Transcript Slide 1
Towards Coupled data assimilation in an intra/seasonal forecast system Oscar Alves CAWCR (Centre for Australian Weather and Climate Research) Australian Bureau of Meteorology Contributors and Collaborators: Patricia Okely, Yonghong Yin, Debbie Hudson, Peter Oke, Terry O’Kane The Centre for Australian Weather and Climate Research A partnership between CSIRO and the Bureau of Meteorology Outline 1. Current data assimilation and ensemble generation strategies 2. What coupled covariances may look like 3. New coupled ensemble generation for multiweek prediction 4. Path towards fully coupled assimilation The Centre for Australian Weather and Climate Research A partnership between CSIRO and the Bureau of Meteorology POAMA-2 Ocean Data Assimilation PEODAS: POAMA Ensemble Ocean Data Assimilation System (Yin et al 2010) Perturb forcing + noise Ocean Model Ocean Observations • • • ASSIM ASSIM Ensemble OI (Oke et al 2005) Covariances from ensemble spread (3D multivariate-time evolving) Poor-persons EnKF: only assimilate into central member Provides an ensemble of initial ocean states (11 ensembles, but 100 member lagged used for covariance calculation) Assimilates in situ ocean temperature and salinity. The Centre for Australian Weather and Climate Research A partnership between CSIRO and the Bureau of Meteorology Example of Ensemble Spread (Estimate of analysis error) Temperature Salinity From Yin et al 2010 The Centre for Australian Weather and Climate Research A partnership between CSIRO and the Bureau of Meteorology Coverting PEODAS to Fully Coupled Assimilation Coupled Model ASSIM Ocean Observations + atmos anals • • • • ASSIM Ensemble OI (Oke et al 2005) Covariances from ensemble spread (3D multivariate-time evolving) Assimilate ocean obs and atmos re-analyses Cross-covariances between ocean and atmos (&ice & land) This will be done with the next version of our model based on UKMO UM coupled to MOM4 What are going to be the issues ? The Centre for Australian Weather and Climate Research A partnership between CSIRO and the Bureau of Meteorology What might coupled co-variances look like Case study: 90 member ensemble forecast from Dec 1996 Estimate covariances from ensemble (e.g. after two months) Patricia Okely and Li Shi Coupled Covariances Ref.: Temp. 100m Colour: Temp. Cont.: Zon. Current 1. Covariances consistent with intra-seasonal activity 2. Non-local covariances (real or not, desirable or not) Ref.: Temp. 100m Colour: SST 3. Large vertical extent (not shown) Vect.: Surf. Wind Ref.: Temp. 100m Colour: SST 4. Time/space covariance aliasing – should we represent this (past event that triggered independent event) Cont.: OLR Patricia Okely The Centre for Australian Weather and Climate Research A partnership between CSIRO and the Bureau of Meteorology POAMA-2 Seasonal and Multi-week systems 1. PEODAS is the bases of ocean data assimilation and ensemble perturbations in our POAMA-2 seasonal prediction system 2. Not suitable for multi-week predictions as no atmospheric perturbations. 3. Atmospheric initial conditions are taken from a atmospheric integration nudged to ERA-40 The Centre for Australian Weather and Climate Research A partnership between CSIRO and the Bureau of Meteorology Coupled Assimilation Step 1: Coupled ensemble generation scheme Coupled Model integrations Central unperturbed analyses: PEODAS and ALI Bred vectors are rescaled and centred to the central analyses 1 day Generates coupled bred perturbations of both the atmosphere and ocean based on the breeding method Rescaling – zonal surface wind spread similar to NCEPERA (Yonghong Yin) The Centre for Australian Weather and Climate Research A partnership between CSIRO and the Bureau of Meteorology MJO Variability Case study: CEI Coupled Analysis 30 days to 1st Mar 1997 EQ, 150E Ensemble Error Variability Coupled Covariances Ref.: Surf. Temp. Ref.: Surf. Temp. Ref.: Surf. Zonal Wind Colour: Surf. Zonal Wind Colour: OLR Colour: Surf. Temp. Conceptual example of real non-local covariances • Suppose you have an MJO error (eg. Speed error or structure error) •Some time later (e.g. 10 days)– there will be non local covariances due to different processes but triggered by the same earlier event Rossby KW MJO error over Brazil The Centre for Australian Weather and Climate Research A partnership between CSIRO and the Bureau of Meteorology Summary • Development of coupled breeding scheme for intraseasonal forecasts is first step towards coupled data assimilation •Co-variance structures capture ~large scale intra-seasonal dynamics •Practical issues: • non local covarariances – real or not • Localisation, especially in the vertical • Ocean and atmosphere on different grids (different time scales) Future •Step 2: Semi coupled (PEODAS ocean, nudge atmos in coupled model) •Step 3: Fully coupled The Centre for Australian Weather and Climate Research A partnership between CSIRO and the Bureau of Meteorology