Title: Hypoxia in Narragansett Bay Workshop Oct 2006
1Hypoxia in Narragansett BayWorkshop Oct 2006
- Modeling
- In the Narragansett Bay CHRP Project
Dan Codiga, Jim Kremer, Mark Brush, Chris
Kincaid, Deanna Bergondo
2Does the word Model have meaning?
- Hydrodynamic
- Ecological
- Research vs Applied
- Prognostic vs Diagnostic
- Heuristic, Theoretical, Conceptual, Empirical,
Statistical, Probabilistic, Numerical, Analytic - Idealized/Process-Oriented vs Realistic
- Kinematic vs Dynamic
- Forecast vs hindcast
3(No Transcript)
4CHRP Program Goals (selected excerpts from RFP)
- Predictive/modeling tools for decision makers
- Models that predict susceptibility to hypoxia
- Better understanding and parameterizations
- Transferability of results across systems
- Data to calibrate and verify models
- ? Following two presentations
5Our approaches
- Hybrid Ecological-Hydrodynamic Modeling
- Ecological model simple
- Few processes, few parameters
- Parameters that can be constrained by
measurements - Few spatial domains (20), as appropriate to
measurements available - Net exchanges between spatial domains from
hydrodynamic model - Hydrodynamic model full physics and forcing of
ROMS - realistic configuration forced by observed
winds, rivers, tides, surface fluxes - Applied across entire Bay, and beyond, at high
resolution - Passive tracers used to determine net exchanges
between larger domains of ecological model - Empirical-Statistical Modeling
- Input-output relations, emphasis on empirical fit
more than mechanisms - Development of indices for stratification,
hypoxia susceptibility - Learn from hindcasts, ultimately apply toward
forecasting
6Heuristic models in research iterative
failure learning
Processes
Conceptual Model
Formulations
Runs that fall short
Parameter values
7But for management models Heuristic goal
less impt Accurate even if not precise
Well constrained coefs Simple (?) (at least
understandable)_____________________________
? Research models
825-30 state vars 70-110 params
- A paradox --
- Realism many parameters
- weakly constrained
- limited data to corroborate
- i.e. Over-parameterized
- (many ways to get similar results)
- . Accuracy is unknown.
- (often unknowable)
9An alternative approach? 4 state variables,
5 processes
Processes of the model
(excluding macroalgae...)
Temp, Light, Boundary Conditions Chl, N, P,
Salinity
O2 coupled stoichiometrically
Productivity
Physics Surface layer - - - - - - - - - Deep
layer - - - - - - - - - Bottom sediment
mixing flushing
Atmospheric
Flux to bottom
Photic zone heterotrophy
deposition
.
N
Sediment
Land-use
organics
Benthic heterotrophy
N P
Denitri- fication
.
10 Corroboration Strength in numbers
Shallow test sites (MA, RI, CT)
11Deep test sites (MA, RI, CT, VA, MD)
Narragansett Bay
Chesapeake Bay
Long Island Sound
12Hydrodynamic Model
Equations Momentum balance x y directions ?u
v??u fv ?f Fu Du ?t
?x ?v v??v fu ?f Fv Dv ?t
?y Potential temperature
and salinity ?T v??T FT DT ?t
?S v ??S FS DS ?t
The equation of state r r (T, S, P)
Vertical momentum ?f - r g ?z
ro Continuity equation ?u ?v ?w 0 ?x
?y ?z
Initial Conditions Forcing Conditions
ROMS Model
Regional Ocean Modeling System
Output
13Hydrodynamic Model
Grid Resolution 100 m Grid Size 1024 x
512 Vertical Layers 20 River Flow USGS Winds
NCDC Tidal Forcing ADCIRC
Open Boundary
14This project Mid-Bay focus
Narragansett Bay Commission Providence Seekonk
Rivers
Mt. Hope Bay circulation/exchange /mixing study.
ADCP, tide gauges (Deleo, 2001)
Extent of counter
Summer, 07 4 month deployment (Outflow pathways)
Bay-RIS exchange study (98-02)
15This project Mid-Bay focus
Narragansett Bay Commission Providence Seekonk
Rivers
Mt. Hope Bay circulation/exchange /mixing study.
ADCP, tide gauges (Deleo, 2001)
Extent of counter
Summer, 08 Deep return flow processes
Bay-RIS exchange study (98-02)
16Model-Data Comparison
Salinity - Phillipsdale
Model
Salinity (ppt)
Data
Time (days)
17Model-Data Comparison
18Hybrid Driving Ecological model with
Hydrodynamic Model
Lookup Table of Daily Exchanges (k)
dP1/dt P1(G-R) - k1,2P1V1 k2,1P2V2 ...
19Modeling Exchange Between Ecological Model
Domains
20Passive Tracer Experiment
21Passive Tracer Experiment
22Passive Tracer Experiment
23Long-term AimsHybrid Ecological-Physical Model
- Increased spatial resolution of ecology approach
TMDL applicability - Scenario evaluation
- Nutrient load changes
- Climatic changes
- Alternative to mechanistic coupled
hydrodynamic/ecological modeling
24Empirical/Statistical ModelingOverall Goals
- Data-orientedcomplements Hybrid less
mechanistic - Synthesize DO variability
- Spatial (Large-scale CTD towed body)
- Temporal (Fixed-site buoys)
- Develop indices
- Stratification
- Hypoxia vulnerability
- First Hindcasts to understand relationship
between forcing (physical and biological) and DO
responses - Long-term Predictive capability for forecasting
and scenario evaluation
25- Candidate predictors for DO
- Biological
- Chlorophyll
- Temperature solar input
- Nutrient inputs (Rivers, WWTF, Estuarine
exchange) - Others
- Physical
- River runoff, WWTF water transports
- Tidal range cubed (energy available for mixing)
- Windspeed cubed (energy available for mixing)
- Others (Wind direction Precip Surface heat flux)
26(No Transcript)
27Strategy start simple develop method
- Start with Bullock Reach timeseries
- 5 yrs at fixed single point (no spatial
information) - Investigate stratification (not DO-- yet)
- Target variable strat sigt(deep)
sigt(shallow) - Include 3 candidate predictor variables
- River runoff (sum over 5 rivers)
- Tidal range cubed (energy available for mixing)
- Windspeed cubed (energy available for mixing)
282001
29Visually apparent features
- Stratification reacts to events in each of
- River inputs
- Winds
- Tidal stage
- Stratification events appear to be
- Triggered irregularly by each process
- Lagged by varying amounts from each process
30Low-pass and subsample to 12 hrsCompare
techniques
- Multiple Linear Regression (MLR)
- No lags
- Optimal lags determined individually
- Static Neural Network
- No lags
- Lags from MLR analysis
- coming soon Dynamic Neural Network
- Varying lags
- Multiple interacting inputs
31Stratification Dst kg m-3
Multiple Linear Regression No lags r20.42
(River alone 0.36)
MLR with lags River 2 days Wind 1 day Tide
3.5 days r20.51 (River alone 0.48)
32Static Neural Net No lags r20.55 (River alone
0.41)
Static Neural Net Lags from MLR r20.59 (River
alone 0.52)
33Advantages/Disadvantagesof Neural Networks
- Advantages
- Nonlinear, can achieve better accuracy
- Excels with multiple interacting predictors
- Dynamic NN input delays capture lags
- Varying lags from multiple interacting inputs
- Transferable conveniently applied to other/new
data - Easy to use (surprise!!)
- Main disadvantage
- opaque black-box can be difficult to interpret
ameliorated by complementary linear analysis,
sensitivity studies, isolating/combining
predictors
34Next steps
- Stratification
- Consider additional predictors
- Surface heat flux precipitation WWTF volume
flux - Different sites (North Prudence, etc)
- Treat spatially-averaged regions
- Apply similar approach to DO
- Finish gathering forcing function data
- Chl solar inputs WWTF nutrients
- Corroborate Hybrid Ecological-Hydrodynamic Model