|
Overview | Background | New Activity
Overview
The Coastal and Inland FLooding Observation and Warning (CI-FLOW) Project
consortium is working with NCSU to couple their existing estuary model,
watershed water quality model, and an estuary water quality model to
the NSSL multi-sensor precipitation estimation system and NWS distribution
hydrologic model. The
proposed activity is to create the generic mathematical couplings necessary
for the investigation of the responses of primary ecosystem parameters
to flooding events. The resulting CI-FLOW demonstration program will facilitate
the evaluation and testing of new technologies and techniques to produce
accurate and timely identification of coastal, estuary and inland floods,
flash floods and their impacts on the coastal ecosystem. Project CI-FLOW
is focused on the Tar-Pamlico River Basin in North Carolina.
NEW! CI-FLOW 5 Year Science Plan (.doc, 2.4 MB)
Download PowerPoint Presentation: CI-FLOW
Project Overview (Dec. 2004) (.ppt, 3.0 MB)
CI-FLOW One Page Project Description (.pdf, 369 kB)
Background
In February 2000, the National Severe Storms Laboratory (NSSL), National
Sea Grant (NSG) College Program, University of Oklahoma, North Carolina
State University (NCSU), and the North and South Carolina Sea Grant programs
established a joint project, centered in North Carolina areas affected
by Hurricane Floyd. The original collaborators were later joined by the
National Weather Service Office of Hydrologic Development and the National
Environmental Satellite, Data and Information Service (NESDIS). The primary
demonstration area was the Tar-Pamlico River basin. This project, called
CI-FLOW, has established a research and demonstration program for the
evaluation and testing of new technologies and techniques to produce
accurate and timely identification of inland and coastal floods and flash
floods.
The previous activities of Project CI-FLOW include 1) implementation in
the Tar River Basin of QPE-SUMS (Quantitative Precipitation Estimation
and Segregation Using Multiple Sensors; Gourley et al. 2001), a cutting
edge multi-sensor precipitation estimation technique, 2) implementation
of Vflo™, a physics-based distributed hydrologic model (Vieux 2001; Vieux
and Vieux 2002), and coupling of Vflo™ with QPE-SUMS, 3) coupling of the
NCSU Estuary-Lower River Flood model (Xie and Pietrafesa 1999) with output
from both QPE-SUMS and Vflo™, and 4) replacement of the Vflo™ model with
the NWS Hydrology Laboratory's distributed model and linking the
model to QPE-SUMS.
In 2003, Vflo™ and QPE-SUMS were producing real-time estimates of precipitation
and river stage for key points along the Tar River Basin. The initial
products are available to researchers, forecasters, and other potential
users through the Internet in real-time. During that time, our
team members at the SC Sea Grant Extension office were able to collect
information on the critical hot spots for flooding along the Tar-Pamlico
River Basin. This information was helpful in determining locations
where river stage and flow forecasts will be needed.
In
2004, the NWS/OH Hydrology
Lab Research Modeling System (HL-RMS) was introduced and coupled
with the FLDWAV channel model to eliminate proprietary software
source code and licensing issues (Koren et al., 2004). HL-RMS performed
very well in NOAA's NWS-sponsored Distributed Model Intercomparison Project
(DMIP) (Smith et al., 2004; Reed et al., 2004). DMIP garnered participation
from 12 leading distributed modeling researchers in Canada, Denmark,
New Zealand, China, and the US. DMIP was the first extensive comparison
of distributed hydrologic models amongst themselves and to traditional
lumped models. Results from DMIP proved that HL-RMS performed very well and
is a scientifically sound tool for I modeling river basin hydrology (Reed et
a1., 2004)
Like HL-RMS, FLDWAV represents the 'state-of-the-science' in one-dimensional
hydraulic channel routing. FLDWAV is the product of a long and intense
development process (Fread, 1992; Fread et a1., 1988), and has been proven
to accurately reproduce observed river stages and discharges during flood
events. Algorithms in FLDWAV solve the complete one-dimensional St. Venant
equations for unsteady flow. In addition, FLDWAV has the capability of
modeling the effects of a growing list of non-standard hydraulic features
including the ability to model dendritic rivers systems and channel networks,
to account for the effects of off-channel storage areas connected to
the waterway or separated by levees, the ability to account for the effects
of hydraulic structures (dams, bridges, levees), and the ability to handle
flows in the subcritical and/or supercritical flow regime. FLDWAV is
being used operationally to model several coastal rivers including the
Columbia, lower Mississippi, and St. Johns rivers. It also serves as
the basis for the generation of several types of forecast flood inundation
maps (Cajina et aI., 2002).
Under last year's CI-FLOW work plan, HL-RMS and FLDWAV were modified to create
a direct linkage between the two. Now, output from HL-RMS can be used as
direct input into FLDWAV, allowing the two models to function as 'stand-alone'
components that do not need to be tied to any current river forecasting
system such as that used by NOAA's NWS. We believe that this combination
is one of the first (if not the first) linkage of operational distributed
models and advanced channel routing models for river forecasting.
An initial coupling of QPE SUMS with
the OH Distributed Modeling System (HL-RMS) occurred by converting
QPE SUMS onto the HRAP grid at 1 km resolution. In addition, the HL-RMS
was coupled with the OH channel routing model. Furthermore, an effort
to connect the channel model and the Estuary model was begun.
return to top
New Activity
Water quality modeling
The newest activity that has just begun includes implementation of a
precipitation-driven watershed water quality model on the Tar-Pamlico
River. The output from the watershed model will be directly coupled with
an estuary water quality model, which will be further linked with the
existing NCSU estuary- lower river flood model. The integrated modeling
system will then be capable to explore responses of primary ecosystem
parameters, such as nutrients, dissolved oxygen, and primary production,
to upland flooding events.
The watershed water quality model is based on a public domain software
program HSPF (Hydrologic Software Program – Fortran)
distributed by the U.S. EPA's Center for Exposure Assessment
Modeling. HSPF is a comprehensive watershed simulation model
designed to simulate a wide range of hydrologic and water quality
processes (Bicknell et al., 2001). The model simulates various
forms of nutrients (e.g. ammonium, nitrate, organic nitrogen
and phosphate) based on time series of precipitation data and
land use characteristics at the river basins.
The estuary water quality model is based on CE-QUAL-ICM from US Army Corps
of Engineers, Waterways Experiment Station (Cerco and Cole, 1994) and HEM3D
(Hydrodynamic Eutrophication Model in Three Dimension) from the Virginia
Institute of Marine Science (Part et al., 1995). Both models have been
heavily used in water quality modeling in Chesapeake Bay and its tributaries
(Cerco, 1995; Lin and Kuo, 2003). Upon receiving the nutrient loadings
from the watershed model, the estuary water quality model simulates spatial
and temporal distributions of multiple components of nutrients, dissolved
oxygen, carbon and phytoplankton.
Model Calibration
Once a model is selected, it must be calibrated for a watershed (James
and Burges, 1982). Calibration is widely recognized as a critical
step in the application of any hydrologic or channel routing model. This
year we will focus on the calibration of HL-RMS and FLDWAV. As seen in
the purple boxes in Figure 1, the models are major components of the
CI-FLOW system. Implicit in the process of
calibration is the collection, quality control, and formatting of all
data needed for the calibration process.
HL-RMS
takes the gridded rainfall observations from QPE-SUMS and converts
the rainfall to runoff and streamflow in each grid. Runoff and streamflow
from each grid move down hill to the next grid, eventually making
their way to the main Tar River channel below Tarboro. For HL-RMS,
calibration involves deriving the best values of the model parameters
in each of the grid cells in the Tar River basin. Here, initial parameters
describing the type of soils, the slope of the land, and the type
of vegetation must be specified. HL-RMS assumes each grid cell contains
a river, so the size, shape, and flow resistance of the river must
be numerically described. In calibration, these initial parameters
are adjusted until the computed flows agree with observed flows measured
by the US Geological Survey. Calibration of distributed models
is an active area of research. In NOAA’s NWS method, traditional
methods of hydrologic model calibration (Smith et al., 2003) are
combined with emerging techniques to adjust the parameters in each
grid cell.
Determine Statistical Biases for HL-RMS
The Hydrometeorology Group of HL will analyze the precipitation input
to HL-RMS as necessary during the hydrologic model calibration process. Such
an analysis might be necessary to determine if statistical biases in
the precipitation fields emerge or change during the calibration period. The
analysis will be a thorough comparison with an independent set of reference
rain gauge reports.
FLDWAV Connectivity and Calibration
FLDWAV
takes all of the runoff and streamflow generated by HL-RMS and routes
the combined inflows through the Tar River system which includes the
Tar River from Rocky Mount, NC to the head of the Pamlico Sound and all
of its major tributaries within the routing reach. Water levels
and discharges are generated at all locations in the river system for
all times. In addition, the computed river levels can be used to
generate flood inundation maps.
In
order to calibrate the river system, FLDWAV
requires the following information (Sylvestre et al., 2002): the river
system defined based on hydraulic conditions; inflows; observed river
levels and discharges (if available) at the gage locations; channel roughness
represented by Manning; cross section data representing the topography;
and a description of any critical hydraulic structures (.e.g., dams,
bridges, levees) in the river system. Rivers with significant backwater
and/or significant attenuation due to storage will be modeled dynamically,
while rivers that simply add to the flow will be modeled as lateral inflow.
Topographic information will be obtained from LIDAR-based Digital Elevation
Model data. The roughness coefficients will be adjusted until the
difference between the computed and observed water level time series
has been minimized at each gage location. To ensure the correct
volume is maintained in the river system, the relationship between the
effective and inactive storage will be adjusted until the difference
between the computed and observed discharges is minimized. The
FLDWAV parameters will be calibrated for both record high flows and low
flows to ensure model stability when running operationally. They
will be verified with an independent flood.
Since the hydrologic and hydraulic model calibrations will be done concurrently,
discharge and water level data will be obtained from the USGS and NWS archives. After
the calibrations have been completed, inflows from HL-RMS will be used. An
assessment will be done to determine the best way to represent the gridded
inflows in FLDWAV.
Improved QPE
A primary goal of CI-FLOW is to develop and identify an optimum set
of techniques and algorithms to serve as a state-of-the-science NOAA
multi-sensor QPE. Currently within NOAA, there are three primary research
activities focused on the production of near-real time high-resolution
multi-sensor QPE.
NSSL: National Mosaic for Quantitative Precipitation
(NMQ/Q2). The
NMQ/Q2 consists of a prototype real-time computing system that generates
three-dimensional mosaics of radar reflectivity and a suite of derived
products including multiple rainfall products. The system has also
been designed to ingest all relevant grids for rainfall estimation purposes
such as multiple radar, rain gauges, satellite imagery, model output,
and lightning flashes. The reflectivity mosaics and derived products
are interpolated to a 1-km CONUS grid mesh on 31 vertical layers and
are updated every 5 min. The system can process radar reflectivity
into rainrates outside the geographic boundaries covered by operational
WSR-88D precipitation products. Reflectivity data are quality controlled
using several reflectivity and velocity-based fields in a neural network.
A technique for correcting rainfall estimates for reflectivity profile
effects has been implemented. A capability for ingesting data from
radar networks other than WSR-88D is under development. Lastly, different
techniques for adjusting rainfall accumulations using collocated rain
gauge reports have been implemented.
OHD: Multisensor Precipitation Estimator (MPE). The MPE function within
the Advanced Weather Interactive Processing System (AWIPS), integrates
rain gauge, radar, and satellite estimates into fields covering the
area of responsibility for individual WFO's and RFC's. MPE includes
a large suite of interactive tools for quality control (QC) of all
inputs, particularly interactive and automated rain gauge QC. The
rainfall estimates are interpolated to a 4-km grid and updated hourly.
An enhanced MPE (EMPE) is under final development; EMPE will create
1-km grids on a subhourly update cycle.
NESDIS: Hydro-Estimator (HE). The HE hourly rainfall product is based
on satellite infrared (IR) signatures for rainfall rate modulated by
precipitable water and lower tropospheric mean relative humidity. An
improved calibration of this algorithm is under development. Since
its inputs come from geostationary satellites and numerical weather prediction
models, coverage by the HE is nearly continuous in time and space across
the conterminous United States and adjacent areas. Presently, HE output
is ingested by MPE.
The CI-FLOW project will leverage these existing activities to create
a research collaboration between NSSL, OHD, and NESDIS focused on the
production of the most accurate multi-sensor QPE updated every 5 to
6 minutes on a 1-km spatial scale. This research collaboration will:
- Document existing QPE techniques and algorithms
- Compare the performance
of these techniques and algorithms in past precipitation events to
actual ground-truth rain gauge reports
- Identify the most accurate
and efficient techniques and algorithms to incorporate into a NOAA
QPE product
- Evaluate the impact of this
NOAA QPE on Tar River basin streamflow simulations produced by advanced
distributed hydrologic models
Additional Radars
By leveraging the previous successes of Project CI-FLOW,
two complimentary proposals (e.g., NOAA HPCC and NSSL DRDF) have been
funded to include both FAA TDWR radar data and Canadian radar data
in the algorithms that estimate the precipitation (Figure
2). The
data from these radars are not only taken closer to the ground (less
chance for errors due to evaporation, etc.), but will also help fill
in gaps between the NWS radars, thus improving overall QPE accuracy
and coverage. Canadian
radar data will provide additional sampling over the Great Lakes and
Great Lake watershed headwaters in Canada. Several TDWRs in the
Tar-Pamlico basin should lead to improvements in the precipitation
estimation in the basin.
Enhanced Coverage
The QPE-SUMS coverage along coastal regions will be expanded
from 230 km to 460 km, increasing the coverage over the ocean to provide
earlier detection and better short-term prediction of precipitation.
Collectively these efforts should improve the QPE used as input to
the NWS and NCSU coupled model system and, therefore, improve the likelihood
for improved flash flood forecasts.
return to top |