Aspects of ECMWF model performance in polar areas

Abstract

Global numerical weather prediction skill over polar areas is assessed, mostly based on the European Centre for Medium‐Range Weather Forecasts (ECMWF) system but also the Met Office, Japan Meteorological Agency (JMA), Environment Canada and National Centers for Environmental Prediction (NCEP) analysis data. Polar forecast verification against analyses shows a similar trend of forecast improvement over the past 12 years compared with improvements at lower latitudes. These improvements are presumably due to increased model resolution and model sophistication, improved data assimilation methods and increased observational data coverage and better data quality. By comparing ECMWF's real‐time forecast skill changes against those from reforecasts initialized from reanalyses, it is possible to quantify how much of the improvement is from system improvements and how much is attributable to weather variability. Ensemble skill also improved over time and, again, consistently across latitudes. The quality of analyses serving for forecast verification and initialization has been investigated further. An intercomparison of The Observing system Research and Predictability Experiment (THORPEX) Interactive Grand Global Ensemble (TIGGE) analyses and forecasts revealed substantial differences for surface parameters, but also at lower levels in the troposphere, where most of the physical processes relevant to weather in the short‐to‐medium range take place over the poles. The differences between the TIGGE analyses were generally much larger than differences between members of the ECMWF 4D‐Var ensemble of analyses generated internally at ECMWF. This suggests that neither the multi‐analysis approach nor ensemble data assimilation may represent polar analysis uncertainty properly. This is particularly visible at the surface and lower levels in the atmosphere. Forecast spread and error match much better north of 65∘N where less atmospheric variability prevails along the entire forecast range, while in areas of significant synoptic activity the spread also appears too low.

Advancing Polar Prediction Capabilities on Daily to Seasonal Time Scales

“…Standard verification has moreover mostly concentrated on midlatitude and tropical regions. Only very recently has the skill of current operational forecasting systems in the polar regions been considered (Bromwich et al 2005;Jung and Leutbecher 2007;Jung and Matsueda 2016;Bauer et al 2016). More work will be needed, especially on the verification of near-surface parameters as well as snow and sea ice characteristics (especially drift and deformation).…”

“…Given that observations are key to producing accurate initial conditions and hence forecasts, relatively sparse observational coverage in polar regions may be one explanation as to why the skill of weather forecasts in polar regions is relatively low (see also Jung and Leutbecher 2007;Jung and Matsueda 2016;Bauer et al 2016). In addition, data assimilation systems are not adequate to optimally exploit the information provided by existing observations, as will be discussed below.…”

“…Over polar areas, shortcomings in all three main data assimilation components (models, observations, and assimilation algorithms) contribute to suboptimal state estimates (e.g., Jung and Leutbecher 2007;Bauer et al 2016) leading to a detrimental impact on forecast skill across all time scales. In the atmosphere in which boundary layer processes and atmosphere-surface interaction-particularly with variable sea ice coverageare shallow and dominant, the small scale of cyclonic systems (e.g., polar lows) and the interaction of the flow with extremely steep orography are currently not well resolved in global models (and observations) and even less so in data assimilation systems (Tilinina et al 2014).…”

Thin Arctic sea ice in L-band observations and an ocean reanalysis

“…This is consistent with the ice being thicker in ORAS5 than in SMOS-SIT: the thicker the ice, the smaller the surface heating by conductive heat flux from the relatively warm ocean water below the ice to the relatively cold surface of the ice. However, different near-surface temperatures in the two reanalyses (JRA-55 and ERA-Interim,) might also play a role (see Bauer et al, 2016), because they will have a direct impact on the implied sea-ice bulk temperature. Note that there is an apparent artefact in the ice surface temperature in the SMOS-SIT product: it has a constant value of around −4 • C for extended periods in November and December.…”

“…These external data dependencies introduce uncertainties that are often difficult to quantify. For instance, near-surface temperature over Arctic sea ice can vary by several degrees between atmospheric analyses from different centres (Bauer et al, 2016). Moreover, different radiative transfer models exist to calculate the L-band emissivity of a given sea-ice slab, and the calculated L-band TBs can vary considerably depending on the model chosen (Maaß, 2013;Richter et al, 2018).…”