Observations assimilated R-factor to adjust model error variances and K-Factor to adjust observation error variances. ETKF employs SST and SLA bias correction via AR() model fit. TABLE Summary of ocean observations assimilated into the Global Climate Model SST RADS altimetry SSS In-situ T/S (WMO GTS) NAVO-AVHRR JASON AQUARIUS Argo AMSR-E CryoSat SMOS CTD AMSR- EnviSat XBT WindSat Altika RAMA PATHFINDER SARAL PIRATA VIIRS Sentinel3 TAO-TRITON TABLE Number of ocean observations and superobs assimilated on 3/ & /7 Date 3/ type # used # discarded # superobs SLA 49 4574 SST 9478 55 4784 TEM 3454 4956 98345 SAL 5398 547 69894 SSS 6335 963 889 Total 65468 499 693 Date /7 type # used # discarded # superobs SLA 83693 7587 9 SST 9548 6776 5959 TEM 43 96 36354 SAL 94544 85884 337366 SSS Total 4474 7383 5536 DFP-NESP workshop May 8 Slide 7 of
Tropics-Extra-tropics (seasonal increments) Comparison of ocean and atmosphere increments averaged over the boreal winter (DJF) along 4 W (ETKF a); EnOI b) and S (ETKF c); EnOI d) a) Temperature increments (DJF) c) Depth (m) \ Pressure (hpa) b) degrees Depth (m) \ Pressure (hpa) d) degrees Latitude Longitude DFP-NESP workshop May 8 Slide 8 of
3: Assimilation statistics ( S- N) EnOI ( analysis, static cov) versus ETKF (96 analyses, flow dependent + SST bias correction) 5 SST (K) 6 4-3 4 5 6 7 8 9 3 4 5 6 5. SLA (m) T (K).5.5.5 3 4 5 6 7 8 9 3 4 5 6 4.5.5 5 -.5 3 4 5 6 7 8 9 3 4 5 6 EnOI bias EnOI MAD ETKF bias ETKF MAD 4 S (psu).. 5 3 4 5 6 7 8 9 3 4 5 6 DFP-NESP workshop May 8 Slide 9 of
Ensemble Prediction System Forecast evolution Bred vector generation Random initial perturbations with prescribed RMS whose amplitude defines the rescaling. Random perturbations Rescaling interval: month Many iterations Bred perturbations Observed state Analysed state Modelled state ΔT=(t-t ) t Control forecast t Ensemble mean forecast Ensemble spread Δt=n(t-t ) =nδt PAST PRESENT Bred perturbations: i.e. the difference between the evolved perturbed forecasts and the control, renormalised via the norm defined by the RMS of the initial perturbations and the length of the rescaling interval. DFP-NESP workshop May 8 Slide of FUTURE Time Control Ensemble forecast Future observed state
BV s versus 8 day forecast errors Comparison of ensemble averaged BV s (shaded) and EnOI/ETKF analysis increment (contour) along sections at 4 W and S. 3 3 3 3 DJF BV vs EnOI inc 4W..5.4.6.3... 5 5. JJA.5.6.4.3... 5 5 5 5. DJF BV vs EnOI inc S........8.9.7.6.5.4.3. JJA..7.6.5.4... 4 8 6 4 8.3.4...4.. 3 3 3 3.3 MAM.4..3... 5 5... SON.3.4.5.6.7.8. 5 5... MAM....3 4 8 6 4 8...... SON.9.5.8.7..6..4.3.3. 4 8 6 4 8 DFP-NESP workshop May 8 Slide of
Scale selection localisation length scales and adjustment of observation impact factors to tune increments to select spatio-temporal scales of relevance to given forecast lead times. mask regions of variance relevant to those chosen spatio-temporal scales such as the in band variance for temperature with an appropriate threshold (.5 RMSE calculated from 5 years of control simulation). - months lat itu 8- months DFP-NESP workshop May 8 Slide of e itud long de 4-48 months 48-96 months
Error growth rates isosurface BV growth rates 3 times larger than S- N BVs 3 a) Obs Multivariate ENSO Index (MEI) (Black overlay) growth rate of average BV per month (- month isosurface) correlation to MEI temp (deg C) depth (m) 5 5 5 3 4 5 6 7 8 9 3 4 5 6 5 time in years.5.5 3 3.5 4 4.5 5 5.5 6.5.5 3 b) growth rate of average BV per month (S-N) correlation to MEI temp (deg C) depth (m) 5 5 5 3 4 5 6 7 8 9 3 4 5 6 5 time in years.4.6.8..4.6.5.5 DFP-NESP workshop May 8 Slide 3 of