ICEX 2018
ICEX is a five-week biennial exercise conducted by the US Navy that allows the Navy to assess operational readiness in the Arctic, increase experience there, advance understanding of the Arctic environment, and continue to develop relationships with other services, allies and partner organizations.
ACCAP was involved in ICEX2018. Brettschnieder’s analog forecasting technique was used for daily forecasts and for historical break-up match for when sea ice would deteriorate to a point where the camp was no longer usable. His technique was used to prepare daily forecasts by matching sea level pressure initialization and 5-day forecast with historical sea level pressure fields using the NCEP/NCAR R1 Reanalysis data set.
This technique was used to generate a 5-day forecast for the ICEX camp location. The program Brettschnieder wrote ingested the 12 UTC data and systematically evaluated the dynamic model initial conditions (hour-0) against the 12 UTC conditions for each day in the R1 reanalysis period (since 1948).
The best analog matches from previous years during each season were determined and a point forecast was generated using the best analog matches. The output was loaded to a publicly available webpage for ICEX members to view or to generate a new forecast using different parameters.
Additionally, a tabular list (Historical Breakup Match) was generated during each forecast run showing the correlation score and percentile match of each day in the numerical guidance versus the known breakup dates in 2014 and 2016 and the dates leading up to the known breakup events.
The daily analog forecast allowed for near real-time ability to compare numerical guidance with historical analogs, the ability to generate user-friendly output for multiple uses on the fly, and ability to contextualize the pattern evolution as shown in the models, with historical pattern evolution. This product really provided a “reality check” for the forecast team when assessing raw numerical output.
Analog-based forecasting captured the evolution of the sea level pressure pattern in the Beaufort Sea nearly every day of the ICEX project. Using monthly data, analogs outperform numerical models north of 70°N quite consistently.
One important thing to note is that the size and shape of the selection domain is critical for identifying analogs that are useful for generating a point forecast. A large domain is ideal for finding analogs that will generate good forecasts several days out – but sacrifices short-term forecast skill. The opposite is also true. For the ICEX project period, the analogs-based forecast did not “beat” the numerical guidance at forecast ranges of one to several days. However, at Days 3, 4, and 5, analogs nearly outperformed the numerical guidance. This is an extremely significant finding. If an analogs approach can outperform numerical guidance, the applications are nearly limitless.
For the second product (the Historical Breakup Match), most of the ICEX staff participated in earlier ICEX programs and have an “institutional knowledge” of the previous breakup events in 2014 and 2016. The historical breakup match tool was particularly useful to the ICEX team for relating their institutional knowledge with the evolution of the numerical guidance. Anecdotal reports from the ICEX team indicate the people at Camp Skate (ICEX Camp) found it to be a popular product that filled a niche need and was widely circulated amongst the team. A product that is used and appreciated by the team is a key selling point.
Daily Forecast Next Steps: ACCAP plans to a further exploration of match domain size/shape, as well as other statistical modifications that should improve the verification score of the analogs project. We would also like to work with the ICEX programming team to automate the forecast process so that it automatically runs every 12 or 24 hours.
Historical Breakup Match Next Steps: We intent to work with the ICEX team to identify all dates where ice integrity deteriorated for input into the database. We will then generate map-based or chart-based graphical output to supplement the existing tabular output.