More information on forecast product

Short-range forecasting products

Overview

We estimate that a 5-day sea-ice edge forecast is in average skillfull on a 20 to 60 km scale compared to the sea-ice charts. The sea-ice drift forecast has a bias of around 2km/day and a root-mean square error 5 to 8 km/day.

Details

A very detailed verification of the forecasting system is provided in this document. Furthermore, the sea-ice edge position forecasts are evaluated in Melsom et al. (2019), and the forecast product quality is monitored in real-time.

Seasonal forecasting products

Overview

The seasonal sea-ice forecasts are skillfull compared to the climatology for a forecast range of 1 week to about 4 months. The skill is strongly dependent on the season. The following table gives an estimate on the mean forecast range we consider the forecast products skillfull:

Forecast Start Month Lead time in weeks
January 6.0
February 4.5
March 4.5
April 10.5
May 15.00
June 21.0
July 18.0
August 12.0
September 8.5
October 8.5
November 6.5
December 4.5

Details

A detailed analysis of different verification methods and the performance of the seasonal forecast system compared to the climatology is provided in Palerme et al. (2019). Below some details are given concerning the quality of the forecasts compared to the climatology (forecast), the seasonal variability, and the accuracy of the sea-ice edge forecast in space.

A. Probability in weeks - Forecasts is better than reference

Number of weeks during which the forecasts outperform the climatology (assessed from the forecasts starting between 1999 and 2014). The climatology is defined as the mean observed sea ice conditions during the 10 years preceding the forecasts. The three columns (25 %, 50 %, and 75 % probability) represent the duration (weeks) during which the probability that the forecasts outperform the climatology is 75 %, 50 %, and 25 %. For example for the forecasts starting on June 1st, the probability that the forecasts outperform the climatology is 75% after 5.25 weeks, 50% after 12.5 weeks, and 25 % after 17.25 weeks.

Forecast Start Month 75% probability 50% probability 25% probability
January 0.75 1.0 3.5
February 0.75 2.0 3.0
March 1.0 1.5 4.25
April 2.0 4.5 11.25
May 3.75 7.0 10.25
June 5.25 12.5 17.25
July 1.75 3.5 14.25
August 0.0 8.5 11.5
September 3.75 6.0 8.25
October 0.75 2.5 7.5
November 1.0 2.5 4.0
December 0.0 1.0 1.25

B. Probability in weeks - Forecasts as good as reference

Number of weeks during which the forecasts outperform or are equivalent to the climatology (assessed from the forecasts starting between 1999 and 2014). The climatology is defined as the mean observed sea ice conditions during the 10 years preceding the forecasts. The three columns (25 %, 50 %, and 75 % probability) represent the duration (weeks) during which the probability that the forecasts are not not worse than the climatology is 75 %, 50 %, and 25 %. For example for the forecasts starting on June 1st, the probability that the forecasts are not worse the climatology is 75% after 11 weeks, 50% after 21 weeks, and 25 % after 27.5 weeks.

Forecast Start Month 75% probability 50% probability 25% probability
January 2.0 6.0 8.5
February 3.75 4.5 8.25
March 2.0 4.5 6.25
April 5.25 10.5 21.25
May 10.5 15.0 28.75
June 11.0 21.0 27.5
July 5.5 18.0 20.25
August 3.75 12.0 17.0
September 4.75 8.5 14.25
October 4.5 8.5 13.0
November 4.75 6.5 8.0
December 2.0 4.5 6.25

C. Predictability on space and time scales

Mean distance between the forecast and observed ice edges (km) depending on the forecast start month and the lead time (weeks). The forecasts starting between 1999 and 2014 have been used for producing this figure. The lead time represents the time since the initialization of the forecast. For example, a lead time of 5 weeks for a forecast starting on January 1st represents the week between January 29th and February 4th. Note that all the forecasts start on the first day of the month.