Deliverable No. 4.1 Initial assessment of the added value of observations in existing long time-series datasets – guidance for dedicated observing system experiments

The assessment of state-of-the-art long-term datasets, such as analyses and reanalyses, and short-range weather forecasts was carried out for Arctic regions. It was demonstrated that these datasets can be used for understanding the Arctic climate system variability. It was shown that the quality of (re)analyses in Arctic regions has increased over the last years, despite the challenges related with the modelling, observation usage and data assimilation encountered in Arctic regions. This further indicates that Numerical Weather Prediction systems can be used for numerical experimentation aimed at informing the design of future observing systems in the Arctic.

The use of atmospheric observations in the Arctic region was further analysed for the ECMWF NWP system. The analysis highlights that for global NWP, the polar regions are a particularly data-rich area in terms of satellite sounding observations from polar-orbiting satellites, whereas conventional observations are sparse north of 70N. However, the use of the satellite radiances is hampered during the winter period due to problems over snow and sea-ice in the forward modelling of the satellite radiances and cloud detection, combined with potentially larger errors in the forecast model. There is evidence of considerable bias, originating from the forecast model (especially around 200 hPa), the observation operator for microwave radiances (especially for surface-sensitive radiances over snow and sea-ice), and the cloud screening applied to infrared radiances. As these biases are not accounted for during the assimilation, biases in the resulting analyses are very likely in polar regions. The crucial role of conventional observations in identifying these biases has been highlighted. In addition, our analysis shows that background errors used in the analysis over the Arctic are likely to be under- estimated for the lower troposphere and the upper-troposphere lower-stratosphere. To improve the use of satellite data in the troposphere, improvements in the forward model and/or cloud detection would be needed, whereas to improve the use of near-surface data, it would be necessary to increase the background errors to avoid limiting the adjustments observations can make to the short-range forecasts in this area during the assimilation.

An initial assessment of the impact of observations has been performed using adjoint-based techniques that estimate the contribution of the observations to the reduction on the overall short-range forecast error. This suggests that conventional Arctic observations appear to contribute the most to reducing the global short-range forecast error during winter, while microwave observations from polar orbiting satellites contribute the most in summer. To further assess the impact of different observation types, particularly on improving the medium-range forecasts and on interactions with the mid-latitudes, observing system experiments (OSE’s) are needed where different observation types are removed from the Arctic and the effect analysed. These will be conducted in task 4.2.2, with a view to provide further guide the design of future atmospheric observing systems.

In terms of Arctic sea ice, fourteen state-of-the-art reanalyses were analysed in order to understand the added value of sea ice concentration (SIC) assimilation for the sea ice thickness (SIT) field. First, it was aimed to identify whether or not reanalyses built with SIC assimilation are closer to observed values of SIT, in contrast to the products built with no sea ice data assimilation. All reanalyses were compared against several sources of observations such as moored upward-looking sonars, submarines, airbornes, satellites and ice boreholes. The best performing reanalyses (the ones which SIT values are closer to the observations) indeed assimilates SIC data (eg. TOPAZ1, C-CLORS05 and PIOMAS). However, the SIC data assimilation does not necessarily improve the reproduction of the SIT fields in all reanalyses. In those cases, the reanalyses do not take the best advantage of the covariances between SIC and SIT. Second, the impact of SIC assimilation in two aspects of the SIT variability, the time and length scales, was analysed. The results clearly show that SIC assimilation plays a role in both scales, which are considerable shorter for reanalyses making use of SIC data assimilation. This is likely due to the fact that when a reanalysis assimilates SIC information, the system is forced towards the assimilated conditions, differently from what occurs with free-running models, likely bringing the scales towards a more realistic values.