lable at ScienceDirect

Environmental Modelling & Software 67 (2015) 26e30

Contents lists avai

Environmental Modelling & Software

journal homepage: www.elsevier .com/locate/envsoft

Short communication

Ocean modeling with flexible initialization for improved coupledtropical cyclone-ocean model prediction

Richard M. Yablonsky*, Isaac Ginis, Biju ThomasGraduate School of Oceanography, University of Rhode Island, 215 South Ferry Road, Narragansett, RI 02882, USA

a r t i c l e i n f o

Article history:Received 18 September 2014Received in revised form3 January 2015Accepted 5 January 2015Available online

Keywords:Ocean circulationNumerical modelCoupled modelParallel computingTropical cyclone (hurricane) forecastingModel initialization

* Corresponding author. Tel.: þ1 401 874 6508; faxE-mail addresses: rmyablon@mail.uri.edu (R.M. Ya

(I. Ginis), bijuthomas@mail.uri.edu (B. Thomas).

http://dx.doi.org/10.1016/j.envsoft.2015.01.0031364-8152/© 2015 Elsevier Ltd. All rights reserved.

a b s t r a c t

This paper introduces the Message Passing Interface Princeton Ocean Model for Tropical Cyclones(MPIPOM-TC), created at the University of Rhode Island (URI). MPIPOM-TC is derived from a combinationof the parallelized version of the Princeton Ocean Model (POM), called the Stony Brook Parallel OceanModel (sbPOM), and URI's non-parallelized POM for Tropical Cyclones (POM-TC), which has been usedfor many years as the ocean component of NOAA's and the U.S. Navy's operational hurricane forecastmodels. In addition to parallelization, the flexible initialization capabilities of MPIPOM-TC and otherelements of its architecture will facilitate further improvements to the ocean component of research-based and operational tropical cyclone (hurricane) forecast models worldwide.

© 2015 Elsevier Ltd. All rights reserved.

Software availability

Software name: MPIPOM-TCDevelopers: Richard Yablonsky, Biju Thomas, and Isaac GinisContact: wrfhelp@ucar.eduSoftware requirements: Linux or Unix, Fortran compiler, mpich 2,

parallel NetCDF, NetCDFProgramming language: Fortran 77/90Availability and cost: Available for free since September 2014 as a

component of the community HWRF hurricane model athttp://www.dtcenter.org/HurrWRF/users/downloads/index.php

1. Overview

Tropical cyclones (hurricanes) are known to generate vigorousresponses in the open ocean, which include upper-ocean currentsof 1e2 m s�1 and sea surface temperature (SST) cooling of severaldegrees. Ocean models that simulate these responses are animportant component of coupled tropical cyclone-ocean predictionsystems. The motivation to couple an ocean model to a hurricane

: þ1 401 874 6728.blonsky), iginis@mail.uri.edu

model is to create accurate sea surface temperature (SST) andcurrent fields for input into the hurricane model. An uncoupledhurricanemodel with a static SST field is restricted by its inability toaccount for SST changes during model integration, which oftencause high intensity prediction bias (e.g. Bender and Ginis, 2000).Similarly, a hurricane model coupled to an ocean model that doesnot include fully three-dimensional ocean dynamics only accountsfor some of the hurricane-induced SST changes during modelintegration (e.g. Price, 1981; Yablonsky and Ginis, 2009, 2013).Finally, two-way hurricane-ocean model coupling allows for real-time prediction of the ocean response under hurricane forcing.

The three dimensional, primitive equation, numerical oceanmodel that has become widely known as the Princeton OceanModel (POM) was originally developed in the late 1970s and 1980s(Blumberg and Mellor, 1987). In 1994, a version of POM available atthe time was transferred to the University of Rhode Island (URI) forthe purpose of coupling to NOAA's Geophysical Fluid DynamicsLaboratory (GFDL) hurricane model to improve 3e5 day hurricaneforecasts (Bender and Ginis, 2000). Since operational imple-mentation of the coupled GFDL/POM model in 2001, additionalchanges to POM were made at URI and subsequently implementedin the operational GFDL model, including improved ocean initiali-zation (Falkovich et al., 2005; Bender et al., 2007; Yablonsky andGinis, 2008). This POM version (now known widely as POM-TC)was also coupled to NOAA's Hurricane Weather Research and

Fig. 1. History of POM and MPIPOM-TC development from the late 1970's to 2014.

Table 1Key differences between the sbPOM code and the MPIPOM-TC code.

File name Changes from sbPOM to MPIPOM-TC

pom.f Only used if uncoupled from hurricane model; if coupled, drive_ocinitialize.f Options added for f-plane, spatial smoothing and desmoothing (Be

wave-dependent turbulence, and non-zero initial currents (geostroinput constants.

advance.f Prescribed surface forcing only applied when uncoupled from themode interaction, external mode, vertical velocity, horizontal advedesmoothing of temperature, salinity, and currents optionally appl

bounds_forcing.f Maximum depth and the number of vertical sigma levels are not san A-grid.

solver.f Missing MPI exchanges added; A-grid used for 1-D only dynamics;parallel_mpi.f None, except for two minor changes when coupled to the hurricanio_pnetcdf.F I/O parameters added; print times and output directory changed; spom.h Many new parameters added (see comments in code for details).pom.nml Ten name list parameters added to control new and enhanced cap

Fig. 2. MPIPOM-TC world

R.M. Yablonsky et al. / Environmental Modelling & Software 67 (2015) 26e30 27

Forecasting (HWRF) model (Yablonsky et al., 2015), which becameoperational in 2007.

Separate from the URI-based changes to POM for hurricaneapplications, both Princeton-based and community-based im-provements to and applications for POM were made between 1994and 2012, including development of pom2k (Mellor, 2004) and theaddition of Message Passing Interface (MPI) routines to allow POMto run in parallel on multiple processors, as in the Stony BrookParallel Ocean Model (sbPOM; Jordi and Wang, 2012) and theAdvanced Taiwan Ocean Prediction (ATOP) system (Oey et al.,2013). With the availability of expanded computing resources atNOAA, URI has now merged the 2012 versions of the sbPOM andPOM-TC and made additional improvements to allow for flexibleinitialization options and a relocatable grid, culminating in the

ean.f used.nder et al., 1993), 1-D (vertical only) dynamics on an A-grid (vs. 3-D C-grid),phic or prescribed) and/or sea surface height; alternative values for

hurricane model; added missing MPI exchanges; omitted lateral viscosity,ction, and horizontal diffusion for 1-D only dynamics; smoothing andied; temperature growth eliminated; model fails if SST is unreasonably high.pecified here; new boundary conditions added for 1-D only dynamics on

wave breaking energy disabled; wave-dependent turbulence optionally applied.e model.yntax fixed.

abilities.

wide ocean domains.

R.M. Yablonsky et al. / Environmental Modelling & Software 67 (2015) 26e3028

MPIPOM-TC for coupled hurricane-ocean models (Fig. 1), whichbecame operational in the GFDL and HWRF hurricane forecastmodels in 2014.

2. Model description

MPIPOM-TC is an end-to-end ocean modeling system, with bothresearch and operational applications, that includes generation of

Fig. 3. HWRF intensity error statistics (kt, upper left), track error statistics (nm, center left)Atlantic hurricane seasons, run with either the 2013 baseline coupled with POM-TC (H131, rblue); courtesy of NOAA's Environmental Modeling Center (EMC). HWRF coupled with MPItrack differences are small. MPIPOM-TC SST (�C) cold wake (upper right), TRMM Microwavewake difference (lower right), generated by Hurricane Katia's forcing through 00 UTC 08 Sgreater than POM-TC, consistent with a reduced positive intensity bias. (For interpretation oof this article.)

an initial ocean condition, a procedure for prescribed hurricanewind forcing, and/or two-way coupling between the ocean and aprognostic hurricane model (e.g. GFDL or HWRF). The MPIPOM-TCcode is built on the parallelized sbPOM framework (Jordi andWang,2012), including the main driver (pom.f), problem setup (ini-tialize.f), code advancement in time (advance.f), boundary condi-tions and forcing (bounds_forcing.f), basic dynamics (solver.f),parallel tasks for communications between processors

, and intensity bias statistics (kt, lower right) using 655 test cases from the 2010e2012ed), as in Yablonsky et al. (2015), or the 2013 baseline coupled with MPIPOM-TC (H134,POM-TC reduces the positive intensity bias compared to HWRF coupled with POM-TC;Imager 3-day averaged observed SST (center right), and MPIPOM-TC e POM-TC SST coldeptember 2011. MPIPOM-TC cooling compares well with observations and is generallyf the references to colour in this figure legend, the reader is referred to the web version

R.M. Yablonsky et al. / Environmental Modelling & Software 67 (2015) 26e30 29

(parallel_mpi.f), parallel I/O (io_pnetcdf.F), model domain size anddecomposition (pom.h), and name list (pom.nml). Key differencesfrom sbPOM, many of which are derived from POM-TC, are sum-marized in Table 1 and commented in the code with the authors'initials.

Beyond the changes summarized inTable 1,MPIPOM-TC includesseparate files to provide specific capabilities during hurricane-model-coupled and/or uncoupled simulations, including pre-scribed hurricane wind forcing (hurricane_wind.f) using NOAA'sTropical Cyclone Vitals Database (Trahan and Sparling, 2012) or theInternational Best Track Archive for Climate Stewardship (Knappet al., 2010), spatial smoothing and desmoothing (smoothing.f),wave-dependent turbulence (calcpwave.f), and initially geostrophiccurrents (geovel.f). For coupled simulations only, eight additionalfiles are required: atm_ocm.f, cmp.comm.f, make_om_grid.f,make_om_msk.f, mp_arr_pros_om.f, mpi_more.f, ocm_atm.f, andpom.comm.f. Like sbPOM, MPIPOM-TC is scalable with a wide vari-ety of domain decompositions (Jordi and Wang, 2012).

3. Model initialization and configuration

An innovative aspect of MPIPOM-TC is the development offlexible, plug-and-play, Fortran-based initial condition modules. Inthe 2014 operational GFDL and HWRF models, the initial conditionmodule in the Atlantic Ocean (fbtr) involves feature-based modi-fications to the U.S. Navy's Generalized Digital EnvironmentalModel (GDEM) monthly temperature (T) and salinity (S) clima-tology (Yablonsky and Ginis, 2008; Teague et al., 1990), followed byassimilation of a daily SST product (Reynolds and Smith, 1994); theinitial condition module in the eastern North Pacific Ocean (gdm3)uses GDEMv3 (Carnes, 2009), assimilated with daily SST (Reynoldsand Smith, 1994). Initial condition modules have also been devel-oped using the stand-alone Navy Ocean Data Assimilation (NCODA)daily T and S fields (ncda; Cummings, 2005; Cummings andSmedstad, 2013) and versions of the HYbrid Coordinate OceanModel (HYCOM) that use NCODA (e.g. hycu; Chassignet et al., 2009).All of these ocean products are available in the public domain forreal time tropical cyclone forecasting. An idealized initial conditionmodule (idel) with user-specified T and S, flat bathymetry, and noland is available too.

To simulate tropical cyclones worldwide, each MPIPOM-TCinitial condition module is relocatable to regions around theworld. Currently, these regions include the Transatlantic, East Pa-cific, West Pacific, North Indian, South Indian, Southwest Pacific,Southeast Pacific, and South Atlantic domains (Fig. 2), as well as theidealized domain. Domain overlap helps to prevent loss of oceancoupling. To avoid domain-specific code (pom.h), all worldwidedomains are set to the same size: 869 (449) longitudinal (lat-itudinal) grid points, covering 83.2� (37.5�) of longitude (latitude)and yielding a horizontal grid spacing of ~9 km. Horizontal domaindecomposition is 3 � 3, yielding 291 (151) local grid points on eachof 9 processors. There are 23 full sigma levels from the sea surfaceto the sea floor, with higher vertical resolution in the mixed layerand upper thermocline.

4. Model evaluation and concluding remarks

The primary purpose of MPIPOM-TC development is to facilitatefurther improvements to both operational and research versions ofhurricane forecast models (e.g. GFDL, HWRF, and the U.S. Navy'sversion of the GFDL model [GFDN]). Initial operational MPIPOM-TCimplementation in GFDL and HWRF was completed in 2014 for theTransatlantic and East Pacific basins after (1) rigorous evaluation ofnumerous hurricane track and intensity forecasts using eitherPOM-TC or MPIPOM-TC (e.g. Fig. 3), (2) scaling and performance

tests to ensure that the forecast completes within the operationaltime window, and (3) comparison of individual forecasts to bothremotely-sensed and in situ ocean observations (e.g. Fig. 3;Yablonsky et al., 2015). Worldwide MPIPOM-TC operationalimplementation in GFDL, HWRF, and GFDN is planned after furthertesting and evaluation. The MPIPOM-TC framework described herewill allow for more rapid improvements and transition of researchto operations (R2O) in the future, including: development andevaluation of alternative initialization modules; physics improve-ments (especially to the turbulence closure model to properly ac-count for wave-dependent turbulence); three-way couplingbetween MPIPOM-TC, a hurricane model, and a wave model (e.g.WAVEWATCH III, Tolman, 2014); and high-resolution nesting,especially at the coastline to forecast storm surge. While these newcapabilities may improve forecasts of hurricanes and their impacts,R2Omust consider a balance between the best model configurationand the computational efficiency required to produce timely fore-casts with available resources. Nonetheless, advanced MPIPOM-TCconfigurations that cannot be immediately transitioned to opera-tions may be used for research and further model development,including future R2O. In addition, simplified MPIPOM-TC configu-ration options (e.g. 1D only dynamics and idealized initial condi-tions) can isolate specific physical processes to guide future modelimprovements.

Acknowledgments

We thank Ligia Bernardet's HWRF team at the DevelopmentalTestbed Center (DTC) for facilitating operations to research (O2R)by making MPIPOM-TC available as part of the 2014 communityHWRF system. We thank Vijay Tallapragada's NOAA/NWS/NCEP/EMC/HWRF team and Morris Bender's NOAA/OAR/GFDL team forfacilitating 2014 operational MPIPOM-TC implementation. Finally,we thank Antoni Jordi for sbPOM assistance. This work was fundedby two NOAA HFIP grants and one NOAA JHT grant awarded to URI(NA12NWS4680002, NA14NWS4680023, and NA13OAR4590192),as well as a DTC Visitor Program Award to Richard M. Yablonsky.

References

Bender, M.A., Ross, R.J., Tuleya, R.E., Kurihara, Y., 1993. Improvements in tropicalcyclone track and intensity forecasts using the GFDL initialization system. Mon.Weather Rev. 121, 2046e2061.

Bender, M.A., Ginis, I., 2000. Real case simulation of hurricane-ocean interactionusing a high-resolution coupled model: effects on hurricane intensity. Mon.Weather Rev. 128, 917e946.

Bender, M.A., Ginis, I., Tuleya, R., Thomas, B., Marchok, T., 2007. The operationalGFDL coupled hurricane-ocean prediction system and a summary of its per-formance. Mon. Weather Rev. 135, 3965e3989.

Blumberg, A.F., Mellor, G.L., 1987. A description of a three-dimensional coastal oceancirculation model. In: Heaps, N. (Ed.), Three-Dimensional Coastal Ocean Models.Am. Geophys. Union 4, 1e16.

Carnes, M.R., 2009. Description and Evaluation of GDEM-V 3.0. Naval ResearchLaboratory, Stennis Space Center. MS 39529, 21 pp. [NRL/MR/7330e09-9165].

Chassignet, E.P., Hurlburt, H.E., Metzger, E.J., Smedstad, O.M., Cummings, J.,Halliwell, G.R., Bleck, R., Baraille, R., Wallcraft, A.J., Lozano, C., Tolman, H.L.,Srinivasan, A., Hankin, S., Cornillon, P., Weisberg, R., Barth, A., He, R., Werner, F.,Wilkin, J., 2009. US GODAE: global ocean prediction with the hybrid coordinateocean model (HYCOM). Oceanography 22, 64e75.

Cummings, J.A., 2005. Operational multivariate ocean data assimilation. Q. J. R.Meteor. Soc. 131, 3583e3604.

Cummings, J.A., Smedstad, O.M., 2013. Variational data assimilation for the globalocean. In: Park, S., Xu, L. (Eds.), Data Assimilation for Atmospheric, Oceanic andHydrologic Applications, vol. II. Springer, pp. 303e343.

Falkovich, A., Ginis, I., Lord, S., 2005. Ocean data assimilation and initializationprocedure for the coupled GFDL/URI Hurricane prediction System. J. Atmos.Ocean. Technol. 22, 1918e1932.

Jordi, A., Wang, D.-P., 2012. sbPOM: a parallel implementation of Princeton OceanModel. Environ. Model. Softw. 38, 59e61.

Knapp, K.R., Kruk, M.C., Levinson, D.H., Diamond, H.J., Neumann, C.J., 2010. Theinternational best track archive for climate stewardship (IBTrACS): unifyingtropical cyclone best track data. Bull. Am. Meteor. Soc. 91, 363e376.

R.M. Yablonsky et al. / Environmental Modelling & Software 67 (2015) 26e3030

Mellor, G.L., 2004. Users Guide for a Three-dimensional, Primitive Equation, Nu-merical Ocean Model (June 2004 Version). Prog. in Atmos. and Ocean. Sci.,Princeton University, 56 pp.

Oey, L., Chang, Y.-L., Lin, Y.-C., Chang, M.-C., Xu, F., Lu, H.-F., 2013. ATOP e theadvanced Taiwan ocean prediction system based on the mpiPOM. Part 1: modeldescriptions, analysis and results. Prog. Atmos. Ocean. Sci. 24, 137e158.

Price, J., 1981. Upper ocean response to a hurricane. J. Phys. Oceanogr. 11, 153e175.Reynolds, R.W., Smith, T.M., 1994. Improved global sea surface temperature analyses

using optimum interpolation. J. Clim. 7, 929e948.Teague,W.J.,Carron,M.J.,Hogan,P.J.,1990.Acomparisonbetweenthegeneralizeddigital

environmental model and Levitus climatologies. J. Geophys. Res. 95, 7167e7183.Tolman, H.L., 2014. User Manual and System Documentation of WAVEWATCH III

Version 4.18. Tech. Note 316, NOAA/NWS/NCEP/EMC/MMAB, 282 pp. þAppendices.

Trahan, S., Sparling, L., 2012. An analysis of NCEP tropical cyclone vitals and po-tential effects on forecasting models. Weather Forecast. 27, 744e756.

Yablonsky, R.M., Ginis, I., 2008. Improving the ocean initialization of coupledhurricane-ocean models using feature-based data assimilation. Mon. WeatherRev. 136, 2592e2607.

Yablonsky, R.M., Ginis, I., 2009. Limitation of one-dimensional ocean models forcoupled hurricane-ocean model forecasts. Mon. Weather Rev. 137, 4410e4419.

Yablonsky, R.M., Ginis, I., 2013. Impact of a warm ocean eddy's circulation on hur-ricaneeinduced sea surface cooling with implications for hurricane intensity.Mon. Weather Rev. 141, 997e1021.

Yablonsky, R.M., Ginis, I., Thomas, B., Tallapragada, V., Sheinin, D., Bernardet, L.,2015. Description and analysis of the ocean component of NOAA's operationalhurricane weather research and forecasting (HWRF) Model. J. Atmos. Ocean.Technol. 32, 144e163.