Spatial Modeling of Mosquito Densities Using MODIS Enhanced Vegetation Index EVI and Near Ground Hum

1
Spatial Modeling of Mosquito Densities Using
MODIS Enhanced Vegetation Index (EVI) and Near
Ground Humidity Indexes adult female Culex
tarsalis and Aedes vexans clustering in Colorado
and Louisiana

• A PRESENTATION TO THE SUMMER COLLOQUIUM ON
  CLIMATE AND HEALTH
• JULY 23, 2004, NCAR, BOULDER COLORADO
• RUSSELL BARBOUR PH.D.
• VECTOR ECOLOGY LABORATORY
• YALE SCHOOL OF MEDICINE
• NEW HAVEN CT.

2
MODELING VERSUS INTERPOLATION

• LINEAR MODELING ATTEMPTS TO IDENTIFY FACTORS THAT
  INFLUENCE THE PARAMETERS OF INTEREST AND EXPLAIN
  OBSERVED VARIATION
• SPATIAL MODELING OR INTERPOLATION USE THE
  MATHEMATICAL PROPERTIES OF THE DATA ITSELF TO
  ESTIMATE VALUES AT UNKNOWN LOCATIONS

3
LECTURE OUTLINE

• Review basic concepts of spatial auto-correlation
• Demonstrate application of these methods to
  estimate mosquito vectors of West Nile Virus

4
BASIC CONCEPTS OF SPATIAL AUTO-CORRELATION

• Toblers first law of geography
• Everything is related to everything else, but
  near things are more related than distant things
• Auto- Correlation violates the assumption of
  independence in that is made in most statistical
  tests
• Ordinary Least Squares Regression (OLS) for
  example, will tend to Type I Error ( falsely find
  significant relationships) if auto-correlation is
  present
• Auto-Correlation can be used to estimate values
  at un-sampled locations

5
QUANTIFYING AUTO-CORRELATION

• Morans I
• Gearys C ratio
• Anselins Local Index of Spatial Autocorrelation
  (LISA) 0R Local Morans I

6
Morans I

• Similar to Pearsons correlation coefficient,
  values between 1.0 and 1.
• Index for dispersion/random/cluster patterns
• Indices close to zero, indicate random pattern
• Indices above zero indicate a tendency toward
  clustering
• Indices below zero indicate a tendency toward
  dispersion/uniform
• Most commonly reported indicator of spatial
  auto-correlation
• Differences from correlation coefficient are
• one variable only, not two variables
• Incorporates weights (wij) which index distance
  between the locations

7
MORANS I CONTINUED

• GLOBAL MORANS I
• Estimates the level of aggregation of values
  or clustering in space for all observations
• Correlogram
• MoransI calculated for observations grouped
  into specific distances

8
TYPES OF SPATIAL STRUCTUREDETECTED BY POSITIVE
MORANSI VALUES

• CLUSTERS
• DATA FOUND IN CLOSE PROXIMITY
• TRENDS
• A GRADIENT USUALLY CAUSED BY A GEOGRAPHIC
  FEATURE (NON- STATIONARITY)
• AUTO-CORRELATION
• SIMILARITY OF OBSERVATIONS CLOSE TO EACH
  OTHER. A CLUSTER MAY OR MAY NOT HAVE
  AUTO- CORRELATION

9
STATIONARITY IN SPACE

• FIRST ORDER (STRICT) STATIONARITY
• A property of a spatial process where all of the
  spatial random variables have the same mean and
  variance value.
• INTRINSIC (WEAK) STATIONARITY
• An assumption that the data comes from a random
  process with a constant mean, and a semivariogram
  that only depends on the distance and direction
  separating any two locations.

SOURCE U. OF ARIZONA
10
PURPOSE OF LOUISIANA SPATIAL MOSQUITO ESTIMATES

• INDICATE AREAS OF HUMAN RISK OF WEST NILE VIRUS
• ASSIST DECISION MAKERS FOR VECTOR CONTROL
  INTERVENTIONS
• ASSESS THE EFFECTIVENESS OF CONTROL MEASURES
• ESTIMATES HAVE NO EXPLANATORY VALUE, STRICTLY A
  PROCESS OF CAPTURING MATHEMATICAL RELATIONSHIPS

11
Aedes vexans
SOURCE SERVICE 1976

• FLOOD WATER MOSQUITO
• STRONG FLIER gt 24 Km/ DAY
• AGGRESSIVE HUMAN BITER
• LOW INFECTION RATES
• HIGH TRANSMISSION EFFICIENCY IF SYSTEMICALLY
  INFECTED (TURELL 2001)

12
Aedes vexans NJ Light Trap Catches 2003 St
Tammany
13
(No Transcript)
14
GLOBAL MORANS I Aedes vexans NJ LIGHT TRAP
CATCHES JUNE 2003 ST. TAMMANY PARISH LA

• Spatial Autocorrelation for Point Data
• ---------------------------------------
•  
• Sample size
  53
•   Moran's "I"
  0.090325
• Spatially random (expected) "I" -0.019231
• Standard deviation of "I"
  0.040462
• Normality significance (Z) 2.707580
  P lt .05
• Randomization significance (Z) 2.952694 P
  lt .05
•  

15
GLOBAL MORANS I VALUES Aedes vexans Catches
NJ Light Traps 2003
16
CORRELOGRAM FOR JUNE NJ LIGHT TRAP CATCHES ALL
SPECIES APRIL 2003 ST TAMMANY PARISH LA.
RANGE
meters
17
ISOTROPIC VARIOGRAM Aedes vexans APRIL 2003
SAMPLE VARIANCE
18
MODIS ATMOSPHERE PRODUCTS

• 1-KM SPATIAL RESOLUTION
• USING THE NEAR-INFRARED ALGORITHM DURING THE DAY,
  1-KM PIXEL RESOLUTION
• THE SOLAR RETRIEVAL ALGORITHM RELIES ON
  OBSERVATIONS OF WATER-VAPOR ATTENUATION OF
  REFLECTED SOLAR RADIATION IN THE NEAR-INFRARED IN
  THE ATMOSPHERE CLOSE TO THE GROUND
• VALUES REPRESENT THE AMOUNT OF WATER PER PIXEL
  THAT COULD THEORETICALLY BE PRECIPITATED OUT OF
  THE ATMOSPHERE

19
MODIS ATMOSPHERE PRODUCT RELATIONSHIP TO Aedes
vexans

• WATER COLUMN PRODUCTS BY NIR AND IR ARE
  APPROXIMATIONS OF ABSOLUTE HUMIDITY AND
  SATURATION DEFICIT IN THE LOWER ATMOSPHERE
• THE IR DATA IS PRODUCED FOR BOTH DAY AND NIGHT..
  INCLUDES DUSK AND DAWN ESTIMATES
• IS AVAILABLE ON A DAILY BASES

20
MEASUREMENT OF Aedes vexans MICRO-CLIMATE
DISPERSAL PARAMETERS
MODIS WATER VAPOR PRODUCTS
RAINFALL
STANDING WATER
HIGH ABSOLUTE HUMIDITY
HATCHING gtEMERGENCE gt DISPERSAL
LIGHT TRAP DATA AND VARIOGRAPHY
CLUSTERING NEAR HOSTS
MORANSI
21
(No Transcript)
22
(No Transcript)
23
RAINFALL ON Ae. vexans CLUSTERING P VALUE .54
CLUSTERING ON MODIS VALUES P VALUE .104
24
SPATIAL FACTORS ASSOCIATED WITH Aedes vexans
DENSITY IN St TAMMANY PARISH

• MODIS WATER VAPOR PRODUCT .38
• URBAN PRESENCE (POPULATION AND LIGHT) .34

25
MODIS INFRARED WATER VAPOR COLUMN MONTHLY DATA
2003
TEMPERATURES ABOVE 90 F
26
Aedes vexans CATCHES in NJ LIGHT TRAPS VERSUS
MODIS HUMIDITY MEASURES 2003
START OF HIGH TEMPERATURE
27
LAKE PONCHATRAIN
28
ESTIMATED DENSITY OF Aedes vexans ADULTS BY
CO-KRIGING LIGHT TRAP AND MODIS HUMIDITY DATA
29
LIGHT TRAP DATA WITH WEAKER SPATIAL STRUCTURE

• Culex tarsalis IS INCRIMINATED IN THE
  TRANSMISSION OF WEST NILE VIRUS TO HUMANS IN THE
  FORT COLLINS COLORADO AREA (NASCI ET AL 2003)
• BREEDS IN ANY SOURCE OF FRESH WATER OTHER THAN
  TREE HOLES . MULTIPLE GENERATIONS
• IRRIGATION DITCHES HIGHLY FAVORABLE BREEDING
  AREAS
• FEEDS ON BIRDS THEN SHIFTS TO MAMMALS AND HUMANS
  AS ABUNDANCE INCREASES

30
Culex tarsalis
31
Culex tarsalis DISPERSALFrom (Reisen 2002)

• SLOW MOVING, 1 Km/ DAY
• WIND DRIVEN
• ACTIVITY RELATED TO VEGETATION COVER

32
Culex tarsalis Clustering near Ft Collins Col.
July 2003
Spatial Autocorrelation for Point
Data ---------------------------------------  
Sample size
70 Moran's "I"
-0.010883 Spatially random (expected)
I -0.014493 Standard deviation of "I
0.020686 Normality significance
(Z) 0.174495 pgt .10
Randomization significance (Z) 0.179128 pgt
.10  
33
SPATIAL RELATIONSHIPS OF C. tarsalis LIGHT TRAP
DATA AND REMOTELY SENSED MICROCLIMATE INDICATORS

• MODIS REMOTELY SENSED VEGETATION INDEX ENHANCED
  VEGETATION INDEX (EVI)
• MODIS WATER VAPOR COLUMN IR DATA

34
APPLICATION OF ARTIFICIAL NEURAL NETWORKING (ANN)
TO IMPROVE ESTIMATES
IRRIGATION, VEGETATION AND WATER VAPOR INDICATORS
COMBINED BY ARTIFICIAL NEURAL NETWORKING (ANN)
RESPONSE SURFACE
ANN RESPONSE SURFACE .61 SPATIAL RELATIONSHIP
WITH C. tarsalis LIGHT TRAP CATCHES
35
MODIS ENHANCED VEGETATION INDEX (EVI) 2003
MAY 9
JUNE 10
JULY 2
36
CLUSTERING OF C. tarsalis in RELATIONSHIP TO
MODIS EVI VALUES
SPATIAL ASSOCIATION WITH EVI VALUES OF 4000-
6000 DURING MOST OF THE 2003 SEASON
37
RESULTS

• AN ASSOCIATION APPEARS TO EXIST BETWEEN C.
  tarsalis AND MODIS EVI AND WATER VAPOR VALUES
  AT THE COLORADO SITE
• A STRONGER ASSOCIATION BETWEEN Aedes vexans AND
  MODIS HUMIDITY DATA WAS FOUND AT THE LOUISIANA
  SITE
• EVEN WEAKLY CLUSTERING SPECIES CAN BE ESTIMATED
  THROUGH APPLICATION OF SPATIAL STATISTICS AND
  ARTIFICIAL NEURAL NETWORKS
• MORE ROBUST INTERPRETATION OF LIGHT TRAP DATA IS
  POSSIBLE
• DAILY MODIS ATMOSPHERIC DATA AVAILABILITY WILL
  ALLOW FORWARD LOOKING MODELS IN THE NEAR FUTURE