Improved Analysis of Tropical Upper Ocean Conditions for Seasonal to Interannual Forecasting

1
Improved Analysis of Tropical Upper Ocean
Conditions for Seasonal to Interannual
Forecasting

• UMD Jim Carton, Gennady Chepurin
• NCEP David Behringer

2
Goal of the research

• Observations have spatially correlated errors
  while ocean models have slowly varying biases.
  Current data assimilation schemes assume zero
  bias.
• Our goal is to identify the forecast bias in the
  current ocean assimilation system and explore
  methods to control its impact on forecasts.

3
Forecast Bias

• Causes
• Errors in forcing
• Errors in initial conditions
• Errors in physics parameterizations
• Errors in numerics

4
NCEP Global Ocean Data Assimilation System
(GODAS)

• Model GFDL MOM3
• Grid Quasi-global domain extending from 75oS to
  65oN zonal resolution is 1o poleward of 30o
  increasing to 1/3o within 10o of the equator 40
  vertical levels 10 meter resolution in the top
  200 meters.
• Physics KPP mixing GM isoneutral mixing of
  tracers nonlinear horizontal viscosity explicit
  free surface variable thickness bottom cell
• Forcing wind stress, heat flux, E-P (NCEP
  Reanalysis 2 or operational GDAS) SST relaxed
  to NCEP SST analysis SSS relaxed to Levitus
  climatology
• Method 3D Var background error variance varies
  geographically and temporally
• Data Temperature (XBTs, Argo floats, TAO/Triton
  moorings), synthetic salinity profiles
• Period 1980 - pres

5
Mean temperature bias
6
Bias in temperature annual cycle
7
Bias in mixed layer depth based on Temperature
8
Mixed layer depth based on temperature
9
Mixed layer depth based on density
10
Forecast bias correction algorithm

• First stage bias correction

Second stage unbiased analysis
where
where
11
95m Temperature bias EOF decomposition
12
Bias model

• Based on EOF analysis we propose the following
  bias model

, and Gi is i-th EOF.
where
13
Reduced space bias correction
If we assume that the forecast bias can be
expressed as a projection on a set of N
principal components
where G is a matrix of size MxN containing the
spatial structure of the principal components
and is a column vector of length N
containing time series then cost function to
minimize is
.
where
14
Test with climatology
Bias added to Levitus climatology. Then the
climatology is sampled at 5 of grid points and
the bias correction algorithm applied.
15
Looking at bias in CFS
16
Seasonal SST
CFS
OBS
Error is in the SH
CFS-OBS
17
Errors in heat content
OBS
CFS-OBS
18
Summary

• We have developed temperature bias model for
  GODAS which includes mean and seasonal biases
• We have developed a two stage bias correction
  algorithm for 3D-var systems. This algorithm has
  been coded for implementation into GODAS and
  intensively tested on a single processor
  computer

Resources needed We
need access to the JCSDA IBM SP computer to
Implement our codes into GODAS.
19
Whats next ?

• Implement algorithm in GODAS
• Make bias correction multivariate apply
  two-stage bias correction algorithm to other
  assimilating parameters (salinity, sea level,.)
• Continue look at bias in the coupled system