ECMWF’s strategic goal of improving tropospheric predictions on timescales from one day to one year requires us to carefully study all potential sources of predictive skill. The stratosphere is one such source, particularly on monthly and seasonal timescales. However, ECMWF’s Integrated Forecasting System (IFS) has large biases in stratospheric temperature and wind as well as numerous more subtle problems related to the stratosphere. These issues also affect atmospheric reanalysis, for which a good representation of the middle atmosphere is important in its own right. The Stratosphere Task Force was set up in November 2016 to bring together scientists from the Research, Forecast and Copernicus Departments of ECMWF to work together to evaluate and improve IFS performance in the middle atmosphere, including analyses, reanalyses and forecasts. This concerted effort has been boosted by the participation of stratosphere experts Professor Ted Shepherd and Dr Inna Polichtchouk from the University of Reading.
Nine meetings were held in the first 12 months, each with around 20 attendees, and the discussions inspired collaborative work to tackle individual problems. Additionally, Ted and Inna each gave a longer seminar during the year. The main achievements of the Task Force so far include:
In the free-running (i.e. long runs without assimilating observations) IFS Cycle 41r1 there was a warm bias of up to 10 K in the upper stratosphere and up to 20 K in the mesosphere. Several changes have led to this being reduced in the current operational cycle (43r3). The figure shows our recent findings that much of the remaining bias can be eliminated by introducing diurnally varying ozone and reducing solar ultraviolet by 7–8% to match observations of the sun. These changes are being considered for a future cycle. Note that part of the mesosphere bias is a side effect of the ‘sponge’ used to prevent waves from reflecting from the model top.
Like most other global models, the free-running IFS has a 5 K cold bias in the polar lower stratosphere. This has been found to be due to a large moist bias in this region due to excessive transport from the troposphere. Experiments in which the humidity seen by radiation is artificially reduced not only eradicate the cold bias but also improve monthly forecast skill (see ECMWF Technical Memorandum No. 816). Work to provide a physically based solution to the excessive humidity transport is ongoing.
From March 2016, the operational analysis was affected by an erroneous mesosphere jet of up to 180 m/s. This problem has been solved by a modification of the climatological part of the background error covariance model in the data assimilation system. It is nevertheless present in the version of the IFS being used to produce the ERA5 reanalysis.
A careful analysis has been performed on the impact of model parametrizations, particularly non-orographic gravity wave drag, on the Brewer-Dobson circulation, the Quasi-Biennial Oscillation and seasonal temperature biases in the stratosphere (see ECMWF Technical Memorandum No. 809). This has provided valuable information for future adjustments to these parametrizations
A wide range of other important issues have been identified and discussed and will be the focus of activities by the Task Force in the coming year. For example, despite simulating the evolution of Sudden Stratospheric Warmings very well, the stratosphere–troposphere coupling in the IFS is too weak. What is needed to improve the coupling, and therefore predictive skill? Mean stratospheric temperature has been found to have a noticeable dependence on resolution: increasing horizontal resolution from TL255 to TCo1279 results in a 1–2 K cooling at 70 hPa unless it is also accompanied by a modest increase in vertical resolution (e.g. 137 to 162 levels). Do we need more vertical levels operationally, or can the resolution dependence be removed with better numerics? We also wish to make prognostic ozone interactive in the radiation scheme, and work is ongoing to address some of the shortcomings of the available linear ozone schemes to make this possible. Finally, whilst ERA5 brings significant improvements over the older ERA-Interim reanalysis in most respects, it has been found that ERA5 has larger stratospheric temperature biases than ERA-Interim from the 1990s to 2006. This is believed to be due to larger model biases in the newer IFS cycle used for ERA5, compounded by less weight being given to radiosonde data in the newer cycle. The model biases were well corrected by data assimilation only when sufficient GPS radio occultation measurements came online. This highlights the need to improve both the model and the data assimilation in the stratosphere.
We have found the Task Force approach very effective for tackling IFS issues that cut across ECMWF sections and departments, and we envisage that it could be used productively for other topics.