Model error issues: microphysics errors

10/18/2011

Youngsun Jung and Ming XueCAPS/OU

with help from Tim Supinie

Model error issues: microphysics errors

Observation error: Non-Gaussianity, inaccurate observations error variance, none-zero observation error correlation, etc.

Observation operator error

Model error

Source of errors

Example: Observation operator error

http://www.radar.mcgill.ca/science/ex-phenomenon/ex-melting-layers.html

In imperfect model experiments, it is observed that model error dominates the error growth in data assimilation cycles.

Despite this, the characteristics of model error are little known and its statistical properties are poorly understood (Dee 1995; Houtekamer et al. 2005).

For convective-scale NWP, microphysics scheme represents one of the most important physical processes.

Background

Various covariance inflation methods (Tim Supinie)

Parameter estimation

Improving microphysical parameterizations

Outline

Inflation methodsMultiplicative inflation (Anderson and

Anderson, 1999)

Relaxation (Zhang et al., 2004)

Adaptive inflation (Whitaker and Hamill, 2010)

Additive noise (Mitchell and Houtekamer, 2000)a

Sensitive to the inflation

factor/size of noise

Inflation factor

By Tim Supinie

Perfect model scenario– Multiplicative: 1.09– Relaxation: 0.44– Adaptive: 0.43

Imperfect model scenario– Multiplicative: 1.12 -> filter divergence– Relaxation: 0.5 -> filter divergence– Adaptive: 0.8

Change in ensemble spread

By Tim Supinie

Change in ensemble spread

By Tim Supinie

Additive vs. Adaptivet = 1500 sec

Additive noise Adaptive

MAX: 30.88Min: -34.56

MAX: 31.17Min: -27.37

Wz=7km

corr(Z, qr)z=2km

Additive vs. Adaptivet = 3600 sec

Additive noise Adaptive

MAX: 37.12Min: -20.20

MAX: 25.68Min: -27.68

efmean

enmean

Additive (0.5 to u, v, T) vs. Adaptive (0.85)

Sky: Additive + multiplicativeOrange: Adaptive

Certain DSD parameters such as the bulk densities and the intercept parameters of hydrometeors greatly influence the evolution of storm through microphysical processes.

Significant uncertainties exist in those parameters.

Several studies have shown that the EnKF method is capable of successfully identifying parameter values during assimilation process and, therefore, may help improve forecast (Annan et al. 2005a,b; Annan and Hargreaves 2004; Hacker and Snyder 2005; Aksoy et al. 2006a,b; Tong and Xue 2008a,b).

Parameter estimation

Parameter estimation (single-parameter)

Perfect observation operator Imperfect observation operator

Tong and Xue (2008)Jung et al. (2010)

√√

√

Parameter estimation (three-parameter)

Perfect observation operator Imperfect observation operator

Tong and Xue (2008)Jung et al. (2010)

Shade: log10(N0r) for the ensemble mean of EXP_DM at z = 100 m AGLContour: ZDR log10(8x105) ≈ 5.9

Parameter estimation

Example of high hail bias29-30 May 2004 supercellMilbrandt and Yau SM scheme

Ensemble mean analysis at z = 100 m and t = 60 min

0.10.1

Example of high hail bias29-30 May 2004 supercellLFO scheme

Ensemble mean analysis at z = 2 km and t = 60 min

Error in the microphysics scheme

By Tim Supinie

Analyzed polarimetric variables vs. observed(MY)(LIN)

excessive size sorting ?

Assimilating ZDR using a SM scheme

z = 2 km

No ZDR With ZDR

Model error becomes a huge issue for real-data cases.

Various covariance inflation methods are found to be helpful but each method has its own limitations. Understanding strength and weaknesses of each method can help make better use of them.

Additional observations can help only if the observations carries information that the model can handle.

Summary

Certain microphysics bias is very hard to treat and can be further deteriorated during data assimilation when the problem is seriously under-constrained by observations.

Observation operator errors can significantly influence the quality of analysis for storm scale DA.

Therefore, there should be continuous efforts to improve the model and the observation operator.

Summary