Testing multicomponent multiphysics climate models Stephen Haddad, Scientific Software Engineer 7 September 2017 www.metoffice.gov.uk © Crown Copyright 2017, Met Office
Introduction & Overview Introduction: Testing to verify reproducible climate model output Overview: Met Office Hadley Centre (MOHC) climate models Initial state of development and testing Model Restartability Future developments www.metoffice.gov.uk © Crown Copyright 2017, Met Office
MOHC Models – HadGEM3 Scientific configurations GA7, GL7, GO6, GSI8, GC3.1 Component applications & libraries UM, NEMO, CICE Auxiliary: XIOS, oasis3-mct Workflow: rose, post-processing, drivers Developed by consortia www.metoffice.gov.uk © Crown Copyright 2017, Met Office
NEMO UM CICE JULES XIOS UM IOS, Stash Main technical components OASIS3- MCT Cycling CICE JULES Archiving, Post-processing X nodes Y nodes Main technical components Embedded components Increased Technicality Rose suites Coupler IO subsystems www.metoffice.gov.uk © Crown Copyright 2017, Met Office
Development hierarchy Global Coupled (GC3) Global Atmosphere (GA7) Unified Model Global Land-use (GL7) Jules Global Ocean (GO6) NEMO Global Sea (GSI8) CICE www.metoffice.gov.uk © Crown Copyright 2017, Met Office
Initial State of Models Rigorous, formal, operational forecast requirements UM, NEMO Informal, research-oriented CICE, XIOS, scientific configurations Disparity between tests and climate experiments Goal: check technical requirements of coupled model Starting point: state of atmosphere model www.metoffice.gov.uk © Crown Copyright 2017, Met Office
Testing Scientific Software Validation vs Verification Scientist – validation: Model correctly describes physical processes SSE – verification Model is correctly implemented Meets technical requirements e.g. performance, restartability www.metoffice.gov.uk © Crown Copyright 2017, Met Office
Bit-comparable output LRUN CRUN NRUN Day TS Day TS Day TS Restarting a model 1 1 1 1 1 1 6 161 6 1 10 320 Simulation broken into cycle Quick and full restart modes Cycle length arbitrary, technical Identical output for different modes Checkpoint restart driven by scientific needs 11 321 11 321 11 1 16 481 16 1 20 640 21 641 21 641 21 1 26 721 26 1 30 960 Bit-comparable output www.metoffice.gov.uk © Crown Copyright 2017, Met Office
Implementing restart tests Component tests (rose stem tests) Small and fast Run frequently Climate experiment (suite tests) Closer to actual configuration More resource-hungry www.metoffice.gov.uk © Crown Copyright 2017, Met Office
Implementation Challenges Getting scientists to understand there is a problem! Missing documentation Additional bugs www.metoffice.gov.uk © Crown Copyright 2017, Met Office
Current state Restart tests in components and model configuration Nightly tests to verify new developments www.metoffice.gov.uk © Crown Copyright 2017, Met Office
Future work Testing additional models & requirements UK Earth System Model (UKESM) Processor decomposition Exascale programme – Lfric Talk by Matthew Hambley www.metoffice.gov.uk © Crown Copyright 2017, Met Office
Conclusion and questions www.metoffice.gov.uk © Crown Copyright 2017, Met Office