Rapid Update Convective Scale NWP Systems for Australia

The Australian Bureau of Meteorology is testing a Rapid Update, Convective-Scale Numerical Weather Prediction (NWP) system targeted for operational use later this year. This presentation will provide an overview of the developments within this system.

Weather and climate modelling within Australia generally uses the Australian Community Climate and Earth System Simulator (ACCESS), which uses the Unified Modelling System (originally developed by the UK Met Office) for the atmospheric components. It is within this modelling framework that the BoM’s convective scale modelling system has been developed. This involves six model domains, centred over Australia’s main population centres and are referred to as ACCESS-City models, however with domains of 10^o latitude by 12^o longitude they cover much of Australia.

The ACCESS-City systems use the most recent software versions of the atmospheric model, variational assimilation, observation processing system and suite scheduler via the shared software repositories of the Unified Modelling Partnership. The basic model configuration corresponds to UM version 10.6, with a number of extra upgrades, in particular those regarding conservation of moisture within the semi-Lagrangian advection. The core forecast grid is a regular 0.0135^o latitude-longitude (1.5km) grid with a 2^o zone at the end of each row(column) where the stretches to 0.036^o (4km). The assimilation system is a 4d-variational system that is capable of ingesting a comprehensive set of satellite data and regular in situ data. These models are designed to be nested in an N1024 (approximately 12km) global model, although for logistical reasons some testing used a 0.11^o (12km) regional model for the lateral boundary conditions. The 18 member ensemble system runs on a similar grid, but based 0.0198^o (2.2km) core grid.

The assimilation system is run each hour over an hour window, i.e. observations from 1130 to 1230 UTC are collected at 1255UTC, with forecasts being available between 1330 and 1400UTC. The variation in forecast length is due to the length of forecast varying throughout the day - with 36 hour forecasts available 4 times a day and other forecasts ranging from 12 to 24 hours. In later versions it is expected that the previous analysis will be rerun to reduce the effect of the short data cut-off.

One of the novel aspects of this system is the direct assimilation of clear air Doppler radial winds. Despite most Australian radars having single polarization, the data is still useful with suitable quality control. A statistical analysis of innovations for both clear air and precipitation-based radial winds show that the two have comparable error characteristics. The limitations of the quality control does however mean that the lowest scan cannot be used, none the less there is enough data in the other scans to provide useful information. This situation is expected to improve as more dual-polarization radars are introduced.

The use of the radial winds is assisted by a new formulation for the background error covariances. Previous versions used empirical vertical modes to define these covariances, with a length scale for each vertical mode. This gave rise to rather large-scale responses to any observation, which removed much of the horizontal structure within the radial winds. A new formulation, where each horizontal mode at each level is related to a vertical length scale provides a greater range of scales in the analysis increments, assisting in the analysis of low-level wind changes. The analysis of low-level wind changes is particularly problematic near the coast, as insects tend not to fly off shore, providing a significant change in observation density on top of an area of strong dynamical gradients. Ultimately, it is hoped that the introduction of hybrid 4DVar, which allows greater inhomogeneity in the error covariances, will allow for greater differentiation between the responses to coastal versus inland observations.

Other than radial winds, there are a number of other observations used in the ACCESS-City system that are not used in the global system. These include high density atmospheric motion vectors and clear sky radiances from Himawari-8 and high frequency automatic weather station data (thinned to 10 minute frequency).