Playing ‘frisbee’ with seasonal predictions

Playing ‘frisbee’ with seasonal predictions

 

 

This blog is written by Timo Kelder (@timokelder), a 3rd year PhD student in climate science at Loughborough University. Here, he writes about the history of frisbees, and the parallels with his recently published paper on using seasonal predictions for detecting recent trends in rare extremes.

Ultimate frisbee is now a well-established sport, with over 70,000 British participants in 2019. The photo shows how this began in the 1970s by skimming disks. But how did this trend start?

Frisbees were invented by Fred Morrison, who loved pie tins, but for throwing not baking. While hovering the pie tin in a park, he soon found out that many others were intrigued. After some tweaks and twists to the design, frisbees became the popular flying disks they are known for today.

Similarly, Van den Brink et al., 2005 used the ECWMF seasonal prediction system in an unconventional way – instead of making predictions of the future state of the weather, the authors employed (re-) forecasts as “alternate versions of the past” to reduce uncertainties in flood estimation for the River Rhine.

A further tweak was added by Thompson et al., 2017, who named the approach the UNprecedented Simulated Extreme ENsemble (UNSEEN). UNSEEN is a great way of capturing the essence of the idea: extreme events exist in the observed record, but what about events we have not seen but could have occurred? The forecast ensemble essentially gives us many parallel worlds and so a much larger sample of plausible extreme events.

It is crucial, however, to assess whether simulated extremes really are plausible UNSEEN events as opposed to unrealistic events. In our recent paper, we provide a framework to statistically evaluate UNSEEN results based on ensemble member independence, model stability and model fidelity.

Ensemble member independence is required for all events in the ensemble to be unique. The model must be stable as not to express any model drift over time. And model fidelity is essential for the large ensemble to be consistent to observations.

Image

We believe there are other opportunities for using this large UNSEEN sample. For example, devastating extreme precipitation events over Norway and Svalbard raised questions about whether the risk of such hazards had changed over recent decades. Using the larger UNSEEN dataset – including events that might not have been seen in records – we are able to detect changes in rare extreme events, which is not possible from observations alone.

Image

During summer 2020, we made a further twist to UNSEEN during the ECMWF Summer of Weather Code (esowc_2020). We developed a reproducible UNSEEN-open workflow using Copernicus C3S seasonal (re-) forecasts so anyone can explore UNSEEN events, globally. We outlined four potential applications:

  1. Estimating extreme values for engineering design, especially in data scarce regions
  2. Improving risk estimation of natural hazards by coupling UNSEEN to impact models
  3. Detecting trends in rare climate extremes
  4. Advancing understanding of the physical drivers of (non-stationary) climate extremes

With these twists and tweaks to traditional seasonal forecasting, we believe that UNSEEN may see many applications across a range of scientific fields. Like the frisbee which originated from the humble pie tin, so UNSEEN applications are adding a novel ‘spin’ to seasonal prediction archives.

 

Read more:

Kelder, T. et al. Using UNSEEN trends to detect decadal changes in 100-year precipitation extremes. npj Clim Atmos Sci 3, 47 (2020). https://doi.org/10.1038/s41612-020-00149-4

Social media photo credit: Daniel Jordahl, no changes made, CCBY2.0

 

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.