Presented at ASCE Conference
North American Water and Environment Congress '96
Anaheim, Calfornia
June 24-28, 1996
Distributed Parameter Hydrologic Modeling and NEXRAD
for River Forecasting: Scale Issues Facing the National Weather
Service
Michael B. Smith
Dong-Jun Seo
Bryce D. Finnerty
Victor Koren
Office of Hydrology
NOAA/National Weather Service
1325 East-West Highway
Silver Spring, Maryland 20910
Abstract
With the advent of NEXRAD (NEXt Generation RADar), the National
Weather Service has the
opportunity to move from current lumped parameter modeling to
more of a distributed parameter
hydrologic modeling approach for river forecasting. However,
this high resolution precipitation
data poses the problem of calibration at one spatial and temporal
scale while using the models
operationally with NEXRAD data at a finer spatial and temporal
scale. Until national NEXRAD
coverage exists for a long enough time period to prove useful for
calibration, understanding must
be gained as to how to adjust calibrated model parameters to
account for runoff volume
differences resulting from using high resolution rainfall
inputs. To examine this, a synthetic
watershed was modeled using various sized computational elements
using 9 months of archived
NEXRAD rainfall data. Results indicate that finer spatial and
temporal scales result in the
generation of more total runoff and fast response runoff. Tests
indicate that hydrologic model
parameters are not transferrable across scales. When calibrating
at a lumped scale and then
disaggregating the basin for operational forecasting, certain
parameters must be adjusted in
order to recalibrate the hydrologic model.
Introduction
The Hydrologic Research Lab of the National Weather Service (NWS)
has embarked on a
program to move towards distributed parameter hydrologic modeling
for flood and long-term
river forecasting. Currently, the 13 River Forecast Centers (RFC)
of the NWS use continuous
hydrologic models with spatially lumped parameters and
precipitation inputs to provide daily
stage/discharge forecasts at over 4,000 locations across the
Nation (Fread, 1995).
Given the high resolution and quality of NEXRAD data, the NWS
has an opportunity to
depart from lumped parameter modeling and more effectively
account for the spatial and
temporal variability of precipitation. For example, current
procedures at RFC's usually involve
the generation of mean areal precipitation (MAP) values derived
from point raingage data.
These average inputs are normally for 6 hour computational time
increments and for watersheds
several hundred square miles in area. These models are
calibrated using up to 45 years of
historical streamflow and precipitation data, with the result
that the hydrologic model parameters
are inherently related to the spatial and temporal scale of
calibration. In contrast, NEXRAD will
provide hourly rainfall measurements over a 4 x 4 km grid,
representing unprecedented
resolution for the United States.
Rather than moving directly to a gridded or other high
resolution distributed parameter
hydrologic model that is based on the NEXRAD grid, a
semi-distributed modeling approach has
been adopted in HRL as a first step towards utilization of the
NEXRAD data (Smith, 1995). In
this format, a basin currently being modeled by an RFC would be
disaggregated into several
constituent sub-basins. Instead of using point raingage
measurements to compute MAP valuesfor each entire basin, MAP values for each sub-basin would be
derived from the gridded
NEXRAD values. Unit hydrographs would be developed from standard
methods or
geomorphological analysis and used to convert runoff volumes to
discharge values. A
Muskingum-Cunge routing operation will be used to translate
hydrographs to the next
downstream computational point. The goal of the overall research
is to provide RFC personnel
with: 1) hydrologic tools to model sub-basins, 2) guidelines as
to what degree to disaggregate
a lumped basin to capture essential spatial rainfall variability,
and 3) guidelines as to the
adjustment of calibrated model parameters to account for finer
operational modeling scale.
Future research will address the development of gridded
distributed parameter models.
Scaling Issues
As with distributed parameter models in general, problems arise
when the issue of calibration
is considered. The conceptual model currently used by the NWS
that would be applied to each
sub-basin is the Sacramento Soil Moisture Accounting Model
(SAC-SMA) (Burnash, 1995).
Practical experience and statistical analyses have shown that
this model requires a minimum of
5 to 8 years of data for proper calibration (University of
Arizona, 1995). Current NWS
calibration procedures call for a 6 hour time step and are
limited to the basin scale. However,
only a few years of NEXRAD data is available as of the current
date. Until enough NEXRAD
data is available for calibration, it is proposed that the
hydrologic model parameters be
calibrated at a lumped basin scale and 6-hour time step and then
uniformly applied to each
constituent sub-basin.
However, in this approach, some adjustment must be made to the
hydrologic model
parameters as they are derived at a basin scale and 6-hour time
step and then used operationally
at sub-basin scale with 1-hour NEXRAD data. Runoff volumes
generated at the calibration scale
will be different than those generated at the operational scale.
Thus, the NWS must understand
how to adjust model parameters when calibrating at one scale and
operationally forecasting at
a different spatial and temporal scale.
Current Research
Modeling efforts have been carried out to investigate the runoff
volume differences resulting
from lumped versus semi-distributed operational scales. Tests
were conducted to identify SAC-SMA model components having the
greatest scale dependency. In the testing, a 64x64 matrix
of NEXRAD gridded precipitation values was obtained for an area
near the Oklahoma-Arkansas
border. This data was collected for a 9-month period. SAC-SMA
model parameters were
obtained from a calibrated basin within the geographic extent of
the 64 x 64 matrix.
In the testing, the 64 x 64 matrix was considered to represent
a synthetic watershed. For
the 9 month period, runoff volumes from the SAC-SMA were computed
for the watershed at 7
different spatial scales as shown in Table 1. At the coarsest
level, the entire 64 x 64 matrix was
considered to be a lumped basin. For each time step, 64 x 64
elements were used to compute
a single mean areal precipitation value as input into the
hydrologic model. At the next level,
the watershed was disaggregated into 4 sub-basins, each
consisting of 32 x32 NEXRAD
elements. At the finest scale, each of the 4056 NEXRAD cells was
considered to be a sub-basin.
Table 1. Sub-Basin Size for 7 Modeling Scales
| Scale |
Size of Sub-basin in
NEXRAD 4-km Cells |
Number of Sub-Basins |
| 1 |
64 x 64 |
1 |
| 2 |
32 x 32 |
4 |
| 3 |
16 x 16 |
16 |
| 4 |
8 x 8 |
64 |
| 5 |
4 x 4 |
256 |
| 6 |
2 x 2 |
1024 |
| 7 |
1 x 1 |
4056 |
Results
Computations for the 7 scales were performed for each of 3 time
steps: 1, 3, and 6 hour.
Runoff volumes from the SAC-SMA components for each scale and
each time step were depth-averaged for the entire 9 month period.
Figure 1 presents the depth-averaged runoff volumes
for each major SAC-SMA component for the 1-hour time step
scenario.
From Figure 1 it can be seen that some of the component runoff
volumes are dependent
on the size of the constituent sub basin. This indicates that
hydrologic model parameters are not
transferrable across scales. In particular, surface flow displays
a marked scale
dependency. At the coarsest scale, no
surface runoff was computed. Surface runoff occurs in the SAC-SMA
model when the two
reservoirs representing the upper soil layer are filled and the
rainfall rate exceeds the rate of
percolation and interflow generation. Tests of the 9-month
NEXRAD data set indicate that only
one rainfall event achieved 100% coverage of the entire 64 x 64
synthetic watershed. Due to
this partial coverage, computations of mean areal precipitation
at the coarsest scale include
many NEXRAD cells with zero rainfall, resulting in smaller MAP
values being input to the
SAC-SMA representing the entire basin. Consequently, the two
reservoirs in the upper soil
layer never fill so as to produce surface flow.
Figure 1. 1-Hour Mean of Runoff Components Vs. Size of
Computational Area using 9
Months of NEXRAD Data.
The greatest surface runoff was computed when the watershed
was modeled at the finest
spatial scale. Thus, as a basin is disaggregated into smaller
sub-basins to capture the spatial
variability of rainfall as measured by NEXRAD, more surface flow
would be generated if the
same hydrologic model parameters were used as in calibration. The
sub-basin representation
would no longer represent a calibrated system; the SAC-SMA
parameters governing the
generation of surface flow would need adjustment. Figure 1 also
shows that parameters
associated with the generation of interflow and supplemental base
flow would need adjustment.
Conclusions and Further Research
From the series of runoff volume tests at different spatial and
temporal scales, it is clear that
hydrologic parameters for the SAC-SMA are not constant across
scales. If parameters are
derived during calibration at a lumped scale, they must be
adjusted when the lumped basin is
disaggregated into a collection of sub-basins. It is envisioned
that parameter adjustment
guidelines will be developed during testing that involves
generating discharge hydrographs from
sub-basins. Five watersheds in Oklahoma have been selected for
initial testing. Historical
streamflow data and precipitation data have been assembled.
Calibration at a lumped scale is
in progress. These basins will be disaggregated to various
levels of sub-division and resultant
discharge hydrographs will be compared to observed streamflow
records.
References
Burnash, R.J.C., "The NWS River Forecast System - Catchment
Modeling," Computer Models of Watershed Hydrology: Edited
by V.P. Singh, Water Resources Publications, 1995, pp.311-366
Fread, D.L., "A Pathway Toward Improving Hydrologic
Predictions," Proceedings: Iowa Hydraulics Colloquium Issues
and Directions in Hydraulics, IAHR's Journal of
Hydraulic Research, June, 1995.
University of Arizona, Department of Hydrology and Water
Resources, (1995). "Progress Report, 1995-1996 Cooperative
Agreement NA37WH0385," Submitted to the Hydrologic
Research Laboratory of the U.S. National Weather Service.
Lindsey, S.D., "Strategy for Utilizing Radar-Based
Precipitation Estimates for River Forecasting," Engineering
Hydrology: Edited by Kuo, C.Y., ASCE, 1993, pp. 940-945.
Smith, M.B., "Distributed Parameter Modeling Project Plan,
Phase 1 - Distributed Inputs"
Hydrologic Research Lab Internal Publication, 1995.
|
www.nws.noaa.gov
