Earth System Modeling Framework ESMF and the Transformation

Earth System Modeling Framework ESMF and the Transformation of Earth System Modeling Sylvia Murphy murphys@ucar. edu http: //www. esmf. ucar. edu 1

Outline § § § What is ESMF? Who is Using it? How is ESMF Transforming Earth System Modeling? Current Development Status Future Directions 2

What is ESMF? ESMF is a software library that consists of coupling tools and model utilities that increase modularity, interoperability, and performance of geoscience applications. 3

What models are using ESMF? Operational Version of the GFS Weather Forecasting Model GEOS-5 Atmospheric GCM HYCOM Ocean Model COAMPS Atmospheric Model CICE-HYCOM MIT GCM WRF Weather Forecasting Model ADCIRC Hydrodynamic model ADCIRC-WASH 123 GFDL FMS MOM 4 UCLA AGCM 58 components are currently using ESMF 4

What programs are using ESMF? NUOPC: National Unified Operational Prediction Capability Consortium (NOAA, Navy, Air Force Weather Agency) Image from NOAA rapid refresh (RR) model, a replacement model for the Rapid Update Cycle (RUC) model. Has announced intentions to base its next generation National Weather Prediction (NWP) system on ESMF BEI: Battlespace Environments Institute Developing integrated forecasting systems across Do. D. MAP: Modeling and Analysis Program Facilitating research and model use of NASA data 5

The Transformation of Earth System Modeling ESMF is changing: § § How models are structured How models are developed The scale of collaborations Governance strategies for developing community scientific software 6

How Models are Structured § Hierarchical component based architecture § Interchangeability § Easy replacement of component elements § Interoperability § Swapping component between applications § Coupling diverse applications 7

How Models are Developed § Large number of components available from across the community § Develop science components for a well-understood community architecture and interfaces § Technological advances are propagated quickly across many applications simultaneously 8

The Scale of Community Collaborations BEI: Battlespace Environment Institute NUOPC: National Unified Operational Prediction Capability CCSM: Community Climate System Model Collaborations across: § Operational centers (e. g CNMOC, COPC, FNMOC) § Research centers (e. g. GFDL, NCAR, NASA Goddard) § Universities (e. g. MIT, University of Hawaii) § Different domains (weather, climate, space weather, hydrology) WRF: Weather Research & Forecasting Model MAP: Modeling Analysis and Prediction Program 9

Governance strategies for developing community scientific software Shared governance model emphasizes: § Distributed, global development § Direct participation of users in setting development priorities § Multi-agency management § Incorporates open-source methodologies into scientific software development esmf_jst@ucar. edu has 300 subscribers who participate in ESMF development 10

ESMF Release Plan 2002 2003 2004 2005 2006 2007 2008 2009 ESMF v 1 Prototype ESMF v 2 Components, VM and Utils ESMF_Grid. Comp. Run() ESMF v 3 Index Space Operations ESMF_Array. Sparse. Mat. Mul() ESMF v 4 Grid Operations ESMF_Grid. Create() ESMF_Field. Regrid() ESMFv 5 Standardization Build, init, data types, error handling, … Last public Last internal ESMF v 2. 2. 2 r. ESMF v 3. 0. 3 11 2010

Current Features (v 3. 0. 3) § § § Component-based architecture Allows serial or parallel execution Components can run concurrently or sequentially Single executable support, limited support for multiple executables Support for model ensembles User friendly uniform, rectilinear, and curvilinear grid representation Unstructured and multi-patch grids represented in index space Regridding via sparse matrix multiply with user defined weights Utilities such as time manager, message logger, and resource file manager are usable and adding features Fortran interfaces and complete documentation, increasing number of C interfaces Distributed with Reference Manual, Users Guide, ~2500 tests and examples ~2000 downloads 12

Current Supported Platforms § § § § § Cray X 1 Cray XT 3/4 HP ZX 6000 Linux intel IBM Bluegene, IBM Opteron Linux pathscale IBM Opteron Linux pgi IBM SP MAC G 5 Darwin xlfgcc MAC G 5 Darwin nag § § § § § MAC G 5 Darwin absoft Pentium III Linux lahey Pentium III Linux pgigcc PC Xeon Cluster Linux absoftintel PC Xeon Cluster Linux g 95 PC Xeon Cluster Linux intel PC Xeon Cluster Linux nagintel SGI Altix Linux nagintel SGI Origin 3800 ESMF runs 2500 tests on all these platforms every night 13

Sparse Matrix Multiply Updated § User supplied weights § Facilitates the widest possible couplings § The methods have the following new features: § Improved performance and reduced memory usage § Support for distributed definition of factors § Support for user-supplied arbitrary sequence indices via 1 D Dist. Grid § Support for ESMF_TYPEKIND_I 4/I 8/R 4/R 8 § Default initialization of destination Array to zero (can be overridden) 14

New Structured Grids Image of HYPSE (Hydrostatic Padua Surface Elevation Model) Dept. of Physics, Univ. of Padua § Support 1 D, 2 D, 3 D and higher dimensions § Supports parallelization across any or all grid dimensions § Supports uniform, rectilinear and curvilinear grids 15

Near Future: Structured Grids § Multi-tile grids (e. g. cube-sphere) § More efficient memory use for non -curvilinear grids § Coordinate halo § A wide variety of different types of connections between tile edges § Automatic generation of topology and coordinates for common shapes (e. g. sphere) Image courtesy Steve Thomas and Rich Loft, NCAR 16

Near Future: General Regridding Curl of bilinear interpolant Exact curl Curl of patch-recovered interpolant Work conducted by David Neckels/ESMF in collaboration with Bill Large of the ocean modeling section at NCAR. 17

Near Future: Unstructured Grids § Uses same infrastructure as general regridding § Representation for hydrology grids and other grids § Future basis for exchange grids and observational data streams Image of rossby waves on a triangular adaptive grid (Alfred. Wegener-Institute for Polar and Meteorological Forecasting) 18

Longer Term: Interoperability & Integration Initiative § Enable groups with compilations of ESMF components to develop integrated modeling systems § Define standards to increase component interoperability § Metadata conventions § Strategies for automation and supplemental tools (e. g. MAPL) http: //esmfi 3. org/wiki/ 19

Questions? http: //www. esmf. ucar. edu 20