Met Office progress with DEMETER retrospective seasonal forecasts

1
Met Office progress with DEMETER retrospective
seasonal forecasts

• Met Office Monthly and Seasonal Prediction Group

2
Models used GloSea

• Coupled ocean-atmosphere model (GloSea)
• based on HadCM3, plus...
• Ocean horizontal resolution enhanced in tropics
  from 1.25o to 0.28o
• Number of ocean levels increased from 20 to 40
• Coastal tiling scheme used to improved resolution
  of coastline
• Ocean data assimilation scheme
• 9-member ensemble generated from a combination
  of perturbed ocean assimilation runs and
  instantaneous SST perturbations applied at
  initialisation time.

3
Models used HadAM3

• Uncoupled AGCM (HadAM3)
• resolution 2.5o latitude, 3.75o longitude, 19
  vertical levels
• forced with predicted SST anomalies and sea-ice
  extent, based on persistence from 1-month period
  prior to initialisation.
• 9-member ensemble, generated from 6hr lag in
  initialisation times.

4
Progress with integrations

• For both GloSea and HadAM3
• 9 years (1987-1995) of 6-month hindcasts have
  been completed for 4 start dates (1st Feb, 1st
  May, 1st Aug, 1st Nov)
• Output is archived in ECMWF MARS
• Production is following that of ERA40
  re-analysis, which is providing initial
  conditions
• Plan is to complete 30 years of hindcasts,
  1969-98
• Verification results for all DEMETER models (and
  the multi-model) is available at
• http//wms.ecmwf.int/research/demeter/index.html

5
Timeseries of All India Rainfall (JJAS)
predictions at one-month lead GloSea
6
As previous slide, but with variance inflation
applied to the ensemble
Ensemble capture rate is 100 when inflation is
applied
7
Timeseries for West Africa rainfall (JJA) at
one-month leadcoordinates of WA region 0oN -
25oN 20oW - 25oE
GloSea
HadAM3
Note improved 1988 prediction with GloSea
8
Summary

• 9-years of hindcasts have been completed with
  GloSea and HadAM3 (also ECMWF, Meteo-France and
  LODYC)
• Results for 1987-1995 rainfall over India and
  West Africa look encouraging (but sample is still
  small)
• There is evidence of skill benefits from coupled
  modelling (over uncoupled)