Technical Discussion

The computer model used to make the initial forecast is based on atmospheric data from November of the previous year through March of the current year. Initial forecasts will be issued about April 5th of each year and subsequent, fine-tuned, forecasts using later information will be issued shortly after the first of the month in May, June, and July. However, it is not expected that these later forecasts will change the forecast more than a few days.

Continuous monitoring of ice conditions and weather forecasts will be started the first of July and forecast updates will be made as required.

Accuracy of the Fairweather Forecast
The initial run of the Fairweather model is completely objective, requiring no subjective evaluation of ice state or other factors. Test runs of the model (hindcasts) using historical meteorological data for the years 1973-2001 show that the accuracy of the forecast has an average error of + - 7 days, a maximum error of 16 days, and a minimum error of 1 day. This forecast is issued about April 5th. For the mathematically inclined, the correlation between the two sets of dates is .7936.

The chart above uses National Ice Center (NIC) breakup dates and data available through March of the forecast year. There are two exceptions. In 1975 NIC reported that the ice did not go at all and a review of Canadian and NIC ice charts does show that the shipping lane did not open that year according to NIC criteria. However, one of the largest sealifts to the North Slope took place that year. A few barges made it around Point Barrow on September the 2nd and the lane closed in again on September the 6th. It opened again on September the 25th and 30 barges made it to Prudhoe Bay, where they promptly froze into the bay. ARCO called this the ?sealift from hell.? We have, rather arbitrarily, used September 25th as the date of breakup for the year 1975. The other exception is 2001 when, from personal knowledge, ships began to transit the route on August 8, but in a report recently issued NIC used August 17th has the opening date for that year. We are using August 8th for that 2001.

Only slight improvements in forecast accuracy are made using April and May data with a correlation of .8345 using November through April data and .8475 using data through May.

The graph below shows a comparison of forecasted and actual dates of the opening of the shipping lane using Nov. through June data. The average error at the end of June is + - 6 days, a maximum error of 13 days, and a minimum of zero. Correlation of the two sets of dates is .8591.

As can be seen in the above discussion, the accuracy of the forecast made at the end of March is improved only slightly using April through June data. There are two major reasons for this: 1) Wind and temperature in the days immediately preceding breakup can advance or impede the time of breakup by several days, and these conditions cannot be forecast accurately more than a week or so in advance and, 2) ice analysis is highly subjective and in the early years was only performed once or twice weekly and thus the actual date of opening of the sea-lane may have a built in error of a few days.

The above graph uses data through July 15th of the forecast year. The use of data through July 15th improves the smoothness of the plot and increases the correlation to .8698 but does little for the absolute error with an average of + - 6 days and a maximum error of 11 days.

The forecast model might be improved with the use of multi-year ice quantities in the southern Beaufort and the use of a smaller grid size (now 2.5 degrees) for the meteorological data. Work is continuing on the forecast model and any improvement will be posted on this web-site.

Data used in these forecasts was obtained from the NOAA-CIRES Climate Diagnostics Center, Boulder, Colorado, USA, from their Web site at http://www.cdc.noaa.gov/

Similar forecasts can be issued for other icy seas using similar techniques. Contact Chuck Samuels if you have any questions. chuck@fairweather.com