Second International Conference on Subseasonal to Seasonal Prediction (S2S) and
Second International Conference on Seasonal to Decadal Prediction (S2D)

Daily news

Day 4: Frontiers in weather and climate research and needs of stakeholders

The fourth day of the S2S and S2D conferences showcased a variety of topics. Before joining back together for a joint plenary session, the S2S theme focused on ocean and land initialization and processes, while the S2D participants wrapped up their conference by discussing frontiers in Earth system predictions. Once the two groups reunited, they approached topics on model initialization, initialization shock, and errors; research to operations; and timescale interactions. Below are some highlights from the day.

Subseasonal to Seasonal Conference Highlights

While there was much focus on the atmosphere’s role in S2S predictions, throughout the conference, and specifically on this day, the role of land and ocean processes were discussed and their coupled interactions with the atmosphere. Presenters also discussed current and future analysis, modeling, and prediction skill concerning S2S predictability.

Paul Dirmeyer, George Mason University, provided an entertaining talk on the “sweet spots” of weather and climate forecasting that land processes provide a source of predictability. Land anomalies in soil moisture, snow, and vegetation, which vary on low-frequency timescales, can affect the atmosphere from weeks to months and are a source of predictability beyond the deterministic limit for weather forecasts from national weather prediction models. He also noted that reliable, high-quality real-time observations of land surface states are now becoming widely available to better initialize the models. Using an analogy, Dirmeyer compared how freight companies deliver packages or the internet delivers data to how the atmosphere can deliver predictable S2S phenomena in much of the same way via persistent large-scale circulation features. The biggest problems are often in the last mile, where the cost and most of the failures occur, such as if the land surface is poorly initialized or land-atmosphere coupling is not well represented.

Focusing on the ocean, Ramalingam Saravanan, Texas A&M University, discussed how oceanic fronts and mesoscale eddies can be important on the S2S timescale through their influence of storm tracks. Oceanic mesoscale eddies in the mid-latitudes affect surface winds and boundary layer depth and have a non-linear impact on moisture and rainfall. Local and remote atmospheric responses to mesoscale SST anomalies, associated with the oceanic front and eddies, can be properly simulated by models that has sufficient resolution (order of 25 km or less) to resolve the small-scale diabatic heating, so the full effect of mesoscale SST forcing on the storm track can be correctly simulated.

Duane Waliser, NASA Jet Propulsion Laboratory, led a discussion on the role of observations in the modeling and forecast capabilities on the S2S timescale. He described how there are opportunities from the US Decadal Survey of Earth Science and Applications from Space for prioritizing observations, and that there could be better and practical uses of model experiments for observation system development (i.e., develop model experiments with observation systems in mind for the second phase of S2S Prediction Project). Closer communication between the S2S modeling and observation communities is needed. Participants made subsequent suggestions for S2S phase 2 including: (i) addressing the need for observations, (ii) advancing the understanding of the systematic impact of ocean data assimilation on the understanding of underlying physical processes and forecast skill in the S2S range, and (iii) enabling closer connection between the modeling and observation communities, as well as users.

Additional highlights include:

• India is a well-known hotspot of land-atmosphere coupling, where realistic land surface initializations from S2S ensemble model simulations yield small precipitation and wind fields biases over Indian summer monsoon region. (Obbe Tuinenburg, Utrecht University)

• The land surface, initialized from calculated surface variables using reanalysis, fails to capture heat wave events over Europe and Russia in subseasonal to monthly timescales using the UK Met Office GloSea5 seasonal forecast system. (Philip Davis, UK Met Office)

• Ocean observation systems have potential for improving subseasonal predictions by addressing inaccuracies in coupled air-sea interaction processes relevant to the S2S predictability range. However, ocean data assimilation in coupled models can still be challenging, while their influence on subseasonal predictability still needs to be systematically demonstrated, as in the case of sea ice, and underlying processes better understood. (Aneesh Subramanian, Scripps Institution of Oceanography; Chidong Zhang, NOAA Pacific Marine Environmental Laboratory)

Seasonal to Decadal Conference Highlights

Thursday morning saw the last two of the individual-track sessions: “Hindcast and forecast quality assessment,” that carried over from Wednesday, and  “Frontiers in Earth system prediction.”

Where Wednesday’s contributions had focused on technical aspects, such as appropriate statistical tests and skill measures, Thursday’s last two talks looked at skill measures for domain problems, namely forecasts of Atlantic hurricane activity (Louis-Philippe Caron, Barcelona Supercomputing Center) and seasonal sea ice forecasts (Steffen Tietsche, ECMWF). The speakers noted that Atlantic hurricane predictions by global climate models are at least as good as 10-year persistence and statistical forecasts, and that initializing sea ice thickness using observational data generally improves Arctic sea ice extent and edges in hindcast experiments. This improvement could influence the atmospheric variables, such as surface temperatures.

The concluding session was dedicated to “Frontiers in Earth system prediction.” Hongmei Li, Max Planck Institute for Meteorology Germany, opened with a keynote talk on decadal predictability of the ocean carbon uptake variation.

Both Li and and subsequent speaker Tatiana Ilyina, Max Planck Institute for Meteorology Germany, addressed variability on decadal timescales of the observed global mean ocean carbon uptake. Such effects can be reproduced by Earth system models in which only physical variables are assimilated. Retrospective forecasts show a predictive skill in ocean carbon uptake up to two years. Further talks in the session highlighted:

• Biogeochemistry prediction with different models, including corresponding initialization, data assimilation, and their potential utility for marine resources and fisheries;

• Long-lasting prediction skill for global carbon fluxes up to 6–7 years, where drivers of predictability are related to delta pCO2, dissolved inorganic carbon, and alkalinity; and

• Predictability beyond 12 months of regional winter sea ice extent can be enabled by advanced initialization schemes.

Joint S2S and S2D Plenary

During the first session, contributions were given on evaluating and addressing model systematic and initialization errors in the Earth system components and data assimilation. Alicia Karspeck, Jupiter Intelligence, discussed the science and skill of near-term climate predictions over the past decade, and new methods and trends being employed for CMIP6. Decadal prediction has evolved between CMIP5 and CMIP6, although the system initialization has many commonalities between the two exercises. Initialization is one of the main aspects that receives attention, while the forecast drift is considered the main problem affecting the progress of decadal prediction.

On the research to operations front, Debbie Hudson, Australian Bureau of Meteorology, showcased thoughtful engagement with user communities in their country. She indicated that many of their users want a simple outlook, while others (often a more specialized subset) will want the data and finer details. Providing an example, Hudson showed how a rural research and development for profit project – the aim was to help manage impacts to extreme events for the agriculture sector – provided forecast information that was customized to the user needs. In a first step, extensive communication was done with the stakeholders to assess which types of and how to provide data. Next, they developed frameworks for communicating the forecast data and its uncertainties, helping them determine what actions should be taken to reduce their risk. Often the can be done through case studies rather than skill scores. She emphasized that the process with users in iterative and integrated; it’s not a linear process.

Additional highlights include:

• Forecast initialization of models is an important step towards improving the forecast, although big challenges remain. It is possible to produce more skillful predictions at the extended and seasonal range by correcting model bias during forecast phase. (Magdalena Balmaseda, ECMWF)

• Process understanding is required in addition to forecast quality assessment. And this process understanding can be benefited from a better representation of model error during the simulation.

• Model resolution to eddy-resolving grids is considered necessary not only to better simulate ocean-atmosphere interactions but also to have the correct impact of predictable ocean processes on the atmosphere. However, the initialization at those timescales is particularly difficult due to the cost of data assimilation, in addition to the difficulties to obtain adequate observations. (Ben Kirtman, University of Miami)

• The US Navy Earth system model (NESM) is to provide predictions in the range from day to month by 2022. NESM uses a scheme that is informed by process studies that improves the parametrization of convection and the performance for MJO prediction. The model performance for the Pacific North American Oscillation, North Atlantic Oscillation, and sea ice is in in the range of other forecast systems. (Carolyn Reynolds, Naval Research Laboratory)

• Analysis of WRF model runs based on the CFSR reanalysis was used to analyse the interaction of MJO, ENSO, and the diurnal cycle. This approach reduces rainfall biases in the tropics and reveals that ENSO and the Indian Ocean dipole interact with MJO on long timescales. This can also affect the variability of the diurnal cycle of tropical rainfall. (Tieh-Yong Koh, Singapore University of Social Sciences/Nanyang Technological University)

• The surface temperatures in the North America dipole are connected to the large-scale circulation. The upper level jet-stream position is modulated by topography and diabatic surface heating such that the flow often generates a trough and ridge pattern over North America. Additional variability also comes from internal jet fluctuations as well as tropical and extratropical forcings. This circulation is very complex, that it is hard to analyze if the mean jet position or the persistence of the flow, and therefore surface temperature extremes change with global warming. (Simon Wang, Utah State University)