1. Ryan Kang, Wee Leng Tan, Thea Turkington, Raizan Rahmat
Use of ECMWF Subseasonal-to-Seasonal (S2S) Predictions for Extreme Temperature Forecasts over Singapore and the surrounding regions during the April 2016 heatwave episode
Ryan Kang, Wee Leng Tan, Thea Turkington, Raizan Rahmat
Subseasonal and Seasonal Prediction Section
FOCRAII 2019
9 May 2019

2. Outline Climate information of mainland Southeast Asia
Case Study: Apr 2016 “Heatwave”
Decaying phase of a strong El Niño 2016
Model data and methods (Deterministic and Probabilistic)
Case study analyses
Skill assessment
Summary and Conclusions
Future works

3. Climate information of mainland Southeast Asia (MSA)
Extreme temperatures in Southeast Asia caused by El Niño and worsened by global warming (Thirumalai et al., 2017, Nature Communications)
*Heatwave episodes in Singapore from 1979 onwards
Year
Episodes
1983
March 10-14, March 24-28, April 9-21
1998
March 20-29
2010
March 6-9
2016
April 17-19
Surface air temperature climatology in MSA based on the entire CRU data set (1901–2014), which indicates that
April is the warmest month in the region.
* In Singapore, a heatwave is defined as occurring when the daily max temperature is at least 35°C on 3 consecutive days and the daily mean temperature throughout the period is at least 29°C

4. “Heatwave” over Peninsula
CNA, 22 Apr 2016
Straits Times, 22 Apr 2016
Straits Times, 18 Mar 2016
Early 2016 heatwave over Peninsula, April hottest
(Max) temperature > 37˚C for more than 3 days (Perlis and Pahang)
Schools closed in Malaysia, heat stroke conditions
El Niño to blame; week-to-week variations in conditions are of interest

5. Model data and method Subseasonal-to-Seasonal (S2S) Predictions
ECMWF (51 ensemble members), every Mondays and Thursdays
Variable: Weekly average of daily mean temperature (T2M)
Hindcast model calibration with past 20 years (3 start dates, centered within the forecast window of 1-week, 11 x 3 X 20 = 660 re-forecast members)
Lead-dependent model climatology for mean-bias correction of model drifts
Deterministic and Probabilistic Products
Anomaly (Ensemble mean - hindcast mean)
Probability exceeding certain %-tile thresholds (Ranking and counting method)

6. CASE STUDY ANALYSIS

7. Case Study (Apr 2016): ERA Int Anomalies
WK1
WK3
WK2
WK4
Week 1 to week 2
Also very warm for southern China

8. Case Study (Apr 2016): ERA, Zoom in …
Week 1 to Week 2 - Became warmer
Week 2 - Peak of the heat wave over Singapore
WK1
WK2
Week 3 - Some signs of receding warm conditions from the south
Week 4 - Relief at western coast and further receding warm conditions over Singapore
WK3
WK4

9. Case Study (week of 11 Apr ‘16) - Deterministic ECMWF S2S Forecasts
Obs Anomalies
LT1
LT2
Warm week captured up to LT2
More representative in central/north
LT3

10. Obs Percentile (>99%)
Case Study (week of 11 Apr ‘16) - Probabilistic ECMWF S2S Forecasts
Obs Percentile (>99%)
LT1
LT2
A probability of >50% for “Weekly average temperature Above 99% threshold” over Singapore for LT2
LT3

11. Case Study (week of 25 Apr ‘16) - Deterministic ECMWF S2S Forecasts
Obs Anomalies
LT1
LT2
Receding pattern captured up to LT2
Potentially useful forecast for cessation
LT3
LT4

12. Case Study (week of 25 Apr ‘16) - Probabilistic ECMWF S2S Forecasts
Obs Percentile (>99%)
LT1
LT2
Low probability for “Weekly average temperature Above 99% threshold” over Singapore, up to LT2
LT3
LT4

13. SKILL ASSESSMENT

14. Verification data and method
ERA Interim (Re-grid to 1.5° x 1.5° resolution, same as S2S resolution)
Verification Method
No ‘standard’ way unlike seasonal predictions (pentad, 7-day, 10-day?)
Example:
Forecast first 7-day week: Apr
Lead time (LT) 1: 4 Apr; LT 2: 28 Mar, LT 3: 21 Mar, LT 4: 14 Mar
Sample size: Only 20 years (too little?)
Use target “full-month” hindcast verification (for increased robustness, more samples)
Assessment: Anomaly Correlation Coefficient (ACC) and Mean Square Skill Score (MSSS) for anomaly plots

15. Anomaly Correlation (ECMWF vs ERA Int)
LT1
LT2
LT3
LT4

16. MSSS (ECMWF vs ERA Int)
LT1
LT2
LT3
LT4

17. Summary and Conclusions
Warm week of 11 Apr 2016 is predicted by ECMWF S2S model up to a lead time (LT) of 2 weeks
Receding warm spatial pattern conditions, for the week of 25 Apr 2016 was also captured by the model up to a LT of 2 weeks
Relatively high skill for the Peninsula region: MSSS ranges between 0.3 to 0.7 up to a LT of 4 weeks

18. Summary and Conclusions
Demonstrates the ability of the ECMWF model to forecast week-to-week variations in temperature, including the peak and cessation of warmest temperature
Opportunity to provide products for worsening and/or improving extreme temperature conditions
Important implications in public’s preparedness against heat exhaustion between the weather (days) and seasonal (months) timescales

19. Future Works
Generate heat wave index: Probability of daily temperature exceeding 90%, 95% or 99%-tile for at least three consecutive days within a week
Hindcast verification score (ROC) for “Probability exceeding certain percentile thresholds” probabilistic products

21. Case Study (Apr 2016): ERA Int Percentile
WK1
WK3
WK2
WK4

22. ERA-Interim obs dataset
Initial dates for 2016 model runs
Verification e.g.: April, temperature, for each grid
4 Apr
7 Apr
11 Apr
14 Apr
18 Apr
21 Apr
25 Apr
28 Apr
6 x 20 = 120 samples
Wk 1 (4-10 Apr)
Wk 3
Wk 5
Weeks for verification
Wk 2
Wk 4
Wk 6
ERA-Interim obs dataset
ECMWF Forecast
Climo ‘96-’15
2016 Anomaly
Climo ‘96-’15
Anomaly
Climo ‘96-’15
Anomaly
1996
1997 …
...
2014
2015
1996
1997 …
...
2014
2015
1996
1997 …
...
2014
2015
Lead Time 1
Fcsts …
Lead Time 4
Fcsts