Real-time Flood Forecasting in Iowa with a Data Assimilated Hydrological

Model

Scott J. SmallThe Iowa Flood Center

IIHR Hydroscience & EngineeringThe University of Iowa

Real-time flood forecasting in Iowa

Streamflow forecasts are produced: Every hour For 10 days At over 2000

locations

Streamflow forecasts based on the IFC model are publicly available in the Iowa Flood Information System (http://iowafloodcenter.org/)

Flood warnings at points of interest

Streamflow forecasts based on the IFC model are publicly available in the Iowa Flood Information System (http://iowafloodcenter.org/)

Iowa Flood CentersDistributed Hydrological Model

Example of Landscape decomposition (Turkey River)

Parameters describing soil conductivity, infiltration

capacity, slope, and roughness are assigned to

each terrain unit

Example of Landscape decomposition (Turkey River)

Statewide implementationOur distributed hydrological model provides streamflow predictions everywhere in Iowa

About 420,000 individual hillslopes

IFC Modelling Framework

Four ODEs per hillslope

Mass Conservation Equations

Fluxes between control volumes

Channel Routing

Radar rainfall coverage of Iowa

0.004167 degree spatial resolution (about 300m by 400m)

5 minute temporal resolution

Custom numerical solvers produce solutions to the system of differential equations

ASYNCH numerical solver for ODEs with a tree structure

Developed at the Iowa Flood Center for distributed hydrological models

Runs on distributed memory architectures

Written in C

Combining streamflow observations withmodel projections

Observations

USGS Streamflow Gauges Around 200 in Iowa 15 minute resolution

IFC Bridge Sensors Around 200 in Iowa 15 minute resolution Rating curves under

development

USGS Gauges and IFC Bridge Sensors

Given a collection of observations over a period of time, what initial model states best reproduce the observations?

Data assimilation performed through a least squares fitting of model inputs

Find the discharge at each channel minimizing the energy functional

with the linearized model discharges

and discharge sensitivity matrix H and observation vector d.

Calculating the sensitivity matrix requires solving an additional ODE at each link for every upstream link

Links too far from a gauge can be ignored in the sensitivity matrix

Travel times with (high) linear velocity can limit greatly calculations.

Iowa Flood CentersStreamflow Forecasting System

IFIS

Real-time flood forecasting with data assimilation

BackgroundTime

Time ofForecast

1. Model channel discharges are selected at the background time to fit data

Real-time flood forecasting with data assimilation

AssimilationPeriod

BackgroundTime

Time ofForecast

1. Model channel discharges are selected at the background time to fit data

2. Model is advanced to the time of forecast

Real-time flood forecasting with data assimilation

AssimilationPeriod Forecast

BackgroundTime

Time ofForecast

1. Model channel discharges are selected at the background time to fit data

2. Model is advanced to the time of forecast

3. Model based forecast is produced

Iowa River above the city Marengo, IA

7,200 km2drainage area(18,000 hillslopes)

10 available USGSGauges

Forecasts from June 20th, 2015

Future Work and Additions

Expand data assimilation domain to the entire state of Iowa

Incorporate Iowa Flood Center bridge sensor data (requires rating curves)

Adjust hillslope model states based upon streamflow observations

Incorporate observation and model uncertainty

Real-time Flood Forecasting in Iowa with a Data Assimilated Hydrological ModelReal-time flood forecasting in IowaFlood warnings at points of interestIowa Flood CentersDistributed Hydrological ModelSlide Number 5Slide Number 6Slide Number 7Slide Number 8Slide Number 9Slide Number 10Slide Number 11Custom numerical solvers produce solutions to the system of differential equationsCombining streamflow observations with model projectionsObservationsUSGS Gauges and IFC Bridge SensorsSlide Number 17Data assimilation performed through a least squares fitting of model inputsLinks too far from a gauge can be ignored in the sensitivity matrixIowa Flood Centers Streamflow Forecasting SystemReal-time flood forecasting with data assimilationReal-time flood forecasting with data assimilationReal-time flood forecasting with data assimilationIowa River above the city Marengo, IASlide Number 25Future Work and Additions