Research towards the next generation of NOAA Climate Reanalyses
Principal Investigator:
Dr. Arun Kumar, NOAA, National Centers for Environmental Prediction, arun.kumar@noaa.gov
Co-Principal Investigators:
Dr. Gilbert P. Compo, CIRES/NOAA ESRL Lead Investigator, University of, Colorado/CIRES and NOAA ESRL, gilbert.p.compo@noaa.gov
Dr. Jeffrey S. Whittaker, NOAA, Earth System Research Laboratory, jeffrey.s.whitaker@noaa.gov
Dr. Prashant D. Sardeshmukh, University of Colorado/CIRES and NOAA ESRL, prashant.d.sardeshmukh@noaa.gov
Dr. Russell Vose, NCDC Lead Investigator, NOAA/National Climatic Data Center, russell.vose@noaa.gov
Summary: Prepare for 4 streams of atmospheric reanalyses back to 1850 that improve upon the previous generation of NOAA reanalyses.
Abstract
The fidelity of new reanalysis datasets (MERRA, 20CR, CFSR, ERA-Interim) at representing climate variability of the 20th century has enabled significant advances in climate research. In this research proposal, we will investigate known shortcomings of these datasets, while developing a framework for a new NOAA Climate Reanalysis (NCR) system to ameliorate them.
The NCR will eventually have four “streams” to meet the various user needs for reanalysis
information:
Stream 0: Boundary-forced, 1850-present “AMIP” simulation with large ensemble
Stream 1: Historical, 1850-present using only surface data
Stream 2: Modern, 1946-present using only surface and conventional upper air data
Stream 3: Satellite, 1973-present using quality-controlled satellites, Global Positioning
System Radio Occultation, and surface and conventional upper air data.
One of the foci of this research will be to use observing system experiments. In these, the 2000-2010 observing system is reduced to that of selected historical periods to investigate the impact to the time-varying quality and density of the observing system and determine ways to reduce this impact. We will use innovative methods to assess the relative importance and impact of model errors and observational errors on the quality and homogeneity of the reanalysis fields, with particular attention to reducing or eliminating spurious jumps and trends. The framework for the NCR system will leverage recent advances in operational data assimilation for global weather prediction, as well as newly digitized observational datasets and global model improvements. While initially focusing on the atmosphere to develop the NCR framework, this project will serve as the basis for further NCR efforts, incorporating advancements generated by other projects supported by MAPP, such as integration of ocean, chemistry, and land components and the treatment of observational and model biases. International coordination and data sharing with NOAA's reanalysis partners at NASA, ACRE, ECWMF, and JMA and synergies from the NOAA Reanalysis Task Force will be crucial in achieving the project's goals on a limited budget.
Progress:
Work that is currently going on at NCEP:
1. Hybrid EnKF data assimilation system is running on GAEA; The present configuration NCEP is running is at T254/L64;
The diagram below illustrates some problems in porting the EMC operational system to GAEA, mostly having to do with queue wait for batch jobs. The transition to a one job per day format has been accomplished for the 3DVAR system, doubling the throughput of that system to over 20days/real day. This infrastructure is being fitted to the hybrid/ENKF system, and along with some other optimizations, a similar improvement in process is expected.
2. At present, two streams of reanalysis are in progress. One is from 1981 and one is from 1998. The one from 1981 covers the period with more conventional data while the one from 1998 covers a period with larger contribution from the satellite data. With the ingest of AMSU data starting in 1999, transition across 1999 was also problematic, and a sharp discontinuity in time evolution of many variables analysed occured.
The figure below shows comparisons of 2 GAEA 3DVAR runs with CFSR, MERRA(orange), and ERA-I looking at 1hPa arctic temperature during a transition from SSU to AMSU instruments in late Oct1998. The black lines traces the result of the 3DVAR which mistakenly bias corrected SSU#3 during Oct1998. The reaction was surprisingly dramatic. The green shows the same system with the correct methodology applied. The GAEA runs assimilate AMSU ch14 (w/o bias correction) and track ERAI and MERRA fairly well after the transition.
For both the streams, a control 3DVAR and a Hybrid EnKF is being run, and their comparison will facilitate improvements in Hybrid EnKF over the 3DVAR. Both versions will also be compared with the existing CFSR product; a comparison of anomaly correlation scripes for Hybrid ENKF(prhl4), 3DVAR(prhl5) and CFSR are show below.
3. We are also testing a lower resolution version of 3DVAR (e.g., T126/L48) to increase the throughput and assess feasibility of an assimilation system that could replace NCEP legacy reanalyses that are currently operational - R1 & R2;
4. Setting up a stream to test Hybrid EnKF over a time window with sparse data.
Jack Woollen presentation on progress