Seasonal Prediction ACCESS-S2 Hindcast (1981-2018)

Description

ACCESS-S2 is the Australian Bureau of Meteorology's seasonal prediction system used for multi-week and seasonal prediction (operational since October 2021). This data record provides access to the hindcast set i.e., forecasts run retrospectively for dates in the past. The hindcast set consists of a 3- or 9-member ensemble from the 1st, 6th, 11th, 16th, 21st, and 26th of every month for the period 1981-2018, as well as additional ensembles at the middle and end of every month.

ACCESS-S2 uses the UK Met Office GloSea5-GC2 coupled model and Bureau-developed data assimilation and ensemble generation. The global coupled model includes high resolution atmospheric (N216 which is ~60 km in the mid-latitudes; and 85 levels in the vertical) and ocean (0.25 degree with 75 levels) models, as well as multi-layer land and sea-ice models. Hindcasts (retrospective forecasts) are generated from the ACCESS-S2 system using historical data to initialise the forecasts. Forecast outputs on the model grid are available for atmosphere, ocean, land and ice variables.

A subset of variables have been downscaled and calibrated over the Australian land area at 5km resolution. The variables include rainfall, minimum and maximum temperature, vapour pressure, evapotranspiration, solar exposure and 10-m windspeed. The calibration technique is based on quantile-quantile matching and is the same as was used for ACCESS-S1.

The Hindcast set consists of 3 overlapping subsets:

- The "Climatology" set is a 3-member ensemble of 279 days from the 1st, 6th, 11th, 16th, 21st, and 26th of every month for the period 1981-2018. This set is primarily used to bias-correct and calibrate the real-time forecasts.

- The "Multiweek" set is a 9-member ensemble of 42 days for the 1st and 16th including the preceding 2 days for each. This will allow the creation of two 27-member lagged ensembles (3-days and 9 ensemble members on each day) every month. For example, for the mid-month Multiweek ensemble it will have Hindcasts from the 14th, 15th, and 16th to create a lagged 27-member ensemble. For the "start of month" Multiweek ensemble it will have Hindcasts from the 1st of that month and the preceding 2 days (for May that would be April 29th, 30th and May 1st).

- The "Seasonal" set is a 3 member ensemble of 279 days from the 1st, and the preceding 8 days. This will allow the creation of a 27-member lagged ensemble (9-days and 3 ensemble members on each day). For example, for May the 27-member ensemble would include forecasts from April 23rd, 24th, ..., 29th, 30th and May 1st.

Local host (English)

Direct access to the data is available on the NCI servers:

• path: /g/data/ux62/access-s2/hindcast 

Register for local access:

•  https://my.nci.org.au/mancini/project/ux62/join

Access information (English)

Preferred citation:

• Australian Bureau of Meteorology 2023.  Seasonal Prediction ACCESS-S2 Hindcast (1981-2018). NCI Australia (Dataset) doi: https://dx.doi.org/10.25914/60627ad0ba3ee 

NCI THREDDS Data Server:

•  https://thredds.nci.org.au/thredds/catalog/catalogs/ux62/access-s2/hindcast/hindcast.html

If accessing from NCI thredds you can also acknowledge the service:

• NCI Australia (2021): NCI THREDDS Data Service. NCI Australia. (Service) doi: 10.25914/608bfc062f4c7

Technical info

Lineage:

The ACCESS-S2 coupled model (GC2) is the same as that used in ACCESS-S1, apart from the following changes:

- correction to the setting associated with rivers draining into inland lakes (which was known to create errors in ACCESS-S1);
- increasing the ocean-atmosphere coupling frequency from 3-hours to 1-hour (this improves the shallow layers in the Pacific Ocean), and
- correcting the bug associated with the updating of the ozone ancillary.

The ensemble generation process in ACCESS-S2 is the same as that in ACCESS-S1. Details of the coupled model and the ensemble generation can be found in:

Hudson, D., Alves, O., Hendon, H.H., Lim, E., Liu, G., Luo J.-J., MacLachlan, C., Marshall, A.G., Shi, L., Wang, G., Wedd, R., Young, G., Zhao, M., Zhou X., 2017: ACCESS-S1: The new Bureau of Meteorology multi-week to seasonal prediction system. Journal of Southern Hemisphere Earth Systems Science, 67:3 132-159 doi: 10.22499/3.6703.001.

ACCESS-S2 is distinct from ACCESS-S1 in terms of the data assimilation. ACCESS-S1 used ocean initial conditions from the the UK Met Office. For ACCESS-S2, the data assimilation is done at the Bureau using a different method. It uses a weakly coupled daily cycle, with direct replacement of the atmospheric basic variables with ERA-Interim (or the NWP analysis for real-time forecasts). It is weakly coupled because the ocean assimilation sees the background state from the coupled model simulation. The ocean data assimilation uses multivariate ensemble optimum interpolation, where the covariances are derived from a static model ensemble. In situ temperature and salinity observations are assimilated, and corrections to currents are generated based on the ensemble cross covariances with temperature and salinity. The land surface and sea-ice are indirectly initialised through the coupling process (in response to atmosphere and ocean forcing). In ACCESS-S1 the soil moisture was initialised using climatology, whereas in ACCESS-S2 realistic time-varying soil moisture is used to initialise the forecasts.

Additional details