using tigge data to understand systematic errors of atmospheric river forecasts

Using TIGGE Data to Understand Systematic Errors of Atmospheric

River Forecasts

G. Wick, T. Hamill, P. Neiman, and F.M. Ralph

NOAA Earth System Research Laboratory

Physical Sciences Division

Third THORPEX International Science Symposium September 18, 2009

Outline

• Introduction/Motivation• Data and Tools• Model Evaluations• Conclusions

Zhu & Newell (1998) concluded in a 3-year ECMWF model diagnostic study:1) 95% of meridional water vapor flux occurs in narrow plumes in <10% of zonal circumference.2) There are typically 3-5 of these narrow plumes within a hemisphere at any one moment.3) They coined the term “atmospheric river” (AR) to reflect the narrow character of plumes.

Given the above: ARs are very important from a global water cycle perspective.

Relationship to Precipitationand Flooding

Atmosphericriver

Objectives/Research Questions

• Are atmospheric rivers well represented in the model analyses and forecasts within TIGGE?

• Is the width of the features accurate and is this dependent on the model resolution?

• Are the frequency of the events accurately forecast?– Does this vary with forecast lead time?

• Is there any apparent bias in the model forecasts of total precipitable water?

• Is there a benefit in resolving atmospheric rivers through ensemble forecasts

Data

• TIGGE Control Forecasts of Total Column Water– ECMWF– UK Met Office– JMA– CMC– 12 Z initialization– 0, 3, 5, 7, and 10-day lead times

• Satellite retrievals of integrated water vapor from SSM/I– 12-hour grids centered at forecast time– Multiple retrieval algorithms implemented

• Analysis here for October 2007 – March 2008• Northeastern Pacific Region (15-55 N, 110-160 W)

“Bias” Adjustment

• Mean analysis total water adjusted to match observed precipitable water over full Pacific

• Accounts for forecast biases and differences between total column and precipitable water

• Critical for threshold-based identification

Model Bias (cm)

ECMWF 0.31

UKMO 0.49

JMA 0.22

CMC 0.35

Illustrating the Idea

18 hour lead

66 hour lead

114 hour lead

162 hour lead

ECMWF Comparison against observations on 5 June 2007 at 6Z

Atmospheric River Identification

SSM/I Integrated water vapor (cm)

16-Feb-04;p.m. comp.

IWV >2cm:<1000 km wide

IWV >2cm:>2000 km long South coast

North

coast

1000 km

Objective River Identification Procedure

• Isolate top of the tropical water vapor reservoir

• Threshold IWV values at multiple levels and compute gradients

• Cluster points and compute skeleton to estimate river axis

• Identify points satisfying width criteria

• Cluster center points to identify segments of sufficient length

• Determine if river intersects land or is potentially influenced by data gaps

Application to Verification of TIGGE

• SSM/I Derived IWV for 16 November 2007

• ECMWF 120-hour forecast valid on 16 November 2007

Reproduction of Event Frequency

• Results searched to find days with at least one AR within the domain

• Days where observations potentially influenced by gaps in coverage removed from record

• Results expressed as % change from observation

• 100 AR days observed over season

0

5

10

15

20

25

30

35

0 3 5 7 10

Forecast Lead (Days)

Per

cen

t o

f O

bse

rved

AR

Day

s (%

)

ECMWF

UKMO

JMA

CMC

General overestimate observed

Prediction of Specific Events

Probability of Detection

False Alarm Rate

Representation of AR Width

• Compared width on days where 1 event was both observed and predicted

• Average computed over entire length of AR

ECMWF, 0 Lead

Representation of AR Core Strength

• Results similarly computed for average peak core strength along length of AR

ECMWF, 0 Lead

Predictability of Landfalling Events

Probability of Detection

False Alarm Rate

Conclusions to date

• Objective river identification and characterization tool developed to facilitate quantitative results

• Atmospheric rivers generally well-predicted in TIGGE models– Apparent over prediction of frequency– No significant bias in width or core strength

• Models generally similar in performance for predicting atmospheric rivers

• Need to further explore sensitivity to observations• Ultimately can explore potential benefit of resolving

atmospheric rivers through ensemble forecasts

Observational studies by Ralph et al. (2004, 2005, 2006) extend model results:1) Long, narrow plumes of IWV >2 cm measured by SSM/I satellites considered proxies for ARs.2) These plumes are typically situated near the leading edge of polar cold fronts.3) P-3 aircraft documented strong water vapor flux in a narrow (400 km-wide) AR (along AA’).4) Airborne data also showed 75% of the vapor flux was below 2.5 km MSL.5) Moist-neutral stratification <2.8 km MSL, conducive to orographic precip. boost & floods.

400 km

Atmos. river

Accuracy of the Satellite IWV Products

• Recently completed independent accuracy assessment with GPS radio occultation data from COSMIC

• Results suggest rms accuracy around 2.5 mm

Primary Extracted Characteristics

• Number of rivers present in scene

• Location of center points along core of river

• Width of river at all points along axis and average width

• Core IWV values along axis

• Orientation of river at all points along axis