Application of ensemble prediction for rainfall forecasts

1
Application of ensemble prediction for rainfall
forecasts
www.ec.gc.ca
X. Deng, G. Candille, P. Houtekamer and G.
Pellerin Meteorological Service of Canada
Oct. 2006, Chengdu, China
2
Outline

• Introduction
• Verification results of the Canadian EPS
  precipitation forecasts during Summer
• Tropical Storm Ernesto case
• Verification against analyses
• Verification against observations
• Summary

3
Introduction

• The present Canadian ensemble prediction system
  (EPS)
• 16 medium range integrations
• horizontal resolution 1.2 degree
• Ensemble Kalman filter
• Precipitation products from the present Canadian
  EPS
• Probability forecasts and ensemble mean

4
Relative Operating Characteristic (ROC) Curves
for Summer 2005 and 2006
On 13 December 2005, the EPS had major changes in
the EnKF data-assimilation system and in the
configurations of the forecast models
5
Tropical Storm Ernesto (August/September 2006)
03 UTC Sept. 1
03 UTC Aug. 29
6
Ernesto made landfall near Wilmington, North
Carolina at 330 UTC 1 September 2006
7
Precipitation analysis vs. precipitation
observations
24-h accumulation SHEF rain gauge data, (2888
stations east of 85W)
Optimum Interpolation precipitation analysis,
done every 6h, and summed over 24h
12 UTC Aug. 31 12 UTC Sept. 1
8
Probability forecasts for the 24h accumulation
exceeding 10mm valid on Sept. 1, 2006 00 UTC
9
Probability forecasts for the 24h accumulation
exceeding 25mm valid on Sept. 1, 2006 00 UTC
10
Probability forecasts for the 24h accumulation
exceeding 50mm valid on Sept. 1, 2006 00 UTC
11
24h precipitation accumulation, 012-036 h
ensemble mean forecast and verifying analysis
Solid black line ensemble mean Dashed red line
OI PCP analysis
12
Verification against observations (10mm)
pcp 10mm, pc26 (3744)
Bres
Brel
BSS
BSS 1 Bres -Brel
13
Verification against observations (25mm)
pcp 25mm , pc12 (1737)
Bres
Brel
BSS
BSS 1 Bres -Brel
14
Verification against observations (50mm)
pcp 50mm, pc4 (531)
Bres
Brel
BSS
BSS 1 Bres -Brel
15
Histogram for observations and forecasts
16
Predicted vs. observed 24h accumulation
17
Summary

• The ROC scores show improvement of the EPS
  precipitation forecasts from Summer 2005 to 2006.
• The probability forecasts of precipitation have
  value up to 3-4 days.
• The ensemble mean in short range forecasts
  matches the analysis patterns of precipitation.
• The objective verifications show skills up to 3
  days for 10 mm threshold, and also for 25 mm
  threshold with less confidence.
• The EPS shows overforcast for lighter amounts of
  precipitation and underforecast for higher
  amounts. Increasing horizontal resolution should
  improve the forecasts.
• Equal weight ensemble average may not be a good
  approach for heavy rainfall forecasts.

18
www.ec.gc.ca