Acknowledgements: Collaborators : Dr. Judy Curry, Dr. Hai-Ru Chang, Dr. Jun Jian, Dr. Paula Agudelo, Dr. Carlos Hoyos, James I. Belanger Thanks to collaborators.

Acknowledgements:

Collaborators : Dr. Judy Curry, Dr. Hai-Ru Chang, Dr. Jun Jian, Dr. Paula Agudelo, Dr. Carlos Hoyos, James I. Belanger

Thanks to collaborators in South Asia especially A.R. Subbiah, and Tom Brennan, Tim Palmer at ECMWF.

• Each year, 10’s of millions (mostly the poor) face danger and loss due to tropical cyclones and monsoonal flooding

• Flooding is the greatest cause of death and destruction in the developing world (e.g., India 6 million ha inundated, affecting 35-40 million people)

• Our advanced prediction system allows these risks to be forecast 10-14 days and up to 1-3 months ahead.

• Three examples:

1. Tropical cyclones forecasting

2. Flood forecasting in Bangladesh

3. Flooding in Pakistan

OVERVIEW

MAJOR RIVER BASINS IN SOUTH ASIA

Major rivers, fertility and high population density

Region with catastrophic weather events

In: IndusG: GangesB: BrahmaputraI: IrrawaddyM: MekongR: RedJ: JiangY: YangtzeYe: Yellow

1. Tropical cyclones forecasting

Nargis and Myanmar: 2008

Location Country Year Ocean basin Fatalities

1 Bhola Island

Bangladesh

1970

Bay of Bengal

>350,000

2 Hooghly River India 1737 Bay of Bengal 350,000 3: Haiphong Vietnam 1881 West Pacific 300,000 4 Backerganj I Bangladesh 1584 Bay of Bengal 200,000 5 Backerganj II Bangladesh 1876 Bay of Bengal 200,000 6 Chittagong Bangladesh 1897 Bay of Bengal 175,000 7 Nina China 1975 West Pacific 171,000 8 CoxÕs Bazaar Bangladesh 1991 Bay of Bengal 140,000 9 Nargis Myanmar 2008 Bay of Bengal 138,000

10 Bombay India 1882 Arabian Sea 100,000 11 Hakata Bay Japan 1281 Japan 65,000 12 Calcutta India 1864 Bay of Bengal 60,000 13 Barisal Bangladesh 1822 Bay of Bengal 50,000 14 Calcutta India 1942 Bay of Bengal 40,000

Deadliest Tropical CyclonesDeadliest Tropical Cyclones

Almost all of the worst tropical cyclone disasters have occurred in the Bay of Bengal: 17 of the highest 25

The Nargis Anomaly– Bhola Cyclone (1970) no warning or evacuation

At least 350,000 people killed

– Chittagong Cyclone (1991) limited warning or evacuationAt least 138,000 killed; $2.2 billion

in economic damage

– Cyclone Sidr (2007) 3,000 people killed; $1.7 billion

in economic damage

– Cyclone Nargis (2008)

138,000 people dead or missing;

$11 billion in economic damage

So, what went right in Bangladesh in 2007 (Sidr) and what went wrong in Myanmar in 2008?

good forecasts (IMD/CFAB), excellent communications on ground, evacuations

fairly good forecasts, good IMD communication, no surge forecast, limited/zero response in Myanmar

Forecasting North Indian Ocean Tropical cyclones on 1-15 day time scales

• Probabilistic forecasts of tropical cyclone are made routinely by CFAN

• North Indian Ocean-west Pacific particularly predictable.

• Genesis and landfall of Sidr and Nargis predicted 8.5 and 13.5 days in advance and 7.5 and 11 days, respectively

• Forecasts shown are lower resolution than current schemes. Nor has there been the statistical rendering such as CFAN does daily in Atlantic

Tropical Cyclone Nargis 2008

• First Tropical Storm Advisory: 4/27 12 UTC• Landfall: Ayeyarwady Div. of Burma 5/2 12

UTC• Maximum Intensity: 115 kts

Cyclone Nargis on 5/2 4:40 UTC Image courtesy of MODIS Rapid Response Project at NASA/GSFC

Tropical Cyclone Nargis Forecast Animation

FCST APRIL 19, 2008

-13 DAYS TO LANDFALL -8 DAYS TO IMD TC RECOGNITION

FCST APRIL 21, 2008

-11 DAYS TO LANDFALL -6 DAYS TO IMD TC RECOGNITION

FCST APRIL 23, 2008

-9 DAYS TO LANDFALL -4 DAYS TO IMD TC RECOGNITION

FCST APRIL 25, 2008

-7 DAYS TO LANDFALL -2 DAYS TO IMD TC RECOGNITION

FCST APRIL 27, 2008

-5 DAYS TO LANDFALL 0 DAYS TO IMD TC RECOGNITION

FCST APRIL 29, 2008

-3 DAYS TO LANDFALL 0 DAYS TO IMD TC RECOGNITION

FCST MAY 1, 2008

-1 DAYS TO LANDFALL +2 DAYS TO IMD TC RECOGNITION

SUMMERY OF PREDICTIVE SKILL IN THE NORTH INDIAN OCEAN

• Northern Indian Ocean tropical cyclones have the greatest predictability of all tropical oceans

• The probabilistic forecasts have over a week lead on Indian Meteorological Department Forecasts

• Plans produce probabilistic to 15 days for the NIO eventually extending to 30 days (as per NATL)

• As deltas are very susceptible to storm surge, we plan to add a storm surge component

Is it possible to make a difference by implementing a robust forecasting system and risk management strategies of extreme

events?

2. End-to-end prediction in Bangladesh:

Extended probabilistic flood

forecasts on medium and long time scales

Seasonal: 1-6 months STRATEGIC

Intraseasonal: 15-30 days STRATEGIC/TACTICAL

Short-term: 1-15 days TACTICAL

Provide: Overlapping forecasts to allow both strategic and tactical decisions for disaster mitigation, water resource management and agricultural optimization:

AIM: OVERLAPPING 3-TIERED SYSTEM

Produce: o A system that takes developed world technologies and interfaces them with the needs and abilities of developing world infrastructures

o Probabilistic forecasts to allow proper risk assessment

o A system that is useable and adaptable throughout the developing world

Quantitativeinformationfrom the

user community

Combination of probabilistic forecasts of

system plus user information produces easy to understand aggregate risk analysis for decision

making

RENDERING THE FORECAST USEFUL

Probability of the occurrence event determined by ensemble forecasts

Costs of strategies to mitigate impacts determined by user community

Probabilities and costs of occurrence combine to produce an aggregate risk

User community makes decisions based on aggregate risk assessment

COMBINING PROBABILITIES WITH SPECIFIC PROBLEMS TO DEVELOP A MEASURE OF AGGREGATE RISK

Risk = (cost) x (probability)

SITUATION:o Ganges and the Brahmaputra exist within two of the largest

catchment basins in the world

o Large agrarian populations subject to a host of environmental extreme events

Challenge:•Extended stream flow forecasts out to 10-days at entry points of the Ganges (G), Brahmaputra ( B) and Meghna (M) into Bangladesh•Probabilistic forecasts for assessment of flood risk•Forecast communication to union and village level.

Framework: •Europe: ECMWF rainfall and meteorological forecasts•USA: statistical rendering and hydrological forecasts•Bangladesh: Incorporation of forecasts into a national disaster management scheme

Bangladesh

DATA ISSUES:o Hydrological streamflow data collected at the borders of India and

Bangladesh: no upstream Indian data

o Satellite data and derived precipitation products from NASA and NOAA

Question: Is it possible to produce forecasts with no upstream data from within the Brahmaputra and Ganges catchments?

Can ECMWF precipitation forecasts allow the calculation of “synthetic” hydrological data?

DISCHARGE FORECAST SCHEMES: Hopson & Webster 2010 Webster et al. 2010

10-day means: 40 members Daily: 51 members

DISCHARGE FORECAST SCHEMES: Hopson & Webster 2010 Webster et al. 2010

STATISTICAL RENDERING OF ENSEMBLE PRECIPITATION FORECASTS

Using a “quantile-to-quantile” technique, we adjust each ensemble forecast of precipitation using satellite data as the statistical base.

Comparison of the ensemble mean 0-144 hr adjusted precipitation July 20-26 2007 prior (as it turned out) to a major flooding event.

PRELIMINARY 1-10 DAY FORECASTS

• Real-time experimental forecast were made for 2004 and 2006. If successful, would move on to a fully operational system

• Opportunity taken to introduce the concepts of probabilistic forecasts and risk to Bangladesh Government institutions and NGOs.

• During this period, the Ganges has not exceeded flood level. Thus only the Brahmaputra forecasts are shown. Ganges forecasts did not possess any false positives during these years (see http://cafan2.eas.gatech.edu)

B

G

2004 real time 10-day forecasts of the Brahmaputra

• No flooding of Ganges

• Early Brahmaputra flooding with double peak

Ensemble member distribution (day +10)

Probability of exceeding flood level 10-days in advance calculated from ensemble spread

For forecast made July 08 for 17, 2004

Note: If only one forecast had been made, rather than an ensemble of forecasts, any forecast (e.g., “a” or “b” would have been equally likely and there would be no information in the forecast

“a” “b”

flood level

• During 2007 and 2008, the forecasts were used operationally in Bangladesh

• In test regions the forecasts were transmitted (via cell phone) to district leaders and from there to the village level

• Prior training was instigated for evacuation, livestock safety, early harvest, safe drinking water ahead of high probability forecast of floods

• Forecast for 2008-present were produced

RESPONSE TO BANGLADESH FLOOD FORECASTS

Flood Forecasting and Warning Center (FFWC) • Water level forecasts throughout

Bangladesh • Localised flood inundation maps

National level Disaster Emergency Response Group • Emergency response plans, logistics for

preparedness and relief in advance*

NGOs • 1-10 days forecast• Evacuation and response plans to protect

lives and livelihoods Local project partners

• Assessed the risk of flooding based on localised flood inundation maps

District level relief and emergency • Mobilise resources for relief activities

* This is the first time that Bangladesh acted ahead of major flooding!

ECONOMIC ANALYSIS OF THE 2007-2008 MITIGATION

• The 10 day forecasts were found to permit the

following savings:

House $130 (4 months labor*)

Agricultural $190 (6 months)

Household effects $270 (10 months labor)

Livestock $500/cattle (2 years labor)

* Based on $350/year average pay rural Bangladesh

SEASONAL 1-6 MONTH DISCHARGE FORECASTS

• In 2007, ECMWF System-3 coupled ocean-atmosphere model became available

• Jian et al (2009: QJRMS) in a diagnostic study had shown that there was extended predictability, especially for the Ganges.

• System-3: 110 km resolution, 41 member ensembles each month out by 7 months

• Experimental 1-6 months forecasts were run in real-time from April, May, June, July for 6 months.

• System showed extended flood risk at the correct time of the season in 2007 and 2008

DOWNSCALED 2007 FORECASTS

• Downscaled seasonal forecast predict high probability of July-August flood level exceedance

In: IndusG: GangesB: BrahmaputraI: IrrawaddyM: MekongR: RedJ: JiangY: YangtzeYe: Yellow

Asian river basins where CFAB 1-15 day forecast are applicable and seasonal predictability established

During July and early August, torrential rainfall caused widespread floods in the Indus Valley

Many millions of people were affected, 160,000 km2 inundated, 1600 killed immediately, and water borne diseases occurred

2. WHERE THE JULY/AUGUST PAKISTAN FLOODS PREDICTABLE?

WAS 2010 AN EXREME YEAR?

Climatologically, monsoon rainfall decreases across India towards Pakistan

Pakistan on western edge of monsoon

July/early August was an active phase of the monsoon

Webster et al. 2010

Climate MJJA

Northward propagation: active monsoon phase

WAS 2010 AN EXREME YEAR (cont.)?

Monthly Pakistan Rainfall Anomalies

Extreme Rain Events

Webster et al. 2010

• Large interannual variability with rainy and drought seasons

• In terms of extreme events there are a number of years that are quite similar to 2010.

• Over northern Pakistan extreme events in 2010 much larger than other years except 2008

• There have been other major flood years: 1950, 1973, 1976, 1977, 2001, 2008

• Floods in 2010 occurred because of high rain rates over mountainous north. 2009 drought and logging probably exacerbated run-off.

PREDICTABILITY OVER NORTHERN PAKISTAN JULY-AUGUST 2010

• ECMWF operational medium range ensemble forecast:

Observation Forecast IC 07/27/2010July 28-29 Event

• Precipitation forecasts (made 6 and 4 days in advance) of the major northern July precipitation event compare well with satellite observations

PREDICTABILITY OVER NORTHERN PAKISTAN JULY-AUGUST 2010 (cont.)

July Rainfall Predictability (2007-2010)Probability that the predicted rainfall>July Mean + 1 StDev

• Each of the July/August major pulses of rainfall over northern Pakistan is forecasted 6-10 days in advance

Major Weather Phenomena with High Socioeconomic Impact

Major Weather Phenomena with High Socioeconomic Impact

References:

Webster, P. J., Thomas M. Hopson, Carlos D. Hoyos, Jun Jian, Hai-RuChang, Paula A. Agudelo, Judith A. Curry, Timothy N. Palmer, A. R. Subbiah, Robert L. Grossman, 2009: Extended-range probabilistic forecasts of river discharge in large basins: A Bangladesh experience, Submitted to Bull Amer. Meteorol. Soc.

Hopson, T.M., and P.J. Webster, 2009: A 1-10 day ensemble forecasting scheme for the major river basins of Bangladesh: forecasting severe floods of 2003-2007. J. Hydromet. In press

Webster, P. J., T. Hopson, C. Hoyos, A. Subbiah, H-. R. Chang, R. Grossman , 2006: A three-tier overlapping prediction scheme: Tools for strategic and tactical decisions in the developing world. In Predictability of Weather and Climate, Ed. T. N. Palmer, Cambridge University Press. P 645-673.

Webster, P.J., V.E. Toma and H-M Kim, 2010: Were the 2010 Pakistan floods predictable? Submitted to GRL.

Supplemental Slides

Tropical Cyclone Tracking

• Tropical cyclone detection uses a modified tracking scheme of Vitart (1997)

• Variables describe the structure of the system and include:Tropical Cyclone Center: –Local Max 850 hPa Relative Vorticity > 3.5x10-5 s-1

–Local Min of Sea Level Pressure

To Eliminate Extratropical Cyclones:Warm Core Center: –Closest Local Max Temperature in 500-200 hPa layer –Closest Local Max Thickness in 1000-200 hPa layer.

3-days 10-days 15-days

SUMMERY OF PREDICTIVE SKILL IN THE NORTH INDIAN OCEAN

• Northern Indian Ocean tropical cyclones have the greatest predictability of all tropical oceans.

• CFAN probabilistic forecasts have over a week lead on Indian Meteorological Department Forecasts

• CFAN proposes to produce probabilistic to 15 days for the NIO eventually extending to 30 days (as per NATL)

• As deltas are very susceptible to storm surge, we would plan to add a storm surge component