Seasonal Predictions of Very Hot Days in HK using NCEP CFSv2 Francis - PowerPoint PPT Presentation

Seasonal Predictions of Very Hot Days in HK using NCEP CFSv2 Francis Tam 1,2 , Wai Kit Lo 2 , Kunhui Ye 1 1 Guy-Carpenter Asia-Pacific Climate Impact Centre, City University of Hong Kong 2 School of Energy and Environment, City University of Hong Kong ������������������������������������������������������������������������� ��������������� !"

Introduction – I – “ Predicting the statistical summary ” (ECMWF, 2013) � Seasonal forecasts are carried out by major operational centers – Dynamical forecast systems (GCM-based) are commonly used – e.g. rainfall, temperature, tropical cyclones (Chan et al. 1998; Chan and Shi 1999; Zhang et al. 2012, Sohn et al. 2013, Tung et al. 2013) � Seasonal forecast from Hong Kong Observatory since 2007 – Global-Regional Climate Model (G-RCM) from Experimental Climate Prediction Centre (ECPC) Configurations of the G-RCM School of Energy and Environment 1

Introduction – II � Other types of extremes: heat waves – unpleasant feelings, financial costs (e.g. energy), environmental loss (e.g. wildfire), health care issues, and even fatalities � Recent study on heat wave predictability in US (Luo and Zhang 2012) Observations NCEP CFS run initiated in April 2011 School of Energy and Environment 2

Introduction – III � Can heat waves over China be predicted a season ahead? T max >35 o , 3-5d T max >35 o , >5d (from T. Ding et al. 2010) Here we study the seasonal predictability of very hot days in Hong Kong School of Energy and Environment 3

Data: HKO Observations � Historical observations from Hong Kong Observatory (HKO) – Measured at Hong Kong Observatory Hq – 24 times everyday � Air Temperature [1947/01/01 – 2008/06/30] � Relative Humidity [1965/01/01 – 2008/06/30] – Time period � 1982 to 2007 (26 years) � 06/01 to 09/30 (JJAS) School of Energy and Environment 4 ������� ���� � �������� ������������������� ��������������� ������������� ����������������� �����������

Data: CFSR Reanalysis � Climate Forecast System Reanalysis (CFSR) (Saha et al. 2010) – Global, high-resolution � 0.5 degree horizontal resolutions � 64 vertical levels (surface to 0.26hPa) – Coupled atmosphere-ocean-land surface-sea ice system – Includes observed CO 2 variations, changes in aerosols, other trace gases and solar variations – Variables used: � Temperature @ sigma=0.995: TMP_P0_L104_GLL0 � Relative Humidity @ sigma=0.995: RH_P0_L104_GLL0 School of Energy and Environment 5 ������� ���� � �������� ������������������� ��������������� ������������� ����������������� �����������

Data: Hindcast experiments by CFS � Climate Forecast System version 2 (CFSv2) (Saha et al. 2013) – Quasi-global, fully coupled atmosphere-ocean-land model � R2 NCEP/DOE Global Reanalysis: for atm. & land surfaces initial conditions � GODAS: for ocean initial states � GFS: for atmospheric model � MOM3: for the ocean component – New in version 2 � Four level soil model � Interactive three layer sea-ice model – Variables used � Temperature @ 2m: TMP_P0_L103_GGA0 � Specific Humidity @ 2m: SPFH_P0_L103_GGA0 School of Energy and Environment 6

CFS Hindcast Configuration Hindcast Configuration for CFSv2 (T126L64) Jan 2 Jan 1 Jan 3 Jan 4 Jan 5 Jan 6 0 6 12 18 0 6 12 18 0 6 12 18 0 6 12 18 0 6 12 18 0 6 12 18 9 month run 1 season run 45 day run 1st 6th 11th 16th 21st 26th 31st 0 6 12 18 0 6 12 18 0 6 12 18 0 6 12 18 0 6 12 18 0 6 12 18 0 6 12 18 7x4=28 Hindcast Runs initiated in May 28 in total … School of Energy and Environment 7

Measurement of Heat Stress to Human � Various approaches to apparent temperature (AT) (Steadman 1979, Steadman 1984, Steadman 1994) – Wet Bulb Globe Temperature (WBGT) � Developed in the late 1950s for the US Marine Corps Recruit Depot � Designed to be a measure of heat stress for human beings (American College of Sports Medicine 1984, ISO 7243) WBGT = 0.7 T w + 0.2 T g + 0.1 T d where T w the wet-bulb temp., T g the globe temp., T d the dry-bulb temp. [for outdoor] � Approximation by HKO ( 梁延剛 et al. 2009) WBGT = -12.065 + 1.193 T + 0.0688 RH School of Energy and Environment 8

JJAS Mean Anomalies Correlation WBGT WBGT (JJAS mean anomalies, CFSR vs CFSv2) (JJAS mean anomalies, HKO vs CFSv2) � Low correlation betn seasonal mean WBGT – CFSR vs CFSv2: R 2 is 0.0888 – HKO vs CFSv2: R 2 is 0.0233 School of Energy and Environment 9

Climatological Distributions WBGT WBGT WBGT upper 15% upper 15% upper 15% upper 33.3% upper 33.3% upper 33.3% lower 33.3% lower 33.3% lower 33.3% lower 15% lower 15% lower 15% � 26-year WBGT distributions – No. of days are normalized to 122 for easier comparison – Upper and lower 15% and 33.3% thresholds are shown – Compared to HKO, CFSR and CFSv2 are less skewed School of Energy and Environment 10

Determination of very hot days more (37.5%) (15.9%) fewer (4.25%) � Define “ Very Hot Days ” to be days with WBGT > threshold for upper 15% in climatology � Count no. days each year in each of these categories: – WBGT > top 15% – WBGT< bottom 15% – WBGT in between School of Energy and Environment 11

Correlation of no. of day prediction vs obs CFSv2 vs CFSR ������������������������� R 2 = 0.5818 CFSv2 vs HKO ������������������������� R 2 = 0.1964 School of Energy and Environment 12

Definition of “ “ “ Hot Season in JJAS ” “ ” ” ” � A “ Hot JJAS Season ” is one during which “ Very Hot Days ” occur more often than climatology more (37.5%) School of Energy and Environment 13

Hot JJAS seasons Percentage of Very Hot Days in JJAS (above 15% by WBGT) (WBGT above upper 15% threshold) 45 Hot Percentage of days above threshold for 15% HKO: observations 40 Hot CFSR: reanalysis 35 CFSv2: forecast Not Hot Hot 30 Hot Not Hot Hot Not Hot 25 Not Hot 20 15 10 5 0 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 � HKO data gives some obvious spikes (e.g., 83 ’ , 87 ’ , 90 ’ , 93 ’ 98 ’ ) � Both CFSR and CFSv2 show an increasing trend School of Energy and Environment 14

Performance in Classification of Hot / Not Hot Season of JJAS Above 15% HKO Hit Rate 0.769 Hot Not Hot False Alarm Rate 0.308 Hot 10 4 14 CFSv2 False Alarm Ratio 0.286 Not Hot 3 9 12 Hanssen-Kuipers Discriminant 0.462 13 13 26 – 13-13 distribution of “ Hot ” and “ Not Hot ” seasons from HKO � Rough comparison – 14-12 ratio in CFSv2 hindcast � Year-by-year comparison – Performance in Hit Rate (0.769), False Alarm Ratio (0.286) is acceptable, with Hanssen-Kuipers Discriminant of 0.462 School of Energy and Environment 15

Percentage of Very Hot Days Before vs. After Trend Removal – CFSR and CFSv2 tend to suggest more “ hot ” seasons after 1998 � Obvious increasing trend of hot years � Remove linear trends in JJAS mean temperature Percentage of Very Hot Days in JJAS (CFSR) Percentage of Very Hot Days in JJAS (CFSv2) School of Energy and Environment 16

Categorization after trend removal Above 15% HKO Hit Rate 0.700 Hot Not Hot False Alarm Rate 0.313 Hot 7 5 12 CFSv2 False Alarm Ratio 0.417 Not Hot 3 11 14 Hanssen-Kuipers Discriminant 0.388 10 16 26 � Rough comparison (Hot vs. Not Hot) – 10-16 for HKO and 12-14 for CFSv2 – fewer than before trend removal � Slight performance drops after trend removal – Hit Rate: 0.769 � 0.700; – False Alarm Ratio: 0.286 � 0.417 – Hanssen-Kuipers Discriminant: 0.463 � 0.388 School of Energy and Environment 17

Concluding remarks � CFSv2 shows some skill in forecasting the number of “very hot days” in HK – R 2 =0.1964 for WBGT > top 15% threshold from HKO data – R 2 =0.5815 based on CFSR data � CFSv2 gives promising results in capturing the occurrence of a hot season – Hit Rate = 0.796; False Alarm Ratio = 0.286 and Hanssen-Kuipers Discriminant = 0.462 � Hit Rate = 0.7 after removal of temperature trend. Thus the mean temperature trend does not contribute much to the seasonal forecast skill � Where does the seasonal prediction skill comes from? (ENSO?) School of Energy and Environment 18

Regression maps of JJAS mean (DJF) Corr = 0.6 Work is on going! Based on CFSR data School of Energy and Environment 19

Thank you! Q & A