Table of Contents
Physical Science Team (PST) Cover Letter to PMC (click here)
PST Review of the Implementation of FATHOM in Florida Bay (click here)
FATHOM Modeling Group’s Response to PST
Review (click here)
PST Review of the Implementation of RMA10 in
Florida Bay (click
here)
***********************************
TO: Interagency PMC
for the Florida Bay and Adjacent Marine Systems Science Program
FROM: Tom
Lee, Peter Ortner, and DawnMarie Welcher, Physical Science Team
SUBJECT: Review
of ongoing Florida Bay physical modeling projects
Attached please find the final reports the
PST developed, at the specific request of the PMC, on FATHOM and RMA10. Initially, both modeling groups (FATHOM and
RMA10) were asked to furnish the PST copies of their final reports. While the FATHOM group had to generate this
report, the ACoE folks furnished one that was dated last October. Regardless, PST members were asked to
provide substantive reviews of one or the other of these reports or both.
A number of PST members responded and summaries of their responses were created and, along with the individual reviews, sent back to all the PST members to obtain any additional input. The revised summaries and the individual reviews were then sent to both modeling groups to give them an opportunity, should they wish to avail themselves, to provide the PST a 1-2 page response or comment we could additionally attach to the summary report that would be given to the PMC. The FATHOM group did so (also attached), but the WESLAB group (ACoE) did not.
***********************************
Review of
the Implementation of FATHOM in Florida Bay
for the Florida Bay and Adjacent Marine
Systems Science Program
by Members of the Physical Science Team
Background
Everglades National Park (ENP) has supported
the development of FATHOM, a dynamic mass-balance model, for use in modeling
salinity in Florida Bay. Results of
this work are described in the technical report, " FATHOM: Model Description
and Initial Application to Florida Bay," (undated). Copies of these reports were furnished to
the Physical Science Team by the authors of the model (B.J Cosby, W.K. Nuttle
and J.W. Fourqurean) for review. This
report describes the model at an intermediate stage of development. A final product is expected by the end of
1999.
The primary purpose of this review is to
assist the FATHOM modeling group by providing critical comments and insights of
physical scientists with experience working in Florida Bay. Our intent is to provide a critique of model
formulation and application in order to evaluate the model's potential and to
suggest model improvements that will eventually lead toward scientifically
defensible model predictions. These comments also provide the modelers with an
opportunity to inform the PMC of their recent efforts towards these goals. The comments and the response from the
modeling team will be compiled into a single report that will form part of
the PMC's documentation of the overall
modeling program in Florida Bay.
The main body of this report consists of a
brief report on the status and description of the FATHOM model and a summary of comments
received from six reviewers followed by the complete text of each individual
reviewer.
Status of FATHOM (February 1999)
Sponsor: Everglades National Park
Function: FATHOM is primarily a mass-balance, i.e. box, model that is currently capable of simulating the spatial and temporal variation in salinity in inner Florida Bay.
Status: Implementation and testing of the mass-balance solution code is complete. Results of an investigation into the influence of variation in net freshwater supply on salinity are reported in a manuscript submitted for publication. Results of sensitivity studies with FATHOM are reported in the Year-1 report submitted to Everglades National Park.
Objectives: Work for year-2 will improve the calibration of the model, with attention to the period 1989 through 1995, and perform simulations to estimate residence times in each of the basins in Florida Bay. The PMC is requesting simulated salinity output from a series of management scenarios for discussion at the next Florida Bay Science Conference.
Description
of FATHOM
FATHOM simulates variation in salinity in
Florida Bay based on hydrologic inputs for the period 1965 through 1995. Refinement and verification of the solution
routine were completed in FY98.
Preliminary simulation results and comparison to Robblee’s historical salinity
database were delivered to ENP in September.
The annual report for FY98 includes results of a model sensitivity
study. A manuscript describing these
results is in preparation for publication, tentatively planned for the Journal
of Coastal Research.
Work planned for FY99 will investigate the
Bay-wide response of salinity to different hydrologic scenarios, which will
reflect the anticipated changes in Everglades hydrology due to
restoration. Further work also will
include refinements in the bathymetry and salinity input data. The bathymetry GIS data have been analyzed
to provide an objective measure of bank widths. This information parameterizes the hydraulic solution for tidal
mixing in the model. For work completed
in FY98, this information was input manually.
Also, FATHOM will be modified to accept a time-varying salinity boundary
condition along the Keys and the western boundary with the Gulf of Mexico. Results of work in FY98 suggest that this is
a significant source of variation in salinity in Florida Bay.
Summary of PST comments
Several reviewers feel the FATHOM approach
could serve as a useful tool to examine long-term (greater than monthly)
changes in Florida Bay salinity and mass (water) balance. However, one reviewer thinks it’s too early
to evaluate the model performance and another adopts a "wait and see"
attitude. Two reviewers question the
long-term usefulness of the model and another has serious concerns and lists a
number of fundamental weaknesses in the model formulation which are believed to
limit FATHOM as a predictive salinity or hydrodynamic model for Florida
Bay. This reviewer also feels that the
FATHOM Report should include a set of clearly stated objectives, proven model
capabilities, and performance evaluation criteria.
All reviewers feel that the model is
insufficiently calibrated and not ready for examining management alternatives.
Two reviewers feel that the physics in the model are too simplified for a
complex region such as Florida Bay. A
deficiency expressed by several reviewers is the models inability to simulate
the effect of direct wind stress on the water surface for water movement and
sea level set-up or set-down. There is
also concern expressed regarding the return factor formulation as being
inappropriate and fundamentally flawed. More explanation of the model time step
is needed. Several reviewers feel that the Key West sea levels cannot be
applied to all boundaries to account for storm effects. The situation seems to be further
exacerbated by the use of monthly values. There is concern that Manning’s n
should be further adjusted (fudged) in order to calibrate the model. In some areas of Florida Bay, salinity is
controlled mainly by runoff, rainfall, and evaporation. Advection in these areas may be of little
importance because they are basically basins isolated by banks. In these areas FATHOM can be expected to
simulate salinity better than in areas where advection is important. Some
reviewers are concerned that the procedure for specifying boundary tides is not
clear. Also a minimum to maximum tidal variation is known as a tidal range not
an amplitude as presented in the report.
Comments of Reviewer 1
I have reviewed the draft report of the
application of the FATHOM model to Florida Bay and provide herewith some
comments. In general I feel that this approach is quite useful for examining
the long-term salinity data set for the region. In particular, the approach
should be very helpful to park managers in considering different methods of
water delivery to Florida Bay. I am impressed by some of the initial results
and the rapid progress that has been made. It is obvious that considerable
effort has gone into the project. But as pointed out in the report the model
application is far from complete and there is much more to do following the
“next steps” listed on page 57, which seem quite appropriate.
At this stage it is really too early to judge
the model in terms of how well it reproduces the salinity magnitude and
temporal and spatial variability in different parts of the Bay for there are
many refinements yet to be made both to the data and the model. But one can say
that results look promising and that the approach should work. One particular
strength of the approach is that it provides a method for estimating exchange
rates between different basins of the Bay and their residence times. However to
do this requires that the model be calibrated to the available data, which has
not been done yet and that all the available data be utilized, both within the
basins and at the boundaries to reproduce observed variations. This also has not
been completed. In this regard it is important to introduce a realistic time
varying salinity for the Gulf boundary. Recent observations from Un. Miami and
FIU large-scale surveys shows that holding the Gulf salinity constant at 35 psu
is inappropriate and will tend to damp variations in the western regions of the
Bay.
The model when calibrated will essentially
provide a “best fit” to the available data from which exchange rates can be
estimated. This should help account for the poor spatial resolution and the
assumption that each basin is well-mixed and that there may be some
“short-circuiting” of exchange between basins. I think the residence times
computed in this manner will be considerably longer than those given in the
report.
Specific Comments:
Tidal amplitudes listed in Table 2.2 appear
to be too small for the northern Gulf and too large for Biscayne Bay. The table
lists “tidal amplitudes” as (cm, maximum to minimum). By definition this is not
a tidal amplitude but rather the tidal range. As such the tidal range for the
northern Gulf should be closer to 2.0 m for spring tides. Increasing the tidal
forcing at this boundary should help increase the salinity variability there to
be closer to observed. Also the mean tidal range at the Biscayne Bay boundary (which is really the southern part of
Barnes Sound) should be closer to 16 cm as I remember.
Page 32 mentions the use of Key West sea
level data to establish the monthly pattern of water surface elevations. These
data should be useful for the seasonal patterns of temperature and shifts in
the Bermuda high, but probably not appropriate for the passage of storms since
sea level response at Key West to wind forcing will be different than along the
western boundary of Florida Bay due to topographic constraints and changing
coastline orientations. A careful analysis of sea level responses at Naples,
Key West, Flamingo and Long Key CMAN is needed to find the proper sea level
responses along the western boundary. New bottom pressure data collected near
the western boundary of the Bay as part of our circulation study could also be
used for this purpose.
On page 33 a 0.1 value for Mannings friction
coefficient seems high. Maybe this
coefficient should be adjusted in the process of obtaining a best fit to the data.
The model-simulated salinity is compared to
observed in Table 2.3. The east region is almost 5 psu less than observed. This
is a large difference and suggests that part of the problem may be due to poor
estimation of the fresh water input both in magnitude and distribution. Test
cases should be run for times when the fresh water input was well resolved. In
the Central, South and West regions the mean salinity was larger than observed,
suggesting again that realistic salinity values are needed at the western
boundary and these should be allowed to vary as observed. The fact that
observed standard deviations were significantly greater in the South, Central
and West regions than simulated also indicates that holding the Gulf salinity
constant is not appropriate. The only region where the simulated standard
deviations were close to observations was in the East, where one assumes that
variability is more related to P-E+R and less to exchange with adjacent areas.
The finding that the average salinity for all regions agree with the simulation
should not be considered a significant result for it is merely a non-physical
result from combining the overestimation of salinity in some regions with the
underestimations in other regions. The actual comparison was not particularly
strong for any region. The differences in simulated values and observed appears
to be explainable in terms of data input and how the model was used and is not
the result of a random process as suggested in the report.
The salinity comparisons for the 9-year
period and the 31-year period are very similar, which gives one confidence in
the data and use thereof in the model.
Page 48 and 49: I’m not sure that the
comparison of simulated mean velocities through the Atlantic boundary (Keys
tidal channels) with Ned Smiths results is meaningful since Smiths observations
and the simulations were for different time periods.
The turnover times given in Table 2.5 are
much shorter than I would have expected. Is this because it is assumed that the
tidal flux is completely mixed within each basin? And what fraction of the
tidal flux is “new water” i.e., water that had not entered the basin
previously.
The conclusion section suggests that the
model is now ready to examine sensitivity to different management scenarios. I
would agree if the use was restricted to investigation of model sensitivity
only. However, I would caution the investigators that the refinements to data
and the model listed on page 57 should be made before extending much effort on
the management scenarios. Also the model should be adjusted until a “best-fit”
with the data is obtained and then this “calibrated model” could be used to
estimate exchange rates and residence times for the various basins.
The example of a management scenario given on
page 56 indicates that the simulations are quite responsive to freshwater
delivery. The results listed in Table 2.6 also suggest that the exchange rates
are weak between the basins for there was little dilution of salinity in the
South and Central basins compared to the large variation in the East.
Comments of Reviewer 2
I got only to page 35 (the description of the
salinity simulations). That seemed like a logical cut-off point, because the
simulations should not be looked at too critically before the model has been
calibrated carefully.
The authors provide a pretty good critique of
the model themselves, offering both pros and cons. I like that kind of candor.
In general, the report is well written, and it reads easily.
My thoughts will be primarily in the form of
questions, because in reading the text quickly I may have missed some of the
points that you did in fact make. I
don't expect answers, but you may want to take these questions into account in
revising the text at some point.
Am I correct in thinking that by ignoring
direct wind stress effects the model will not allow for the wind-induced set-up
or set-down of water level within the bay?
As I read it, wind forcing can raise or lower water level along the
boundaries, but the only response within the bay would be for water to drain in
(if coastal sea level is rising) or drain out.
The additional set-up and set-down of water levels within the bay
(relative to the boundary) could involve substantial amounts of water, and thus
reduce flushing times. I guess this is
one of the trade-offs mentioned early in the report.
I'm sure you noticed that a citation is
missing in the middle of page 9 of section 1.
The return factor you describe on page 1-9
makes sense when water is moving back and forth with the ebb and flood of the
tide between two sub-basins. But what
happens in the near-tideless northeast corner of the bay, where wind forcing
makes the ebb and flood of the tide highly asymmetrical, and occasionally the
flow persists in one direction for days at a time? It seems that the return factor would be inhibiting the inflow of
dissolved and suspended material when little if any of the arriving water is,
in fact, "returning." Over
these longer time scales, mixing within the giving and receiving basins may be
virtually complete. A time-dependent
return factor may be another compromise.
I don't find a definition for the capital phi
symbol in the right hand side of equation 1.2.
The time step noted at the middle of page
1-10 varies from 1 to 10 minutes. I
would be interested to know what guides the choice of, e.g., 1 minute, as
opposed to 10 minutes.
Of course, the tidal forcing section was of
particular interest. Again, I realize the model is in an early stage, but with
so much water level data from the bay itself, and with such complex tidal
patterns within the bay, what is the rationale for using Miami tides? Perhaps by chance Miami's harmonic constants
are a close approximation of those found at some place(s) in the bay (I was
once surprised to learn that Hampton Roads tides are a very close approximation
of tides along the seaward side of Lee Stocking Island, Exuma Cays, Bahamas),
but it would certainly make a better impression on the reader if you used the
information that is available from the study area itself. The spring-neap sequence you refer to is
only part of the tide story. The
tropic-equatorial sequence arising from the interaction of diurnal tidal constituents
are important along the west side of the bay.
Parenthetically, if Table 2.2 includes "maximum to minimum"
tide levels, these would be tidal ranges, rather than "amplitudes."
On the middle of page 31, there is a
typo. The exchanges with Biscayne Bay
would occur in the northEAST corner of Florida Bay. Also, if you want to define a better boundary condition here,
harmonic constants are available from Jewfish Creek that would be better than
an extrapolation of Miami tides.
Finally, on line 3 of page 31 you refer to a
"baroclinic" set-up of water levels.
I believe this should be "barotropic," unless you are in fact
referring to an increase in water level due to changes in density (which I
suspect would be pretty small).
I wish I had more time to go through
this. I have the impression this is
definitely a viable alternative to RMA-10.
The real proof will come with the validation of the model, of course,
and I look forward to seeing how that turns out.
Comments of Reviewer 3
Although not stated as an explicit objective,
the reported work appears to be an attempt at determining the effects on
Florida Bay (FB) of restoring runoff towards historical levels, pages 1-1 to
1-2. It is never clearly stated what
implications of such restoration are considered. At least three parameter spaces could be directly affected:
salinity patterns and timing (P1), transport magnitudes and timing (P2), and
water quality other than salinity (P3).
The authors seem to focus on P1 although residence times relevant to P2
are also considered.
This review will consider 4 broad areas given
major attention in the report: modeling approach, FATHOM, model application and
data synthesis, and future directions.
I. MODELING APPROACH
Authors expound on their view of a number of
modeling issues, page 1-2 and scattered in the text. A caution is noted against "complex, many-parameter,
physically-based models", which leads them to the selection of a
simplified, physics-based model using an empirical (Manning) approach for
simulating friction flow.
Unfortunately authors's views are subjective,
controversial, and vague, for example (my comments in parens) 1: simplified
methods are strictly empirical (In fact, there are many simplified
physics-based methods); 2: complex
models are difficult and slow to implement (depends on what type of model, who
does the modeling, and how much effort is put into it. Accuracy of results also come into
play); 3: "complex,
many-parameter, physically-based models" have more model parameters than
can be constrained by data (I don't know of any such model, can the authors
give an example?. The models in wide use have very few if any free parameters,
bottom friction is generally not a free parameter, it is a function of bottom
roughness, which is a physical characteristic just like depth which must be
known or estimated); 4: There is no cause and effect in empirical models (So
this would be true for the empirical Manning eq.? Then what causal behavior can
be expected in a hybrid model which uses Manning eq.?).
My sense is that authors feel hard pressed to
justify their approach, but don't have really strong arguments and therefore
are compelled to suggest rather vague or incoherent reasons. They don't even follow their own advice as
explained above and below.
It would have been useful to have: A clear
set of objectives; A list of model capabilities and major incapabilities
(sorry); A set of performance or evaluation criteria.
II. FATHOM
The authors claims that FATHOM is rapid,
efficient, and accurate, while including responses to rainfall, runoff,
evaporation, sea level, tides and wind (P 1-2 last paragraph). Some of these claims are not supported by
the reported work. Before addressing
this in detail some general comments about the modeling approach may be
helpful.
The model is very similar to the link-node
type models developed 25 - 30 years ago.
The features are very similar except FATHOM uses a simplified and
specialized momentum equation for banks.
FATHOM therefore also suffers from the same known problems of this type
of model: subjectivity in schematization, inability to handle local wind
stress, excessive numerical diffusion, and lack of convergence to 2-D
flow. In addition, FATHOM employs an
untested dynamic formulation for flow over banks.
Subjectivity: The basin/bank schematization is difficult
to do objectively and scientifically. In other words establishing a clear set
of guidelines for how to carry out the schematization is difficult (impossible)
and certainly was not done here. (Think about it: how do you specify where a
bank begins and ends, account for it's plan curvature, account for its cross
section, etc. etc.). Therefore the results are not reproducible and the model
in effect has an infinite number of degrees of freedom. The schematization then becomes an expression
of the authors' opinion about the behavior of the system, which is good if
their opinion is right, but bad if they haven't covered all bases. Because of the many degrees of freedom,
calibration may not lead to a model useful for prediction or even diagnosis.
Wind: The effects of
wind are presumed, P 1-8, to be local mixing (no advection) and propagation of
surges (included by forcing model boundaries).
The assumption that the effect of local wind stress is only to mix
waters within a basin conflicts with ENP water level gage data and HBOI current
meter work that shows there definitely is locally wind driven transport between
basins. Just think winter storms! The issue of wind forced response is
important because it is probably the dominant response in a large part of the
bay where tides are tiny.
Numerical diffusion: Because the model instantaneously mixes a
basin every time step, non-physical (usually excessive) diffusion is
introduced. Even in the unlikely event
that FB is diffusion dominated, maybe giving this type of model a chance, it is
near impossible to adjust model diffusion to physical diffusion. One way of visualizing the problem is to
consider the spreading from a source in the middle of a basin under a steady
flow situation. In one (6 min) time step the source material is brought into
all (because of the return factor even into upstream) neighboring basins in
significant quantities. The next time step brings the material into the
neighbor basins' neighbors and so on.
Even setting the return factor
to zero causes excessive diffusion.
Synoptic observations of salinity in the bay
show that the dominant transport mechanism in most areas is advection. Without proper modeling of advective and
diffusive processes, the FATHOM model cannot describe water quality parameters
correctly, and the shorter the time scales the worse the predictions, because
shorter time scales are almost certainly dominated by advection whereas
diffusion becomes less important.
Convergence: The basin/bank
(node/link) models do not converge towards two-dimensional flow patterns. It is not possible to refine these models to
get better approximations to true 2-D flow because of the 1-D flow in links or
over banks. Once a bank is put in
place, the flow must occur in a certain direction across it. This is in conflict with natural flows which
can and do vary in direction. It is
therefore not possible to assure that the chosen schematization adequately
represents the (known) physical characteristics of the system.
Dynamic formulation: Bank flow
formulation is highly simplified and idealized. No validation is provided which could help demonstrate how well
the model simulates a known problem.
For example [1.11] always assumes a 'transition' loss equal to the
velocity head, however, for gradual transitions, that is excessive. The Manning
formula was obtained for uniform flow in open channels. It may not describe the boundary layer
processes for transitional flows over submerged banks. Supercritical flow is assumed to occur, page
1-13 middle paragraph. Yet there is not
mention of the ensuing hydraulic jump.
Have these hydraulic phenomena been observed in FB? The dynamic formula used in FATHOM are
concocted by the authors without proper theoretical or empirical support and no
references.
After raising these serious issues, one might
well ask how some of the FATHOM results can be in reasonable agreement with
data. First, it is not surprising that
water levels can be simulated with some accuracy. Water levels result from flow
and conservation of mass. By adjusting
flows water levels can usually be fit to data.
However, it has been shown time and again that fitting water levels does
not mean that the corresponding flows represent actual flows. Second, because averaging the salinity
observations in space and time is similar to introducing lots of diffusion in
the salinity fields, this will favor comparison with a highly diffusive
model. However, more importantly, in
some areas the primary forcings of salinity variations are derived from runoff,
rainfall, and evaporation. In areas
where these forcings dominate the salinity response, FATHOM can be expected to
predict salinity reasonably well. This would be areas receiving runoff, while
they receive significant runoff! And in
isolated basins. This does not mean
that FATHOM predicts transports well, in fact it cannot! Finally, the intimate familiarity and
knowledge about the system collectively possessed by the authors likely also
played an important role. This
reinforces the desirability of involving modelers with a good familiarity of
the system to be modeled.
Additional detailed comments:
The dynamics of the flow in basins is
completely ignored. The distribution of
flow to and from the various banks cannot be correct. In some basins (large and
shallow) this will be a big error, in others (small and deep) it may be less
important.
With a return factor, FR (eq. [1.1]),
different from zero, there is unrealistic upstream advection. With FR = 0.33, the model grossly violates
the transportive property (a quantity can only be advected downstream), Roache
(1972).
What is the meaning of phi in [1.2].
There does not appear to be a reproducible
way of characterizing bank dimensions, locations, and alignments. The more removed from reality the harder it
is to calibrate the model and to use it for prediction.
The model contains adjustment (fudge) factors
en masse: bank characteristics, friction coefficients, and return factors. This is the type of model the authors
admonished against in their discussion of modeling approach.
There does not appear to be any channels in
the model (Appendix A), only banks.
Channels between banks have been shown to be very important (HBOI data)
in conveying water between basins.
Channels should not be modeled with the flow formula presented in the
report.
III. MODEL APPLICATION AND DATA SYNTHESIS
Section 2 of the report examines an initial
application of the FATHOM model. The
model evaluation covers a 31-year period and includes other very desirable
elements, such as the error analysis.
The "Background" section, page 21,
contains some general observations about estuaries, but contains a curious
statement: "The patterns of salinity in estuaries result from a dynamic
steady state ....". Salinities are
usually dynamic, rarely at steady state, and never dynamic and steady state at
the same time. The authors define FB as
a seasonally hypersaline estuary. While
this may be appropriate for some areas of FB it may not properly describe other
areas, and a general classification of the bay as a whole is probably not
meaningful.
Page 24.
FATHOM simulated hourly values ... Other places in the report, much
smaller time steps (6 min) are mentioned.
In preparation for model application and data
comparison, historic salinity, rainfall, runoff, and evaporation were
assembled, page 25 to 30. This is
potentially very useful information.
Though leaving plenty of room for improvements, this effort is
courageous. It seems the salinity data
from the ENP MMN are missing from the compilation? A better, more complete
explanation of methods used to derive the temporal and spatial distribution of
runoff and rainfall would be helpful. Although they used a spatially weighted
average rainfall in model simulation, it appears a spatially variable rainfall
distribution was available. With proper
rationale these results could well serve as a baseline against which future
improved estimates could be gauged. The
evaporation estimates could probably be significantly improved. The SFWMD has made estimates for their model
simulations based on a more sophisticated analysis.
Effects of wind cannot be included by
specifying water elevations at boundaries page 32. Moreover, monthly sea levels from Key West cannot capture wind
effects since they are most energetic at the shorter, 3-10 day, period
band. Was the same sea level applied at
all boundaries? That would distort most
remote wind forcing.
The specification of boundary tides, page 31
and Table 2.2 was not clear. Showing
the generating formula would be helpful.
Why is there such a big difference between first high tide in north and
south Gulf? The variation in amplitude in the Atlantic seems far greater than
in reality. (Range would be a better
term than amplitude). Are the channels
between the keys modeled as banks?
Is 1 year model spin-up sufficient? page 34
top.
The early onset of the wet season in the
Central region is possibly due to the excessive diffusion in the model, page
37.
The simulated tidal amplitudes appear
reasonable, however, local wind effects are clearly missing.
I had difficulty understanding the comparison
of FB-to-Atlantic channel flows. The
text, last parag. page 48, indicates that the model calculated velocity for
Channel 2 was used to normalize all results, but, figure 2.23 seem to indicate
data was normalized by data average and model was normalized by model average
for Channel 2. How did actual magnitudes compare? Also what period was used for averaging data and model?
Are the results, inflow north and outflow
south, shown in figure 2.23 possibly due to the specified (questionable)
variation in tidal amplitude in the Atlantic (table 2.2) ? Were any model adjustments made to achieve
the results shown in fig. 2.23? How did
unaveraged model results compare to data?
What were the magnitudes of Chan 2 velocities, model and data? Comparing velocity is a poor choice, total
discharge is the critical measure.
The water budget, exchange patterns, and
residual circulation are solely model results and lack validation against
data. As previously pointed out the
model has severe shortcomings and lacks direct wind forcing. Model input, for example sea level and
tides, have similar problems.
The conclusion that "the use of the
model to examine system behavior is justified", page 55 1st par., is
inappropriate because of lack of model validation and other fundamental
shortcomings detailed above.
The inferred residual circulation and
simulated turnover times for the interior of FB are entirely unvalidated. A critical forcing, local wind stress on the
surface is completely missing from model and analysis.
The problems of recirculation and different
flow paths described in the bottom par. of page 55 are exactly the reasons that
FATHOM has no chance of providing the needed answers. An approach properly
accounting for advective and diffusive fluxes is required. To conclude that using FATHOM is the best
solution for estimating retention times sadly ignores everything about how this
system works. One should not be misled
by salinity simulations for a few very isolated basins where advection and
diffusion are of little importance most of the time. That is not the case for
the majority of the FB system.
In summary FATHOM may address P1 in some
isolated basins or small basins receiving a significant amount of runoff. P2 is not achievable because of the limiting
(bank-basin) model formulation. P3 is
impossible because of missing P2 and unrealistic numerical diffusion.
IV.
FUTURE DIRECTIONS
The authors clearly want to continue their
FATHOM modeling, and even indicate so in the report title. A list of items to do is stated on page
57. The refinement items 1 to 7 are
desirable although the indicated methods may not be the best. These all involve input data to any FB model
and are not specific to FATHOM. Work in
these areas by qualified researchers is strongly endorsed.
The structural changes suggested for FATHOM
are futile. How do you split one large
basin into three smaller subbasins hydrodynamically? How do you decide on bank
alignment and characteristics? This is
the trip down the non-convergent path mentioned earlier, refinements do not
lead toward real behavior. Even if an
objective reproducible method for deriving a "bank width" could be
derived, the dynamics of FATHOM are too far removed from reality and would
require endless model adjustment. Worse
predictive capability is compromised.
The agreement of FATHOM computed salinities with observations can undoubtedly be improved by tuning the multitude of model parameters, however, fundamental discrepancies will remain, predictive ability is unattainable, and understanding of the underlying processes will be lost. Predictive capabilities will be limited and will not include circulation (P2) or water quality (P3). This situation is likely to get worse as better data become available for forcing and validation.
Roache, P.J., 1972: Computational Fluid
Dynamics. Hermosa, Albuquerque,
New Mexico.
Wang, John D. July 1, 1999
Comments of Reviewer 4
This is a fascinating approach to develop a
tool for examining long-term (> monthly) changes in Florida Bay salinity and
mass (water) balance. It extends a
strictly empirical approach to include some of the physically-based processes
that the authors feel are paramount in Florida Bay *primarily frictional effects over the shallow banks between
basins that are assumed well-mixed). In
general, I have two points to make:
I.
The authors have done a good job of assembling historical data bases for
forcing and evaluating (particularly salinity) the model; and
II. While the results of these initial
calculations are encouraging, I think the strict structural approach of FATHOM
(based on the physiography of the Bay) and the subsequent simplified physical
assumptions (*) are going to limit its usefulness in the long run.
Specific comments follow, in order of
importance:
1.
The treatment of the "wind-driven" forcing (i.e. low frequency
forcing at the boundary as a water level boundary condition) is weak for
several reasons. It allows for no
direct wind-driven movement of water in Florida Bay (a large omission) and as
it is formulated it does not account
for the important synoptic scale wind forcing
(regional space scale and the 2-5 day time scale).
The use of the Key West water level record
does not capture the wind-driven set up and set down (what the authors refer to
as stochastic passages of storm systems [month-to-month "noise"] on
p32 - please note that storm system time scales are not monthly; they are in
the synoptic
[2-5 day] band) that a truly coastal station
(e.g. Naples) does, although Key West does represent the seasonal steric
effects as well as the long term trend.
What are meant by
"wind-tides"? Diurnal sea
breeze?
2.
The tidal forcing is oversimplified, as acknowledged. Both the NOAA (POM) and ACoE (ADCIRC)
regional models have demonstrated that they can provide more detailed tidal
boundary conditions that represent the mixed tide on the Gulf side and the more
pure semidiurnal tide on the NE Atlantic side.
3.
It's worth noting in the salinity results that the model underestimates
the variance by about 50%, except in the eastern basin. I would suggest that
this may, in part, be due to the lack of true
synoptic wind forcing (#1 above).
I think the model does better in the east, as
stated, because this basin is closer to the sources of runoff and further
removed from the larger uncertainties of the west and south boundary
conditions. It's possible too that the
constraints of well-mixed basins in the west and south are
even less valid there where the basins are
larger. For example, on p37 (Figure
2.13) the authors state that the model reproduces the hypersaline seasonal
(average) signal but in fact the model misses this spring event in both
amplitude and phase.
4. On
p56, where the runoff is doubled, it's curious that the model is so insensitive
to this away from the eastern basin, whereas the historical observed records of
runoff (Figure 2.6) and salinity (Figure 2.4) suggest that the central area
salinity does respond to changes in runoff.
5. On
p22 it is stated that the lack of density stratification limits the
tidally-driven and baroclinic exchanges.
Please note that tidally-driven exchanges in FL Bay are barotropic and
that horizontal density stratification in FL Bay can be significant.
6.
Where's basins #'s 1-3? Is this
why, if there are 41 basins, I see #'s 43, 43, and 44?
Comments of Reviewer 5
Florida Bay is morphologically unique when
compared to other eastern US estuaries in that is a broad lagoon with diffuse
freshwater inputs, has a low input:volume ratio, a high evaporation:input
ratio, and is partitioned by shallow carbonate mud banks which restrict and
direct flows in the complex basin mosaic.
The FATHOM model for Florida Bay is, in
essence, a simple mass balance box model.
What is different from other estuarine box models (e.g. Officer) is the
necessary use of Manning's equation to determine flow over the
banks/shoals. This is a relatively
simple extrapolation and obviates the need for a full-blown hydrodynamic model.
I feel the most important benefit of the
FATHOM model (as with other ecological models) is its heuristic
performance. This first run of the
model is more valuable in pointing out what important data are missing/not
available than in providing useable results and predictions. I think the authors and users should remain
aware of this inherent learning attribute in their future efforts to modify the
model and in their recommendations for future monitoring/measurement efforts.
Assumptions:
The model relies on certain assumptions on
the physical nature of the estuary which I think could be better defined:
1. The endmember salinities of the Gulf of
Mexico and Atlantic Ocean are held constant in the model at 35 (psu). We know this is not the case from our data
from 4 years of quarterly monitoring.
For the 9 stations in the gulf closest to the model boundary: mean
salinity was 35, SD = 1.79, min.=26, max.= 39, n = 126. For the offshore/reef
Atlantic stations from Tavernier Creek to Seven Mile Bridge: mean salinity =
35.9, SD = 0.05, min. = 33.7, max. = 36.9, n = 145. For the inshore
Atlantic stations: mean salinity = 35.9, SD = 0.10 (double the offshore
stations), min. = 32.5, max. = 38.4 (hypersaline), n = 141. I believe this is one of the most important
reasons why the model salinities, especially in the western zone, were so far
off the mark. I think it would behoove
the authors to run a sensitivity analysis by varying endmember salinities in an
effort to determine the magnitude of this effect.
2. Estimated evaporation rate (unmeasured)
has a seasonal component but no interannual variability. The sensitivity analysis showed that the
eastern zone was most sensitive to fluctuations in evaporation rate.
Results:
1. Differences between observed and predicted
salinity are the sum of all errors (precipitation, evaporation, runoff,
groundwater, and flux) and cannot be partitioned. Sensitivity analysis showed that salinity was equally sensitive
to precipitation, evaporation, and Manning's n.
2. Turnover time calculations are confounded
by back flow of solute into the basins which had previously been transported
out. In other words, the water at the
boundary of each basin is not independent (autocorrelated) of the basin itself.
3. One of the most useful results of the
model is the calculation of annual net water movement (Fig. 2.25). This net
flux of higher salinity water from the west mixing with freshwater inputs in
the northeast and resultant flow out of the estuary towards the southwest is
borne out empirically with the data from the SERC monitoring program.
Comments of Reviewer 6
With respect to the simulation of the
temporary inter-basin exchange of water across the mud banks using a
specialized momentum equation, there is now modeling technology that can handle
this problem. The layer(s) of fluid are considered to exist everywhere but the
layer thickness collapses to zero when/where a mud bank emerges. The model equations remain valid at all grid
points but there is no mass flux (V * thickness = 0) in regions where the layer
becomes "massless". Later, the layers
re-inflate with water and the cross-bank transport is automatically
reactivated. (Think of squishing Jello between layers of Saran Wrap using your
fist.)
This technology is employed in the Miami
ocean model (MICOM) at UM-RSMAS to simulate isopycnic layer
"outcropping" at the ocean surface. The key is to use a
high-performance advection scheme with low numerical diffusion (like FCT or
Smolarkiewicz' MPDATA) which can accurately predict where the mud banks should
emerge. (These numerical algorithms also resolve steep gradients and hydraulic
jumps.) This allows the natural, realistic flow to dynamically determine
when/where the mud banks should surface or submerge. The formulation avoids the
problems caused by FATHOM's static mud bank structure (cited by another
reviewer).
***********************************
Response to:
"Review of the Implementation of FATHOM in Florida Bay" by the
Members of the Physical Science Team of the Florida Bay and Adjacent Marine
Systems Science Program
Prepared by B.J. Cosby and J.W. Fourqurean
The PST review comments reflect a great deal
of effort on the part of the reviewers and have been very useful to our group
in planning the next steps in the development and application of FATHOM. In
addition to these reviews, we have had a number of individual conversations
with members of the PST regarding the FATHOM model. Many of the comments have
been incisive and have led to changes in the way FATHOM is being applied to
Florida Bay. It is also clear from the comments that shortcomings in the data
used to drive the model (and to evaluate the model) can be a cause for as much
concern/discussion as the model itself, and to that end we have undertaken to
obtain better data where needed. Some
(but not all) of these problems with the model and data were recognized by us
and pointed out when we prepared the report. We are grateful to the members of
the PST who provided ideas toward resolving the problems we knew about as well
as the new problems that were raised. This review process has apparently worked
- you spoke and we're trying our best to hear. From some of the comments,
however, it is evident that some confusion still exists about the model, its
conceptual basis, its intended purposes, and its mathematical structure. In
some concepts and details, the reviewers have come away from reading the same
report with distinctly different impressions of the model, what it is capable
of, and what it has been asked to do so far.
Dr. Lee has asked us to respond to the
reviews in 1-2 pages, so we will not address here any specific comments of the
reviewers. Rather we will: 1) address briefly what we perceive as the need to
get FATHOM "on the street" so that its capabilities and shortcomings
can be understood and evaluated by potential users, and we can begin working
toward a resolution (if possible) of remaining difficulties; and 2) attempt to
provide a brief explanation of the approach and philosophy we adopted in
developing the model, and what we consider its role to be in ecological and
physical investigations in Florida Bay.
With respect to the specific comments and criticisms presented in the
reviews, they have been noted and we are working to resolve them.
1) Getting FATHOM "on the street"
There seems to be, at the moment, a problem
arising from the fact that FATHOM is currently known only in broad outline.
Prior to the presentation to the PST (and the follow-up report) there had been
little formal presentation of the model and its capabilities or shortcomings.
This resulted from the fact that the funds available for the project have been
spent mostly on the development of the model and the data bases needed to
implement and evaluate the model. The model (and its appropriate uses in
Florida Bay) must be judged on merit (what does it do? and how well does it do
it?), and to that end, we are trying to get as much exposure as is possible for
the model and its current capabilities and shortcomings.
The way forward for the FATHOM model is to
get some simulation results out in front of the appropriate audience so that
the strengths and weaknesses of the model can be evaluated. We are firmly of
the opinion that as much is learned (if not more) when a model
"breaks" as when it "succeeds". The initial feedback we have received has been gratifying, and we
appreciate the time and effort expended on discussion, comments, and
suggestions. The model can run on PC's, and we have made a demonstration
version of the model available on CD-ROM along with the simulation results from
the PMC scenario workshop held in August. We are preparing another report
similar to the one reviewed by the PST and are planning two more papers for
submission to peer-reviewed journals.
We (in all likelihood) will be calling on some of you in the near future
as these activities move forward. Further suggestions on how we can get folks
involved in using and criticizing the model are welcome.
While it may be the opinion of some that
FATHOM is "not ready for prime time", our opinion is that the current
state of the model represents a significant step forward in the development of
a tool for examining physical and ecological processes in Florida Bay. It may
be that current uses are primarily heuristic (as several reviewers point out)
and that the reliability of the model's predictive capabilities has yet to be
established. However, if the model is not "exercised", discussed, and
dissected, that predictive reliability can never be determined.
2) FATHOM - Approach and Philosophy
There seems to be a general misconception
that FATHOM was developed as an alternative to
the RMA-10 model or to hydrodynamic models in general. In fact, that is
not the case. FATHOM was intended from the outset as an aggregated mass-balance
model designed to utilize the basin/bank arrangement of the bay to define
regions within the bay that would be useful in ecological studies requiring
information about the movements of water and solutes between individual basins.
The comparison of FATHOM simulations to observed salinity was seen as a way to
partially "validate" the solute transport results from the model.
The aggregated nature of the model was
adopted so that the model could be more easily used in heuristic studies -
"What if...? experiments". It was also envisioned that the model, once
adequately calibrated and validated, could also be used to provide
quantitative estimates of physical properties important to ecological studies
(basin residence times, salinity variations, water levels, etc.). Again, the need for a model that could be
readily implemented to derive these estimates was an a priori consideration.
Finally, we intended that the model be configured such that it could be used to
examine potential effects of changes in freshwater inputs to the bay and thus
have heuristic and perhaps predictive utility in management contexts as well as
scientific studies.
The expected applications of the model do not
require simulation of continuous velocity fields within the bay (rather exchanges
between basins is perceived to be the critical process) and so we did not adopt
the hydrodynamic approach. To criticize the FATHOM model, therefore, because it
is "empirical" and not hydrodynamic misses the point concerning: 1)
its intended uses; and 2) its ability to reproduce important behavior in the
Bay. FATHOM is a rather successful refinement of a linked-node, mass-balance
box model, and not a failed hydrodynamic flow and transport model. The
tide-driven flows across banks in FATHOM provide estimates of exchange fluxes between well-mixed basins in a way
that is tightly constrained by the available bathymetric data for the bay. The simulation outputs from FATHOM have much
potential utility for ecological and physical investigations, providing model reliability and confidence in the model can be established.
Whether or not this latter condition can be satisfied is the focus of large
part of our anticipated future activities.
Our approach to building models is to add model complexity in a step-wise fashion and compare model results to data at each step. Using this paradigm, we are better able to discriminate between those factors that influence variation in bay characteristics (by their contribution to the incremental improvement in model simulation success) and those factors that appear to have little effect. We are well aware that the FATHOM model potentially has a large number of degrees of freedom that can be exploited through an optimization scheme to maximize the model fit to observed data. However to do so would defeat one of our favorite tactics - "breaking" the model to learn something new. At each step of improvement in model structure or input data, we have learned something about the relative importance of each change on overall model behavior. Therefore, we have so far left the model “uncalibrated” (by purposefully assuming, for example, uniform bottom roughness, when in fact we could assign a different roughness to each bank and possibly improve the simulation results relative to observations). If reliable observations of variables or processes in the Bay are not currently available, we apply "Occam's razor" and choose the simplest assumption for the missing information, and then evaluate how well or poorly the model performs. We consider that this modelling approach makes the best use of the available data in Florida Bay, especially where existing data may not currently be adequate to support the calibration and validation of a hydrodynamic flow and transport model (for instance there is very little data on tide-driven currents within the bay to use for flow calibration).
As the emphasis of FATHOM applications moves
to predictive exercises, it will become increasingly important to have the
"best" input data to drive the model and we will implement a
calibration of the model to maximize model success in simulating salinity.
While the reviewer's are correct in asserting that there is more to be done, it
is also correct that progress has been made. We look forward to continuing this dialogue with the PST and other
interested parties.
***********************************
Review of the Implementation of RMA10 in Florida Bay
for the Florida Bay and Adjacent Marine
Systems Science Program
by
members of the Physical Science Team
The Waterways Experiment Station (WES), U.S.
Army Corps of Engineers (ACoE) has calibrated its RMA10 hydrodynamic model for
use in Florida Bay. Results of this
work are described in the technical report, “Field and Model Studies in Support
of the Evaluation of Impacts of the C-111 Canal on Regional Water Resource,
South Florida, Part IV: Florida Bay Hydrodynamic Modeling,” dated October
1998. The Jacksonville District Office
of the ACoE furnished copies of this report to the PMC in December 1998. The PMC subsequently forwarded the report to
its Physical Science Team for review. A
later version of this report which addresses some of the issues raised here may
be available, however, this later report was not available for review.
The primary purpose of this review is to
assist the WES modeling team by providing critical comments and insights of
physical scientists with experience working in Florida Bay. Our intent is to provide a critique of model
formulation and application in order to evaluate the model's potential and to
suggest model improvements that will eventually lead toward scientifically
defensible model predictions. These comments also provide ACoE with an
opportunity to inform the PMC of their recent efforts towards these goals.
This report consists of a brief report on the
status of the RMA10 model, a condensed description of the model, a summary of
comments received from five reviewers, and the full text of each individual
review.
Status of RMA10 (February 1999)
Sponsor: U.S. Army Corps of Engineers
Function: RMA10 is a 3-D hydrodynamic model that has been implemented for Florida Bay in 2-D format for Florida Bay and the near-shore region of the Florida Shelf north to the 10,000 Islands and through the Keys onto the reef tract down to the lower Keys. RMA10 provides detailed information on tides and currents in response to tides on the boundaries, wind, rainfall, runoff, and evaporation. RMA10 also has the capability of simulating salinity.
Status: Initial development and implementation of RMA10 is complete. The draft final report provides results of salinity simulations for three runoff scenarios.
Objectives: The PMC is requesting simulated salinity output from a series of management scenarios for discussion at the next Florida Bay Science Conference.
Description
of RMA10
(Note: This section is based on source material
from the WES web site and the project report, "Field and Model Studies in
Support of the Evaluation of Impacts of the C-111 Canal on Regional Water
Resources, South Florida", dated October 1998)
The two-dimensional model includes dynamic
coupling between the salinity and hydrodynamic fields. Model development and verification were
accomplished using extensive data sets taken by WES and data supplied by other
agencies and organizations, such as the National Park Service, USGS, Harbor
Branch Oceanographic Institute, NOAA, and the South Florida Water Management
District.
The model verification period is that period for
which WES data were taken (March 1996 to April 1997). The model was initialized using August 1995 salinity
distributions from the USGS surveys.
Freshwater inflows from USGS gauged flows into northeast portion of the
Bay and net freshwater (rainfall - evaporation) values derived from three WES
meteorological stations were included in the model. Verification (calibration) of the model was attained by adjusting
bottom roughness in the model to provide the best match to tides, currents and
salinity variation (in that order) over a relatively short period of time (20
days in September 1996 and February 1997).
The first model calculations examined the
salinity of the Bay under varying wet season freshwater inflows. Three flow scenarios were considered. The base flow is representative of the flow
in a wet year. The NSM (Natural System
Model) flow represents a conjectured flow through the everglades before human
intervention. Alternative D represents
a proposal for future water usage. The
calculations for Florida Bay based on these three flow scenarios used tides,
net rainfall, and winds for the year 1996.
Calculations were made for the three-month period, September through
November. Results of the three inflow
calculations show that diverting more freshwater into the northeastern part of
the Bay through the Taylor Slough system has these effects:
· Lowers the salinity of the northeast
corner of the Bay in approximate inverse proportion to the increase in
freshwater flows;
· Pushes these salinity changes out to mid
Bay;
· Creates a sharper gradient of salinity in
the mid-Bay area.
The model provides flow fields for a WES
water quality study of Florida Bay. To
enable the model hydrodynamics to be used in conjunction with a water quality
model with different spatial resolution, a unique projection technique was
developed through a partnership between WES researchers and the University of
Texas at Austin. This projection tool enables two-dimensional hydrodynamic
results to be applied to water quality modeling without the two efforts having
the same computational mesh, so that the ideal resolution for each portion of
the modeling effort can be used.
Summary of PST comments
All five reviewers recommend that further
work is needed before the RMA10 model can be accepted as calibrated and
verified. The reviewers recommend using
quantitative measures of model performance in future calibration and
verification work. Each expressed
frustration with trying to make judgements about model performance from plotted
output. Performing harmonic analysis on
the modeled water levels near measurement sites would provide information
directly comparable to known values of tidal amplitude and phase angles for 4-5
of the major tidal constituents for the interior of Florida Bay and at selected
sites along the boundary. More attention
needs to be paid to how well the model predicts the response of the Bay to
sub-tidal (long-term) forcing from winds. This could be evaluated for subtidal
sea level changes and water flows at the boundaries and between interior basins
at those sites where measurements have been made. Further work on calibration/verification is needed to build
confidence in the capability of the RMA10 model to predict water levels, flows,
and transports in Florida Bay and at the boundaries.
The reviewers raised fundamental issues
related to the boundary conditions and transport processes in the
interior. Errors in the tide
predictions of the ocean model, used to provide the ocean boundary condition,
should be estimated and, when known, their propagation into the RMA10 domain
should be investigated. Reviewer 1
raises the issue of whether the coastal ocean model captures the net water
slope and distribution of residual flows that are know to occur across the
boundaries of the Bay. Wind is expected to drive circulation and mixing in the
interior of the Bay; therefore, it is inappropriate to assume a uniform wind
field over the Bay without a sensitivity study to determine the errors inherent
in this assumption. Reviewer 4 points out
that it is not clear if the sea level boundary conditions at the open ocean
boundaries were provided by the NOAA POM model or by ACoE’s ADCIRC model.
Reviewers were confused by several aspects of
the simulation runs that were performed to investigate the sensitivity of the
model to changing freshwater flow. In
particular, why were simulations initialized in August 1995 but run with
inflows from a different year? Several
reviewers pointed out that magnitudes of freshwater inflows appear to be much
higher than observed (by a factor of 10 or more) and, as a result, modeled
salinity values are too low in the northeast Bay and off Shark River. The
general pattern of salinity seemed reasonable but the values were largely
different from observations due to the uncertainty of the freshwater inflows. Also,
the results of the simulations were difficult to judge without knowing how long
the effects of initial conditions persist to influence the calculated salinity
field. Tests should be run to determine the spin-up time for the model and
effect of initial conditions.
Comments of Reviewer 1
The model verification section was of primary
interest to me, and I must say I would like to see this approached with tables
rather than with figures. First, model
verification is documented with analog plots for the most part (Plates
1-30). Model simulation results are
plotted along with observations from the same time and place. I would prefer to see a comparison of
harmonic constants. Given the
importance of tidal forcing in the Bay, results of the 20-day simulations could
be subjected to the same harmonic analyses used for observations to quantify
the amplitudes and phase angles of the 4-5 major tidal constituents. Actually, only 15 days are needed for this,
so the data are probably already available.
Second, the model's ability to reproduce
hourly transport along the western boundary is of less importance than the
model's ability to reproduce the net inflow and outflow over longer time
scales. Two of the most important
features of the regional circulation are the inflow through the northern and
central parts of the open western boundary and the outflow through the southern
part of the boundary. Did the model
reproduce these features? From Plates
5-7 we have no idea. Another major
feature of the regional circulation is the long-term net outflow through the
major tidal channels of the Middle Keys.
Plate 9 contains data from Long Key Channel, but how do these hourly
water levels translate into the long-term average 250-300 m3/s net
outflow that has been calculated for this channel? Similarly, we have values for half-tidal-cycle inflows and
outflows for all of the major tidal constituents and for all of the major
channels. How do model results match
up?
A really convincing validation would come
from a series of co-amplitude and co-phase charts for the major tidal
constituents. Is the model reproducing
the negligible tidal conditions in the interior of the Bay? I don't think anyone is expecting the model
results to mirror the observations exactly, but I have no clear picture of how
good the simulations are. If the model
is to be used as a management tool, shouldn't we know how good it is?
The report (page 9, bottom) notes that a
user-friendly graphical user environment has been developed to facilitate the
post-processing of model outputs. If
so, then a closer look at the output should be a relatively easy matter (easy
for me to say, anyway).
From the beginning, I have wondered about the
decision to use ADCIRC model output for the boundary conditions that will serve
as input to the RMA10 model. Have the
ADCIRC-derived boundary conditions been tested against observations to quantify
the errors that are being fed into RMA10?
Also, do the ADCIRC boundary conditions
include a net transport into Florida Bay?
This would seem to be very important if RMA10 is going to force water
out through the tidal channels on the other side of the Bay.
Comments of Reviewer 2
The objectives are succinctly stated, however,
no performance criteria are given by which the results can be judged.
The authors' objective (a) can be said to have
been achieved, though model calibration and validation is inadequate. Additional quantitative model verification
and sensitivity analyses are needed before an assessment of the model performance
can be made.
Fundamental issues related to transport
processes, their type and parameterization in the model are not addressed as
detailed below.
Chapter 4 of the report attempts to address the
second objective, (b). Without proper
model verification little confidence can be attached to the results. The freshwater inflows seem inappropriately
high and, certainly, are far beyond the range of values used previously with
the Model. The results are shown for
two times on Nov 28 and 29.
Curiously, these times were chosen only 34 hours
apart and the patterns are expectedly virtually identical. The statements on p. 23, line 10,
"Figures 45 and 45 are given to illustrate the fact that the salinities,
particularly in the northwest corner of the Bay, are changing very little. This reflects the nearly constant inflows
into the northeast part of the system for this time period" demonstrate a
poor understanding of how this system behaves, even accepting that "northwest"
is a typo for "northeast".
The comparison between base flow and NSM flows
shows that increased inflow of freshwater from Taylor Slough results in lower
salinities in eastern Florida Bay. Hardly a surprising result. However, the reliability of the computed
values depends on transport and diffusion processes, whose magnitudes in the
model are unknown and untested.
In summary, the best one can say is that the
reported work raises many questions but that answers are hard to come by. Because of the lack of proper calibration
and verification, as detailed below, confidence in the reported model or its
ability to predict transport and mixing in Florida Bay is low. In this respect, the data collected for
model calibration and validation seems seriously deficient in the interior of
Florida Bay. To test the viability and reliability of the model, a set of
performance criteria based on a set of relevant objectives and associated data
must be developed and the model should be evaluated against these criteria.
Specific questions or comments:
P 2. The
process in which model is iteratively adjusted to fit data is usually called
calibration not verification.
P 3, L 11 bot.
These are crucial statements that one would expect support for. Data in terms of stratification, and either
model results or analysis in terms of circulation impacts.
No support is provided.
P 6, L 4 bot.
Circulation in the interior of Florida Bay is controlled by the
wind. Yet all the comparisons focus on
tides. What about wind?
P 9 bot.
It's interesting that the authors reference only two of the four model
studies performed simultaneously. Were
they not aware of the two other ones?
P 10, Figs 5 and 6. There are very abrupt changes in element sizes within the
grid. Does this require extra numerical
smoothing to control numerical high wave number noise?
P 10 bot. What is the numerical diffusion? How
much eddy viscosity was applied.
P 11.
Using a large-scale tidal model to compute boundary conditions is not a
certain solution to a difficult problem.
In the present situation, the proximity of the offshore boundaries to
inlets, the complex bottom topography, and the inability of the large-scale
model to properly represent these features can cause mismatch. Therefore, careful validation must be
made. The comparisons, e.g. channel
tidal flows, show that there is a serious mismatch.
P 11, Figs 11 to 18 show prescribed tides with
no comparison to actual tides. There
is, therefore, no way to verify the model forcing directly.
P 11. A
high degree of correlation does not justify making the wind uniform. This must be validated or sensitivity should
be established. There are important
phase shifts within the domain in the peak energy band for wind. After all, the major wind systems move
through the area.
P 11. At
what height was the wind sensor? Were
wind speeds corrected to standard height?
P 12 bot.
Was salinity zero or was it assumed to be zero? What is the significance of averaging over
three days. If salinity is not zero,
not all discharge would be freshwater.
P 12 - 13.
The explanation of how freshwater discharges were obtained was
incomprehensible to me. How were the
estimates mentioned on P. 12, 14 bot. determined? What are these estimates?
Was a 72-hour average used to drive model? Where was the Taylor River gage
located? What is the justification for
using that to determine discharges at all points?
P 13.
Since net rainfall is highly uncertain, a sensitivity analysis could
have been made.
P 14.
The choice of initial salinity condition, August 95, for the so called
verification calculations is peculiar since the model simulation starts
September 96. The issue of how long the
model needs to be run for results to become independent of the initial
conditions is
never addressed.
To test the effect of initial salinity
conditions, the model could have been run with different initialization. It would be very valuable to ascertain that
the model can reach a realistic salinity field after sufficient time if started
from arbitrary salinity; however, that may take longer runs than have been made
so far.
P 15.
The term verification is used to describe model calibration and
validation. Most investigators
distinguish between calibration in which model parameters are adjusted and
verification (or validation) where model performance is tested for one or more
independent data sets not used in the calibration procedure.
P 17.
The comparison of water levels would be more helpful if it addressed at
least three areas: boundary condition errors, tidal fluctuations, and subtidal
variations. The reported comparison
consisting of time series plots is purely qualitative and very difficult to
evaluate. The water surface comparisons
reveals large tidal discrepancies, which are bound to cause major discrepancies
in flows that primarily depend on the surface slopes. Equally important is the
model's apparent failure to describe the low frequency variations in water
level which are apparent in the figures.
It would have been useful to plot low passed water elevation
comparisons, especially since these likely would be caused by wind, and wind
probably is a major forcing in eastern Florida Bay.
P 18-19.
Velocities. Again the
comparisons are strictly qualitative.
Terms such as "make sense", "compares well", and
"agreement is good" are used subjectively. How well does the model
describe harmonic constants, direction of subtidal flow, magnitude of subtidal
flow? The importance of subtidal flows
seems totally lost as the authors' focus exclusively on tides. Plates 31 to 36 seem to indicate very
different magnitudes and phases between data and model. The text for Plates 32 and 33 does not seem
to agree with what is shown in the figures.
Since tidal flows in the channels result from differences in tides
inside Florida Bay and in the Atlantic, the discrepancies in channel flows
indicate that model calculated tides in the Bay and/or offshore tides have
problems. Similarly, since subtidal water level variations differ significantly
from observed values, model calculated wind driven flow is probably unreliable.
P 19-20.
Salinity. A simulation was made
to determine salinity using estimated freshwater inflows. No mention is made of the magnitude and
distribution of these inflow estimates.
The model uses salinity from August 1995 as initial condition, but the
model is started Sept 14, 1996. Results are only shown at points near the
Atlantic where the salinity remains near 35 to 36 no matter what, or for a
point in Joe Bay which remains fresh at all times. No results are shown for the interior of Florida Bay. It is interesting that no freshwater was
routed into Joe Bay, an almost enclosed bay, yet the salinity there is near
zero. How does Joe Bay become fresh, or
is the entire region fresh?
A simulation was also made for Feb. with an 80%
reduction in inflow. The reason for
this reduction is not explained. What
was used for initial conditions? The
resulting salinities, only one station shown, appear to be completely different
from observed.
Finally, a discussion of longer calculations made
with inflows based on USGS data is given without any results shown. The discussion tests the reader's
imaginative power. A couple of figures
would have been helpful even though only qualitative results are given.
The nebulous conclusions of the model
verification part of the report aptly represent the results.
Chapter 5.
Summary and Conclusions
Conclusions are too subjective and
qualitative. Some are trivial or so
general that no new information is obtained, (e.g. "The results of the
calculations were that increasing the flows into the northeast part of the
Florida Bay model will result in a freshening of the northeast to mid Bay
areas.").
What is wind blow down? Why was it not mentioned before?
Comments of Reviewer 3
1.
Model Characteristics:
2D vertical average Model -
RMA10, finite element includes evaporation, precipitation, salinity,
temperature and sediment transports including erosion and deposition.
Evaporation: 27.5" to
30.0" 3/96 - 4/97 with small spatial variation.
Has a Graphical User
Environment (GUE) allows user to visualize model results.
Bathymetry based on nautical
charts (NOAA #11451 example).
Boundary Conditions (B. C.)
- ocean salinity, tide, wind, net rainfall and freshwater inflow.
Tidal - uses global ocean
model (Luettich et al., 1991 and 1992) & POM (Aikman). These models include subtidal wind effects.
Winds -29 km Eta model and
CMAN stations.
Winds used for B.C.
development are considered as spatially uniform i.e., no variation with
longitude.
B.C. are set for two 20-day
periods Sept. 96 and Feb. 97.
Fresh water inflows are most
problematic, for they consist of complicated inflows from many small creeks
that are not well measured.
Quantitative values of fresh
water inflows were not available for initial verification so estimates were
used for wet (Sept.) and dry (Feb.) seasons.
Compared model data to
intensive WES surveys in 9/96 and 2/97.
Also use USGS fresh water
inflows: weak in summer and fall 1996.
During high fresh water flows the discharge to Florida Bay is
complicated by increased overland sheet flow and multiple ungauged channels.
Taylor River gage showed a
net inflow to Bay and it was used to construct inflows from other points, i.e.
Shark River Slough system. The Shark River data contain water level only, no
flows.
Taylor River flows were
observed near the mouth from 8/96 to 2/97
Total range was 0 to
120 cfs (0 to 3.4 m3/s).
Typical flows ranged
from 10 to 50 cfs (.3 to 1.4 m3/s)
Maximum flows were
approximately 120 cfs (3.4 m3/s) in mid Oct. 1996
Weak flows of 10 to 30
cfs (.3 to 0.84 m3/s) occurred from Dec 96 to Feb. 97
Rainfall and
evaporation are included in the model as a net freshwater input.
Initial salinity set
to USGS salinity map of 8/95.
Verification proceeds from tides to
velocity to salinity. Model is adjusted
to agree with data at each step by adjusting bed roughness.
2. Model comparison with data - Results
generally similar for Sept. 96 and Feb. 97:
Tides
Model tides (sea level) generally greater
than observed for ocean tides.
Model tides were much smaller than
observed on the Bay side of the Keys.
Poor agreement between model and observed
tides in Long Key Channel.
Model tides were generally greater than
observed inside Florida Bay.
Discharges (volume transports)
In interior channels (Jewfish, Dusenburg
Ck. And Grouper Ck.) the model flows led observed flows by approximately
1/2-day.
Whale Harbor Channel: phasing is poor but
magnitude OK.
Channel 5: phasing poor but magnitude OK.
Long Key: phasing poor but magnitude OK.
Salinities
Salinity verification uses two sets of
freshwater inflows (historical and later measured); salinity has 1 to 2 ppt offset after rather brief spin-up period
for gages near Keys passages.
Only interior long-term salinity gage is
at Joe Bay where the model salinity is 10 to 20 ppt less than observed.
Using USGS inflow data and their 8/95
salinity map as initial conditions, the general pattern of the model salinity
field compares reasonably well with the observed. However, major disagreement occurs in the northeast portion of
the Bay and offshore of Shark River where the model salinities are much fresher
than observed. Also in the northwest portion of the Bay near East Cape the
model buldges the 30 psu isohaline toward the west more than observed.
The results indicate that better
freshwater inflow data is needed before quantitative comparisons with
observations can be made.
3. Experimentation:
Three freshwater inflow scenarios derived
from coupled surface-ground water model.
The SFWMD 2X2 coupled surface-ground water model provided head values to
use in the hydrodynamic model. The
Hydrologic model calculates surface flow values a few miles inland from the
coast (due to uncertainties in defining topography in the coastal swamps).
Three scenarios: 1995 (wet year), NSM
(Natural System Model, i.e. before man) and Alt-D (proposed for future water
usage).
For Taylor Slough the NSM fresh water
discharge is about twice the wet year or about equal to Alt-D in the full wet
season.
For Shark River discharge all three
scenarios are about the same.
Observed fresh water discharges in Taylor
Slough range from 0 to 3.4 m3/s
Taylor Slough model discharges
Dry Season: 500 to 1500 cfs (14 to 42 m3/s)
which is about a factor of 10 greater than observed.
Wet Season: 2000 to 4000 cfs (56 to 113 m3/s)
which is a factor of 40 to 80 times observed.
Shark River model discharges
Dry Season: 1000 to 3000 cfs (28 to 85 m3/s)
which is about 5 to 17 times previous estimates.
Wet Season: 6000 to 8000 cfs (170 to 226 m3/s)
which is 30 to 40 times larger than previous estimates.
The model calculated salinity in Florida
Bay based on these three fresh water input scenarios using tides, net rainfall
and winds for the three month period Sept. to Nov 1996. However, the freshwater inflows were
calculated for the year 1995.
The resulting model salinity patterns are
much too fresh in northeast Florida Bay and off Shark River because the modeled
inflows were 10 to 80 times larger than observed.
4. Conclusions:
1) Model
not verified as yet.
2) Model
needs to be compared to a wider range observations and quantify results of comparisons.
3) Need
to use correctly modeled inflows of freshwater or observations.
4) Need
to run the model for same time period as inflows.
5) Need
to compare model salinity values and field patterns to observations.
6) Need
to compare model time series of transports to observations at boundaries
and
in channels connecting interior basins.
7)
Need to compare model sea level to observations for tidal and subtidal
variations.
8) Need to compare model drifter trajectories to
real drifter tracks through western Florida Bay.
9) Need
to compute and quantify differences and errors.
10) Should
use the new high-resolution bathymetry data set for Florida Bay.
Comments of Reviewer 4
This report documents the development and use
of the 2D RMA10 hydrodynamic model for Florida Bay, wherein the
"verified" model system was used to experiment with three different
freshwater scenarios to examine the salinity response in the Bay. In general, I liked the report. It is written well and presented clearly and
I found it to be pretty thorough. It
follows a logical progression from describing the major inputs and the region
(FL Bay prototype), to describing the model and its verification, to describing
the numerical experiments and results.
However, I do have a few comments.
I.
Boundary Conditions
Water level: The tidal boundary conditions
look reasonable (Figures 11-18) although it is not clear which of the two
sources (the NOAA's POM or ACoE's ADCIRC) of model-derived water level boundary
conditions were actually used in the simulations presented. Both of these water level sources also
include wind effects, which can be substantial, especially in winter on the
west FL shelf. Were the effects of the
wind-driven subtidal boundary condition effects evaluated (for example, in
terms of set up and set down or in terms of x-FL Bay pressure differences)?
Wind: A spatially uniform wind field is used
to drive RMA10, based on combining the three WES interior FL Bay weather
stations (Figure 2). Of course these
three stations are well correlated, they're only 10s of kilometers apart! Why weren't other wind sources considered
(e.g.
C-MANs) to look wider afield for
correlations? Assuming the wind over
the west FL shelf part of the RMA10 domain to be the same as inside the bay
could be erroneous and wind-driven responses on the west FL shelf can propagate
into the bay.
Freshwater inputs: Agreed, freshwater inflows
are the single most problematic factor.
II. Verification
Calibration of the bed roughness: The process
of tuning a model using the bed roughness is described but none of these steps
are illustrated for RMA10 applied to Florida Bay? It would be good to see some of this and understand how much
tuning had to be done, vis-a-vis tidal elevations, currents, and salinity.
Tides: The model tides do look pretty good,
testament to the RMA10 model itself but more so to the quality of the tidal
boundary conditions used.
Subtidal water levels: A northerly wind event
is referred to on p18 but no analysis is done.
It is important to examine the subtidal response of the model to
wind-driven events. The tides should be
removed (low pass filtering is fine) from both the observed and simulated water
levels and the resultant records compared.
Velocities: The tidal velocities look pretty
good. Although the authors do not
acknowledge that the N-S component of the model flow at ST1 looks pretty bad
(phase and amplitude) after 9/30.
Discharge: It's frustrating to only have ~1/2
day of observed discharge data to compare with the model discharge, although I
understand what it takes to get such measurements. Indeed, the model seems to lead the observed data by ~1/2
day. Could some of the simulated
synoptic time
scale fluctuations (Figures 32 and 33) be
related to wind? Flow through these
channels is not always to the south, as stated on p19.
Salinity: Why is the 0-40 ppt range used in
all these plots? Doing this doesn't
allow us to see what is happening. The
model doesn't seem to capture events in the observed salinity record at TG1 on
~days 22, 30, and 33. Overall, given
the acknowledged uncertainties in the freshwater
inflows and boundary salinities, a 1-2 ppt
difference is reasonable. However, I
would argue that the model does not appear to parallel the observed salinity
records, even after the spin up. Plot
these data at a proper scale (not 0-40 ppt) and this will become quite evident. I think the 10 ppt offset at JBTS speaks for
itself and the two records have nothing in common.
Bay-wide model salinity patterns look
qualitatively reasonable; however, the actual salinity values can be way off.
In summary:
1.
RMA10 seems well calibrated for the tides, both elevations
and currents;
2.
Subtidal water level fluctuations need to be
examined in both the model results and observed data. There can be considerable barotropic effects in Florida Bay due
to wind-driven set up and set down and due to large-scale wind-driven pressure
differences across the Bay. These might
explain some of the synoptic time scale changes simulated in the channel
discharges (Figures 32 and 33).
3.
The tuning of the model needs to be illustrated and
quantified;
4.
The large uncertainties in fresh water inputs and
salinity boundary conditions lead to large discrepancies between simulated and
observed salinities. Qualitatively, the
large-scale simulated salinity patterns look reasonable.
Comments of Reviewer 5
1.
The reviewers point out that the circulation in the
interior of Florida Bay is controlled by the wind, yet a spatially uniform wind
field is used. Since surface wind stress forcing is so crucial in setting up
water slope and driving currents, especially in shallow water, a wind field
which contains realistic horizontal gradients should be used instead.
Even a simple "regime
catalog" (sea breeze, hurricane, cold front passage) of wind fields
created from analytical textbook equations would be an improvement. The
hurricane or frontal gradient could be moved around (similar to what the Storm
Surge Unit does at the National Hurricane Center) in model sensitivity tests.
Confusion: was a high-resolution wind field,
constructed from CMAN stations and NCEP's 29 km Eta model, used as boundary
conditions? This should provide
adequate horizontal gradients.
2.
Instead of just calibrating the model with real
data, the model could be gradually "nudged" to the real data using a
Newtonian relaxation scheme. This would be simple to implement [one term in the
governing equation for each variable being nudged: dV/dt + ... + alpha*(V-Vobs) = 0, for example],
computationally insignificant, and would help overcome simplifications/deficiencies in the model's
physical parameterizations (bottom friction, etc.) and structure (lack of
channels/gaps).