Jamie Vaudrey, Department of Marine Sciences, UConn CHRP – EPA Workshop Model Skill in Predicting Hypoxia in Narragansett Bay July 7, 2015 Using EcoGEM to Predict Hypoxia

Motivation, Rationale & Objectives As part of a NOAA Coastal Hypoxia Research Program project: Apply a novel, reduced complexity, parsimonious ecological model to predict hypoxia in Narragansett Bay. Implement within a fast running, coarse boxed scheme linked to a fine resolution hydrodynamic model. Simulate responses to nutrient reductions and climate change. Make the model available for direct use by managers.

Highly resolved hydrodynamic model (ROMS) linked to a coarse boxed scheme. figure courtesy of Dave Ullman, URI linking ROMS to EcoGEM

figure courtesy of Dave Ullman, URI Dye initialized in an element at 12am each day, allowed to mix for 24h, fate of dye yields exchange coefficient for the day. Gross Exchange Matrix (GEM) is a 3-d array detailing exchange among all elements for each day. time = 0:00time = 24:00 linking ROMS to EcoGEM

1 / dmany / d Structure of program: boxes are mixed at start of each day biology state variables are formulated as differential equations biology is allowed to integrate for 24 h (Runge-Kutte 3,4 integration scheme)

 O 2 N P N Land-use Atmospheric deposition N P Productivity Temp, Light, Boundary Conditions Chl, N, P, Salinity Phytoplankton.. Physics Surface layer Deep layer Bottom sediment Flux to bottom Photic zone heterotrophy Benthic heterotrophy Denitri- fication O 2 coupled stoichiometrically BZI ƒ(Chl 2d ) ƒ[Chl] ƒ(OM,T) % Processes of the model & basis for formulations: mixing flushing only 17 constants and coefficients Benthic C

TDN TDP DIN DIP organic nutrients inorganic and organic nutrients enter via rivers and groundwater 10% of dissolved organic nutrients are available (labile)* within the estuary, results are compared to inorganic nutrients model coverts organic to inorganic organic nutrients in estuarine waters are largely refractory

organic nutrients Valiela, I., S. Mazzilli, J.L. Bowen, K.D. Kroeger, M.L. Cole, G. Tomasky, and T. Isaji ELM, An Estuarine Nitrogen Loading Model: Formulation and Verification of Predicted Concentrations of Dissolved Inorganic Nitrogen. Water, Air, & Soil Pollution 157(1-4): groundwater: 10% of initial DON is labile atmospheric deposition: 40% of initial DON is labile

Perry, R. and J.M.P. Vaudrey (2015, in prep) Fate of organic N in a small, shallow estuary. technical report, Dept. of Marine Sciences, UConn. organic nutrients

Perry, R. and J.M.P. Vaudrey (2015, in prep) Fate of organic N in a small, shallow estuary. technical report, Dept. of Marine Sciences, UConn. organic nutrients DON refractory in estuarine water 10% lability for DON may be high

forcing conditions Forcing Conditions ROMS forcing functions described by Dave Ullman light from Eppley Labs, Newport, RI wind from T.F. Green Airport for exchange of oxygen with the atmosphere precipitation from T.F. Green Airport for nutrient loads from wet and dry deposition water temperature from ROMS

Boundary Conditions assume salt = 0 ppt in rivers benthic carbon is not transferred among elements N and P = river concentration * flow (m 3 ) phytoplankton = surface buoy data at boundary oxygen = surface buoy data at boundary boundary conditions

N and P concentration from NBC data ( ) or WWTF data – linear interpolation between dates for missing days flow from ROMS model, partitioned based on USGS river flow data

boundary conditions width of arrows indicate relative magnitude of volume input

boundary conditions rivers – from Phillipsdale buoy, surface

Skill Assessment Data Discrepancy Model predicts a box-wide, daily average. Field data are not strictly comparable: buoys – a fixed location, 15 minute data (can get daily average) CTD surveys – many locations, but infrequent (can get box-wide average) nutrient and productivity surveys – few locations, few dates A “perfect skill” is not expected and is highly unlikely. skill

Relative Operating Characteristic (ROC) curve used for evaluating models in many fields compares the “hit rate” to the “false alarm rate” > 0.5 = skilled; 1 = perfect skill evaluates all elements at once – whole model Mason, S. J., and N. E. Graham Conditional probabilities, relative operating characteristics, and relative operating levels. Weather and Forcasting 14: Sheng, Y. P., and T. Kim Skill assessment of an integrated modeling system for shallow coastal and estuarine ecosystems. J. Mar. Syst. 76: hit rate = mod + obs + thresholds every 5.4 ppt, yielding 20 evaluations false alarm rate = mod + obs - salinity 0.97 skill

O 2 buoy 0.89 O 2 CTD 0.80 Phyto buoy 0.69 N survey 0.88 Skill – ROC, whole model > 0.5 = skilled; 1 = perfect skill skill

Skill – by element (just O 2 shown) skill

Water Column Production skill

Water Column Respiration skill

Sediment Rates blue = model green = core data only 25 data points, from 2005 & 2006 model data are from corresponding element on corresponding day data from: Fulweiler, R.W., S.W. Nixon, B.A. Buckley, and S.L. Granger Reversal of the net dinitrogen gas flux in coastal marine sediments. Nature 448, no. 7150: skill

Predicting Hypoxia want to compare the model prediction of a box-wide daily average to a minimum daily value for the box use the buoy data to develop this relationship

regression equation relates average daily O 2 to minimum O 2 using buoy data Surface and Bottom for Buoy Data: (1/2) - BR (3/4) - BR Bullocks R. (5/6) Conimicut P. (5/6) N. Prudence (7/8) (9/10) – CP + PP Greenw. B. (11/12) Sally Rock (13/14) Mt. View (15/16) Pop. Pt. (17/18) Mt. Hope B. (19/20) Quons. Pt. (21/22) (23/24) - PP

2006 (281) Predicted # of days below 2.9 mg/L. Estimated using regression relationship between buoy daily minimum and buoy daily average.

2007 (198) 2006 (281)

2009 – worst (194) 2009 – best (138)

What EcoGEM can and cannot do… Cannot utilize new physical forcings to get at effects of changes in river flow or stratification – needs ROMS for this and 2007 were good years in terms of differences – so able to predict new years by “bracketing” (as for 2009). -- need to apply appropriate nutrient reduction, relative to ‘06-’07 -- if boundary conditions for the year are available, use This model can be used to explore the effects of nutrient reductions and temperature changes associated with climate change.

Publications and Products Vaudrey, J.M.P., M.J. Brush, and D. Ullman (in prep) Chapter 5: Modeling estuarine hypoxia in Narragansett Bay, R.I. with an intermediate-complexity approach. in: Modeling Coastal Hypoxia: Numerical Simulations of Patterns, Controls and Effects of Dissolved Oxygen Dynamics, eds: J. Dubravko, K.A. Rose, R.D. Hetland, K. Fennel. Springer Publishing. Ganju, N.K., M. J. Brush, B. Rashleigh, A.L. Aretxabaleta, P. del Barrio, M. Forsyth, J.S. Grear, L.A. Harris, S.J. Lake, G. McCardell, J. O’Donnell, D.K. Ralston, R.P. Signell, J.M. Testa, and J.M.P. Vaudrey (in press) Progress and challenges in coupled hydrodynamic-ecological estuarine modeling. Estuaries and Coasts. Kremer, J.N., J.M.P. Vaudrey, D.S. Ullman, D.L. Bergondo, N. LaSota, C. Kincaid, D.L. Codiga, and M.J. Brush (2010) Simulating property exchange in estuarine ecosystem models at ecologically appropriate scales. Ecological Modelling. 221: Vaudrey, J.M.P. and J.N. Kremer (2015) Narragansett Bay EcoGEM Model. technical report. Department of Marine Sciences, University of Connecticut. Vaudrey, J.M.P. (2015) EcoGEM – Hybrid Physical-Ecological Model of Narragansett Bay, RI. NarrBayEcoGEM.exe Vaudrey, J.M.P. (2015) User’s Guide to NarrBayEcoGEM.exe