Modelling tick dynamics using volunteer data (2017)

Modelling tick dynamics
using volunteer data
Irene GARCIA-MARTI
20th June 2017

This session
 Present a use case of GIS modelling for Public Health:
 data collection  modelling  production of results
 Volunteered Geographic Information (VGI) plays a key role
 Relate with course objectives:
 Visualize spatial patterns of a disease
 Identify risk factors of a disease
2

Distribution of Lyme disease
http://en.wikipedia.org/wiki/Lyme_disease
Lyme, CT
4

Evolution of Lyme disease in the Netherlands
Source: RIVM
5

6
0
10
20
30
40
50
60
70
80
90
100
1994 2001 2005 2009 2014
#ofcases
Thousands
Tick bites and Lyme cases
1994 – 2014
Tick Bites Lyme Disease
Sources:
 Dutch National Atlas
 Press releases
 (Hofhuis et al, 2015)
 Highlights:
 Decrease in tick bite
consultations
 Increase (gentler) in the
cases of Lyme disease
 Causes:
 More awareness and more
people self removing ticks
 More carelessness in
people due to higher tick
densities
 More ticks means more
Lyme cases, but there is
no increase on infection
rate

Serious matter
Actions in the Netherlands
Source: 7

Serious matter
 Lyme disease has a high cost:
 For individuals:
 Long-lasting sequels:
muscles, joints, brain
deterioration
 For public health agents:
 Treatment for a potentially
chronical disease
 Population at risk:
 Children and elder
9

Why is this happening?
Causes
 Global changes
 Weather dynamics
 Wildlife ecosystems
 Socio-economic changes
 Heavy urbanization
 Fragmentation
 Human leisure
Consequences
 Ecological:
 Longer season
 Higher densities
 New suitable habitats
 Spatial:
 Northwards expansion
 More human-tick
contact
 How to monitor ticks?
 What variables influence
the number of ticks?
 Can we predict tick
dynamics for each point in
NL?
Challenges
10

Why is this happening?
Causes
 Global changes
 Weather dynamics
 Wildlife ecosystems
 Socio-economic changes
 Heavy urbanization
 Fragmentation
 Human leisure
Consequences
 Ecological:
 Longer season
 Higher densities
 New suitable habitats
 Spatial:
 Northwards expansion
 More human-tick
contact
 How to monitor ticks?
 What variables influence
the number of ticks?
 Can we predict tick
dynamics for each point in
NL?
Challenges
11
Do we even have data
to start figuring out
these mechanisms?

Volunteered data collection
 Since 2006:
 Group of volunteers sample 17
locations in NL on a monthly
basis
 Count ticks in its different life
stages (i.e. larvae, nymph,
adults)
 First citizen science project of
its kind!
Source: WUR
12

13
 Volunteer flagging dataset:
 Three types of habitat:
coniferous, deciduous and
grasses/bushes
 One-day tick counts per
month between 2006-2014
(currently, around 3.000
samples)
Volunteered data collection

Motivation
Now we know the evolution of tick counts in the time series
and we can link it to environmental variables
to train models that predict tick activity…
…and understand main drivers of the phenomenon
14

Important factors on tick dynamics
15
From “Lyme disease: The ecology of a complex system”
R.Ostfeld (2012)

16
From “Lyme disease: The ecology of a complex system”
R.Ostfeld (2012)

• Start questing season
• Survival through winterTemperature
• Increases tick survival
• Prevent tick dessicationPrecipitation
• Keeps soil moisture high
• Prevent tick dessicationVegetation
• Sustains tick populationWildlife
17

Where are the data coming from?
• KNMITemperature
• KNMIPrecipitation
• Official product: Land Cover by WUR
• Remote sensing: MODIS imagery in GEEVegetation
• No dataWildlife
Mostly, raster layers
18

Modelling
Is tick dynamics a linear phenomenon?
19
Source: (Dantas-Torres & Otranto, 2013)
A variable alone, does not predict the tick dynamics accurately

Modelling
Is tick dynamics a linear phenomenon?
20
Source: (Dantas-Torres & Otranto, 2013)
Yet, there should be a non-linear relationship out there!

Modelling
 Machine Learning algorithms:
 Non-linear models
 Not based on any previously known assumption
 They learn from data
 Capable of working with a huge number of variables
21

Modelling: Random Forest
 Ensemble learning method for
classification and regression
 Based in multiple decision trees
(a forest)
 Data are randomized and passed
to each tree in the ensemble
 Robust and stable algorithm
22

Data and underlying function Single regression tree
10 regression trees Average of 100 regression trees
Source: Quantitative Economics Tartu Blog

Modelling: Random Forest (Regression)
𝑦 = 𝑓(𝑥1, 𝑥2, … , 𝑥 𝑛)
1) Given a set of predictor variables X
2) Given a continuous response variable Y
Build a model to predict the value of Y for a new array of X
24

Modelling: Random Forest (Regression)
𝑦 = 𝑓(𝑥1, 𝑥2, … , 𝑥 𝑛)
1) Given a set of predictor variables X:
 Weather predictors: T, P, EV, RH, SD, VP
 Vegetation: NDVI, EVI, NDWI
 Land use and land cover
2) Given a continuous response variable Y
 The number of active questing ticks
Build a model to predict the value of Y for a new array of X
25

Modelling: The recipe
for each flagging_site:
for each sampling_date:
# Time to get our environmental predictors!
Get mean weather in the previous 7 days before the date of the sampling
Get vegetation indices (NDVI, EVI, NDWI) for a site and date of sampling
Get land cover of the site
Build a table with 3.000 rows and 7 columns
Model with the Random Forest for Regression
Visualize results in the geographic space
Tools
Python
GDAL
scikit-learn
matplotlib
QGIS/ArcGIS
26

Modelling: General performance
27

Visualizing: Back to geographic space
28
Predicted tick activity
June 1st, 2014
 Trained model is applied to each
pixel with forest of the Netherlands
 Interpretation:
 Provinces of Drenthe and
Groningen presented high activity
of ticks on that day
 Randstat area presented the
lowest tick activity

Visualizing: Back to geographic space
29

Conclusions
 Collective effort of volunteers can be used to devise models capturing tick
dynamics at the country level
 Contribution intended to help designing public health campaigns:
 Reduce the incidence of Lyme disease
 Increase the awareness of citizens to this problem
 Deliver results in platforms such as tekenradar.nl
30

Discussion
 Do any of you have experience in modeling species?
 What kind of spatial modelling is done in your countries?
 What else would you include in the analysis?
31

Modelling tick dynamics using volunteer data (2017)

More Related Content

Similar to Modelling tick dynamics using volunteer data (2017) (20)

Recently uploaded (20)

Modelling tick dynamics using volunteer data (2017)