Benchmark seasonal prediction skill estimates based on regional indices

Abstract

Abstract. Basic statistical metrics such as autocorrelations and across-region lag correlations of sea ice variations provide benchmarks for the assessments of forecast skill achieved by other methods such as more sophisticated statistical formulations, numerical models, and heuristic approaches. In this study we use observational data to evaluate the contribution of the trend to the skill of persistence-based statistical forecasts of monthly and seasonal ice extent on the pan-Arctic and regional scales. We focus on the Beaufort Sea for which the Barnett Severity Index provides a metric of historical variations in ice conditions over the summer shipping season. The variance about the trend line differs little among various methods of detrending (piecewise linear, quadratic, cubic, exponential). Application of the piecewise linear trend calculation indicates an acceleration of the winter and summer trends during the 1990s. Persistence-based statistical forecasts of the Barnett Severity Index as well as September pan-Arctic ice extent show significant statistical skill out to several seasons when the data include the trend. However, this apparent skill largely vanishes when the data are detrended. In only a few regions does September ice extent correlate significantly with antecedent ice anomalies in the same region more than 2 months earlier. The springtime “predictability barrier” in regional forecasts based on persistence of ice extent anomalies is not reduced by the inclusion of several decades of pre-satellite data. No region shows significant correlation with the detrended September pan-Arctic ice extent at lead times greater than a month or two; the concurrent correlations are strongest with the East Siberian Sea. The Beaufort Sea's ice extent as far back as July explains about 20 % of the variance of the Barnett Severity Index, which is primarily a September metric. The Chukchi Sea is the only other region showing a significant association with the Barnett Severity Index, although only at a lead time of a month or two.

Summer predictions of Arctic sea ice edge in multi-model seasonal re-forecasts

“…Bonan et al (2019) found evidence for this loss in predictive capacity in GCMs in the Arctic marginal seas between June and May starts by analyzing correlation between sea ice area and sea ice volume from previous months of CMIP5 preindustrial control runs. This springtime "predictability barrier" is also consistent with evaluations of empirical forecasts based on observational data (Walsh et al 2019). Moreover, initialized predictions often perform at substantially lower skill levels than those estimated in potential predictability studies (Guemas et al 2016;Bushuk et al 2019).…”

A Mechanism for the Arctic Sea Ice Spring Predictability Barrier

“…The study of Brunette et al (2019), which used winter coastal divergence as a predictor of summer Laptev SIE, found maximum skill when coastal divergence was integrated up to the first week of May and a notable drop in skill when integrated to the first week of April. Other statistical prediction systems report skillful detrended SIE predictions for forecasts initialized after 1 May, but not prior to this date, consistent with a spring predictability barrier (Kapsch et al, 2014;Lindsay et al, 2008;Liu et al, 2015;Petty et al, 2017;Schröder et al, 2014;Walsh et al, 2019;Williams et al, 2016;Yuan et al, 2016). Similarly, while not necessarily mentioning a spring barrier, other studies documenting the detrended Arctic SIE prediction skill of dynamical prediction systems display a barrier-like skill structure corresponding to initialization month May (Dirkson et al, 2019;Merryfield et al, 2013;Msadek et al, 2014;Sigmond et al, 2013;Wang et al, 2013).…”

Prospects for Seasonal Prediction of Summertime Trans-Arctic Sea Ice Path

“…The 5-yr climatology follows a changing climate but does not assume linear change, which may not be warranted, especially if there is a natural variability contribution to Arctic sea ice decline (Swart et al 2015;Zhang 2015;Ding et al 2017). We expect a persistence forecast to be better than the climatological forecast for short lead times based on the observed persistence of sea ice extent anomalies (Walsh et al 2019) and the improvement in short-lead-time forecast skill accompanying improvement in sea ice concentration initialization (Zhang et al 2022). The forecast reverts to climatology over longer time scales by weighting the anomaly persistence contribution to decline exponentially with an MIP anomaly autocorrelation decay time scale determined from observations.…”