Tuesday, April 25, 2017

Workshop on High-Resolution Coupling and Initialization to Improve Predictability and Predictions in Climate Models

September 30 - October 2, 2015

NCWCP Conference Center
5830 University Research Court
College Park, Maryland

Supercell thunderstorm and tornadoes in the GFDL Super High-Resolution Atmosphere Model (Super HiRAM). Learn more...

Organizers: Brian Medeiros (NCAR), Shian Jiann Lin (NOAA/GFDL), Bill Putman (NASA), Travis O Brien (LBNL), Steve Klein (LLNL), Jim Kinter (COLA/GMU), Steve Penny (UMD/NCEP), Annarita Mariotti (NOAA/CPO), Renu Joseph (DOE)


Executive Summary


There is a growing demand for reliable climate predictions (intra-seasonal to decadal) and projections (decadal and longer, including secular trends) at regional and local scales. In recent decades, weather prediction skill has dramatically improved due to a combination of factors, including the use of higher spatial resolution, improved physics, and better methods in data assimilation. In contrast, seasonal prediction skill has modestly improved compared to our understanding of the processes underpinning predictability.

Coarse resolution climate models do not properly represent potentially important coupled phenomena, such as interactions between tropical cyclones and their wakes and coupling between low clouds and small-scale ocean temperature gradients, as well as interactions between the land surface and the atmosphere, the sea surface and the atmosphere, among many others. There are indications that resolving such phenomena would improve the simulated climate, improve simulation of extremes, and increase confidence in climate predictions and projections. Research has shown, however, that high-resolution is not a panacea for all problems and needs to be accompanied by improved physical process modeling at the appropriate scale.

There have been a number of recent efforts that use initialized climate models to address issues from model development to decadal predictability. For example, the North American Multi-Model Ensemble (NMME) system has made significant progress in improving seasonal to interannual predictions. The last Coupled Model Intercomparison Project (CMIP5) had a major component dedicated to evaluating climate models in a decadal hindcast mode. In addition, climate models used for climate predictions and projections are often configured in weather prediction mode to identify climate biases and test new configurations since many long-term biases manifest within a few days and initialized simulations are valuable when comparing to process-level observations collected in field-campaigns (e.g. the Transpose-AMIP component used in CMIP experiments or DOE’s Cloud-Associated Parameterization Testbed - CAPT). In all these cases, models are typically run at their standard, relatively coarse, resolution (i.e. 1-degree horizontal grid spacing or coarser) and none of them specifically addresses the potential advantages or challenges of using higher-resolution coupled systems. The communities involved in these modeling efforts are exploring the use of high-resolution coupled systems, but these communities are quite separate and enhanced communication among them could help move these efforts forward.

The need to improve climate prediction and its realism at local to regional scales, and the need to identify and reduce biases in the coupled system in climate models motivate the need to explore climate prediction techniques in coupled models at higher resolution. The majority of the work done to date in prediction and the use of initialized climate models focuses on the benefits of higher resolution in atmosphere-only models (e.g. research using the CAPT framework at multiple DOE labs). There has been relatively less work done to explore the benefits of higher resolution in coupled models, and even less in the context of using realistically-initialized coupled simulations. Pioneering work in this area is ongoing at NOAA’s GFDL, by COLA in collaboration with ECMWF, and at the UK Met Office, among others. Previous work in the context of the CAPT, has shown the benefits to understanding atmospheric processes and biases in the atmospheric model component alone. However the importance of understanding the relative roles of process-level representation and resolution on biases in climate models in a coupled framework has received less attention. Results suggest that coupled processes and associated predictability in a given modeling system may be altered by better resolving the relevant processes even as important deficiencies remain in representing atmospheric and oceanic processes. Experiments have been proposed as part of CMIP6 to examine the impact of increasing resolution in Atmospheric Model Intercomparison Project (AMIP) style experiments (HR-MIP) and also in long-term non-initialized coupled simulations. However, an experimental design to investigate the impact of resolution on initialized coupled simulations used for climate has so far not been included.

To examine the benefits to prediction of both modeling and initializing at higher resolution, several scientific and practical questions will need to be addressed by the various communities involved using initialized climate modeling. These are best tackled first by individual modeling groups but over time it may be worthwhile to consider a multi-model setting, which requires multi-center coordination, so as to yield theoretical and practical results that are applicable to the wide array of models. A common core experimental framework for experiments and analysis, carefully defined with scientific community involvement could facilitate evaluation and comparison of model strengths and weaknesses. Initial exchanges of ideas could identify for what purposes a common experimental modeling framework would be useful. These discussions would include how to design appropriate numerical experiments that can reliably test hypotheses given the necessary trade-offs in resources (especially high-performance computing; HPC) between resolution, length of predictions, ensemble size and complexity of model processes.

A Workshop on High-Resolution Coupling and Initialization to Improve Predictability and Predictions in Climate Models is proposed to enhance communication among the initialized simulation communities, summarize the current status of the research, and to probe a potential experimental framework that would optimally address major pressing questions in the context of available computing resources. The main questions will include, but are not limited to:

  1. How does prediction skill and fidelity change when resolution is increased in combination for the various components of the prediction system? How can we diagnose and address model behaviors that lead to the sensitivity? Are there specific processes in the coupled system that drive both prediction error and simulation bias?
  2. Under what conditions is an initialized coupled modeling system (like the CAPT, or Transpose-AMIP) a useful framework for identifying the processes that lead to long-term systematic biases in coupled models and hinder their use for climate prediction? Do initialized coupled model simulations reveal the fast coupled processes, such as rapid development of flux errors, contributing to long-term biases? What timescales should be targeted by such efforts? What initialization methods should be used to address predictability issues?
  3. What is the ideal size of the ensemble needed at various time scales for this effort, both for prediction and for understanding coupled processes and biases?
  4. What resolution is feasible given the state-of-art HPC systems available to the U.S. community? How will increasingly high-resolution data be stored and shared for community research?
  5. How can observations best be used to assess and initialize high-resolution coupled-model simulations for prediction? Should new or different observations be collected to aid these efforts?

DOE and NOAA managers and key modeling center and community representatives, involving other relevant US agencies and scientists, aim to organize the proposed workshop to include about 30 participants.

Envisioned workshop outcomes are:

  1. Interaction and sharing of knowledge between two communities that conduct similar work but for different final outcomes (predictability vs. prediction)
  2. An assessment and synthesis of the current status of the communities using realistically-initialized climate models;
  3. Establishing under what conditions and for what purposes it would be suitable to design a multi-model experimental framework to systematically and optimally address major questions about the use of high-resolution in initialized coupled climate models.

A workshop report is expected within 30 days of the workshop.

The Modeling, Analysis, Predictions, and Projections (MAPP) Program's mission is to enhance the Nation's capability to understand and predict natural variability and changes in Earth's climate system. The MAPP Program supports development of advanced climate modeling technologies to improve simulation of climate variability, prediction of future climate variations from weeks to decades, and projection of long-term future climate conditions. To achieve its mission, the MAPP Program supports research focused on the coupling, integration, and application of Earth system models and analyses across NOAA, among partner agencies, and with the external research community.

Learn more...

Download our program brochure (pdf). 

Upcoming Events

4/26/2017 11:00 AM - 12:00 PM

MAPP Webinar Series: NOAA Unified Modeling Task Force Overview

The NOAA CPO Modeling, Analysis, Predictions, and Projections (MAPP) program will host a webinar on the topic of an overview of the NOAA Unified Modeling Task Force on Wednesday, April 26, 2017. The announcement is provided below.