|
Report to MD Dept. of Environment
Development of an integrated modeling system of watershed and
estuarine ecology for management of the Patuxent basin
Center for Environmental Science
Horn Point Laboratory (HPL)
Chesapeake Biological Laboratory (CBL)
Appalachian Laboratory (AL)
University of Maryland |
Engineering Research and
Development Center (ERDC)
US Army Corps of Engineers |
Thomas R. Fisher, lead PI, HPL
Michael Kemp, HPL
Raleigh Hood, HPL
Michael Williams, HPL
Gregory Radcliffe, HPL
Walter Boynton, CBL
Keith Eshleman, AL
Dan Fiscus, AL
Carl Cerco, ERDC
Sung Chan Kim, ERDC
Table
of Contents
Introduction
Progress Report
Bathymetry of the Patuxent
Estuarine Transect Line
Estuarine Modeling Grid
Marshes in the Patuxent
Watershed Model Validation
Weather Data
Patuxent Nutrient Budgets
Diffuse Source Models
Miscellaneous
References
Introduction
(back to table of contents)
The University of Maryland Center for Environmental Science
(UMCES), jointly with the Waterways Experiment Station (WES), is developing
a coupled modeling system for nutrient management of the Patuxent basin.
Funds for this project became available from MD Department of the Environment
(MDE) in January 2002, and this report covers the first year of progress
on this project (Jan. 2002 - Dec. 2002). The main goal of this research
effort is to deliver to MDE by December 2004 a watershed model (HSPF)
spatially linked to a 3-dimensional estuarine circulation model (CH3D)
and water quality model (CE-QUAL-ICM). This coupled modeling system of
the Patuxent will be used by MDE to develop TMDLs in the first half of
2005 with assistance from UMCES and WES. All models are those in current
use by the EPA Chesapeake Bay Program (CBP) using existing Bay Program,
USGS, and other available data.
The original intent of this project was an independent calibration of
the watershed and estuarine models using all available data. Following
calibration to minimize errors in model output, validation with a reserved
set of data was proposed to be done to generate independent estimates
of model errors in segments of the terrestrial basin. However, at the
request of MDE in the summer of 2002, six months after the project began,
MDE expressed concern about the legal ramifications of more than one set
of watershed model output (CBP and MDE), and they requested that we use
exactly what the CBP uses for watershed model output. We have agreed to
work only with the existing watershed model output, and to focus our efforts
on validation of the watershed model and on the investigation of possible
improvements to future versions of the watershed model (specifically Phase
5). More details on these changes are summarized in a recent letter to
MDE, which initiated the development of an addendum to the MOU between
MDE and UMCES that describes these changes requested by MDE.
The estuarine models will be run as described in the original proposal.
We have improved the original modeling grid with a higher resolution version
(see below) in order to improve the simulation of the effects of bathymetry
and shoreline morphology. The new estuarine grid is embedded in the current
grid of the entire Bay in order to simulate and estimate the role of exchange
with the mainstem Bay as a net nutrient source or sink for the Patuxent.
To support the modeling efforts, watershed and estuarine data are being
compiled in GIS databases as needed for modeling and mapping of model
output.
In addition to the main goal of delivering the coupled modeling system
to MDE, we also have four research objectives. After we have validated
the models in routine use by the Chesapeake Bay Program, we will do the
following
- compare these models with alternative approaches
(watershed: landuse-specific yield coefficients, the Patuxent Landscape
Model, GWLF, SPARROW; estuary: box and statistical models).
- compare the importance of various sources of N and
P in the basin (direct atmospheric deposition, diffuse and point sources,
main stem Bay advective inputs).
- investigate the processing of the watershed inputs
within the estuary (conversion of inputs to phytoplankton and sediments).
- explore net estuarine retention within the Patuxent
vs. export to the mainstem Bay.
Progress Report
(back to table of contents)
This progress report consists of a series of task descriptions.
These have been conducted collectively and by individual principal investigators
and associates and have been underway for the last year in order to achieve
the main goal of this project. The descriptions are neither complete nor
comprehensive, and the material shown is a sample of the activities which
are underway to give examples of some of the project accomplishments.
Bathymetry
of the Patuxent (back
to table of contents)
One of the first tasks undertaken in the project was the construction
of a bathymetric database for the Patuxent. Using data collected by MDE
in August, 2001 (prior to the start of the project), as well as NOAA point
soundings, we created a point coverage and grid file in ArcGIS. The point
coverage was the actual sounding data with spatial coordinates, whereas
the grid file was extrapolated from the point coverage to provide a spatially
continuous and smoothed depth field for the development of the new estuarine
voxel grid for estuarine modeling (see below). The point coverage of the
original data is illustrated below in Fig.
1.
Estuarine
Transect Line (back
to table of contents)
An estuarine transect line is a convenient reference for location within
the estuary in terms of distance from the mouth. We created an arc coverage
which ran through the deepest parts of the estuary from the mouth to above
Western Branch. The line was segmented at 1 km intervals from the mouth
to km 105 in the little Patuxent River. The origin at the mouth of the
estuary corresponds to the origin used by the estuarine circulation and
water quality models, and also corresponds to the origin used by Cronin
and Pritchard (1975). Fig. 1 shows
the estuarine transect line, with the 1 km points along the line.
Estuarine
Modeling Grid (back
to table of contents)
One of the more important improvements to the estuarine modeling was the
development of a more detailed voxel grid. The goal of this effort was
to attempt to match the model voxels with the true bathymetry and shoreline
as much as possible. The voxel grid in use for the Patuxent at the beginning
of this project was developed from the perspective of Chesapeake Bay as
a whole, and had relatively low resolution (136 by 96 by 19). The spatial
fidelity relative to shorelines and bathymetry was low when viewed from
within the Patuxent, particularly in the upper estuary (see old grid in
upper estuary in Fig. 2). The bathymetry
data described above and a shoreline arc file created from USGS topographic
maps were used to develop a higher resolution voxel grid (152 by 166 by
19) which preserved more of the true shoreline and bottom bathymetry,
and which also incorporated more of the tributary structure of the Patuxent,
including Western Branch in the upper estuary. Note the additional tributaries
and the more realistic rendition of Point Patience in the lower Patuxent
(lower right corner of Fig. 2),
as well as the more accurate representation of the upper Patuxent (upper
right in Fig. 2). Not shown in
Fig. 2 is the more realistic bathymetry
which is rendered by subsurface voxels by extracting data from the NOAA
NGDC Coastal Relief Model into the new grid. It is hoped that the more
realistic bathymetry and shoreline in the voxel grid will improve the
simulation of the estuarine circulation and mixing, and initial model
runs have already been made successfully with the improved grid.
Marshes
in the Patuxent (back
to table of contents)
Marshes are an important landscape component that have not been included
in previous versions of the Bay Program’s models. Intertidal and
non-tidal marshes were not considered as landscape sinks for TSS, N, and
P in either the watershed or estuarine modeling. The omission of intertidal
marshes from the estuarine water quality model may be responsible for
the significant overestimation of simulated N and P concentrations compared
to those observed in the upper estuary where the marshes are very abundant
(see also nutrient budget section below). In order to include these important
landscape components in the coupled Patuxent modeling system, we have
created a detailed polygon coverage of wetlands from USGS topographic
maps. Three kinds of polygons were defined: (1) those which exchange laterally
with the estuarine voxels, (2) those which occur at the end of a creek
and exchange with the end of a voxel, and (3) those which are surrounded
by land and interact with the land only (see Fig.
3). In addition to the attributes normally assigned by ArcGIS, each
intertidal marsh polygon has attributes consisting of the adjacent estuarine
voxel id with which it exchanges water and materials as well as the number
of the nearest estuarine km from the estuarine transect line described
above in Fig. 1. In 2003 we will
complete the marsh polygon coverage, adding additional attributes of NWI
class, dominant species, CNP plant biomass, CNP burial rates, and denitrification
rates, using both observed and extrapolated data.
Watershed
Model Validation (back
to table of contents)
The modification of the objectives of watershed modeling
have changed our primary focus to validation of the watershed model. Below
we give some preliminary examples of the current validation errors for
water discharge, total N, total P, and total suspended solids in the phase
4.3 watershed model.
Simulating water discharge accurately is essential for the coupled modeling
system. Using data from Chesapeake Bay Program model version 4.3 obtained
from Gary Shenk, and USGS data, we compared model output to observed flow
on a daily basis. Scatter plots of model vs observed flow for Bowie and
Laurel USGS stations are shown in Figs. 4
and 5. The model output for the
Bowie segment shows an unbiased relationship to observed flow (Fig.
4), with rms errors of " 82% (ave. root-mean-square difference
between predicted and observed relative to observed). This error reflects
the scatter about the 1:1 line in Fig.
4) and indicates that on a daily basis most of the predicted flow
values are within a factor of two of the observed flows.
The Laurel model output agrees less well with observations (Fig.
5). There is more scatter and many zero values for flow (not shown,
but about 1400 out of 5100 daily data points). This difference between
stations is probably due to the fact that Bowie was one of the calibration
stations, whereas Laurel was not, which makes Fig.
5 a true validation test of the model output.
Daily data were also aggregated to annual runoff for 1984-1997. At Bowie,
this comparison showed good agreement between the average observed and
predicted runoff (36 vs 38 cm 1/y, respectively), with annual rms errors
of + or - 18% (see Fig. 6). This
indicates that model precision improved from daily predictions (rms errors
= 82%) to annual predictions (rms errors = 18%), as model errors tend
to cancel at longer time scales. For the entire 14 year period (1984-1997),
the cumulative error declined still further to 6%. For Laurel, the average
observed and predicted annual runoff values were 21 and 17 cm 1/y, respectively,
with annual rms errors of 38% and a cumulative error of 19%. The comparison
of annual runoff values suggests no substantial model bias for the Bowie
station (approximately equal scatter about the 1:1 line in Fig.
6), with 18% rms errors. However, for Laurel there was a negative
bias with larger rms errors of 38% (Fig.
6).
The Bowie segment is the only one with no predicted flow values of zero.
For Bowie, spectral analysis was done on both the observed and predicted
daily time series for the 1984-1997 period (Fig.
7). The results indicate that the phase 4.3 model does a reasonable
job of matching the frequency domain and scaling behavior of the actual
data; however, the biggest difference was in the high frequency portion
of the spectrum (time scales <7 days, right vertical line in Fig.
6 at log f = -0.85), where the spectral slope
for the observed data was 1.98 and the model was 1.47. This suggests less
attenuation of higher frequency variance in the model relative to actual,
which may be due to a hydrograph recession constant which is too small
or a time resolution problem. The model output also does not have the
same peak at the 1-year period as the actual data (left vertical line
in Fig. 6 at log f = -2.56). We
plan additional work to clarify the results of the frequency analysis,
identify the causes, and propose corrections to the phase 5 watershed
model.
Simulation of N concentrations in streams is another
essential feature of watershed modeling. To validate the model output,
we have compared the daily average predicted total N (TN, mg L-1) at Bowie
with water chemistry observations by USGS in grab samples (Fig.
8). In this case, the data points scatter about the 1:1 line, indicating
little apparent bias and rms errors of 43%. This can be seen more clearly
in Fig. 9, which is a frequency
distribution of the % difference between predicted and observed TN values
relative to observed TN at Bowie. Errors are approximately log normally
distributed and range from -70% (low predicted values) to +150% (high
predicted values). The median error is only 0.7%, but the mode is -15%,
indicating a slight tendency of the model to underestimate TN. Several
measures of validation errors for TN are given in Fig.
9, but it is clear that most predicted values fall within "50%
of the observed ones.
TP and TSS estimates by the phase 4.3 watershed model
were weaker than for TN. In Fig. 10,
the comparison of modeled and observed TP shows evidence of model bias
(slope of 0.33) and considerable scatter (rms errors of 68%). As for TN
(Fig. 9), the % errors for TP were
log normally distributed and ranged from -90% to +360%, with most falling
between "100% (not shown). The mode of the distribution was -25%,
and median errors were -8%, both again suggesting a tendency to underestimate
TP. For TSS (total suspended solids), even larger errors were found (Fig.
11). There was more scatter for TSS (rms errors of 240%), with evidence
of model bias (slope of 0.50). In a large fraction of the paired values
in Fig. 11, there is a large range
of observed TSS values (2-500 mg L-1) for which modeled values range only
over 7-30 mg L-1. This may be indicative of mismatches in the timing of
discharge events (e.g., modeled water discharge lags or precedes observed
discharge), or some other model inaccuracy. We plan to investigate the
causes of the apparent biases in modeled values of both TP and TSS, and
for phase 5 we hope to provide suggestions to reduce the scatter in all
predicted watershed values compared to observations (Figs. 4-11).
Weather
Data (back
to table of contents)
Initially we were compiling weather and other input data for the watershed
calibration. This activity occurred only in the first half of 2002 prior
to the MOU change described above. While these data will not now be used,
a considerable archive was acquired, particularly for weather data. A
total of 25 stations for precipitation, temperature and other climate
data were identified (Fig. 12).
The 4 WBAN stations (excluding Sterling, VA) are the ones that have all
the data required by the Chesapeake Bay HSPF model. The other co-op stations
may also have useful data for precipitation and temperature.
Patuxent
Nutrient Budgets (back
to table of contents)
For the past several years efforts ave been underway to develop
updated nutrient budgets for the Patuxent River estuary. These efforts
follow completion of an N and P budget published by Boynton et al. (1995)
based largely on information available for the late 1980’s. Results
from that effort indicated the following:
- the Patuxent was moderately loaded with both N and
P relative to other estuarine systems
- both point and diffuse sources were important
- direct atmospheric deposition of N was less important
- burial and denitrification were
important N losses
- burial in sediments was the primary loss term for
P
export to the Chesapeake Bay was small for N and almost
zero for P
As part of the Patuxent TMDL project, these budgets are
being updated. Using a nutrient budget framework, this re-evaluation uses
the following new information: (1) the 1985-2000 time series of input
data which contains both wet, normal and dry years; (2) measurements of
long-term nutrient burial (based on 210Pb profiling) and denitrification
(based on the new N2 /Ar method) in subtidal sediments; (3) an evaluation
of N and P losses due to burial and denitrification in the tidal marshes
developed by Merrill (1999); (4) a box model of the tidal Patuxent that
enables estimates of nutrient transport at key portions of the system
(Hagy et al. 2000); (5) availability of several watershed models estimating
diffuse source nutrient inputs; (6) a system-wide “experiment”
of attempted N reductions, concluded with the completion of biological
nitrogen reduction (BNR) technology at all of the major sewage treatment
plants in the basin, which allows us to see where and how much loads have
been reduced.
This activity is still underway, and here we only present a summary of
TN loads and a preliminary examination of diffuse source load estimates
from three different approaches. Average annual TN inputs from all sources
were organized for the period 1985-2000 and the years having the highest
and lowest overall TN loading rates were extracted and results depicted
in Fig. 13. The loads and estimated
transport from the middle to lower estuary and from the estuary to Chesapeake
Bay are shown for the years with the lowest (1991) and highest (1996)
loads. The fall line load was compiled by USGS measurements of flow and
nutrient concentration. This approach uses statistical modeling and includes
point, diffuse and direct atmospheric deposition of TN to the river surface.
The loads to the middle estuary included direct atmospheric deposition
to estuarine surface waters, estimated septic inputs, point source inputs,
and all other diffuse source inputs as estimated from the Chesapeake Bay
Program land use model. Inputs to the lower basin were the same as for
the middle basin except that there were no significant point source inputs.
Transport from the middle to the lower basin and from the lower basin
to Chesapeake Bay was estimated using the box model referenced above.
There was just over a factor of two difference between the lowest and
highest load years on record. This represents a very substantial difference,
considering that nutrient management goals for Chesapeake systems generally
aim for smaller reductions than indicated here for inter-annual variability.
The loads at the fall line, middle and lower basins were all higher during
the wet year of 1996, but especially so in the middle basin. These increased
loads occurred after full implementation of BNR at the major sewage treatment
plants indicating that diffuse sources are of key importance, even in
a basin typically thought to be point source dominated. This finding is
further supported by the observation that the lowest loads occurred in
1991, prior to full implementation of BNR technology.
Losses within the estuary were significant in this new budget. Of the
nitrogen that entered the tidal Patuxent, a relatively small proportion
(21-23%) was transported out of the estuary into Chesapeake Bay, with
little difference between wet and dry years. However, there were very
significant N losses during transit from the head of tide (river km 75,
see Fig. 1) to the beginning of the mesohaline estuary (river km 40).
During the dry year about 34% of all N entering the estuary upstream of
the mesohaline zone was removed, while in the wet year 45% of all inputs
upstream of the mesohaline zone were removed. These losses appear to be
related to denitrification and long-term burial of N in both sub-tidal
and tidal marsh sediments. Direct measurements of these losses have yet
to be summarized, but preliminary estimates suggest that measured rates
are sufficient to account for these large estimated losses in the budget.
Diffuse
Source Models (back
to table of contents)
One of the key issues related to the preparation of a TMDL concerns accurate
estimation of nutrient loading rates. For some of the sources typical
of estuarine situations, estimates are either quite accurate (e.g. point
source discharges) or are relatively small (e.g., direct atmospheric deposition
to surface waters ) and therefore not centrally important. In cases where
most of the drainage basin is above the head of tide, excellent estimates
can be developed from high frequency flow and concentration measurements
(e.g., USGS data sets). However, in those cases where a significant portion
of the estuarine drainage basin is below the head of tide, other methods
need to be utilized to estimate nutrient inputs. One of the most common
methods is to use a watershed model to compute loads.
In the case of the Patuxent, about 60% of the drainage basin is located
below the head of tide. Hence there is an important need for modeling
nutrient inputs from these tidally influenced, primarily coastal plain
portions of the basin. As is the case with numerous other variables of
environmental interest, the Patuxent has a relatively long history of
diffuse source modeling. Models have been developed by the Chesapeake
Bay Program (HSPF model), University of Maryland scientists (Costanza
et al. 2001, spatially explicit, mechanistic model), USGS (statistical
models of load at the head of tide), and researchers from the Smithsonian
Environmental Research Center (a large collection of small watershed monitoring
with statistical modeling of loads). Several other models have been initiated
in the past, but current status of those modeling activities or results
are not known at this time.
To compare the load estimates from different investigations, TN loads
for the head of tide (e.g., Bowie, MD) from three programs are summarized
on a seasonal basis in Fig. 14.
There appears to be general correspondence among all three estimates in
terms of seasonal pattern with highest loads associated with winter or
spring and smallest loads associated with summer or fall. The coherence
between the empirical USGS load estimate and the CBP model results is
tighter than with the Costanza et al. (2002) results. In general, the
Costanza model estimates higher loads than either of the other models.
While on the one hand the differences between these approaches raises
a certain amount of concern, we plan to gain a more comprehensive understanding
of processes regulating diffuse source inputs to this estuarine system
by examining these approaches more closely in this project.
Miscellaneous
(back to table of contents)
A research project with nine scientists at four locations generates additional
research activities as well as administrative overhead. Brief summaries
of some of the more pertinent ones are provided below.
An additional activity was created when a coalition of land owners from
Island Creek in the Patuxent approached us concerning water quality problems.
They are concerned about increased boat traffic (especially jet skis and
ski boats associated with a waterfront bar) and the effects on shoreline
erosion and water clarity in Island Creek. With some ideas from us, the
landowners applied for and received WRAS funds from DNR to investigate
this issue. This offshoot of our main project will get underway this summer.
There is also collaboration between this project and one headed by Dr.
Tom Jordan at the Smithsonian Environmental Research Center (SERC). A
letter of endorsement was given to Jordan for his proposal to NOAA, which
addressed nutrient inputs to the Patuxent and which also included funds
for workshops to compare various sources of data, as in Fig.
14 above. This effort was recently funded, and we anticipate considerable
collaborative efforts between the two projects over the next two years.
Funds within our existing budget were internally rearranged to create
a postdoctoral research position. After an extensive search, Dr. Michael
Williams of the Ecosystem Center at Woods Hole was hired as of January,
2003. Michael commutes to Horn Point Lab for a week each month, and will
be in residence full-time in June (after his daughter completes first
grade). His primary activities are validation of the watershed and estuarine
models.
To maintain communication between project participants, we have held quarterly
project meetings. We have met both in person at meetings of opportunity
and on the UM Interactive Video Network (IVN) on the following dates:
Nov. 6, 2001 (ERF meeting); Dec. 3, 2001 (CBP Modeling Subcommittee meeting);
Dec. 17, 2001 (IVN); Jan. 23, 2002 (IVN); May 15, 2002 (IVN); Aug. 29,
2002 (IVN); and Dec. 20, 2002 (IVN). The next project meeting is planned
for Mar. 24, 2003 on IVN. Summaries of all meeting have been previously
provided to MDE, along with brief monthly activity reports through the
end of 2002.
References
(back to table of contents)
Boynton, W. R., J. H. Garber, R. Summers, and W. M. Kemp. 1995. Inputs,
transformations, and transport of nitrogen and phosphorus in Chesapeake
Bay and selected tributaries. Estuaries 18: 285-314
Costanza, R,. A. Voinov, R. Boumans, T. Maxwell, F. Villa, L.Wainger,
and H. Voinov. 2002. Integrated ecological economic modeling of the Patuxent
River watershed, Maryland. Ecological Monographs 72: 203-231
Cronin, W. B. and D. W. Pritchard. 1975. Additional statistics on the
dimensions of the Chesapeake Bay and its tributaries: cross-section widths
and segment volumes per meter depth. CBI Spec. Rep. 42 Johns Hopkins University,
475 pps.
Hagy, J. D., W. R. Boynton, and L. P. Sanford. 2000. Estimation of net
physical transport and hydraulic residence times for a coastal plain estuary
using box models. Estuaries 23: 328-340
Merrill, J. Z. 1999. Tidal freshwater marshes as nutrient sinks: particulate
nutrient burial and denitrification. PhD thesis, University of Maryland,
342 pps.
|