Implementation of an Adaptive Bias‐Aware Extended Kalman Filter for Sea‐Ice Data Assimilation in the HARMONIE‐AROME Numerical Weather Prediction System

Abstract

Abstract Sea ice surface temperature is an important variable for short‐range numerical weather prediction systems operating in the Arctic. However, when provided by numerical sea ice models, this variable is seldomly constrained by the observations, thus introducing errors and biases in the simulated near‐surface atmospheric fields. In the present study a new sea ice data assimilation framework is introduced in the HARMONIE‐AROME numerical weather prediction system to assimilate satellite sea ice surface temperature products. The impact of the new data assimilation procedure on the model forecast is assessed through a series of model experiments and validated against sea ice satellite products and in‐situ land observations. The validation results showed that using sea ice data assimilation reduces the analyzed and forecasted ice surface temperature root mean square error (RMSE) by 0.4 °C on average. This positive impact is still traceable after 3 h of model forecast. It also reduces the 2 m temperature RMSE on average by 0.2 °C at the analysis time with effects persisting for up to 24 h forecast over the Svalbard and Franz Josef Land archipelagos. As for the 2 m specific humidity and 10 m wind speed, no effect was observed. Possible impact on the upper‐air fields was assessed by comparing the model forecast against the radiosonde soundings launched from Spitsbergen, with no clear improvement found. Implications of using a coupled surface‐atmosphere data assimilation technique in HARMONIE‐AROME are discussed.

Keywords

• sea ice
• data assimilation
• extended Kalman filter
• numerical modeling