Table 1 Top-5 variable importance rankings from the permute-and-predict method.
From: Seasonal Arctic sea ice forecasting with probabilistic deep learning
(a) | ||||||
|---|---|---|---|---|---|---|
Rank | Apr intitialisation (5-month lead time) | Jun intitialisation (3-month lead time) | Aug intitialisation (1-month lead time) | |||
1 | Sept linear trend SIC forecast | (−0.56%) | Jun upw. solar radiation anomaly | (−1.32%) | Aug SIC | (−22.94%) |
2 | Apr upw. solar radiation anomaly | (−0.51%) | Sept linear trend SIC forecast | (−0.74%) | Aug sea level pressure anomaly | (−2.55%) |
3 | Apr SIC | (−0.46%) | Jun SIC | (−0.63%) | Aug 500 hPa gpt. height anomaly | (−1.58%) |
4 | Apr 2 m air temperature anomaly | (−0.22%) | Aug linear trend SIC forecast | (−0.45%) | Aug upw. solar radiation anomaly | (−0.66%) |
5 | Feb 500 hPa air temperature anomaly | (−0.22%) | May 2 m air temperature anomaly | (−0.39%) | Jun 2 m air temperature anomaly | (−0.29%) |
(b) | ||||||
|---|---|---|---|---|---|---|
Rank | Oct intitialisation (5-month lead time) | Dec intitialisation (3-month lead time) | Feb intitialisation (1-month lead time) | |||
1 | Mar linear trend SIC forecast | (−0.28%) | Mar linear trend SIC forecast | (−0.50%) | Feb SIC | (−8.69%) |
2 | Apr linear trend SIC forecast | (−0.18%) | Feb linear trend SIC forecast | (−0.34%) | Feb 500 hPa gpt. height anomaly | (−0.58%) |
3 | Feb linear trend SIC forecast | (−0.14%) | Dec sea level pressure anomaly | (−0.30%) | Feb upw. solar radiation anomaly | (−0.40%) |
4 | Apr SIC | (−0.12%) | Apr linear trend SIC forecast | (−0.22%) | Apr linear trend SIC forecast | (−0.24%) |
5 | Oct sea level pressure anomaly | (−0.08%) | Apr SIC | (−0.09%) | Feb sea level pressure anomaly | (−0.21%) |
