Abstract
Abstract. Nowadays many seasonal forecasting centres provide dynamical predictions of sea ice. While initializing sea ice by assimilating sea ice concentration (SIC) is common, constraining initial conditions of sea ice thickness (SIT) is only at its early stages. Here, we make use of the availability of Arctic-wide winter SIT observations covering 2011–2016 to constrain SIT in the ECMWF (European Centre for Medium-Range Weather Forecasts) ocean–sea-ice analysis system with the aim of improving the initial conditions of the coupled forecasts. The impact of the improved initialization on the predictive skill of Arctic sea ice for lead times of up to 7 months is investigated in a low-resolution analogue of the currently operational ECMWF seasonal forecasting system SEAS5. By using winter SIT information merged from CS2 and SMOS (CS2SMOS: CryoSat2 Soil Moisture and Ocean Salinity), substantial changes of sea ice volume and thickness are found in the ocean–sea-ice analysis, including damping of the overly strong seasonal cycle of sea ice volume. Compared with the reference experiment, which does not use SIT information, forecasts initialized using SIT data show a reduction of the excess sea ice bias and an overall reduction of seasonal sea ice area forecast errors of up to 5 % at lead months 2 to 5. Using the Integrated Ice Edge Error (IIEE) metric, we find significant improvement of up to 28 % in the September sea ice edge forecast started from April. However, sea ice forecasts for September started in spring still exhibit a positive sea ice bias, which points to too slow melting in the forecast model. A slight degradation in skill is found in the early freezing season sea ice forecasts initialized in July and August, which is related to degraded initial conditions during these months. Both the ocean reanalyses, with and without SIT constraint, show strong melting in the middle of the melt season compared to the forecasts. This excessive melting related to positive net surface radiation biases in the atmospheric flux forcing of the ocean reanalyses remains and consequently degrades analysed summer SIC. The impact of thickness initialization is also visible in the sea surface and near-surface temperature forecasts. While positive forecast impact is seen in near-surface temperature forecasts of early freezing season initialized in May (when the sea ice initial conditions have been observationally constrained in the preceding winter months), negative impact is seen for the same season when initialised in August month when the sea ice initial conditions are degraded. We conclude that the strong thinning by CS2SMOS initialization mitigates or enhances seasonally dependent forecast model errors in sea ice and near-surface temperatures in all seasons. The results indicate that the memory of SIT in the spring initial conditions last into autumn, influencing forecasts of the peak summer melt and early freezing seasons. Our results demonstrate the usefulness of new sea ice observational products in both data assimilation and forecasting systems, and strongly suggest that better initialization of SIT is crucial for improving seasonal sea ice forecasts.
Cited by
1 articles.
订阅此论文施引文献
订阅此论文施引文献,注册后可以免费订阅5篇论文的施引文献,订阅后可以查看论文全部施引文献