2023 First report complement

Arctic Sea Ice Information Weekly Report on Arctic Sea Ice

June 16, 2023

This page describes the prediction method used for the 2023 First report and the accuracy of the prediction.

Consideration of the effects of sea ice transport up to the prediction date

The First reports until 2022 were based on sea ice conditions as of April 30. However, sea ice actually continues transporting after that time. Figure 1 shows the sea ice distribution on September 10, 2022 (observation) overlaid with the sea ice edge (sea ice concentration 20%) predicted in the 2022 Third report. Especially in the East Siberian Sea, sea ice remains in the west (left) side compared to the prediction. One of the reasons for that is sea ice transports westward after the time of prediction.

Sea ice distribution in summer is strongly influenced by sea ice age distribution. To improve the prediction method, it is necessary to estimate the ice age distribution on a forecast target date. For this purpose, we calculated sea ice transport from May 1 to the target date by using the mean sea ice drift velocity since May over the past four years. Figure 2 shows the calculated sea ice age distribution (Figure 2 middle) and the observed sea ice age distribution (Figure 2 right) at the end of August 2022. Comparing the two figures, the old ice distributions is almost the same. Considering these results, we estimated the sea ice age distribution on the target date assuming average sea ice movement and used the derived distribution for the prediction of sea ice distribution.

Figure 1: Predicted ice edge in the 2022 Third report (green line) and observed ice distribution (white area)

Figure 2: Sea ice age distribution at the end of April 2022 (left), and predicted ice age distribution by considering sea ice movement after May (middle) and observed distribution (right) at the end of August 2022

Parameters used for prediction

We repredicted the sea ice distribution in 2022 using the following four sets of parameters and considered the best performing set of parameters. Parameter sets (3) and (4) consider the “sea ice movement up to the target date” in the above explanation.

(1) sea ice redistribution from winter to spring and sea ice age in spring
(2) sea ice redistribution from winter to spring, sea ice age in spring, and mean divergence of sea ice
(3) sea ice redistribution from winter to spring and sea ice age in spring (considering movement up to summer)
(4) sea ice redistribution from winter to spring, sea ice age in spring (considering movement up to summer), and mean divergence of sea ice (considering movement up to summer)

Figure 3 overlays the prediction results (lines) with the observed ice distribution. While predictions by all methods are almost the same for the first half of August, there is variation among methods from late August onward. Figure 4 shows the total prediction error of sea ice extent accumulated from August 1 to September 10. Red and blue indicate overprediction and underprediction respectively, and the smaller the sum of the two (the height of the graph), the higher the accuracy. Errors in (3) and (4) are smaller in A1 area including the Beaufort Sea, while those in (1) and (2) are more accurate in A2 area including the East Siberian Sea. This is believed to be the only exceptional case for this year, when old and brittle ice was widespread; (3) and (4) predict that old ice will remain in the Chukchi Sea, as can be seen in the September 1 panel of Figure 3. Overall, method (4) has the best prediction accuracy, so this method was used for the 2023 First report.

Figure 3: Ice edge positions (lines) re-predicted using four different parameter sets overlaid on the observed sea ice distribution for August and September 2022. Light blue, orange, green, and yellow lines indicate the results by parameter sets (1) (2) (3) and (4), respectively. Solid line indicates sea ice concentration 20%.

(a) Error in predicted sea ice extent　　　　　(b) Areas A1 to A5

Figure 4: Difference between observed and predicted values by the five methods (a) in areas A1 to A5 (b). Red and blue indicate the overprediction and underprediction of the ice extent respectively, and the smaller the sum of the two, the higher the accuracy.

Certainty of predicted sea ice distribution

In Figure 3, areas with large variations in predicted ice edge can be considered as areas where prediction errors are likely to be large.

Figure 5 is an animation overlaying the results by methods (2) and (3) with the predicted ice distribution of the 2023 First report. The First report predicts that sea ice on the continental side of the Beaufort Sea will remain from around late July to late August, but the prediction results vary depending on the method used. Therefore, this area is considered to be an area where the reliability of forecasts is relatively low. Also, the prediction results that old ice will remain in the same area after September vary depending on the prediction method. On the other hand, for the East Siberian Sea on the Russian side, all methods show the same results, indicating that the predictions are highly reliable.

Figure 5: Animation of three patterns of predictions overlaid on each other. Sea ice concentration 20% or greater is shown, and where the non-transparent white areas are predicted to remain in all patterns.

Use example of the prediction

The Arctic Sea Ice Information Center provides a chart overlaying the predicted sea ice distribution and planned shipping route as shown in Figure 6 as part of the support for this year’s R/V Mirai Arctic cruise. By comparing the planned route and the predicted ice distribution, it is possible to assess whether the route can be safely navigated and to use this as a reference for reconsidering the route.