Experimental Prediction of Climate-related Malaria Incidence

Krishnamurti, Chakraborty, V. Mehta, A. Mehta Monsoon and Impacts Workshop - Ahmedabad, India 7 February 2007

Experimental Prediction of Climate-related Malaria Incidence

Outline* Climate Variability and Malaria in India

* The FSU Super-ensemble Technique with 13 Coupled Climate Models for Rainfall Prediction: An Experiment for Malaria Incidence Prediction in Botswana, Southern Africa * Next Steps for a Malaria Early Warning System in (Western) India

T.N. Krishnamurti and Arindam ChakrabortyFlorida State University

Tallahassee, Florida, U.S.A.

Vikram M. MehtaThe Center for Research on the Changing Earth System

Columbia, Maryland, U.S.A.

Amita V. MehtaNASA-Goddard Space Flight Center

andUniversity of Maryland-Baltimore County

Greenbelt, Maryland, U.S.A.


Poverty and Health: Malaria, an Example of Vector-borne Diseases Influenced by Climate Variability and Change

Malaria around for 4,000 years, influenced human history to a great extent

According to the WHO’s World Malaria Report 2005, 3.2 billion people lived in areas at risk of malaria transmission at the end of 2004

350 to 500 million clinical episodes of malaria every year

At least one million deaths every year due to malaria

Potential destabilization of socio-economic-political systems, triggering national/international security problems


Influence of El Niño-La Niña Climate Variability on Indian Rainfall and Malaria Incidence

More rain and more malaria cases in western and northwestern India during La Niña (1996; left)

Less rain and fewer malaria cases in western and northwestern India during El Niño (1998; right)

Rain and malaria prediction 2-3 months in advance possible

El Niño-La Niña Climate Index (gray) and annual number of malaria cases in India (blue)

March-April-May

June-July-August

1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000

3 million cases

1.8 million cases

Jun-Jul-Aug 1996 La Niña Jun-Jul-Aug 1998 El Niño


Seasonal Rainfall and Malaria Incidence Prediction in Botswana in Southern Africa: A Case Study

• Malaria incidence dependent on rainfall, temperature, humidity, winds, land cover-use, topography, and other local conditions

• Accurate seasonal prediction of rainfall, temperature, other hydro-meteorological variables, and land cover-use very useful for early warning of malaria risk and decision-making about prevention/mitigation

• Application of the FSU multi-model synthetic super-ensemble technique with 13 coupled climate models in predicting malaria incidence a season in advance in Botswana, using only rainfall prediction


Atmospheric Component Oceanic Component

Model Res. Initial Condition Model Res. Initial Condition

ANR (FSU) FSUGSM with Arakawa-Schubert convection and new radiation (band model) T63L14 ECMWF with Physical

Initialization HOPE global 5o longitude, 0.5-5o

latitude, 17 levels Coupled assimilation relaxed to observed SST

AOR (FSU) FSUGSM with Arakawa-Schubert convection and old radiation (emissivity/absorptivity based)

T63L14 ECMWF with Physical Initialization HOPE global 5o longitude, 0.5-5o


KNR (FSU) FSUGSM with Kuo convection and new radiation (band model) T63L14 ECMWF with Physical



KOR (FSU) FSUGSM with Kuo convection and old radiation (emissivity/absorptivity based) T63L14 ECMWF with Physical



CCM3 (NCAR)

CCM3 atmospheric model T63L18 ECMWF SOM 2.4 o x 1.2 o -2.4o Coupled assimilation

POAMA1 (Australia)

BMRC Atmospheric Model (BAM3) R47L17 From latest atmosphere and ocean conditions from GASP

ACOM2 2o x 0.5o-1.5o, 25 levels

From ocean assimilation which was based on optimum interpolation (OI) technique.

CERFACS (France)

ARPEGE T63L31 ERA40 ERA40 2o x 2o, 31 Levels Forced by ERA40

ECMWF (Europe)

IFS T95L40 ERA40 HOPE-E 1.4o x 0.3o-1.4o, 29 levels Forced by ERA40

INGV (Italy) ECHAM-4 T42L19 Coupled AMIP type OPA 8.1 2o x 0.5o-1.5o, 31 levels Forced by ERA40

LODYC (France)

IFS T95L40 ERA40 OPA 8.2 2o x 2o, 31 levels Forced by ERA40

MPI (Germany)

ECHAM-5 T42L19 Coupled run relaxed to observed SST MPI-OMI 2.5o x 0.5o-2.5o, 23

levels Coupled run relaxed to observed SST

MetFr (France)

ARPEGE T63L31 ERA40 OPA 8.0 182 x 152 GP, 31 levels Forced by ERA40

UKMO (England)

ARPEGE 2.5 x 3.75, 19 levels ERA40 GloSea OGCM

HadCM3 based 1.25o x 0.3o-1.25o, 40 levels Forced by ERA40

Particulars of 13 Coupled Climate Models


Total Forecast Data Sets Available for the Present Study

(13 x 4 x 9 x 6 =) 28086 (starting every 3 months)

9Each of 7 DEMETER models (CERFACS, ECMWF, INGV, LODYC, MPI, MeteoFrance, UKMO)

(13 x 12 x 9 = ) 14043 (starting every month)

1POAMA1


1CCM3


1Each of 4 FSU models (ANR, AOR, KNR, KOR)

Total forecasts during 1989-2001

Length of forecasts (in

months)

Forecasts per month

Model

Total Forecasts = 23400 ( = 468 x 5 + 1404 + 2808 x 7)


Methodology

• Seasonal, super-ensemble rainfall forecasts from 13 coupled atmosphere-ocean models from March 1989 to February 2002 over 17.5o-30.0oE, 27.5oS-17.5oS

• Three months’ lead time forecasts for the peak malaria incidence month of March


Adjusted Malaria Incidence and Rainfall

December to February 1981-82 to 2001-02

• Adjusted log (malaria incidence) (AMI, per 1000 people) in Botswana related to summer (December-February) rainfall (and other factors)

• Initial increase in AMI with increase in rainfall and decrease in very heavy rainfall because mosquito breeding areas washed away

Empirical relationship between AMI and rainfall P (mm/day) in Botswana:

AMI = -0.2541 P2 + 1.9558 P - 3.2823


Super-ensemble (SSE) provides more accurate rainfall forecasts compared to the ensemble mean (EM).

Step 1: December Forecast of December-January-February Rainfall

Rainfall Forecasts made in December for December-January-February Season usingSuper-ensemble and Ensemble-mean Techniques


SSE Provides Better Malaria Prediction (Correlation = 0.19, RMSE = 0.52) Compared to EM (Correlation=-0.47, RMSE=0.68).

Step 2: Forecast of March Malaria Incidence from the December Rainfall Forecast

0.25

1

4

Malaria Cases per 1000People

More AccurateForecasts ofLarger Outbreaks


Next Steps to Develop a Malaria Early Warning System in Western India

* A network of government and private health professionals, hydro-meteorological specialists, and climate prediction specialists

* A malaria observing system consisting of weather observing stations, malaria data gatherers, and a central/distributed data archive system

* A very-high resolution (~10 kms.) seasonal climate prediction system for Western India; synergy with agricultural and water resources impacts prediction

* Quantification of relationships between malaria incidence, hydro-meteorological variables, land cover-use, and other local factors at a very-high spatial resolution

* Quantitative assessments of monsoon climate variability’s, including extreme weather events’, impacts on vector-borne diseases, especially malaria, regional economies, and other societal matters


Thank youThank you


Multimodel FSU Conventional Super-ensemble The superensemble forecast is constructed as,

S ai(Fi F i) O i1

N

are the ith model forecasts.

are the mean of the ith model forecasts over the training period.is the observed mean of the training period.are the regression coefficient obtained by a minimization procedure during the training period. Those may vary in space but are constant in time.is the number of forecast models involved.

iF

iF

O

ia

where,

N

The coefficients ai are derived from estimating the minimum of function,

G (Si

i 1

Ntrain

Oi)2 the mean square error.

E 1

N(Fi F i) O

i1

N

Multimodel bias removed ensemble is defined as,

In addition to removing the bias, the superensemble scales the individual model forecasts contributions according to their relative performance in the training period in a way that, mathematically, is equivalent to weighting them.


Generating Synthetic Data Using EOF

Multimodel Synthetic Ensemble/Superensemble Prediction System

N - Actual Data Sets

Fi(x,T) Fi, n(T).n

i, n(x)

Observed Analysis

O(x,T) Pn(T).n

n(x)

PC EOFTraining Forecast

i = model

n = mode

Estimating Consistent Pattern

What is matching spatial pattern in forecast data Fi(x,t), which evolves according to PC time series P(t) of observed data, O(x,t) ?

P(t) i,nFn,i

i, n(t) (t)

Fireg (T) i,n

n

Fi,n (T)Forecasts

Normalized Weights

N - Synthetic Data Sets )().(),( , xTFTxF nn

regni

syni

Observation

Obs

Experimental Prediction of Climate-related Malaria Incidence

Documents