An Open-Access, Multi-Decadal, Three-Dimensional, Hydrodynamic Hindcast Dataset for the Long Island Sound and New York/New Jersey Harbor Estuaries

Journal of Marine Science and Engineering

Article An Open-Access, Multi-Decadal, Three-Dimensional, Hydrodynamic Hindcast Dataset for the Long Island Sound and New York/New Jersey Harbor Estuaries

Nickitas Georgas 1,*, Lun Yin 1, Yu Jiang 1, Yifan Wang 1, Penelope Howell 2, Vincent Saba 3, Justin Schulte 1, Philip Orton 1 and Bin Wen 1

1 Davidson Laboratory, Stevens Institute of Technology, Hoboken, NJ 07030, USA; [email protected] (L.Y.); [email protected] (Y.J.); [email protected] (Y.W.); [email protected] (J.S.); [email protected] (P.O.); [email protected] (B.W.) 2 Marine Fisheries Division, Connecticut Department of Energy and Environmental Protection, Old Lyme, CT 06371, USA; [email protected] 3 NOAA, National Marine Fisheries Service, Northeast Fisheries Science Center, Geophysical Fluid Dynamics Laboratory, Princeton University, Princeton, NJ 08540, USA; [email protected] * Correspondence: [email protected]; Tel.: +1-201-216-8218

Academic Editor: Richard P. Signell Received: 17 July 2016; Accepted: 11 August 2016; Published: 16 August 2016

Abstract: This article presents the results and validation of a comprehensive, multi-decadal, hindcast simulation performed using the New York Harbor Observing and Prediction System´s (NYHOPS) three-dimensional hydrodynamic model. Meteorological forcing was based on three-hourly gridded data from the North American Regional Reanalysis of the US National Centers for Environmental Prediction. Distributed hydrologic forcing was based on daily United States Geologic Survey records. Offshore boundary conditions for NYHOPS at the Mid-Atlantic Bight shelf break included hourly subtidal water levels from a larger-scale model ran for the same period, tides, and temperature and salinity profiles based on the Simple Ocean Data Assimilation datasets. The NYHOPS model’s application to hindcast total water level and 3D water temperature and salinity conditions in its region over three decades was validated against observations from multiple agencies. Average indices of agreement were: 0.93 for storm surge (9 cm RMSE, 90% of errors less than 15 cm), 0.99 for water temperature (1.1 ◦C RMSE, 99% of errors less than 3 ◦C), and 0.86 for salinity (1.8 psu RMSE, 96% of errors less than 3.5 psu). The model’s skill in simulating bottom water temperature, validated against historic data from the Long Island Sound bottom trawl survey, did not drift over the years, a significant and encouraging finding for multi-decadal model applications used to identify climatic trends, such as the warming presented here. However, the validation reveals residual biases in some areas such as small tributaries that receive urban discharges from the NYC drainage network. With regard to the validation of storm surge at coastal stations, both the considerable strengths and remaining limitations of the use of North American Regional Reanalysis (NARR) to force such a model application are discussed.

Keywords: Long Island Sound; New York/New Jersey Harbor Estuary; NYHOPS model; multi-decadal hydrodynamic hindcast; North American Regional Reanalysis

1. Introduction Every year since 1976 has had an average global temperature warmer than the long-term average. Over the 1979–2013 period, global temperature warmed at an average of 0.26 ◦C per decade over land and 0.10 ◦C per decade over the global ocean [1]. The recently signed Paris Agreement [2], adopted by 195 countries, has a long-term goal of keeping the increase in global average temperature to well below

J. Mar. Sci. Eng. 2016, 4, 48; doi:10.3390/jmse4030048 www.mdpi.com/journal/jmse J. Mar. Sci. Eng. 2016, 4, 48 2 of 22

2 ◦C above pre-industrial levels, and aims to limit the increase to 1.5 ◦C, as doing so is expected to significantly reduce risks and the impacts of climate change. Yet, the Northeast US shelf waters have experienced higher warming rates than the global ocean, deduced by the Sea Surface Temperature (SST) satellite record. Pershing et al. [3] reported that SST rose by 0.30 ◦C per decade between 1982 and 2013 in the Gulf of Maine. The sole long-term observation record for water temperatures within the Long Island Sound (LIS) estuary, a US Estuary of National Significance, at a location near Millstone CT [4] has measured a much more rapid increase in LIS water temperatures than the global average: an alarming 0.44 ◦C per decade between 1979 and 2013, over four times higher than the global average rate. Over coastal Connecticut counties, on LIS’s northern coast, surface air temperatures for the same time period increased by 0.33 ◦C per decade; that rate was double if only the 1992–2012 period was considered, but has decreased somewhat since, to 0.28 ◦C per decade (1979–2015; [1]). Over these last few decades, the LIS ecosystem has undergone profound changes. Ocean warming is suggested to be the most important factor associated with the observed shifts in the mean center of biomass in Northeast U.S. fisheries [3,5–7]. However, understanding what controls the observed trends in the Northeast U.S., and how such processes affect the LIS ecosystem, has been limited due to the paucity of available three-dimensional, physical data. In 2013, the New York and Connecticut Sea Grants and the US EPA Long Island Sound Study joined forces to fund a multi-disciplinary project to address this deficiency, spearheading collaborative research involving numerical modelers, climate scientists, and fishery biologists. The work evaluated conditions and identified warming, freshening, and estuarine circulation trends in Long Island Sound over the past three decades. This research also explored how global climate contributes to long-term and inter-annual variability in the LIS physical environment and its Living Marine Resources [8]. This research article, presented at the 14th Estuarine and Coastal Modeling Conference (ECM14, http://ecm.github.io/ECM14/), focuses on the validation and results of a comprehensive, multi-decadal, hindcast simulation performed using the New York Harbor Observing and Prediction System (NYHOPS) hydrodynamic model that generated a continuous, three-dimensional dataset for a coastal aquatic region that includes Long Island Sound and the New York/New Jersey Harbor (NYNJH) Estuaries, between 1981 and 2013. Section2 describes the data and methods used to set up the multi-decadal hindcast and the data and methods used to validate it for LIS and NYNJH and to have it serve as an open access dataset. Section3 presents validation results. Section4 puts the importance of the validation in perspective and discusses identified or verified trends and climatologies based on the validated model. Conclusions are outlined in the last section. Supplementary material in the form of a comprehensive PowerPoint presentation configured in two parts is also provided. An online THREDDS Data Server (http://colossus.dl.stevens-tech.edu/thredds/catalog.html) was set up to serve the NYHOPS model’s results in oceanographic NetCDF format over the web using the OPENDAP protocol, enabling open access to daily averaged or monthly averaged time series for all the gridded hindcast physical variables in or over the NYHOPS region (including LIS and NYNJH). Simulated climatologies (mean simulated climate conditions averaged over the three decades of the NYHOPS hindcast period) for two- and three-dimensional fields such as water temperatures and salinities, were also generated, and included in THREDDS. The use of the validated results of the model to research global climate teleconnections to the LIS ecosystem and its living marine resources will be presented in subsequent papers that are presently under preparation.

2. Materials and Methods The completed multi-decadal high-resolution three-dimensional hindcast simulation for LIS and NYNJH was based on a nested modeling concept utilizing two hydrodynamic domains (Figure1): The Stevens North Atlantic Predictions model (SNAP) [9–11] and the New York Harbor Observing and Prediction System model (NYHOPS, www.stevens.edu/NYHOPS) [9,11–18]. Both domains were simulated in 3D with the Stevens Estuarine and Coastal Ocean Model code (sECOM) [11,14,19], a derivative of the Princeton Ocean Model [20]. The SNAP model was run ﬁrst, in a diagnostic mode J. Mar. Sci. Eng. 2016, 4, 48 3 of 23

J.domains Mar. Sci. Eng.were2016 simulated, 4, 48 in 3D with the Stevens Estuarine and Coastal Ocean Model code (sECOM)3 of 22 [11,14,19], a derivative of the Princeton Ocean Model [20]. The SNAP model was run first, in a diagnostic mode (clamping temperature and salinity at the initial condition), at its regular 5 km (clampingresolution temperaturegrid for the andcomplete salinity 1979–2013 at the initial simulation. condition), SNAP at its wave regular and 5 kmwater resolution level results, grid for along the completewith observation 1979–2013‐based simulation. temperature SNAP and wave salinity and fields, water were level then results, used along to derive with observation-basedNYHOPS offshore temperatureboundary conditions and salinity and ﬁelds, force were the then usedNYHOPS to derive prognostic NYHOPS hindcast offshore boundarysimulation conditions on its andvariable force‐resolution the NYHOPS grid prognostic (4 km to 25 hindcast m horizontal simulation resolution, on its variable-resolution 10 vertical sigma layers). grid (4 kmThis to is 25 the m horizontalsame nesting resolution, concept 10 used vertical operationally sigma layers). for Thisthe ensemble is the same‐based nesting Stevens concept Flood used Advisory operationally System for the(www.stevens.edu/SFAS ensemble-based Stevens [11]). Flood Advisory System (www.stevens.edu/SFAS[11]).

(a)

(b)

Figure 1.1. Hydrodynamic simulation domains used in this work: ( a)) The Stevens North Atlantic Prediction model (SNAP) domain embedded within the NCEP North-AmericanNorth‐American Regional Reanalysis meteorological model;model; ( b)(b The) The New New York York Harbor Harbor Observing Observing and Prediction and Prediction System model System (NYHOPS) model domain,(NYHOPS) embedded domain, withinembedded the SNAP within model the SNAP domain. model The domain. insert highlights The insert NYHOPS highlights variable NYHOPS grid resolutionvariable grid in theresolution NYNJH in and the Western NYNJH LIS.and ColorsWestern show LIS. bathymetry, Colors show in bathymetry, meters. in meters.

Surface meteorological forcing to both SNAP and NYHOPS was based on gridded data from Surface meteorological forcing to both SNAP and NYHOPS was based on gridded data from the North American Regional Reanalysis (NARR [21]) created for the US National Centers for the North American Regional Reanalysis (NARR [21]) created for the US National Centers for Environmental Prediction (NCEP). NARR (Figure 1) has been shown to have good skill for regional Environmental Prediction (NCEP). NARR (Figure1) has been shown to have good skill for regional climate studies [22], but may be deficient for strong Atlantic precipitation events and Atlantic climate studies [22], but may be deficient for strong Atlantic precipitation events and Atlantic hurricanes [21]. Three‐hourly surface meteorological variables provided by NARR at its 32 km grid hurricanes [21]. Three-hourly surface meteorological variables provided by NARR at its 32 km grid were interpolated to the SNAP and NYHOPS grids and used to force the two models throughout J. Mar. Sci. Eng. 2016, 4, 48 4 of 22 the hindcast. Winds at 10 m above surface and barometric pressure reduced to mean sea level, total cloud cover, relative humidity and air temperature at 2 m above ground were used to compute locally dynamic surface heat flux terms, surface stress terms and surface wave growth terms with the methodology described in [14] and [23] as progressed by Orton et al. [17] based on internally calculated surface wave fields and explicit wave-steepness by Taylor and Yelland [24]. In the construction of the three dimensional NYHOPS hindcast, great care was put into creating high-fidelity lateral and internal boundary conditions for hydrodynamic forces included in the NYHOPS model, to complement the surface meteorological forcing provided by NARR. SNAP model results were used to provide hourly offshore boundary conditions to NYHOPS at the Mid-Atlantic Bight shelf break for surface waves and offshore tidal residuals (storm surge), the latter being used to provide the subtidal part to the tidal NYHOPS water level boundary conditions as in [14,15]. An attempt was made to account for steric and mean sea level rates across the NYHOPS simulations by adding the spatially and-seasonally averaged residual errors across SNAP coastal station predictions within the NYHOPS domain to the NYHOPS offshore water level boundary conditions [14,15]; average rates are listed in [25]. The observed water level records used in this step were tied to the geodetic NAVD88 datum. Further, to provide offshore temperature and salinity profiles at the continental shelf break to NYHOPS for the hindcast, monthly data from the Simple Ocean Data Assimilation (SODA) climatology for water temperature (T) and salinity (S) were acquired beginning in 1959 on a global 0.5 degree geographic resolution grid with 40 standard depth levels in the vertical [26]. Some issues were identified with the continuity and versioning of the available SODA datasets that required significant effort in order to create a consistent monthly climatological dataset for the complete NYHOPS hindcast period of 1979–2013. A unified set of NYHOPS boundary conditions was created using SODA version 2.1.6 from 1979–1999, then SODA version 2.2.4 from 2000–2010, then results from a ROMS model run [18] generated at Rutgers University nested within a global HYCOM model for 2011–2012, and finally HYCOM global model results for 2013 [27]. To decrease climatologically relevant discrepancies between the last two datasets and SODA, bias correction for the last two datasets was performed both for the T/S means and their range. The native ROMS results for 2011–2012 were bias-corrected based on mean and range anomalies between the SODA version 2.2.4 datasets and the ROMS-with-HYCOM datasets for the common years of 2005–2008. HYCOM for 2013 was similarly bias-corrected based on the same debiasing factors. The assumption was that if the ROMS and HYCOM models were biased compared to SODA years 2005–2008 (shifted and inflated/deflated), they would continue being biased in a similar fashion in subsequent years. This assumption was validated by comparing the debiased datasets against SODA 2.2.4 for the last two SODA years, 2009–2010. Both means and ranges were significantly closer to SODA after debiasing (not shown). After the unified monthly T/S dataset from 1979 to 2013 was created at SODA resolution, it was interpolated in space and time along the NYHOPS offshore boundary, checked for vertical density stability, and used to force the NYHOPS 3D hindcast simulation. Distributed hydrologic forcing to the NYHOPS estuarine model was based on daily United States Geologic Survey (USGS) records [25] with comparable but expanded results to other regional published studies [28,29]. As part of this work, a fluvial temperature study for rivers with long temperature time series across the Mid-Atlantic was completed to aid with the assignment of daily temperatures to riverine discharges in the NYHOPS hindcast (NYHOPS includes an extensive hydrologic input network [14]). That work was presented at the annual Mid-Atlantic Bight Oceanography and Meteorology Meeting (MABPOM 2013). Results indicated that river temperatures, and associated thermal inputs to Mid-Atlantic waters increased—similarly to, though somewhat less rapidly than, the regional air temperature trends—with variation in the positive rate values between different MAB watersheds (Figure2). River flows also increased considering the 1979–2013 period as a whole. Based on linear trends estimated from the USGS discharge data, major freshwater river inflow rates to the LIS and NYNJH increased significantly: the Connecticut River at Thompsonville (USGS station J. Mar. Sci. Eng. 2016, 4, 48 5 of 22 J. Mar. Sci. Eng. 2016, 4, 48 5 of 23 totalID 01184000)freshwater by flow 17%—the into LIS—the Connecticut Housatonic River River contributes at Stevenson about 75%(USGS of totalstation freshwater ID 01205500) flow by into 21%,LIS—the and the Housatonic Hudson River River at at Green Stevenson Island (USGS by 33% station (USGS ID station 01205500) ID 01358000). by 21%, and The the completed Hudson River daily at timeGreen series Island of estimated by 33% (USGS river stationflows and ID 01358000).temperatures The from completed 1979 to daily 2013 time were series used of as estimated distributed river dischargeflows and forcing temperatures in the fromNYHOPS 1979 to3D 2013 hindcast. were used Ungauged as distributed tributaries discharge in NYHOPS forcing inare the included NYHOPS through3D hindcast. basin‐ Ungaugedarea scaling tributaries of observed in NYHOPS hydrographs are included from through proximal basin-area gauged scaling rivers. of Finally, observed distributedhydrographs end from‐of‐pipe proximal Point gaugedSource forcing rivers. Finally,is also included distributed in end-of-pipeNYHOPS model Point Sourceruns based forcing on is monthlyalso included climatologies in NYHOPS of waste model water runs treatment based on plant monthly effluent climatologies and power of waste plant waterintake/outfall treatment pairs plant [14].effluent and power plant intake/outfall pairs [14]. TheThe model model hindcast hindcast simulation simulation period startedstarted inin 1979 1979 and and completed completed in in 2013. 2013. The The first first two two years yearswere were considered considered spin-up spin‐up years years for for 3D 3D hydrodynamics. hydrodynamics. Therefore, Therefore, for for consistency, consistency, results results will will be bepresented presented from from 1981 1981 on. on. This This hypothesis hypothesis was testedwas tested in LIS in by LIS considering by considering different different initial conditions initial conditionsupdated everyupdated five every years five from years the hindcast. from the Ithindcast. was found It was that found the NYHOPS that the solution NYHOPS for solutionT and S forwithin T andthe S LIS within estuary the LIS would estuary converge would well converge within well two yearswithin from two initiation.years from initiation.

Figure 2. Linear temperature trends (◦C per decade) for different NYHOPS river stations between 1992 Figureand 2012 2. Linear shown temperature in the Mid-Atlantic trends (°C Region per map.decade) The for trend different over the NYHOPS same time river period stations at the between LIS basin 1992near and Millstone, 2012 shown CT in is the also Mid shown‐Atlantic for comparison Region map. (“M” The intrend the over insert the to same the right). time period Also highlightedat the LIS basinare thenear Hudson Millstone, River CT at Poughkeepsieis also shown (“H”) for comparison and the Delaware (“M” Riverin the at insert Trenton, to NJthe (“D”). right). Monthly Also highlightedNARR-based, are watershed-area-averaged,the Hudson River at Poughkeepsie surface air (“H”) temperature and the linear Delaware trends River are also at includedTrenton, inNJ the (“D”).right Monthly panel as NARR “AIRT-‐wbased,” where watershed watershed‐areaw‐averaged, [DR, HR, surface CR, WLIS, air temperature CLIS, ELIS] linear = [Delaware trends River,are alsoHudson included River, in the Connecticut right panel River, as “AIRT Western‐w” LIS,where Central watershed LIS, Eastern w ϵ [DR, LIS]. HR, 95% CR, conﬁdence WLIS, CLIS, intervals ELIS] = are [Delawareincluded River, as vertical Hudson bars, withRiver, central Connecticut estimates River, shown Western as circles. LIS, Central LIS, Eastern LIS]. 95% confidence intervals are included as vertical bars, with central estimates shown as circles. The NYHOPS model’s application to hindcast total water level, and 3D water temperature and salinityThe NYHOPS conditions model’s in its region application over three to hindcast decades total was validatedwater level, extensively and 3D water against temperature various available and salinityobservational conditions datasets. in its Hourlyregion totalover waterthree levelsdecades collected was validated by the Nationalextensively Ocean against Service various (NOS) availableat 12 coastal observational stations withindatasets. the Hourly NYHOPS total domain water levels between collected 1979 by and the 2013 National were used Ocean to Service quantify (NOS)the model’s at 12 coastal performance stations to within storm surgethe NYHOPS after subtraction domain between of the NOS-predicted 1979 and 2013 astronomical were used tideto quantify(Figure 3a).the Near-surface model’s performance and near-bottom to stormT and Ssurgegrab samples after subtraction taken with variable of the frequency—weekly NOS‐predicted astronomicalto biweekly, tide and (Figure mostly 3a). during Near summer—from‐surface and near a New‐bottom York T City and DepartmentS grab samples of Environmental taken with variableProtection frequency—weekly (NYC DEP) boat between to biweekly, 1981 and and 2012 mostly were usedduring to quantifysummer—from the skill ofa theNew model York for CityT and DepartmentS at NYC DEP of Environmental stations in NYNJH Protection (Figure3 (NYCb). Vertical DEP) CTD boat casts between from cruise1981 surveysand 2012 conducted were used for to the quantifyLong Island the skill Sound of the Study model between for T 1991and S and at NYC 2012 as DEP provided stations by in the NYNJH Connecticut (Figure Department 3b). Vertical of EnergyCTD castsand from Environmental cruise surveys Protection conducted (CT DEEP) for the were Long used Island to quantify Sound the Study skill ofbetween the model 1991 for andT and 2012S in as LIS provided(Figure3 byc). the Near-bottom Connecticut temperatures Department collected of Energy on and a regular Environmental grid that covers Protection most (CT of the DEEP) LIS bottom were usedby CTto DEEPquantify ﬁsheries the skill as part of the of themodel Long for Island T and Sound S in Trawl LIS (Figure Survey 3c). between Near 1992‐bottom and temperatures 2013 were used collectedto test whether on a regular the model’s grid that skill covers in simulating most of the bottom LIS bottom water temperature by CT DEEP showed fisheries signs as part of drift of the over Long Island Sound Trawl Survey between 1992 and 2013 were used to test whether the model’s skill

J. Mar. Sci. Eng. 2016, 4, 48 6 of 23 J. Mar. Sci. Eng. 2016, 4, 48 6 of 22

in simulating bottom water temperature showed signs of drift over the years in LIS (Figure 3d). theFinally, years an in LISobservations (Figure3d).‐based Finally, monthly an observations-based three‐dimensional monthly temperature three-dimensional and salinity temperature climatology anddataset salinity called climatology MOCHA dataset version called MOCHA‐2 created version-2 at createdRutgers at RutgersUniversity University in inNew New Jersey ((http://tds.marine.rutgers.edhttp://tds.marine.rutgers.edu/thredds/catalog/other/climatology/mocha/catalog.htmlu/thredds/catalog/other/climatology/mocha/catalog.html; [30];[30 ] asas updatedupdated in MOCHA’sMOCHA’s 2nd2nd versionversion inin 2012)2012) waswas usedused toto quantifyquantify thethe skillskill ofof thethe NYHOPSNYHOPS hydrodynamichydrodynamic modelmodel for forT and T Sandagainst S climatology.against climatology. MOCHA isMOCHA a three-dimensional is a three climatological‐dimensional analysisclimatological of the analysis temperature of the and temperature salinity, with and a salinity, 0.05 degree with (~5 a 0.05 km) degree grid in (~5 the km) horizontal grid in andthe 55horizontal standard and depths 55 standard in the vertical depths thatin the covers vertical the that MAB, covers from the 45 MAB,◦ N to from 32◦ N,45° 77 N◦ toW 32° to N, 64◦ 77°W. W It to is derived64° W. It from is derived all in situ from data all in available situ data from available the NODC from Worldthe NODC Ocean World Database Ocean 2005 Database and the 2005 NOAA and Norththe NOAA East FisheriesNorth East Science Fisheries Center Science database. Center Comparisons database. Comparisons to MOCHA wereto MOCHA only made were in only LIS asmade the NYNJHin LIS as is the not NYNJH well resolved is not well in its resolved grid. in its grid.

(a) (b)

FigureFigure 3.3. ObservationsObservations used to validatevalidate the model in this study. ((aa)) 1212 NOSNOS coastalcoastal stations;stations; ((bb)) NYCNYC DEPDEP HarborHarbor Survey Survey stations stations throughout throughout the years;the years; (c) Long (c) Long Island Island Sound Sound Study cruiseStudy stations cruise throughstations thethrough years; the (d )years; Long ( Islandd) Long Trawl Island Survey Trawl Site Survey grid. Site grid.

Adopting NOS guidelines and prior literature used for validating the operational NYHOPS Adopting NOS guidelines and prior literature used for validating the operational NYHOPS forecast model [11,15,16,31], the following metrics were used to quantify model skill in the Results forecast model [11,15,16,31], the following metrics were used to quantify model skill in the section: Results section:  Bias or mean error, M.E.: The mean error between model and observations. • BiasRMSE or: meanThe square error, M.E.root :of The the mean average error error between between model model and and observations. observations squared. • RMSER‐square,: The R square2: The square root of of the the average correlation error coefficient between model between and model observations and observations. squared. 2 • R-square,Willmott SkillR : The or Index square of of Agreement, the correlation W.I. coefﬁcient [32]: A non between‐dimensional model andmeasure observations. of how close the • Willmott Skill or Index of Agreement, W.I. [32]: A non-dimensional measure of how close the model’s results are to observations. Values are between 0 and 1, with 1 being a perfect “skill model’s results are to observations. Values are between 0 and 1, with 1 being a perfect “skill core.” score.” • Central frequency of error, C.F.: The percent of errors that are below a given threshold that is  consideredCentral frequency high for of operational error, C.F.: The use. percent The larger of errors the C.F. that, the are better below the a given model. threshold NOS usually that is considered high for operational use. The larger the C.F., the better the model. NOS usually

J. Mar. Sci. Eng. 2016, 4, 48 7 of 22

considers a C.F. ≥ 90% as accepted model performance against the following thresholds: 15 cm for total water level, 3.0 ◦C for T, and 3.5 psu for S [31]. • Taylor diagram, [33]. The Taylor Diagram provides a visual statistical summary of how well the model and observation patterns match each other in terms of their correlation coefﬁcient (R), their root mean square error (RMSE), and their standard deviation (σ). • Brier Skill Score, B.S.S. [34]: The Brier Skill Score essentially compares the magnitude of the difference between a model (NYHOPS here) and observations to that achieved by a reference model (the monthly MOCHA climatology here). The B.S.S. is written as

Ni Ni 1 2 1 2 BSS = 1 − ∑ di,j − pi,j / ∑ di,j − ri,j (1) Ni j=1 Ni j=1

where the vector di,j contains the j = 1, 2, ... , Nj measurements (in situ observations). Similarly, pi,j are the model predictions at the same time and location of the data, di,j, and the vector ri,j contains the predictions of the reference climatology (MOCHA here). BSS compares the ratio of the variance in the observations not explained by the model to that not explained by the reference climatology. If BSS > 0, then the model is in better agreement with the data than the reference model. Conversely, if BSS < 0, it is not as good. NYHOPS model results were interpolated from the native NYHOPS grid to the location and time of individual observations, before making comparisons. It is important to note that there are always discrepancies in the observations due not only to the precision of the instruments and analyses methods, but also due to the difference in the property simulated (the average over a model cell’s volume and output time step) and that measured (sometimes a few samples from a bottle). Even a perfect model, therefore, should not be expected to have BSS = 1, (nor will W.I. and R2 be equal to unity for that matter). Given however the comparison to average climatology, BSS for a skillful model should be consistently higher than 0, and hopefully closer to unity. Finally, binned histograms of model errors against observations were also used to show uncertainty in model results, and whether that uncertainty grew or decreased over the hindcast years.

3. Results The model’s grand-mean temperature and salinity bias against all Long Island Sound Study observations in the Sound was found to be 0.18 ◦C warmer for water temperature, and 1.31 psu saltier for salinity; these biases were assumed constant in time and space and were removed from all raw NYHOPS model gridded time series results. The local tidal correction procedure described in [16] was applied to the raw NYHOPS water level results for the 12 coastal stations. The astronomical tide predicted by NOS at these stations was used in that correction, and then removed from the total water level signals to calculate storm surge. The results used and shown below are after these treatments were put into effect.

3.1. Storm Surge Validation Figure4 shows example storm surge results at The Battery, NY, highlighting the storm surge validation that was performed at each of the 12 NOS stations. The highest storm surge (SS) and total water level (TWL) values at The Battery for the 1981–2013 time period shown were simulated by NYHOPS to have occurred during Hurricane Sandy in 2012. Although this is consistent with the observed record, the NARR-forced NYHOPS hindcast under-predicted surge during that storm by about 2 feet or ~20%. Table1 lists NYHOPS hindcast performance metrics for TWL and SS using the methods of Georgas and Blumberg [15] against NOS observations at 12 coastal stations. RMSE varied between 5.7 cm at Montauk, NY and 14.1 cm at Reedy Point, DE, the latter being in a region (Delaware Bay) not well J. Mar. Sci. Eng. 2016, 4, 48 8 of 22

J. Mar. Sci. Eng. 2016, 4, 48 8 of 23 resolved by the NYHOPS model’s horizontal grid. R2 for TWL that includes astronomical tide varied tide varied between 0.95 at Montauk and 0.99 at New Haven, Kings Point, and Bridgeport where the between 0.95 at Montauk and 0.99 at New Haven, Kings Point, and Bridgeport where the tide range tide range is larger. Excluding astronomical tide, R2 for SS was lowest where RMSE was highest. In is larger. Excluding astronomical tide, R2 for SS was lowest where RMSE was highest. In addition addition to Delaware Bay, this relative degradation in storm surge prediction occurred in to Delaware Bay, this relative degradation in storm surge prediction occurred in west-central LIS at west‐central LIS at Bridgeport and Kings Point. As expected, W.I. and R2 for SS were positively Bridgeport and Kings Point. As expected, W.I. and R2 for SS were positively correlated and showed correlated and showed similar qualitative results. W.I. for TWL was higher than 0.98 everywhere similar qualitative results. W.I. for TWL was higher than 0.98 everywhere (not shown). Similar to (not shown). Similar to the operational NYHOPS forecast validation [16], 8 out of the 12 stations had the operational NYHOPS forecast validation [16], 8 out of the 12 stations had C.F. < 15 cm over 90%, C.F. < 15 cm over 90%, with one more (The Battery, NY) coming also very close at 89.9%. Overall the with one more (The Battery, NY) coming also very close at 89.9%. Overall the results of the NYHOPS results of the NYHOPS hindcast for water level and its storm surge component were good, hindcast for water level and its storm surge component were good, providing conﬁdence that the providing confidence that the model was able to reproduce hydrodynamics reasonably well. model was able to reproduce hydrodynamics reasonably well.

Figure 4.4. ObservedObserved (red)(red) andand NYHOPS-simulatedNYHOPS‐simulated (blue)(blue) SSSS (=TWL—astronomical(=TWL—astronomical tide) time seriesseries forfor thethe completecomplete 1981–20131981–2013 series (left),(left), as wellwell asas fourfour regionallyregionally signiﬁcantsignificant events.events. Correlograms, and theirtheir statistics,statistics, for for both both SS SS and and TWL TWL within within the the complete complete 1981–2013 1981–2013 period, period, are are also also shown shown on theon bottomthe bottom left. left.

Table 1.1. NYHOPS hindcast hindcast performance performance metrics metrics for for total total water water level level (TWL) (TWL) and and storm storm surge surge (SS) (SS) at at12 12 NOS NOS stations. stations.

2 2 Station Name Dates RMSERMSE, , cm CF ≤ 15 cm, % R R WISS Station Name Dates CF ≤ 15cm, % R2SSSS TWLR2TWL WISS cm Lewes, DE 1979–2012 8.6 91.9 0.78 0.97 0.94 Lewes,Reedy DE Point, DE1979–20121996–2013 8.6 14.191.9 71.7 0.78 0.63 0.960.97 0.88 0.94 Reedy Point,Cape DE May, NJ1996–20131979–2013 14.1 7.871.7 94.6 0.63 0.80 0.980.96 0.95 0.88 Cape May,Atlantic NJ City, NJ1979–20131985–2013 7.8 6.694.6 97.2 0.80 0.86 0.980.98 0.96 0.95 Sandy Hook, NJ 1979–2013 9.0 91.1 0.76 0.97 0.93 Atlantic City, NJ 1985–2013 6.6 97.2 0.86 0.98 0.96 The Battery, NY 1979–2013 9.3 89.9 0.74 0.97 0.93 Sandy Hook,Kings NJ Point, CT1979–20131998–2013 9.0 12.891.1 76.4 0.76 0.65 0.990.97 0.90 0.93 The Battery,Bridgeport, NY CT1979–20131996–2013 9.3 11.689.9 79.8 0.74 0.65 0.980.97 0.89 0.93 Kings Point, CT 1998–2013 12.8 76.4 0.65 0.99 0.90 Bridgeport, CT 1996–2013 11.6 79.8 0.65 0.98 0.89 New Haven, CT 1999–2013 9.1 90.6 0.75 0.99 0.93

J. Mar. Sci. Eng. 2016, 4, 48 9 of 22 J. Mar. Sci. Eng. 2016, 4, 48 9 of 23

Montauk, NY 1979–2013 5.7 Table 1. Cont.98.9 0.85 0.95 0.96 New London, CT 1979–2013 7.1 96.5 0.78 0.96 0.94 2 2 Newport, RIStation Name1979–2013 Dates 7.4RMSE , cm CF95.6≤ 15 cm, % R0.72SS R TWL 0.97WISS 0.92 New Haven, CT 1999–2013 9.1 90.6 0.75 0.99 0.93 Montauk, NY 1979–2013 5.7 98.9 0.85 0.95 0.96 3.2. Water TemperatureNew London, and CTSalinity1979–2013 Validation 7.1 96.5 0.78 0.96 0.94 Newport, RI 1979–2013 7.4 95.6 0.72 0.97 0.92 3.2.1. Against New York City Department of Environmental Protection (NYC DEP) Data in the New York/New3.2. Water Jersey Temperature Harbor and (NYNJH) Salinity ValidationEstuary

3.2.1.The correlograms Against New York shown City in Department the top row of of Environmental Figure 5 summarize Protection the (NYC comparison DEP) Data of in the the NYHOPS hindcastNew results York/New against Jersey all Harbor observations (NYNJH) for Estuary T and S taken by NYC DEP between 1981 and 2012 at the surface andThe bottom correlograms of the shown NYNJH in theat all top Harbor row of Survey Figure5 stations.summarize For the water comparison temperatures, of the NYHOPS most points fell alonghindcast the results 1:1 line. against Although all observations this was also for T true and Sfor taken S, there by NYC were DEP several between observations 1981 and 2012 that at were not wellthe surface‐captured and bottomby the ofmodel, the NYNJH with atmost all Harbor of these Survey values stations. being For over water‐predicted temperatures, by the most model. Thoroughpoints station fell along‐by the‐station 1:1 line. and Although event‐by this‐event was alsostudy, true summarized for S, there were in the several Discussion observations section that and in the Supplementarywere not well-captured Material, by the revealed model, withthat mostthese of discrepancies these values being are over-predictedmostly associated by the with model. sewer overflowThorough events station-by-station and other wet and‐weather event-by-event non‐point study, source summarized contributions in the Discussion at small section tidal and tributaries. in the Nevertheless,Supplementary considering Material, the revealed whole that estuary these discrepanciesand NYC DEP are dataset, mostly associated M.E. was with only sewer −0.1 overﬂow°C for T and 0.0 psuevents for andS, RMSE other wet-weather was 1.2 °C non-pointfor T and source2.3 psu contributions for S, W.I. atwas small 0.99 tidal for tributaries.T and 0.95 Nevertheless, for S, while the considering the whole estuary and NYC DEP dataset, M.E. was only −0.1 ◦C for T and 0.0 psu for S, central frequency of error was as high as 99% for T and 93% for S, revealing a very skillful hindcast. RMSE was 1.2 ◦C for T and 2.3 psu for S, W.I. was 0.99 for T and 0.95 for S, while the central frequency The Taylor Diagrams on the bottom row of Figure 5 further show that correlation coefficients were of error was as high as 99% for T and 93% for S, revealing a very skillful hindcast. The Taylor Diagrams higheron than the bottom 0.9 for row both of FigureT and5 S further, and reveal show that that correlation the model coefﬁcients was able wereto capture higher the than range 0.9 for of both variationT bothand forS T, and and reveal S as that shown the model by the was concentric able to capture dotted the rangeblack of circles variation of bothstandard for T and deviationS as shown on the diagram.by the Further, concentric the dotted model’s black error circles standard of standard deviation, deviation approximated on the diagram. by Further, the RMSE the model’s (also errordepicted withstandard the green deviation, circles in approximated the Taylor Diagram), by the RMSE was(also overall depicted smaller with than the the green standard circles in deviation the Taylor of the signalDiagram), it simulated. was overall smaller than the standard deviation of the signal it simulated.

(a) (b)

Figure 5. Cont.

J. Mar. Sci. Eng. 2016, 4, 48 10 of 22 J. Mar. Sci. Eng. 2016, 4, 48 10 of 23 J. Mar. Sci. Eng. 2016, 4, 48 10 of 23

(c) (d) (c) (d) Figure 5. Correlograms (a and b), and Taylor Diagrams (c and d), summarizing the comparison FigureFigure 5.5. Correlograms (a and bb),), andand TaylorTaylor DiagramsDiagrams ((cc andand dd),), summarizingsummarizing thethe comparisoncomparison between the NYHOPS results and NYC DEP Harbor Survey observations for T (a,c) and S (b,d). The betweenbetween the NYHOPS results results and and NYC NYC DEP DEP Harbor Harbor Survey Survey observations observations for forT (a T,c ()a and,c) and S (b, Sd). (b The,d). Taylor Diagrams summarize RMSE (dotted green circles with the observations “OBS” as origin), TheTaylor Taylor Diagrams Diagrams summarize summarize RMSERMSE (dotted(dotted green green circles circles with with the the observations observations “OBS” “OBS” as origin),origin), correlation coefficient (dotted blue radials with 0 origin), and standard deviation (black dotted circles correlationcorrelation coefﬁcient coefficient (dotted (dotted blue blue radials radials with with 00 origin),origin), andand standardstandard deviationdeviation (black(black dotteddotted circlescircles with 0 origin). withwith 0 0 origin). origin). Figure 6 shows example time series for simulated surface and bottom T and S against observed FigureFigure6 6 shows shows example example time time series series for for simulated simulated surface surface and and bottom bottom TT andand SS againstagainst observedobserved surface and bottom T and S at Harbor Survey station N5, the station nearest to The Battery, NY. The surfacesurface and bottom TT andand SS atat Harbor Harbor Survey Survey station station N5, N5, the the station station nearest nearest to The to TheBattery, Battery, NY. NY.The model is seen to capture well the ranges and seasonal signals in both T and S and shows good skill in Themodel model is seen is seen to capture to capture well well the theranges ranges and and seasonal seasonal signals signals in both in both T and T and S and S and shows shows good good skill skill in responding to events in the record. Surface measurements taken at station N5 in the early 1980s, inresponding responding to toevents events in in the the record. record. Surface Surface measurements measurements taken taken at at station station N5 N5 in in the early 1980s,1980s, mid‐1990s, and 2002, revealed somewhat higher salinities during peak ocean salt intrusion summer mid-1990s,mid‐1990s, andand 2002,2002, revealedrevealed somewhatsomewhat higherhigher salinitiessalinities duringduring peakpeak oceanocean saltsalt intrusionintrusion summersummer seasons than those simulated by the model. Higher salinities were also observed but not simulated seasonsseasons thanthan thosethose simulatedsimulated byby thethe model.model. HigherHigher salinitiessalinities werewere alsoalso observedobserved butbut notnot simulatedsimulated in 2002 near the bottom at station N5. inin 20022002 nearnear thethe bottombottom atat stationstation N5. N5.

(a) (b) (a) (b) Figure 6. Time series of observer (red) and simulated (blue) T (a) and S (b) at NYC DEP Harbor Figure 6. Time series of observer (red) and simulated (blue) T (a) and S (b) at NYC DEP Harbor FigureSurvey 6.stationTime seriesN5 near of observerThe Battery, (red) NY. and simulated (blue) T (a) and S (b) at NYC DEP Harbor Survey stationSurvey N5 station near N5 The near Battery, The NY.Battery, NY. 3.2.2. Against Connecticut Department of Energy and Environmental Protection (CT DEEP) Data in 3.2.2. Against Connecticut Department of Energy and Environmental Protection (CT DEEP) Data in Long Island Sound (LIS) Long Island Sound (LIS)

J. Mar. Sci. Eng. 2016, 4, 48 11 of 22

J. Mar. Sci. Eng. 2016, 4, 48 11 of 23 3.2.2. Against Connecticut Department of Energy and Environmental Protection (CT DEEP) Data in Long Island Sound (LIS) Figure 7 shows correlograms summarizing the comparison of the NYHOPS hindcast results againstFigure all CTD7 shows casts correlograms during CT DEEP summarizing Long Island the comparisonSound Study of cruises the NYHOPS between hindcast 2001 and results 2012. againstResults are all CTDsummarized casts during by LIS CT management DEEP Long basin. Island Most Sound points Study fell along cruises the between 1:1 line, 2001and the and model 2012. Resultsresults throughout are summarized LIS were by LIS overall management reasonable. basin. For Most T, RMSE points was fell alongclose to the 1 1:1°C line,in all and three the basins, model ◦ resultsW.I. was throughout greater than LIS 0.99, were and overall model reasonable. errors were For Tless, RMSE than was3 °C close over to 99% 1 Cof in the all time three (C.F. basins, ≥ 99%,W.I. wasFigure greater 7). Salinity than 0.99, RMSE and was model 0.5 errorspsu in were the western less than and 3 ◦ Ccentral over 99% management of the time basins, (C.F. ≥ but99%, reached Figure 1.07). Salinitypsu in theRMSE easternwas basin, 0.5 psu largely in the because western of and a residual central managementM.E. contribution basins, there but of reached 0.6 psu 1.0 that psu was in not the easternfound for basin, the other largely two because basins. of Although a residual theM.E. centralcontribution frequency there was of greater 0.6 psu than that 99% was for not salinity found for in theall three other basins, two basins. the W.I. Although was significantly the central lower frequency in the waseastern greater basin than too, 99% at 0.77. for salinityIt appears in allthat three the basins,removal the of W.I.the grandwas signiﬁcantly‐mean bias of lower 1.31 inpsu the from eastern all raw basin NYHOPS too, at 0.77. hindcast It appears results that made the the removal eastern of thebasin grand-mean of LIS slightly bias offresher 1.31 psu than from it should. all raw This NYHOPS result hindcast will be resultsdiscussed made further the eastern in the basin Discussion of LIS slightlysection. fresher than it should. This result will be discussed further in the Discussion section.

Figure 7.7. Correlograms,Correlograms, summarizing summarizing the the comparison comparison between between the NYHOPSthe NYHOPS results results and all and CT all DEEP CT LongDEEP IslandLong Island Sound Sound Study Study observations observations taken taken between between 1991 1991 and 2012and 2012 for T forand T andS at S the at threethe three LIS managementLIS management basins. basins.

3.2.3. Against MOCHA Climatology in LIS Figure 88 comparescompares the the NYHOPS NYHOPS model’s model’s performance performance to to hindcast hindcast LISS LISS water water temperatures temperatures and andsalinities salinities against against MOCHA MOCHA monthly monthly climatology. climatology. The The top top panels panels of of the the Figure Figure compare compare overall performance throughout the LISS record using Taylor Diagrams,Diagrams, while year-to-yearyear‐to‐year relative performance between 1991 and 2012 is depicteddepicted inin thethe lowerlower panelpanel asas aa B.S.S.B.S.S. timetime series.series. The Taylor diagrams show that the NYHOPS model (“M”) is closer to the observed conditions (“O”) compared to the MOCHA climatology (“C”) for the correlation coefﬁcientcoefficient and the RMSE (Figure 8). Thus the dynamic model exhibits significantly higher correlation and lower error in describing the data. Note that the correlation coefficient scale is logarithmic and that the correlation between the NYHOPS model and the observations is significantly higher especially for salinity

J. Mar. Sci. Eng. 2016, 4, 48 12 of 22

(Figure8). Thus the dynamic model exhibits significantly higher correlation and lower error in describingJ. Mar. Sci. Eng. the 2016 data., 4, 48 Note that the correlation coefficient scale is logarithmic and that the correlation12 of 23 between the NYHOPS model and the observations is significantly higher especially for salinity where it grewwhere from it grew less than from 0.7 less (C) tothan over 0.7 0.9 (C) (M). to The over standard 0.9 (M). deviation The standard captured deviation by the observation-basedcaptured by the monthlyobservation MOCHA‐based climatology monthly appearsMOCHA closer climatology to the one appears based on closer LISS observationsto the one than based the NYHOPSon LISS model,observations consistently than the with NYHOPS the Taylor model, Diagrams consistently for NY Harborwith the (Figure Taylor5 ):Diagrams In both thefor NYNJHNY Harbor and LIS,(Figure the model 5): In seemsboth the to somewhat NYNJH and over-predict LIS, the model the overall seems observed to somewhat range inover water‐predict temperature the overall and under-predictobserved range the in overall water rangetemperature in salinity. and under Even‐ so,predict the NYHOPSthe overall model range wasin salinity. overall Even significantly so, the closerNYHOPS to observations model was that overall the significantly climatology: closer the Taylor to observations dot for the modelthat the (M) climatology: is closer to the the Taylor Taylor dot dot forfor observations the model (M) (O) is than closer the to Taylor the Taylor dot for dot MOCHA for observations climatology (O) (C). than the Taylor dot for MOCHA climatology (C).

(a) (b)

FigureFigure 8. 8.Taylor Taylor Diagrams Diagrams and timeand seriestime ofseriesB.S.S. of, comparing B.S.S., comparing the skill of the the NYHOPSskill of the hydrodynamic NYHOPS modelhydrodynamic (M) to the model climatological (M) to the MOCHA climatological model (C) MOCHA in describing model CT (C) DEEP in describing Long Island CT SoundDEEP Long Study observationsIsland Sound (O) Study taken observations between 1991 (O) and taken 2012 between for T ( a1991) and andS ( b2012) in LIS.for T Taylor (a) and Diagrams S (b) in LIS. summarize Taylor RMSEDiagrams (dotted summarize green circles RMSE with (dotted the observations green circles “O” with as the origin), observations correlation “O” coefficient as origin), (dotted correlation blue radialscoefficient with (dotted 0 origin), blue and radials standard with deviation 0 origin), (black and dottedstandard circles deviation with 0 (black origin). dotted circles with 0 origin). The lower panels in Figure8 also show that the hydrodynamic NYHOPS model was a better The lower panels in Figure 8 also show that the hydrodynamic NYHOPS model was a better predictor of the year-to-year water temperature and salinity observed in LIS than the static predictor of the year‐to‐year water temperature and salinity observed in LIS than the static observation-based MOCHA climatology: During the whole LISS observation record from 1991 to 2012, observation‐based MOCHA climatology: During the whole LISS observation record from 1991 to the relative B.S.S. was always higher than 0, meaning NYHOPS had better skill than climatology each 2012, the relative B.S.S. was always higher than 0, meaning NYHOPS had better skill than year.climatology Peaks close each to year. the optimal Peaks close 1.0 value to the in optimal relative 1.0BSS valueare in seen relative in some BSS years are seen that in were some anomalously years that coldwere or anomalously hot compared cold to climatology,or hot compared such to as climatology, during 1993 such and as 2012, during respectively. 1993 and During2012, respectively. these years theDuring monthly these mean years MOCHA the monthly climatology mean MOCHA would not climatology have been would a good not descriptor have been of a what good happened descriptor in LIS,of what unlike happened the NYHOPS in LIS, model unlike that the included NYHOPS three-hourly model that heat included flux forcing three based‐hourly on heat NARR. flux Given forcing the largebased seasonal on NARR. signal Given in temperature the large causingseasonal higher signal correlation in temperature coefficients causing even higher between correlation MOCHA climatologycoefficients andeven observations, between MOCHA the relative climatologyB.S.S. for and water observations, temperature the ranged relative more B.S.S. from for yearwater to yeartemperature than the ranged one for more salinity, from as year seen to inyear the than smaller the oney-axis for salinity, for salinity as seenB.S.S. in the(Figure smaller8). Similarly,y‐axis for thesalinity 1991–2012 B.S.S. mean (FigureB.S.S. 8). Similarly,for water temperaturethe 1991–2012 (0.48) mean was B.S.S. also for smaller water than temperature the one for (0.48) salinity was (0.80). also Forsmaller salinity, than the theB.S.S. oneranged for salinity largely (0.80). between For 0.6salinity, and 0.9, the since B.S.S. the ranged NYHOPS largely hydrodynamic between 0.6 model and 0.9, that includedsince the daily NYHOPS freshwater hydrodynamic flows was, asmodel expected, that included much more daily able freshwater to reproduce flows spatiotemporal was, as expected, salinity much more able to reproduce spatiotemporal salinity variations compared to the monthly climatological means of MOCHA. No statistically significant trend was found for either water temperature or salinity B.S.S. time series.

J. Mar. Sci. Eng. 2016, 4, 48 13 of 22 variations compared to the monthly climatological means of MOCHA. No statistically signiﬁcant trend wasJ. Mar. found Sci. Eng. for 2016 either, 4, 48 water temperature or salinity B.S.S. time series. 13 of 23

4.4. Discussion Discussion TheThe skillskill ofof thethe comprehensively-forcedcomprehensively‐forced multi-decadalmulti‐decadal NYHOPSNYHOPS HindcastHindcast presentedpresented inin thethe previousprevious paragraphparagraph waswas overalloverall excellent:excellent: across all stations considered, the average Willmott IndexIndex ofof Agreement for storm storm surge surge (tidal (tidal departure) departure) alone alone against against NOS NOS hourly hourly observations observations was was 0.93, 0.93, (9 (9cm cm Root Root-Mean-Square-Error,‐Mean‐Square‐Error, RMSERMSE, 90%, 90% of of errors errors less less than than 15 15 cm). cm). For For water temperature againstagainst availableavailable NYC DEP and CTDEEP observations, observations, W.I.W.I. waswas 0.99 0.99 (1.1 (1.1 °C◦C RMSERMSE, ,99% 99% of of errors errors less less than than 3 3°C).◦C). For For salinity salinity against against NYC NYC DEP DEP and and CTDEEP CTDEEP observations, observations, W.I.W.I. waswas 0.86 0.86 (1.8 psu RMSE, 96% ofof allall errorserrors lessless thanthan 3.53.5 psu).psu). ForFor waterwater levels,levels, modelmodel resultsresults werewere reasonablereasonable overall,overall, thoughthough modelmodel errorserrors againstagainst hourlyhourly observationsobservations increased increased as as expected expected during during major major storm storm surge surge events events and and Atlantic Atlantic hurricanes hurricanes (Figure (Figure9); Figure9); Figure9 can 9 can also also be comparedbe compared to [to16 [16]] (Figure (Figure7) and7) and [ 11 [11]] (Figure (Figure5). 5). This This error error increase increase with with surge surge magnitudemagnitude isis inin partpart duedue toto thethe resolutionresolution inin timetime (3-hourly)(3‐hourly) andand spacespace (36(36 km)km) ofof thethe NARRNARR datasetdataset usedused toto provideprovide windswinds andand barometricbarometric pressurepressure toto thethe NYHOPSNYHOPS model.model. MesingerMesinger etet al.al. in 20062006 [[21]21] statedstated expectations for relatively relatively poorer poorer NARR NARR skill skill during during Atlantic Atlantic Hurricanes. Hurricanes. This This may may in inpart part be bedue due to toa presumed a presumed increase increase of of the the number number of of observations observations fed fed into into NARR during the reanalysisreanalysis processprocess inin laterlater yearsyears comparedcompared toto thethe beginningbeginning ofof thethe NARRNARR recordrecord inin 1979.1979. StormStorm surgessurges duringduring somesome early events, events, the the Nor’Easter Nor’Easter of of March March 1984 1984 and and Hurricane Hurricane Gloria Gloria in 1985 in 1985being being examples, examples, were wereunder under-predicted.‐predicted. The storm The stormsurge surgefrom the from quick the quicktransit transit of Hurricane of Hurricane Bob that Bob devastated that devastated New NewEngland England in 1991 in 1991 was was also also not not captured. captured. For For example, example, the the surge surge built built from from 0 toto overover 5 5 feet ft in in 4 4 h atat Newport,Newport, RI,RI, butbut thethe 3-hourly,3‐hourly, 36 36 km km NARR NARR record record was was not not able able to to capture capture that that hurricane hurricane well. well.

(a) (b)

FigureFigure 9.9. ((aa)) M.E.M.E. andand ((bb)) RMSERMSE asas a a function function of of storm storm surge surge threshold threshold for for negative negative ( −(−)) andand positivepositive (+)(+) surges betweenbetween differentdifferent 5-year5‐year periodsperiods inin thethe NYHOPSNYHOPS hindcast.hindcast.

The residual water level errors during significant storm surge events appeared to decrease The residual water level errors during signiﬁcant storm surge events appeared to decrease toward toward the second half of the simulation (Figure 9), even though these later years included some of the second half of the simulation (Figure9), even though these later years included some of the highest the highest storm surge peaks in the region’s history during Hurricanes Irene and Sandy. It is also storm surge peaks in the region’s history during Hurricanes Irene and Sandy. It is also important to important to note however that the NARR‐forced NYHOPS still under‐predicted Hurricane Sandy’s note however that the NARR-forced NYHOPS still under-predicted Hurricane Sandy’s peak surge peak surge at The Battery by ~2 ft, or ~20% of the observed ~10 ft surge (Figure 4). During the actual at The Battery by ~2 feet, or ~20% of the observed ~10 feet surge (Figure4). During the actual event event in October 2012 the NYHOPS OFS forecast forced by the deterministic North American in October 2012 the NYHOPS OFS forecast forced by the deterministic North American Mesoscale Mesoscale (NAM) model at 12 km resolution under‐predicted Sandy’s surge at The Battery by (NAM) model at 12 km resolution under-predicted Sandy’s surge at The Battery by approximately approximately 3 ft, while it was within 1 ft when, after the fact, the same model was forced with a 3 feet, while it was within 1 foot when, after the fact, the same model was forced with a more accurate more accurate forecast [9] or a high‐fidelity reanalysis [10]. forecast [9] or a high-ﬁdelity reanalysis [10]. Another contributing reason to the apparent increase in the model’s storm surge simulation skill over time may be due to changes in bathymetry and coastline over time, or other morphodynamic or anthropogenic (targeted channel dredging) changes not accounted for in the NYHOPS historic hindcast: the model’s bathymetry is held fixed, and is based on a variety of

J. Mar. Sci. Eng. 2016, 4, 48 14 of 22

Another contributing reason to the apparent increase in the model’s storm surge simulation skill over time may be due to changes in bathymetry and coastline over time, or other morphodynamic or anthropogenic (targeted channel dredging) changes not accounted for in the NYHOPS historic hindcast: the model’s bathymetry is held fixed, and is based on a variety of datasets described in Georgas [14], meant to represent the configuration of the MAB estuaries in the beginning decade of the 21st century. Blumberg and Georgas [13] showed that water levels are quite sensitive to bathymetric uncertainties in the region. Further research is needed in creating a quantitative timeline for such changes, so that future model run versions can account for them. Table2 summarizes some skill metrics of the NYHOPS historic hindcast against water temperature and salinity datasets at different estuarine regions, and compares that skill to the overall skill of the NYHOPS OFS forecasts from Georgas [14] and Georgas and Blumberg [15]. The multi-decadal NYHOPS hindcast appears to have comparable or better skill than the validated NYHOPS OFS. Note however that the periods compared and the stations used to aggregate errors are not the same, nor is the forcing methodology: the NYHOPS Hindcast used estimates of daily river flows based on observed hydrographs by USGS and three-hourly heat fluxes based on the 36 km NARR, while the NYHOPS deterministic OFS used six-hourly NOAA AHPS river discharge forecasts and three-hourly heat fluxes based on the NAM 12 km forecasts.

Table 2. NYHOPS Hindcast Performance metrics summary for water temperature (T) and salinity (S) at regions within LIS and NYNJH.

◦ ◦ ◦ Water Temperature Region Bias, C RMSE, C W.I.T CF ≤ 3 C, % Long Island Sound Western Basin −0.2 1.0 0.99 99 CT DEEP Central Basin +0.3 1.0 0.99 99 LISS Survey Eastern Basin +0.0 0.9 1.00 100 Western LIS −0.4 1.1 0.99 98 New York Harbor Upper East River −0.3 1.1 0.99 99 NYC DEP Inner Harbor +0.0 1.3 0.99 98 Surface/Bottom data Lower NY Bay +0.1 1.3 0.99 98 Jamaica Bay −0.2 1.2 0.99 98 NYHOPS OFS 1 NYHOPS 2 +0.0 1.4 0.98 95

Salinity Region Bias, psu RMSE, psu W.I.S CF ≤ 3.5 psu, % Long Island Sound Western Basin +0.1 0.5 0.91 100 CT DEEP Central Basin +0.0 0.5 0.90 100 LISS Survey Eastern Basin −0.6 1.0 0.77 100 Western LIS +0.5 1.3 0.77 99 New York Harbor Upper East River +0.8 2.2 0.94 93 NYC DEP Inner Harbor −0.1 2.8 0.93 89 Surface/Bottom data Lower NY Bay −0.9 2.2 0.91 91 Jamaica Bay +0.0 1.8 0.71 96 NYHOPS OFS 1 NYHOPS 2 +0.0 2.8 0.77 87 1 Georgas and Blumberg [15]; 2 All stations considered within the NYHOPS model domain, 2007–2009.

Overall hindcast results are well within NOAA standards and skill metrics for T and S (CF > 90%; Table2). Both water temperatures and salinities were very well predicted. Given the greater seasonality in estuarine temperature compared to that of salinity, and the three-hourly meteorological forcing compared to the daily hydrological forcing, the relative RMSE (RMSE divided by the expected range as in Georgas [14]), was higher for salinity than temperature, at an across-station median of 30.9% versus 4.7%, respectively (Figure 10). Figure 10 highlights spatial differences in model skill (as described by the relative RMSE) within NYNJH, by comparing the quartiles of relative RMSE between the hindcast time series and observed time series at NYC DEP Harbor Survey stations. Even though the Hudson River is one of the most dynamic regions in the estuary, it is also one of the best predicted: J. Mar. Sci. Eng. 2016, 4, 48 15 of 22

J. Mar.larger Sci. Eng. circles 2016, in 4, Figure48 10 show that the relative RMSE is at its lowest quantile for both T and S there.15 of 23 On the other hand, the model had less skill simulating T and S within some small tributaries such salinity.as Newtown These tributaries Creek and Flushingreceive Storm Creek inWater the lower flows and and upper occasional East River, Combined respectively, Sewer and severalOverflows fromtributaries the NYC indrainage Jamaica Baysystem, (Figure non 10‐point), especially sources for that salinity. are presently These tributaries not directly receive simulated Storm Water by the NYHOPSflows andmodel occasional but rather Combined estimated Sewer Overflowsas ungagged from thewatersheds. NYC drainage The system, model’s non-point skill there sources would that are presently not directly simulated by the NYHOPS model but rather estimated as ungagged potentially benefit greatly through coupling to hydraulic forecast models developed for NYC DEP, watersheds. The model’s skill there would potentially benefit greatly through coupling to hydraulic although, for a multi‐decadal hindcast, changes in the City’s drainage system over time may also forecast models developed for NYC DEP, although, for a multi-decadal hindcast, changes in the City’s needdrainage to be accounted system over for: time upgrades may also needto two to‐ betimes accounted dry‐weather for: upgrades‐flow toby two-times the City’s dry-weather-flow treatment plants, sewerby separation the City’s treatment in some plants,areas, sewerand increase separation in retention in some areas, storage and increasein others, in for retention example, storage affect in the spatialothers, and for temporal example, affectdistribution the spatial of and storm temporal water distribution and combined of storm watersewage and discharged combined sewage from the hundredsdischarged of the from City’s the interconnected hundreds of the City’spipes interconnectedand outfalls. pipes and outfalls.

(a) (b)

FigureFigure 10. Quartiles 10. Quartiles of relative of relative RMSERMSE forfor (a (a)) water water temperature and and (b ()b salinity) salinity in thein the NYNJH NYNJH for the for the NYHOPSNYHOPS hindcast hindcast at NYC at NYC DEP DEP Harbor Harbor Survey Survey stations between between 1981 1981 and and 2012. 2012.

The Themodel’s model’s dynamics dynamics werewere able able to captureto capture more variabilitymore variability than the observation-basedthan the observation MOCHA‐based MOCHAv2 climatology v2 climatology as evident as inevident relative in Brier relative Skill ScoresBrier (SkillBSS) Scores that were (BSS positive:) that 0.48were for positive: temperature 0.48 for temperatureand 0.8 for and salinity 0.8 sound-wide.for salinity Thesound only‐wide. exception The was only for salinityexception in the was Sound’s for salinity eastern basin,in the where Sound’s easternthe period-averagebasin, where the relative periodBSS‐averagewas −0.29, relative indicating BSS that was the −0.29, MOCHA indicating v2 climatology that the had MOCHA higher v2 climatologyskill than had the debiased higher skill NYHOPS than model the debiased there (not NYHOPS shown). Even model for that there region’s (not salinity, shown). however, Even for the that region’smodel’s salinity, total RMSEhowever, was the only model’s 1 psu, of total which RMSE the remaining was only bias1 psu, was of 0.6 which psu, thethe averageremaining index bias of was agreement 0.77, and the model’s results were less than 3.5 psu away from observed more than 99% of 0.6 psu, the average index of agreement 0.77, and the model’s results were less than 3.5 psu away the time. As a comparison, the NOS C.F. standard for simulated salinity is so that errors should be from observed more than 99% of the time. As a comparison, the NOS C.F. standard for simulated within 3.5 psu for at least 90% of the time [31]. It is important to note here that results in the Eastern LIS salinitybasin is asso well that as errors the Lower should NY Baybe within appear to3.5 have psu been for at degraded least 90% after of debiasing the time (Table [31].2 ).It Theis important salinity to note biashere correction that results in the in NYHOPS the Eastern model LIS is consistentbasin as with,well thoughas the Lower slightly NY smaller Bay than, appear earlier to models have been degradedin the after region: debiasing Both the (Table LISS 2.0 2). and The SWEM salinity models bias correction used climatological in the NYHOPS boundary model conditions is consistent but with,subtracted though slightly 2.0 psu. smaller It is possible than, that earlier the SODA models climatology in the region: is biased Both high the forLISS salinity, 2.0 and but SWEM the results models usedfrom climatological the two regions boundary closest toconditions the open boundarybut subtracted in Table 2.02 may psu. indicate It is possible otherwise. that Also, the like SODA climatologyearlier models, is biased the NYHOPS high for Hindcast salinity, does but not the include results the from precipitation-evaporation the two regions closest imbalance to the over open boundarythe Sound’s in Table waters, 2 may submarine indicate groundwater otherwise. Also, discharge, like earlier or other models, aquifer-related the NYHOPS freshwater Hindcast sources. does not Althoughinclude thesethe diffuseprecipitation sources‐evaporation have been estimated imbalance to contribute over the a total Sound’s freshwater waters, ﬂow oversubmarine the Sound that is an order of magnitude smaller than the Connecticut River [35,36], they may in any case groundwater discharge, or other aquifer‐related freshwater sources. Although these diffuse sources contribute somewhat to the high-salinity bias in the models. Further research is needed to quantify have been estimated to contribute a total freshwater flow over the Sound that is an order of these contributions, and explore the reason for that consistent bias in models of the region. magnitude smaller than the Connecticut River [35,36], they may in any case contribute somewhat to the high‐salinity bias in the models. Further research is needed to quantify these contributions, and explore the reason for that consistent bias in models of the region. A very important finding of the research is that the NYHOPS model’s skill in simulating water temperature, validated against historic data from the LIS bottom trawl survey, did not drift over the years. The error histograms and PDFs included in Figure 11 for four consecutive periods during both the (a) spring and (b) fall survey periods, show that there was no clear indication of an earlier period having greater error than a later one, or vice versa, unlike what was shown earlier for storm surge. A similar analysis against NYC DEP Harbor Survey data in NYNJH for subsequent five‐year periods

J. Mar. Sci. Eng. 2016, 4, 48 16 of 22

A very important finding of the research is that the NYHOPS model’s skill in simulating water temperature, validated against historic data from the LIS bottom trawl survey, did not drift over the years. The error histograms and PDFs included in Figure 11 for four consecutive periods during both J. Mar. Sci. Eng. 2016, 4, 48 16 of 23 the (a) spring and (b) fall survey periods, show that there was no clear indication of an earlier periodfrom 1983 having to 2012 greater also errordid not than indicate a later drift one, in orskill vice for versa, surface unlike or bottom what T was nor shownS (Supplementary earlier for stormMaterial, surge. Pages A similar 10–18). analysis This is againsta significant NYC DEPand Harborencouraging Survey finding data in for NYNJH multi‐ fordecadal subsequent model five-yearapplications periods used from to identify 1983 to and 2012 research also did climatic not indicate trends drift and incausalities. skill for surface Figure or11 bottomalso showsT nor thatS (Supplementarythe median error Material, in simulated Pages bottom 10–18). Thistemperature is a significant in LIS andwas encouraging low for both finding the spring for multi-decadal (+0.1 °C) and modelthe fall applications (−0.5 °C) surveys, used to identify with only and few research predictions climatic being trends further and causalities. than ±1.5 Figure °C from 11 also observed shows ◦ thatsamples. the median error in simulated bottom temperature in LIS was low for both the spring (+0.1 C) and ◦ ◦ the fallThe (− 0.5lackC) of surveys, model drift with onlywas fewfurther predictions supported being through further comparison than ±1.5 C of from the observed simulated samples. water temperatureThe lack time of model series driftto the was one further and only supported long‐term through water comparisontemperature of record the simulated for Long waterIsland temperatureSound taken time at seriesa near to‐surface the one location and only just long-term outside water the temperatureMillstone Power record Plant for Long at Millstone, Island Sound CT taken(Figure at a12). near-surface The central location estimate just outsidefor the thelinear Millstone trend Powerof the Plantobservations at Millstone, at Millstone CT (Figure (0.439 12). ◦ The°C/decade) central estimatewas somewhat for the higher linear trendthan ofthe the linear observations trend of atthe Millstone NYHOPS (0.439 hindcastC/decade) there (0.313 was ◦ somewhat°C/decade), higher and the than simulated the linear annual trend of range the NYHOPS was somewhat hindcast higher there (0.313than observedC/decade), (Figure and the12), simulatedhowever the annual two range trends was were somewhat not different higher at than the observed95% confidence (Figure level.12), however Figure 12 the shows two trends that water were nottemperature different at was the simulated 95% confidence very level.well throughout Figure 12 shows the hindcast that water period. temperature It also was shows simulated clearly very that welltemperatures throughout have the hindcastbeen increasing period. Itat also the showssite and clearly that that2012 temperatures was the warmest have been year increasing in both the at thesimulated site and and that observed 2012 was record. the warmest year in both the simulated and observed record.

(a) (b)

FigureFigure 11.11. ErrorError histogramshistograms ((toptop)) andand probabilityprobability densitydensity functionsfunctions ((bottombottom)) betweenbetween NYHOPSNYHOPS HindcastHindcast bottom temperature temperature results results and and LIS LIS Trawl Trawl Survey Survey bottom bottom temperature temperature samples samples for four for foursubsequent subsequent spring spring (a) and (a) andfall ( fallb) survey (b) survey periods periods from from 1992 1992 to 2013. to 2013.

J. Mar. Sci. Eng. 2016, 4, 48 17 of 22 J. Mar. Sci. Eng. 2016, 4, 48 17 of 23

FigureFigure 12. 12. MonthlyMonthly mean mean water water temperature temperature ( (toptop)) and and water water temperature temperature anomaly anomaly ( (bottombottom;; after after removalremoval of of seasonal seasonal signal) signal) time time series series near near the the Millstone Millstone Nuclear Nuclear Power Power Plant Plant outfall outfall location location at at Millstone,Millstone, CT, CT, as as observed observed (OBS), (OBS), and simulated (MOD).

NYHOPSNYHOPS model model results results were were then then used used to to quantify quantify linear, linear, water water temperature, temperature, salinity, salinity, and and stratificationstratification trends trends for for LIS LIS in in the the hindcast hindcast period period between between 1981 1981 and and 2013. 2013. Spatially Spatially averaged averaged trends trends overover the the three three Long Long Island Island Sound Sound management management basins basins (east, (east, central, central, west), west), and and the the Sound Sound as as a a whole whole areare shown shown in in Figure Figure 13,13, with with confidence confidence intervals, intervals, while while the the spatial spatial variation variation on on the the NYHOPS NYHOPS grid grid cellcell level level is is seen seen in in Figure 14.14. Statistically Statistically significant significant warming warming and and freshening freshening trends trends (Figure (Figure 13),13), nonnon-stationary‐stationary trends trends in in volumetric volumetric fluxes fluxes across across the the western western and and eastern eastern basins basins of of the the Sound Sound (Figure(Figure 15),15), and and an an associated associated statistically statistically significant significant increase increase in in stratification stratification (Figures (Figures 1313 andand 14)14) havehave all all occurred occurred within within the the hindcast hindcast period. period. Based Based on on the the NYHOPS NYHOPS hindcast, hindcast, in in the the Long Long Island Island SoundSound basin,basin, surface surface air temperatures,air temperatures, contributing contributing river temperatures, river temperatures, and receiving and LIS-basin-wide receiving LISwater‐basin temperatures‐wide water (0.34 temperatures± 0.08 ◦C per(0.34 decade) ± 0.08 have°C per all decade) seen significant have all increases seen significant between increases 1981 and between2013, more 1981 so and on the 2013, shallower more so north on the shore shallower and western north Soundshore and than western the south Sound shore than similar the tosouth [37]. shoreThe basin similar wide to [37]. average The isbasin comparable wide average to the is 0.30 comparable◦C per decade to the rate0.30 reported°C per decade for SST rate at reported the Gulf forof MaineSST at the [3]. Gulf The of increase Maine in[3]. major The increase freshwater in major rivers freshwater mentioned rivers in Section mentioned2 may in have Section led to2 may the haveSound led overall to the becoming Sound overall somewhat becoming fresher somewhat (a statistically fresher significant (a statistically trend ofsignificant 0.12 ± 0.05 trend psu/decade), of 0.12 ± 0.05especially psu/decade), near river especially mouths near at the river surface mouths (Figure at the 14), surface increasing (Figure stratification 14), increasing and changing stratification long-term and changingvolumetric long transport‐term fluxesvolumetric in the transport basin in a statisticallyfluxes in the significant, basin in nonstationary a statistically way significant, (Figure 15). nonstationaryFurther research way is needed(Figure to 15). deduce Further whether research these is trends needed are to expected deduce to whether continue these into thetrends future. are expected to continue into the future.

J. Mar.J. Mar. Sci. Sci. Eng. Eng.2016 2016, 4,, 484, 48 18 of18 23 of 22 J. Mar. Sci. Eng. 2016, 4, 48 18 of 23

FigureFigure 13. 13. Decadal Decadal‐averaged‐averaged linear linear trends trends inin waterwater temperaturetemperature ((toptop),), salinitysalinity (middle),), andand Figure 13. Decadal-averaged linear trends in water temperature (top), salinity (middle), and stratification stratificationstratification (bottom (bottom), for), for the the three three LIS LIS basins, basins, and and the the whole whole Sound Sound withinwithin thethe 1981–2013 NYHOPS (bottom), for the three LIS basins, and the whole Sound within the 1981–2013 NYHOPS Hindcast HindcastHindcast period. period. Circles Circles show show central central estimates estimates whilewhile verticalvertical barsbars showshow thethe 95% confidence period. Circles show central estimates while vertical bars show the 95% conﬁdence intervals. intervals.intervals.

Figure 14. Spatial map of locally computed decadal-averaged linear trends in water temperature (left),

salinity (center), density (right), and stratiﬁcation (bottom) within the 1981–2013 NYHOPS Hindcast period. Surface and bottom are from the surface-most and bottom-most NYHOPS cells, mean is for the vertically averaged trends, and stratiﬁcation is computed from surface to local bottom. J. Mar. Sci. Eng. 2016, 4, 48 19 of 23

Figure 14. Spatial map of locally computed decadal‐averaged linear trends in water temperature (left), salinity (center), density (right), and stratification (bottom) within the 1981–2013 NYHOPS Hindcast period. Surface and bottom are from the surface‐most and bottom‐most NYHOPS cells, mean is for the vertically averaged trends, and stratification is computed from surface to local J. Mar.bottom. Sci. Eng. 2016, 4, 48 19 of 22

Figure 15.15.NYHOPS NYHOPS model model transects transects across across which which linear linear trends intrends volumetric in volumetric cross-channel cross transport‐channel weretransport quantified. were quantified. Transects for Transects which statistically for which significantstatistically trends significant in volumetric trends in fluxes volumetric between fluxes 1981 andbetween 2013 were1981 calculatedand 2013 were (at the calculated 95% confidence (at the level) 95% areconfidence highlighted level) with are blue highlighted squares. with blue squares. 5. Conclusions 5. ConclusionsThe NYHOPS model’s application to hindcast storm surge and 3D water temperature and salinity conditionsThe NYHOPS in its region model’s over three-and-a-half application to decadeshindcast was storm validated surge against and 3D observations water temperature from multiple and agencies,salinity conditions as well as climatology:in its region theover average three‐and index‐a‐ ofhalf agreement decades forwas storm validated surge against alone was observations 0.93 (9 cm ◦ RMSEfrom multiple, 90% of agencies, errors less as than well 15 as cm); climatology: for water the temperature, average index it was of 0.99agreement (1.1 C forRMSE storm, 99% surge of errorsalone ◦ lesswas than0.93 3(9 C);cm and RMSE for, salinity, 90% of it errors was 0.86 less (1.8 than psu 15RMSE cm);, 96%for water of errors temperature, less than 3.5 it psu). was The0.99 model’s(1.1 °C skillRMSE in, simulating99% of errors bottom less than water 3 °C); temperature, and for salinity, validated it was against 0.86 historic (1.8 psu data RMSE from, 96% the of Long errors Island less Soundthan 3.5 bottom psu). The trawl model’s survey, skill did in not simulating drift over bottom the years, water a significant temperature, and validated encouraging against finding historic for multi-decadaldata from the Long model Island applications Sound bottom used to trawl identify survey, climatic did trends.not drift However, over the theyears, validation a significant revealed and residualencouraging biases finding in some for areas, multi including‐decadal smallmodel tributaries applications that used receive to urban identify discharges climatic from trends. the NYCHowever, drainage the validation network. Withrevealed regard residual to the biases validation in some of storm areas, surge including at coastal small stations, tributaries both that the considerablereceive urban strengths discharges and from remaining the NYC limitations drainage of thenetwork. use of With NARR regard to force to such the validation a model application of storm weresurge discussed. at coastal stations, both the considerable strengths and remaining limitations of the use of NARRThrough to force the such comprehensively a model application forced, were and discussed. herein extensively validated, multi-decadal simulation for LongThrough Island the Sound’s comprehensively physical environment forced, and performed herein usingextensively Stevens validated, Institute of multi Technology’s‐decadal NYHOPSsimulation hindcast for Long model, Island temperature Sound’s physical increases environment in Long Island performed Sound over using the pastStevens three-and-a-half Institute of decadesTechnology’s have NYHOPS been confirmed hindcast and model, have beentemperature found to increases be statistically in Long significant. Island Sound The over linear the trends past ◦ havethree‐ alsoand‐ beena‐half found decades to be have quite been high confirmed (0.34 ± 0.08 andC have per decade) been found and comparable to be statistically to ones significant. in the Gulf of Maine [3]. Further research is needed in identifying the cause for this increase in temperatures. Source allocation and further sensitivity runs using the NYHOPS model may aid in testing hypotheses in the future. After extensive model validation and debiasing, an online THREDDS Data Server (http://colossus.dl.stevens-tech.edu/thredds/catalog.html) was set up to serve the NYHOPS model’s J. Mar. Sci. Eng. 2016, 4, 48 20 of 22 results in NetCDF format over the web using the OPENDAP protocol, enabling similar analyses through open access to daily averaged or monthly averaged time series for all the gridded hindcast physical variables in or over the NYHOPS region that includes NYNJH and LIS. Fisheries management has traditionally sought to reduce harvesting levels in response to low stock biomass, in its goal to maintain long-term fishery productivity [38]. More recently, accounting for non-stationary environmental forcing has started being considered in stock assessment and management ([39], pp. 131–147) with the realization that a failure to account for shifts in climate that can alter population dynamics can lead to stock collapse as with the Gulf of Maine cod fishery [3]. This progress has been facilitated by new population dynamics models that consider temperature-dependencies and an improved understanding in climate-fisheries teleconnections brought about through advances in environmental modeling. NYHOPS Hindcast results are in demand to support this kind of fisheries research through coupling to habitat suitability indices, population models, water quality models, or to provide boundary conditions to higher resolution embayment or tributary circulation models. The datasets are also presently used to inform ongoing research in climate teleconnections and ecosystem change through exploratory statistical analyses linking regional fisheries abundance to global climate indices. The authors are encouraged by the early interest in these datasets. Simulated climatologies (mean simulated climate conditions averaged over the three decades of the NYHOPS hindcast period) for two- and three-dimensional fields such as water temperatures and salinities, were also generated, and included in THREDDS. Included on the same THREDDS server is a “daily anomaly” dataset, comparing the latest operational forecast of the NYHOPS OFS model to the 1981–2013 debiased daily climatology, so that interested parties can see how different yesterday and the next three days are predicted to be from the climatological average, enabling near-term tracking of anomalous patterns in Long Island Sound. Fish-surveying strategies may also be improved through the use of the NYHOPS model forecasts: NYSDEC and NMFS routinely already use the model’s predictions for adaptive sampling in the Hudson River estuary and the Mid-Atlantic Bight Apex. Visualization is easy with off-the-shelf free software, such as NASA’s Panoply, accessing the datasets over the web.

Supplementary Materials: The following are available online at www.mdpi.com/2077-1312/4/3/48/s1, Comprehensive Validation and LIS Trend Presentation. Acknowledgments: Financial support was provided by the US EPA Long Island Sound Study and NY and CT Sea Grants through the project R/CE-33-NYCT. Long-term NYHOPS model support has been provided through the NOAA Integrated Ocean Observing System program and many other projects. The authors appreciate the help of CT DEEP personnel and Nicholas Kim from HDR. Inc. in providing CT DEEP and NYC DEP temperature and salinity observations used in this work. Author Contributions: Nickitas Georgas, Penny Howell, and Vince Saba conceived and designed the experiments. Yu Jiang, Larry Yin, and Nickitas Georgas performed the experiments. Larry Yin, Yifan Wang, Penny Howell, Yu Jiang, Bin Wen, and Nickitas Georgas analyzed the data; Philip Orton contributed the SNAP grid and provided model input guidance; Nickitas Georgas wrote the paper. Conﬂicts of Interest: The authors declare no conﬂict of interest. The founding sponsors had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, and in the decision to publish the results.

References

1. NCDC. Climate at a Glance. Time Series. Available online: http://www.ncdc.noaa.gov/cag/time-series/ (accessed on 15 July 2016). 2. United Nations. Paris Agreement. United Nations Treaty Collection. Available online: https://treaties. un.org/pages/ViewDetails.aspx?src=TREATY&mtdsg_no=XXVII-7-d&chapter=27&lang=en (accessed on 15 July 2016). 3. Pershing, A.; Alexander, M.A.; Hernandez, C.M.; Kerr, L.A.; le Bris, A.; Mills, K.E.; Nye, J.A.; Record, N.R.; Scannell, H.A.; Scott, J.D.; et al. Slow Adaptation in the Face of Rapid Warming Leads to Collapse of the Gulf of Maine Cod Fishery. Science 2015.[CrossRef][PubMed] J. Mar. Sci. Eng. 2016, 4, 48 21 of 22

4. Dominion Resource Services. Monitoring the Marine Environment of Long Island Sound at Millstone Power Station. Annual Report 2002. Millstone Environmental Laboratory, April 2003. Available online: http://pbadupws.nrc.gov/docs/ML0409/ML040930039.pdf (accessed on 12 August 2016). 5. Nye, J.A.; Link, J.S.; Hare, J.A.; Overholtz, W.J. Changing spatial distribution of fish stocks in relation to climate and population size on the Northeast United States continental shelf. Mar. Ecol. Prog. Ser. 2009, 393, 111–129. [CrossRef] 6. Howell, P.; Auster, P. Phase shift in an estuarine finfish community associated with warming temperatures. Mar. Coast. Fish. Dyn. Manag. Ecosyst. Sci. 2012, 4, 481–495. [CrossRef] 7. Kleisner, K.M.; Fogarty, M.J.; McGee, S.; Barnett, A.; Fratantoni, P.; Greene, J.; Hare, J.A.; Lucey, S.M.; McGuire, C.; Odell, J.; et al. The Effects of Sub-Regional Climate Velocity on the Distribution and Spatial Extent of Marine Species Assemblages. PLoS ONE 2016.[CrossRef][PubMed] 8. Georgas, N.; Howell, P.; Saba, V.; Schulte, J.; Blumberg, A.F.; Orton, P.; Yin, L.; Jiang, Y.; Wang, Y. Analyzing History to Project and Manage the Future: Simulating the Effects of Climate on Long Island Sound’s Physical Environment and Living Marine Resources. Final Report. NOAA Sea Grant Publications, Long Island Sound Study, 2016. Available online: http://longislandsoundstudy.net/wp-content/uploads/2013/08/Georgas-et- al-R-CE-33-NYCT-CR-Final-Report.pdf (accessed on 15 July 2016). 9. Georgas, N.; Orton, P.; Blumberg, A.F.; Cohen, L.; Zarrilli, D.; Yin, L. The impact of tidal phase on Hurricane Sandy’s flooding around New York City and Long Island Sound. J. Extreme Events 2014.[CrossRef] 10. Marsooli, R.; Orton, P.M.; Georgas, N.; Blumberg, A.F. Three-dimensional hydrodynamic modeling of coastal flood mitigation by wetlands. Coast. Eng. 2016, 111, 83–94. [CrossRef] 11. Georgas, N.; Blumberg, A.F.; Herrington, T.; Wakeman, T.; Saleh, F.; Runnels, D.; Jordi, A.; Ying, K.; Yin, L.; Ramaswamy, V.; et al. The Stevens Flood Advisory System: Operational H3-E flood forecasts for the greater New York/New Jersey metropolitan region. Int. J. Saf. Secur. Eng. 2016, 6, 1–15. 12. Georgas, N.; Blumberg, A.F.; Herrington, T. An operational coastal wave forecasting model for New Jersey and long Island waters. Shore Beach 2007, 75, 30–35. 13. Blumberg, A.F.; Georgas, N. Quantifying Uncertainty in Estuarine and Coastal Ocean Circulation Modeling. J. Hydraul. Eng. 2008, 134, 403–415. [CrossRef] 14. Georgas, N. Establishing Confidence in Marine Forecast Systems: The Design of a High Fidelity Marine Forecast Model for the NY/NJ Harbor Estuary and Its Adjoining Coastal Waters. Ph.D. Thesis, Stevens Institute of Technology, Hoboken, NJ, USA, August 2010; p. 272. Available online: http://web.stevens.edu/ ses/documents/fileadmin/documents/pdf/PhD-dissertation_signed-o.pdf (accessed on 15 July 2016). 15. Georgas, N.; Blumberg, A.F. Establishing confidence in marine forecast systems: The design and skill assessment of the New York harbor observation and prediction system, Version 3 (NYHOPS v3). In Proceedings of the 11th International Conference in Estuarine and Coastal Modeling (ECM11), Seattle, WA, USA, 4–6 November 2010; pp. 660–685. 16. DiLiberto, T.; Colle, B.A.; Georgas, N.; Blumberg, A.F.; Taylor, A.A. Verification of a Multi-Model Storm Surge Ensemble Around New York City and Long Island During the Cool Season. Weather Forecast. 2011, 26, 922–939. [CrossRef] 17. Orton, P.M.; Georgas, N.; Blumberg, A.F.; Pullen, J. Detailed modeling of recent severe storm tides in estuaries of the New York City region. J. Geophys. Res. 2012, 117, C09030. [CrossRef] 18. Wilkin, J.L.; Hunter, E.J. An assessment of the skill of real-time models of Mid-Atlantic Bight continental shelf circulation. J. Geophys. Res. Oceans 2013, 118, 2919–2933. [CrossRef] 19. Blumberg, A.F.; Georgas, N.; Herrington, T.; Yin, L. Street Scale Modeling of Storm Surge Inundation along the New Jersey Hudson River Waterfront. J. Atmos. Ocean. Technol. 2015, 32, 1486–1497. [CrossRef] 20. Blumberg, A.F.; Mellor, G.L. A description of a three dimensional coastal ocean circulation model. In Three-Dimensional Coastal Ocean Models; Heaps, N.S., Ed.; American Geophysical Union: Washington, DC, USA, 1987; Volume 4, pp. 1–16. 21. Mesinger, F.; Dimego, G.; Kalnay, E.; Mitchell, K.; Shafran, P.C.; Ebisuzaki, W.; Jovic, D.; Woollen, J.; Rogers, E.; Berbery, E.H.; et al. North American Regional Reanalysis. Bull. Am. Meteorol. Soc. 2006, 87, 343–360. [CrossRef] 22. Choi, W.; Keuser, A.; Becker, S. Identification of mid-latitudinal regional and urban temperature variabilities based on regional reanalysis data. Theor. Appl. Climatol. 2012, 107, 87–98. [CrossRef] J. Mar. Sci. Eng. 2016, 4, 48 22 of 22

23. Bhushan, S.; Blumberg, A.F.; Georgas, N. Comparisons of NYHOPS Hydrodynamic Model SST Predictions with Satellite Observations in the Hudson River Tidal, Estuarine and Coastal Plume Region. In Proceedings of the 11th International Conference in Estuarine and Coastal Modeling (ECM11), Seattle, WA, USA, 4–6 November 2010; pp. 11–26. 24. Taylor, P.K.; Yelland, M.A. The dependence of sea surface roughness on the height and steepness of the waves. J. Phys. Oceanogr. 2001, 31, 572–590. 25. Georgas, N. Analyzing History to Project and Manage the Future: Simulating the Effects of Climate on Long Island Sound’s Physical Environment and Living Marine Resources. Supplementary Matter to Year 1 Report. Available online: http://www.stevens.edu/ses/documents/ﬁleadmin/documents/pdf/R-CE-33-NYCT% 20PR1%20supporting%20year%201%20PR.pdf (accessed on 15 July 2016). 26. Carton, J.A.; Giese, B.S. A reanalysis of ocean climate using Simple Ocean Data Assimilation (SODA). Mon. Weather Rev. 2008, 136, 2999–3017. [CrossRef] 27. Chassignet, E.P.; Hurlburt, H.E.; Metzger, E.J.; Smedstad, O.M.; Cummings, J.A.; Halliwell, G.R.; Bleck, R.; Baraille, R.; Wallcraft, A.J.; Lozano, C.; et al. US GODAE: Global Ocean Prediction with the HYbrid Coordinate Ocean Model (HYCOM). Oceanography 2009, 22, 64–75. [CrossRef] 28. Seekell, D.A.; Pace, M.L. Climate change drives warming in the Hudson River estuary, New York (USA). J. Environ. Monit. 2011, 13, 2321–2327. [CrossRef][PubMed] 29. Kaushal, S.S.; Likens, G.E.; Jaworski, N.A.; Pace, M.L.; Sides, A.M.; Seekell, D.; Wingate, R.L. Rising stream and river temperatures in the United States. Front. Ecol. Environ. 2010, 8, 461–466. [CrossRef] 30. Fleming, N.E.; Wilkin, J. MOCHA: A 3-D climatology of the temperature and salinity of the Mid Atlantic Bight. Eos Trans. AGU 2010, 91, PO35G-08. Available online: http://abstractsearch.agu.org/meetings/2010/OS/ PO35G-08.html (accessed on 15 August 2016). 31. NOAA Technical Report NOS CS 17. NOS Standards for Evaluating Operational Nowcast and Forecast Hydrodynamic Model Systems. Available online: www.nauticalcharts.noaa.gov/csdl/publications/TR_ NOS-CS17_FY03_Hess_OFS-Standards.pdf (accessed on 12 August 2016). 32. Willmott, C.J. On the validation of models. Phys. Geogr. 1981, 2, 184–194. 33. Taylor, K.E. Summarizing multiple aspects of model performance in a single diagram. J. Geophys. Res. 2001, 106, 7183–7192. [CrossRef] 34. Von Storch, H.; Zwiers, F.W. Statistical Analysis in Climate Research; Cambridge University Press: Cambridge, UK, 1999; p. 494. 35. O0Donnell, J.; Wilson, R.E.; Lwiza, K.; Whitney, M.M.; Bohlen, W.F.; Codiga, D.; Fribance, D.B.; Fake, T.; Bowman, M.; Varekamp, J. The Physical Oceanography of Long Island Sound. In Long Island Sound. Prospects for the Urban Sea; Springer: New York, NY, USA, 2014; pp. 79–158. 36. Garcia-Orellana, J.; Cochran, J.K.; Bokuniewicz, H.; Daniel, J.W.R.; Rodellas, V.; Heilbrun, C. Evaluation of 224Ra as a tracer for submarine groundwater discharge in Long Island Sound (NY). Geochim. Cosmochim. Acta 2014, 141, 314–330. [CrossRef] 37. Lee, Y.J.; Lwiza, K. Interannual variability of temperature and salinity in shallow water: Long Island Sound, New York. J. Geophys. Res. 2005, 110, C09022. [CrossRef] 38. Mahon, R.; McConney, P.; Roy, R.N. Governing ﬁsheries as complex adaptive systems. Mar. Policy 2008, 32, 104–112. [CrossRef] 39. Northeast Fisheries Science Center Reference Document 14-04. 58th Northeast Regional Stock Assessment Workshop (58th SAW) Assessment Report. Available online: http://www.nefsc.noaa.gov/publications/crd/ crd1404/crd1404.pdf (accessed on 15 August 2016).

© 2016 by the authors; licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC-BY) license (http://creativecommons.org/licenses/by/4.0/).