Richard KiangNASA Goddard Space Flight CenterGreenbelt, MD 20771

Malaria Modeling for Thailand & KoreaMalaria Modeling for Thailand & Korea— NASA Techniques and Call for Validation Partners

AcknowledgementAcknowledgementAFRIMS Dr. Jame Jones WRAIR Dr. Russell Coleman

Dr. R. SithiprasasnaDr. Gabriella Zollner

USU Dr. Donald Roberts NDVECC Dr. David ClabornDr. Richard AndreDr. Leon RobertMs. Penny Masuoka

NGA Mr. John Doty DOS Mr. Andrew Herrup

UC Davis Dr. John Edman Cornell Univ. Dr. Laura Harrington

Mahidol Univ. Dr. S. Looareesuwan Thai MOPH Dr. J. SirichaisinthopDr. P. Singhasivanon Mr. S. NutsathapanaDr. S. LeemingsawatDr. C. Apiwathnasorn

RTSD Gen. Ronnachai Thai Army Lt. P. SamipagdiDr. Kanok

Mekong Malaria & FilariasisMekong Malaria & Filariasis

Richard.Kiang@nasa.gov

Malaria Cases

Tak

Kanchanaburi

Ratchaburi

Narathiwat

Ban KongMong Tha

Test SitesTest Sites

Ikonos

Filariasis poster Field work / Mahidol Field work / AFRIMS

Kanchanaburi

Ban Kong Mong Tha

Source: SEATMJ

DECISION SUPPORT

Vector Habitat Identification:• Determine when and where to apply larvicide and insecticide

Identification of Key Factors that Sustain or Intensify Transmission: • Determine how to curtail ongoing transmission cost effectively

Risk Prediction:• Predict when and where transmission may occur and how intense it may be

VALUE & BENEFITS

• Increased warning time

• Optimized utilization of pesticide and chemoprophylaxis

• Reduced likelihood of pesticide and drug resistance

• Reduced damage to environment

• Reduced morbidity and mortality for US overseas forces and local population

Mekong Malaria and Filariasis

Richard.Kiang@nasa.gov

Data

- temperature- precipitation- humidity- surface water- wind speed & direction- land cover- vegetation type- transportation network- population density

MEASUREMENTS

• Ikonos• ASTER• Landsat• MODIS• etc.

MODELS

• Vector Habitat Model• Malaria Transmission Model• Risk Prediction Model

HABITAT IDENTIFICATION

V & V

TRANSMISSIONPREDICTION

V & V

RISKPREDICTION

V & V

RISKASSESSMENT

SURVEILLANCE

MONITORING

CONTROL • Vector Control• Personnel Protection

PROJECT OBJECTIVESPROJECT OBJECTIVESINTEGRATED PEST

MANAGEMENT FOR DODINTEGRATED PEST

MANAGEMENT FOR DOD

Objectives, Approaches & Preliminary Results

Habitat identification Identifying key factors that sustainor intensify transmission Risk prediction

Primary schizogony

Asexual erythrocytic

cycle

Hypnozoites relapses

Gametocytes

HUMANVECTOR

PARASITE

Fertilization

Oocysts

Sporozoites

blood meal oviposition eggs larvae pupae adults destroyed

pre-patent incubation delay treatment infectious relapse immunity

Textural-contextual classifications significantly increase landcover mapping accuracy using highresolution data such as Ikonos.

Discrete Wavelet Transform is used to differentiate confusion vegetation types.

Local environment

Landcover

Satellite & meteor.data

Population database

Dwelling

Vector control

Microepidemiologydata

Medical careVector ecology

Host behaviors

Evaluated Thail military airborne data and establishedneural network rectification capability.

Richard.Kiang@nasa.gov

Nonparametric model computes the likelihood of disease outbreak using meteorological and epidemiological time series as input.

Wavelet Transform and Hilbert-Huang Transform Empirical Mode Decomposition identify the driving variables that lead to disease outbreaks and provide more accurate predictions.

Spatio-temporal distribution of disease cases

4000

3500

3000

2500

2000

1500

1000

500

0

Nu

mb

er

of

Pf

& P

v C

as

es

1251007550250Month Number

Pf casesTemperature (deg C) x 100Rainfall (mm) x 5 + 1000

Tak

10

0

-10

1992.51990.01987.51985.0

Mode 1

-10

0

10

1992.51990.01987.51985.0

Mode 2

-5

0

5

1992.51990.01987.51985.0

Mode 3

Mode 1

Mode 2

Mode 3

Bamboo CupsKanchanaburi

Washington, D.C.Washington, D.C.

Space Imaging’s Ikonos imagery

Steps in Performing Discrete Wavelet Transform

Steps in Performing Discrete Wavelet Transform

image

low passon rows

high passon rows

down

sample

cols

low passon cols

high passon cols

down

sample

rows

approx

vertical edges

horizontal edges

diagonal edges

Textural Feature Extraction using Discrete Wavelet Transform

Textural Feature Extraction using Discrete Wavelet Transform

HorizontalEdges

VerticalEdges

DiagonalEdges

H

V DA square

neighborhoodin the imagery

data

A square neighborhoodin the imagery

data

n-D entropyvector

n-D entropyvector

Approx

Class Separability with Textural Features extracted by Discrete Wavelet TransformClass Separability with Textural Features extracted by Discrete Wavelet Transform

Entropy Derived from DWT as Textural Measure to Aid Classification

Entropy Derived from DWT as Textural Measure to Aid Classification

Last 8x8 neighborhood Its WC from DWT

Largest entropy 2nd largest entropyCombined withpanchromatic

Ikonos

1m resolution

North Korea – Malaria TransmissionNorth Korea – Malaria Transmission

Camp Greaves and Surrounding AreaKyunggi, South Korea

Camp Greaves and Surrounding AreaKyunggi, South Korea

Space Imaging’s Ikonos imagery

kr4_truecolor_brightened.jpg

Pseudo Ground TruthPseudo Ground Truth

Kr34_pseudogt.jpg

(R+G+B)/3 (N+R+B)/3Panchromatic

Intensity

Space Imaging’s Ikonos imagery

 From Cook et al. “Ikonos Technical Performance Assessment” 2001 SPIE Proceedings, Algorithms for Multispectral, Hyperspectral, ..., p.94. 

1.00

0.95

0.90

0.85

0.80

0.75

0.70

0.65

0.60

Ove

rall

Cla

ssif

ica

tio

n A

ccu

racy

38363432302826242220181614121086420Case Number

originalwith & withoutpanchromatic

sharpenedper pixel

only

sharpenedplus

original

sharpenedplus neighborhood mean

sharpenedplus

neighborhoodvariance

sharpenedplus

neighborhoodmean &variance

sharpenedplus

medianfilter

sharpenedplus low pass filter

sharpenedplus

dilation

sharpenedplus mean,

variance& median

filter

Classification Accuracy using Pan-SharpenedIkonos Data ( 1 meter resolution)

Classification Accuracy using Pan-SharpenedIkonos Data ( 1 meter resolution)

Detection of Ditches using 1-meter Data(Larval Habitats of An. sinensis)

Detection of Ditches using 1-meter Data(Larval Habitats of An. sinensis)

NDVI from AVHRR MeasurementsNDVI from AVHRR Measurements

NDVI = Normalized Difference Vegetation Index

AVHRR = Advanced Very High Resolution Radiometer

Compiled by NOAA/NESDISfor Feb. 13, 2001

NDVI = (near infrared – red) ÷ (near infrared + red)

Can be used to infer ground cover and rainfall.

Can be derived from other sensors as well.

 

Post-Processing with Class Frequency FiltersPost-Processing with Class Frequency Filters

From a Beechcraft B200 Super King Air Effective surface resolution approx. 1.5m

Sample Image of Royal Thai Survey Department’s Airborne Instrument

Figure 2. (a) Raw, and (b) rectified images for Flight 1. Dark pixels are training samples. Open squares are target locations.

Figure 3. (a) Raw, and (b) rectified images for Flight 2.

(b) (a)

(a)

(b)

Simulated MeasurementsGenerated by Scanner Model

Rectified

open squares = real positionsshaded squares = fitted positions

Using Neural Networkto Rectify AircraftMeasurements

Using Neural Networkto Rectify AircraftMeasurements

Ban Kong Mong ThaSanghlaburi, Kanchanaburi, Thailand

Ban Kong Mong ThaSanghlaburi, Kanchanaburi, Thailand

Anopheles dirus forest; shaded pools; hoofprints in or at the edge of forests; with increasing deforestation, adapting to orchards, tea, rubber and other plantations.

An. minimus forest fringe; flowing waters (foothill streams, springs, irrigation ditches, seepages, borrow pits, rice fields); shaded areas; grassy and shaded banks of stable, clear, slow moving streams.

An. maculatus seepage waters; streams pools; pond edges; ditches and swamps with minimal vegetation; sunlit areas.

An. dirus

An. minimus

TRANSMISSION MODEL

Local EnvironmentLandcoverSatellite & Meteor.

Data

Population Database

Dwelling

Vector Control

MicroepidemiologyData

Medical Care

Vector Ecology

Host Behaviors

Primary Schizogony

Asexual Erythro.

Cycle

Hypnozoites Relapses

Gametocytes

HUMANVECTOR

PARASITE

Fertilization

Oocysts

Sporozoites

blood meal oviposition eggs larvae pupae adults destroyed

pre-patent incubation delay treatment infectious relapse immunity

Spatio-Temporal Distribution of Disease Cases

hx, hy, hproof

rsex, rage, rimmune, revout, rgamet

bx, by tegg, tlarva, tpupa, tmate, tovi, tspor

wbtoh, whtoh, whtob

mage, mspor

tincub, twait, tgamet, theal, tpost, trelapse

100

80

60

40

20

0

500400300200100

NRWELL

NRINF

30

20

10

0

500400300200100

NINFBITE

NBITE

12

10

8

6

4

2

500400300200100

RATESPOR

RATEGAM

200

150

100

50

0

500400300200100

NMWELL

NMINF

2-Year Prediction of Malaria CasesBased on Environmental Parameters

(temperature, precipitation, humidity, vegetation index)

Tak, Thailand

5000

4000

3000

2000

1000

0

20001999199819971996199519941993199219911990

CASES

FITTED

CASES

PREDICTED

Landsat TM Image over Mae LaLandsat TM Image over Mae La

Mae La CampMae La Camp

Sources: CDC DVBID Rutgers Univ. Entomology Dept./NJMCA

Airborne Remote Sensing

Airborne Remote Sensing

ER-2 Fleet

In late 19th Century …

Proteus HeliosAltair

Neural Network Classification of GER 63-channel Scanner DataNeural Network Classification of GER 63-channel Scanner Data

Architecture Training Acc.Rel. Classif.

Acc.

1 hidden layer

with 1 node 88.41 85.52

1 hidden layer

with 3 nodes 99.07 97.93

1 hidden layer

with 5 nodes 98.86 97.52

2 hidden layers

each with 3 nodes 99.07 97.62

2 hidden layers

each with 5 nodes 99.38 97.83

1985-1999 SIESIP ½°×½° temp, precip

1985-2003 NCEP2½°×2½° rel. humidity

1985-2000 AVHRR PF8×8 km² NDVI

1999-2003 MODIS 8×8 km² NDVI

1999-2003 MODIS 5×5 km² surface temp, lifted index, moist., etc.

2000-2003 SIESIP ½°×½° temp, precip

1998-2003 TRMM ½°×½° precip

Time-Frequency Decompositions Dengue Cases – Kuala Lumpur Fourier Transform

Hilbert-Huang Transform Wavelet Transform

RISK PREDICTION MODEL

Nonparametric model computes the likelihood of disease outbreak using meteorological and epidemiological time series as input.

Wavelet Transform and Hilbert-Huang Transform Empirical Mode Decomposition identify the driving variables that lead to disease outbreaks and provide more accurate predictions.

4000

3500

3000

2500

2000

1500

1000

500

0

Nu

mb

er

of

Pf

& P

v C

as

es

1251007550250Month Number

Pf casesTemperature (deg C) x 100Rainfall (mm) x 5 + 1000

Tak

10

0

-10

1992.51990.01987.51985.0

Mode 1

-10

0

10

1992.51990.01987.51985.0

Mode 2

-5

0

5

1992.51990.01987.51985.0

Mode 3

NASA Goddard Space Flight CenterNASA Goddard Space Flight Center

Landsat-1 MSSLandsat-1 MSS Space Imaging’s Ikonos imagery

NASA/GSFC – Close-UpNASA/GSFC – Close-Up

Pan: 1m MS: 4m Space Imaging’s Ikonos imagery

2-Year Prediction of Malaria CasesBased on Environmental Parameters

(temperature, precipitation, humidity, vegetation index)

Ratchaburi, Thailand

800

600

400

200

0

20001999199819971996199519941993199219911990

CASES

FITTEDPREDICTED

CASES

Ban Kong Mong ThaSanghlaburi, Kanchanaburi, Thailand