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

Daily news

Day 5: Future directions for subseasonal to decadal prediction research

After a week of invigorating discussions and a wealth of information presented, conference attendees listened to a few final talks on the interactions of timescales from subseasonal to decadal, highlights from throughout the week, and final remarks on where there is convergence, what questions remain unanswered, and ideas for future research directions. The energy from the week was evident as attendees cued up at the microphones to share their final thoughts with the meeting organizers. Below are the highlights from the final day.

Session on S2S and S2D timescale interactions

Masahide Kimoto, Tokyo University, was the keynote presenter and discussed the predictability of blocking and tropical cyclone activities. He has investigated the predictability of high-impact weather events (mid-latitude blocking and tropical cyclones) and their variation on the S2S timescale, using large ensemble experiments performed with the Japanese Meteorological Research Institute’s global climate model. The model is able to identify ENSO modulation on subseasonal variability, which he used to investigate the influence of El Niño on blockings over the Kamchatka peninsula and tropical storms off the coast of Japan. For blocking, he showed that the number of events centered over the Kamchatka peninsula correlate with the Niño3 index, and during La Niña years the blocking pattern is shifted further to the west. He also used accumulated cyclone energy to study the link between El Niño and tropical cyclone activity, and showed that this accumulated energy correlated with El Niño. He concluded that large ensembles are useful to explore forced modulations of subseasonal variability of phenomena, such as blockings, tropical cyclones, and other high-impact events.

Tim Woollings, Oxford University, related the predictability of the North Atlantic Oscillation (NAO) during winter to jet-stream variability. He detected interannual and decadal variability in the NAO, and associated it with variations in the jet-stream position and intensity. Changes in the jet latitude and speed influence the NAO in different ways and through different mechanisms. The NAO strength is affected by the variability of the jet intensity: strong jets are more stable, while weaker jets are more variable. In strong-jet situations, the NAO predictability is higher (physically, the jet speed influences Rossby-wave breaking, and thus the stability and intensity of the NAO). Variations in the jet positions influence the frequency of the NAO and, thus, its interannual variability.

Antje Weisheimer, ECMWF/Oxford University, presented the multidecadal variability in seasonal predictive skill of the Winter NAO, using seasonal forecasts from the ECMWF coupled reanalysis of the 20th century (CERA-20C). She showed that there is a strong multidecadal variability skill in predicting winter NAO, both in seasonal ensembles run in uncoupled and coupled (3-dimensional ocean and sea ice) ECMWF models. There is a stronger correlation of NAO seasonal predictions for the last 30 years than the average over the last century, due to the decreased skill in the period between 1955–1979, which may be associated with a weakening of the tropical forcing with the NAO. The increased predictability of the NAO during the most recent decades could be linked to the Arctic amplification mechanism. These results indicate that care must be taken when drawing conclusions on the predictability of low-frequency patterns, such as the NAO, using only a few decades.

On the topic of tropical/extratropical interactions, Stephanie Henderson, University of Wisconsin-Madison, showed how a Linear Inverse Model (LIM) can approximate the evolution of a dynamical system and identify initial-time patterns that project on final-time structures. In one case, LIM was used to trace back the initial pattern that over a 15-day period would project onto the Pacific/North American teleconnection pattern (PNA). Results showed that initial condition patterns in the tropics, linked to heating due to active Madden-Julian Oscillation, and in the extratropical circulation are important for PNA pattern growth.

Nirupam Karmakar, Florida State University, investigated the impact of intraseasonal oscillations on the onset and the demise phases of the Indian summer monsoon rainfall. He identified two types of intraseasonal oscillations, one with a low (20–60 day) and a second with a high (10–20 day) frequency. Using eight phases, he showed how the onset or demise of a monsoon phase occurs, and that 59% of the onset and demise phases are explained by the intraseasonal oscillations.

Final Conference Discussion

Gokhan Danabasoglu, NCAR, chaired the final remarks and discussions for the conference. Andrew Robertson, IRI/Columbia University, and Doug Smith, UK Met Office, presented the key takeaways from the S2S and S2D concurrent conferences, respectively (see other day highlights for outcomes from the conferences). Some topics identified as areas of research to be addressed include: the adverse impacts of persistent model biases on predictability and prediction skill; the need for more process-level understanding; taking advantage of the improved skill with land initialization; high-resolution modeling efforts are showing mixed messages, so identifying the challenges and sources of agreement; and for most applications, a modeling ensemble of greater than 10 members is needed.

The general discussion with the audience elevated a number of messages.

• There has not been enough progress at solving the adverse impacts of persistent model biases. Groups working at different timescales should team up to solve outstanding model problems. Overall, more people and resources could be dedicated to the S2S and S2D efforts to address these issues.
• We can’t expect postdocs and early career scientists to solve outstanding model problems if their jobs are not secured and properly rewarded. We need to identify opportunities to help support early career researchers in this field. For example,organizing a summer school may be a good step forward to stimulating early career scientists’ minds to work on S2S and S2D model improvement.
• There is a disconnection between the expected forecast and the correct forecast. We need probabilistic methods of verification.
• There may be some value in diagnosing and correcting model errors early in time before they leap forward. However, these are difficult to diagnose and correcting them may not necessarily improve the long-term forecast.
• We have to be clean in the methodology that gets used. Different methodologies cause difficult comparison and model improvement. It would be very useful to have coordinated toolkits and databases.
• There is a need to improve how we communicate our science, interact with users, and simplify the access to the data.
• The so-called “signal-to-noise paradox” suggests the real world is more predictable than previously thought. We need to be precise in how we communicate the problems we are trying to address.
• Much more work in process understanding has to be done. And in order to progress in process understanding, the model outputs should contain enough fields and keep the data as open as possible.
• It is necessary to keep joint frameworks and conversations ongoing between research and operational communities.

Conference by the numbers

• 347 Participants
• 224 Poster Presentations
• 144 Oral Presentations
• 92 Early Career Scientists
• 38 Countries
• 5 Days
• 2 Conferences

Many thanks to all the organizers, sponsors, volunteers, presenters, and participants who made this conference a success! Presentations and videos will be made available in the coming weeks.