. This study presents two different frameworks aimed at (1) understanding the

influence of the heterogeneity of human response to flood warnings and (2) real time updating of

flood extent maps using volunteered geographic information (VGI), such as social media and

emergency calls. To understand the impact of flood warning information on human behavior, a

distributed human response model is coupled with a residential evacuation model to simulate the

residential evacuation process given probabilistic flood warnings in flood events. The value of

flood warning given different human decision forcing strategies is then evaluated using scenario

analysis approach. The real time update of the probabilistic flood extents using VGI data is done

by using a modified image segmentation algorithm that reflects the hydrologic and hydraulic

processes. The proposed method, named Hydro Region Growing Algorithm (HRGA), estimates

a probabilistic flood extend around a region of influence from a VGI data point. Thus, the

proposed approach allows the estimation of probabilistic flood extents in areas where no models

and/or gage data are operational. Preliminary results for both of these studies highlight the

importance of the social dimension in flood management and its added value to warning systems.

---------------------------------------------------------------------------------------------------------------------

Generating risk map from probabilistic inundation map (real-time risk analysis)

Morteza T. Marvi

University of Illinois at Urbana-Champaign, Urbana, Illinois

Abstract:

There are several types of uncertainty inherent in flood delineation that has to be transferred

appropriately into flood risk analysis. Discharge in a reach is the first uncertain factor. In real and

near-real time, variety of discharge forecasted and estimated such as what HEPEX reports based

on ECMWF ensemble weather forecast. In long-term risk analysis, discharge of a specific return

period is uncertain due to lack of observation, measurement error, etc. Another set of uncertainty

comes from terrain and its feature. Terrain elevation, manning coefficient, expansion-induction

coefficient, and etc. are some examples of uncertain factors affecting hydraulic analysis and

flood delineation. All of these entail using probabilistic flood inundation map as the base of risk

analysis. In this project, I try to use ensemble flood inundation map for real/near-real flood risk

analysis. I hope this project catch the attention of first responders to better find the vulnerable

community during a flood event.

---------------------------------------------------------------------------------------------------------------------

Observed historical flood classification for hydraulic model validation

Bradford Bates1, Keighobad Jafarzadegan2, Shakiba Morvarid3

1 University of Alabama, Tuscaloosa, Alabama 2 Purdue University, West Lafayette, Indiana 3 University of Texas at Dallas, Dallas, Texas

Abstract:

There are many models to identify the extent of flooding, however it seems that there is not any

appropriate source for calibration, evaluation and comparison of their performance. This project

focuses on the flood extent of historical floods by extracting the remote sensing images

corresponded to the flood events. An automated procedure is developed to determine historical

flood date ranges for a specified list of USGS gauging stations; after storing these date ranges,

Landsat images corresponding to these dates are downloaded. In addition to Landsat products,

Synthetic Aperture Radar is useful for water boundary delineation due to its high spatial

resolution and its ability to penetrate cloud cover. In this study, a combination of Sentinel-1

(SAR) and Landsat 8 images are classified to determine inundation extents, which may be used

to validate outputs from hydraulic models.

---------------------------------------------------------------------------------------------------------------------

Development of tools for including information from the Iowa Flood Center to the National

Water Center model

Felipe Quintero

Iowa Flood Center, University of Iowa, Iowa City, Iowa

Abstract:

The Iowa Flood Center (IFC) produce real-time discharge simulations for the State of Iowa over

a fine resolution network describing hillslopes with approximately 0.04 km2. Here is produced a

set of data structures that will allow the matching of that channel representation with the one

obtained with the NHDPlus network. This product is a key component for comparing simulations

of the models executed by the IFC with the results obtained through the National Water Center

(NWC) modeling system. In addition, this enables the assignation of NHDPlus unique channel

identifiers to about 200 water stage sensors maintained by the IFC, so these can be incorporated

within the NWC models.