NERSCPowering Scientific Discovery Since 1974

NERSC Initiative for Scientific Exploration (NISE) 2011 Awards

Improving Algorithms and Resolution for the Ocean-Atmosphere Reanalysis for Climate Applications OARCA (1850-2012)

Gil Compo, University of Colorado at Boulder

Associated NERSC Project: Surface Input Reanalysis for Climate Applications (SIRCA) 1850-2012 (m958)

NISE Award: 7,000,000 Hours
Award Date: March and June 2011

Long-term records of the global weather and climate variations from the 19th to 21st century are urgently needed for assessing the strengths and weaknesses of the next-generation of coupled climate models being used to project the effects of anthropogenic greenhouse gas emissions for the upcoming Fifth Assessment Report of the Intergovernmental Panel on Climate Change (IPCC AR5). Importantly, such records must have quantified uncertainties to allow a quantitative assessment. The Ensemble Kalman Filter method to be used as part of the Ocean-Atmosphere Reanalysis for Climate Applications (OARCA) provides the needed uncertainty estimates. In this project, new algorithms for improving these estimates will be tested in the atmosphere and ocean components of the system. Increased resolution in the horizontal and vertical may reduce the uncertainty in the dataset and will also be tested. The expected outcome is progress towards an OARCA system to generate a record of the state of the global atmosphere and ocean back to 1850 at six-hourly resolution in the atmosphere and 5-day resolution in the ocean.

With our previous grants of NERSC and OLCF cpu hours, we have assembled the 20th Century Reanalysis dataset (20CR; Compo et al., Quart. J. Roy. Met. Soc., 2011) spanning 1871-2008. The 20CR is the longest record of global weather and its uncertainty for use in meeting national and international goals for climate assessment and climate model validation. In a companion effort, our colleagues at Texas A&M University have generated a version 2 of the Simple Ocean Data Assimilation (SODA) dataset of 5-day averaged ocean states over the same period, using the 20CR as input. While they represent a significant achievement that is already seeing results in important areas such as understanding El Nino (Giese et al., Bull. Amer. Met. Soc., 2009), the 1930’s US Drought (Cook et al., Clim. Dyn., 2010), and Arctic warming (Wood and Overland, Int. J. Climatol., 2010), both datasets could be improved and extended back in time with revisions to their algorithms. Newly recovered observations of the historical atmosphere and ocean (e.g., via citizen-science through, professional international initiatives such as and ) will also enhance and extend future versions of both datasets. Improvements in the Ensemble Kalman Filter (EnKF) algorithm used in 20CR will allow us to extract more information from the available observations. Additionally, by coupling the 20CR and SODA systems, further improvement may be possible. Under the new DOE/NSF Earth System Modeling project Ocean-Atmosphere Reanalysis for Climate Applications (DOE PI- Gil Compo, University of Colorado; NSF PI-Ben Giese, Texas A&M) such improvements have been proposed as a way to meet national and international calls for reanalysis datasets extending to the start of the instrumental record. To test the new 20CR EnKF system, and the innovative ocean coupling concept, significant computational resources are needed on the reliable HPC systems of NERSC. This testing and algorithm development will allow us to answer some important scientific and technical questions, and to prepare the software system for the major new revision of the reanalysis dataset in 2012.

Four sets of experiments are planned in this NISE project. 1) We will test the 2011 version of the coupled atmosphere/land/ice Global Forecasting System (GFS) model from the National Centers for Environmental Prediction. Since the 2008 GFS used in 20CR, significant improvements have been made to the treatment of radiation and both shallow and deep convection. Additionally, we will contrast the 20CR EnKF performance using the new GFS at the same low-resolution as in 20CR (total spherical wavenumber “T62” ,~2 degree longitude by ~2 degree latitude and 28 vertical levels) with the new GFS at quadruple the resolution in the horizontal and more than twice the resolution in the vertical (total spherical wavenumber 254 “T254”, ~0.5 degree latitude by longitude and 64 vertical levels).

We expect the higher resolution to improve the representation of extratropical storms, tropical cyclones and convection, and the stratospheric circulation. We have found the T254 is the minimum resolution for a realistic simulation of tropical cyclones. One year of reanalysis at T62 and T254 will be generated for testing. 2) New algorithms for representing model and sampling uncertainty will be tested at the 20CR EnKF system. These tests will include the effect of uncertainty in the specified sea surface temperatures (SST), which is an important component of the total uncertainty. An adaptive method for accounting for model uncertainty through “covariance inflation” will be tested in the 20CR EnKF system using only surface pressure observations (One year of assimilation, 2 inflation algorithms, UK Met Office Hadley Centre HadISST and Texas A&M SODAv2 boundary conditions, T254 resolution). 3) Using the best of the algorithms examined in 2, we will test the incorporation of historical ship and buoy based surface wind observations, along with pressure observations in the high-resolution system. In addition to possibility improving the reanalysis fields, successful incorporation of wind observations may make historical reanalysis back to the 18th century possible, as much of the available observations before 1850 are of wind. (One year of assimilation, 1 algorithm, HadISST and SODAv2 boundary conditions, T254 resolution). 4) An experimental SODA version 3 dataset for the one year period with most promising of the above experiments will be generated. We will then repeat that experiment using the new SODAv3 boundary conditions to further analyze the effect of the uncertainty in the boundary conditions on the atmospheric reanalysis fields. We will also generate an additional SODAv3 dataset using this second version of the improved high-resolution 20CR EnKF fields to determine the likely improvements resulting from a fully-coupled system.