Supplemental Information Supplementary Methods

piloturuguayanAI and Robotics

Oct 15, 2013 (3 years and 11 months ago)

90 views

Supplemental Information


Supplementary Methods


We

used an ensemble
-
modeling approach to forecast the impacts of climate change on
the geographic distribution of henipavirus hosts
(
1
,
2
)
.
We used georeferenced museum
specimen data
coupled with field
-
collected GPS and satellite telemetry data
to model th
e
current and future distribution
of

thirteen

species of bats
reported in the literature as
reservoirs

of

either NiV or HeV
(Table
S1
).

Due to insufficient locality data, two species
known to have had antibodies to henipavirus were excluded from this study
:
Pteropus
lylei

and
Eidolon dupreanum
.
For each
host
species, we obtained specimen localities

from GBIF and the National Museum of
Natural History
, Smithsonian Institution
(NMNH). Georeferenced

museum specimen data were filtered for inconsistencies using
IUCN range distribution maps

to
generate
an
occu
r
rence dataset for bat
h
enipa
virus

reservoirs

(
3
)
.

In addition,

GPS and satellite telemetry data from published and
unpublished field studies were used

for inde
pendent model validation, incorporating
records on
P. giganteus, P. vampyrus
,

P. hypomelanus
, and
Eidolon helvum
(
4
-
8
)
.
Nineteen bioclimati
c variables at
2.
5 arc
-
minute resolution (~
5
km
2
) averaged across
1950
-
1999

were used to generate current bioclimatic niche
s

[Table S2,
(
9
)
]
. In order to
rigorously explore future climate conditions, we used these same 19 bioclimatic variables
derived from downscaled monthly temperature and precipitation outputs of 20 global
circulation models

(GCMs)
(
10
)

(Table S4).
The 4
th

Assessment Report of the
Intergovernmental Panel on Climate Change

(IPCC)

ba
sed their impact

assessments on
the results of almost tw
o
-
dozen
GCMs
run under various scenarios of greenhouse gas
emissions

(
10
)
. The Coupled Model
I
ntercomparison Project of the World Climate
Research Programme provides

a comprehensive, coordinated set of GCM experiments
for impact assessments
(
11
)
. Recent advances in the validation of global circulation
models have embraced the
use of multi
-
model ensembles as

a method of reducing
uncertainty in climate model applications

(
12
,
13
)
. For this example, w
e focus
ed

on a
single midcentury time slice and only the

A2 greenhous
e gas emissions scenario
, which
assumes ‘business as usual’ continued emissions throughout this century
(
14
)
.
For a
midcentury time interval, the A2 scenario spans the high and low end of most emissions
scenarios used to parameterize climat
e model simulations produced for IPCC impact
assessments. The aim of this approach was not to compare the influence of alternative
emissions scenarios on the future distribution of the bat henipavirus reservoirs, but to
demonstrate the utility

of
ENM
for
forecasting potential

range shifts of
the hosts of
directly transmitted pathogens
.

All niche models were generated using Maxent v3.3.3e, a machine learning
algorithm that uses the principle of maximum entropy to estimate a set of rules
correlating environm
ental variables and species occurrences to approximate the potential
geographic distribution of the modeling target
(
15
,
16
)
. Maxent was chosen because of its
established performance with presence
-
only data relative to alternative niche modeling
techniques
(
17
)

and its built
-
in capacity to deal with multi
-
colinearity in the
environmental variables
(
15
)
.
Maxent’s algorithm is related to Bayesian theory that
considers redundant information without penalizing models by over fitting, eliminating
the need to apply variable reduction techniques before running the models (but see e.g.,
(
18
)

for an alternative parsimony
-
based view of variable selection in this application).

To
internally calibrate the models and avoid spatial autocorrelation, we created
subsets of each species’ locality observations using a spatially struct
ured partitioning
procedure [Figure S1
(
19
)
]. For each model based on a subsample, we report the area
under the curve (AUC), a measure that summarizes the model’s overall performance over
every possible th
reshold
(
20
)

(Figure S2). T
he AUC as a metric
for
evaluating model
performance has been criticized for
, among other things,

b
eing

highly sensitive to the
total extent over which models are carried out

(
21
)
, and for its equal weighting of errors
of omission and commission

(
20
)
.

Despite this,
we employ the AUC metric
here as it still
provides a relative measure of variability among the repetitions within a species
.


E
ach resulting simulation

for the current conditions was converted into

a
presence/absence map based on
an inbuilt Maxent threshold rule (minimum training
presence) that

maximiz
es

overprediction

(commission error)
over underprediction

(omission error). In our application, overprediction is preferable to underprediction due to
the relative costs of these two types of error. For example, financially and social
ly, it is
likely less costly to be prepared for an event that does not occur (e.g., cheap to have a
preparedness and response plan and implement surveillance in areas of absence), than be
underprepared for an event that does (e.g., outbreak in an unexpecte
d area with no
capacity to detect or manage it). Preventative measures for pandemic preparedness will
also have collateral benefits in dealing with other infectious diseases. Each binary map
was then evaluated using an
independent
field
-
derived
dataset
co
nsisting of
GPS
locations and satellite telemetry
-
derived locations

for

bat

roosts and foraging locations
, to
measure and report sensitivity and false negative rates (Table S3). These independent
data included NiV prevalence and seroprevalence (i.e. prior
NiV exposure) for these bat
populations. Subsets from these populations were collared for satellite telemetry (Table
S1).
The presence/absence maps
for each species
were combined into a final current
niche model representing the sum of all 100 iterations.

Each one of the final current
bioclimatic niches for the bat species were then projected into the midcentury u
sing 20
alternative

GCMs. Each simulation was converted to a presence/absence map using the
same threshold rule as the
current conditions models
(see above)
, and then combined into
a single output representing the sum of
the

presence/absence maps for each

of the 100
simulations for each of the 20 GCMs
.

To
produce a summary
forecast

of

the

potential

future distribution of
each
h
enipavirus
bat
host,
we
next combined the 20 future projections for each species into a
single map, where presence was defined as a pixel that was predicted to contain the
species in at least 50% of the 20 GCMs. Contrasting the current modeled distribution and
the midcentury p
rojections, we calculated the expansion and contraction in the
distribution of suitable bioclimatic (Figures S3


S28).

A

synthesis map
integrating the
results across

all bat host species’ distributions
yield
ed

an ensemble estimate of

midcentury climate ch
ange induced distributional expansion/contraction for the
henipavirus bat host complex (Figure 2).



1.

Araujo M & New M (2007) Ensemble forecasting of species distributions.
Trends
in Ecology & Evolution

22(1):42
-
47.

2.

Buisson L, Thuil
ler W, Casajus N, Lek S, & Grenouillet G (2010) Uncertainty in
ensemble forecasting of species distribution.
Global Change Biology

16(4):1145
-
1157.

3.

Fitzpatrick MC
, et al.

(2011) Forecasting the future of biodiversity: a test of
single and multi species
models for ants in North America.
Ecography

34:836

847.

4.

Plowright RK
, et al.

(2008) Reproduction and nutritional stress are risk factors for
Hendra virus infection in little red flying foxes (
Pteropus scapulatus
).
Proceedings of the Royal Society B
-
Biol
ogical Sciences

275(1636):861
-
869.

5.

Plowright RK
, et al.

(2011) Urban habituation, ecological connectivity and
epidemic dampening: The emergence of Hendra virus from flying foxes (
Pteropus
species).
Proceedings of the Royal Society B
-
Biological Sciences

278:3703
-
3712.

6.

Wacharapluesadee S
, et al.

(2009) A longitudinal study of the prevalence of
Nipah virus in
Pteropus lylei

bats in Thailand: Evidence for seasonal preference
in disease transmission.
Vector borne and zoonotic diseases (Larchmont, N.Y

[Epub ahead of print].

7.

Richter HV & Cumming GS (2006) Food availability and annual migration of the
straw
-
colored fruit bat (
Eidolon helvum
).
Journal of Zoology

268(1):35
-
44.

8.

Tidemann CR, Vardon MJ, Loughland RA, & Brocklehurst PJ (1999) Dry season
camps of flying foxes (
Pteropus

spp.) in Kakadu World Heritage Area, north
Australia.
Journal of Zoology

247:155
-
163.

9.

Hijmans R, Cameron S, Parra J, Jones P, & Jarvis A (2005) Very high resolution
interpolated climate surfaces for global land areas.
Int
ernational Journal of
Climatology

25(15):1965
-
1978.

10.

IPCC ed (2007)
Climate Change 2007: The Physical Science Basis. Contribution
of Working Group I to the Fourth Assessment Report of the Intergovernmental
Panel on Climate Change

(Cambridge University P
ress, Cambridge, United
Kingdom and New York, NY, US).

11.

Meehl GA
, et al.

(2007) The WCRP CMIP3 multimodel dataset.
Bull. Am.
Meteorol. Soc

88:1383

1394.

12.

Collins M (2007) Ensembles and probabilities: a new era in the prediction of
climate change.
Phi
losophical Transactions of the Royal Society A:
Mathematical, Physical and Engineering Sciences

365(1857):1957.

13.

Tebaldi C & Knutti R (2007) The use of the multi
-
model ensemble in
probabilistic climate projections.
Philosophical Transactions of the
Royal Society
A: Mathematical, Physical and Engineering Sciences

365(1857):2053.

14.

Nakicenovic N & Swart R (2000) Emissions scenarios. Special report of the
Intergovernmental Panel on Climate Change (IPCC). (Cambridge, U.K.).

15.

Phillips SJ, Anderson R
P, & Schapire RE (2006) Maximum entropy modeling of
species geographic distributions.
Ecological Modelling

190(3
-
4):231
-
259.

16.

Phillips SJ & Dudik M (2008) Modeling of species distributions with Maxent:
new extensions and a comprehensive evaluation.
Ecog
raphy

31(2):161

175.

17.

Elith J & Leathwick J (2009) Species distribution models: ecological explanation
and prediction across space and time.
Annual Review Ecology Evolution and
Systematics

40:677
-
697.

18.

Parolo G, Rossi G, & Ferrarini A (2008) Toward i
mproved species niche
modelling:
Arnica montana

in the Alps as a case study.
Journal of Applied
Ecology

45(5):1410
-
1418.

19.

Araujo MB & Guisan A (2006) Five (or so) challenges for species distribution
modelling.
Journal of Biogeography

33(10):1677–1688.

20.

Lobo JM, Jiménez

Valverde A, & Real R (2008) AUC: a misleading measure of
the performance of predictive distribution models.
Global Ecology and
Biogeography

17(2):145
-
151.

21.

VanDerWal J, Shoo LP, Graham C, & Williams SE (2009) Selecting pseudo
-
absen
ce data for presence
-
only distribution modeling: How far should you stray
from what you know?
Ecological Modelling

220(4):589
-
594.





Table
S1
:
Bat species known to be the natural reservoirs of
Hendra or Nipah virus.

The
“Number of localities”

represent
s the georeferenced occurrence records for each species
available at both GBIF and the USNHM. These localities were used to
parameterize the
models.

* Represents localities confirmed by GPS and telemetry that were used for
validation.























Family

Sp
ecies

Pathogen


Number of
localities

Pteropodidae

Cynopterus brachyotis

Nipah virus

231

Pteropodidae

Eidolon helvum

Nipah virus


69

Pteropodidae

Eonycteris spelaea

Nipah virus


142

Pteropodidae

Pteropus alecto

Hendra virus

128

Pteropodidae

Pteropus
conspicillatus

Hendra virus

32

Pteropodidae

Pteropus giganteus
*

Nipah virus

13

Pteropodidae

Pteropus hypomelanus
*

Nipah virus

44

Pteropodidae

Pteropus poliocephalus

Hendra virus

118

Pteropodidae

Pteropus scapulatus

Hendra virus

176

Pteropodidae

Pteropus tonganus

Hendra virus

34

Pteropodidae

Pteropus vampyrus
*

Nipah virus

16

Rhinolophidae

Hipposideros larvatus

Nipah virus

23

Vespertilionidae

Scotophilus kuhlii

Nipah virus

50

Table
S2:

List
of the
19 bioclimatic

variables used in the analysis.




Table S
3
. Sensitivity and False Negative Rate values for the three species
-
distribution models. These values were calculated using confirmed GPS and
telemetry data.



Species

True
positives

False
Negatives

Sensitivity

False
Negative
Rate

Pteropus hypomelanus

35

0

1

0

Pteropus vampyrus

5

3

0.625

0.375

Pteropus giganteus

25

0

1

0

Table
S4
:

List of the 20 Global Circulation Models (GCM’s) used to model future
distribution.


Code

Name

BCCR_BCM2_0

Bjerknes

Centre for Climate Research, Norway, BCM2.0 Model

CCCMA_CGCM3_1

Canadian Centre for Climate Modelling and Analysis, CGCM3.1 Model, T47 resolution

CCCMA_CGCM3_1_T63

Canadian Centre for Climate Modelling and Analysis, CGCM3.1 Model, T63 resolution

MIROC3
_2_HIRES

CCSR/NIES/FRCGC, Japan, MIROC3.2, high resolution

MIROC3_2_MEDRES

CCSR/NIES/FRCGC, Japan, MIROC3.2, medium resolution

CSIRO_MK3_0

CSIRO Atmospheric Research, Australia, Mk3.0 Model

CSIRO_MK3_5

CSIRO Atmospheric Research, Australia, Mk3.5 Model

UKMO_HADCM3

Hadley Centre for Climate Prediction, Met Office, UK, HadCM3 Model

UKMO_HADGEM1

Hadley Centre for Climate Prediction, Met Office, UK, HadGEM1 Model

INGV_ECHAM4

INGV, National Institute of Geophysics and Volcanology, Italy, ECHAM 4.6 Model

INMCM3_0

Institute for Numerical Mathematics, Russia, INMCM3.0 Model

IPSL_CM4

IPSL/LMD/LSCE, France, CM4 V1 Model

IAP_FGOALS1_0_G

LASG, Institute of Atmospheric Physics, China, FGOALS1.0_g Model

MPI_ECHAM5

Max Planck Institute for Meteorology, Germany,
ECHAM5 / MPI OM

CNRM_CM3

Meteo
-
France, Centre National de Recherches Meteorologiques, CM3 Model

MIUB_ECHO_G

Meteorological Institute of the University of Bonn, ECHO
-
G Model

MRI_CGCM2_3_2A

Meteorological Research Institute, Japan, CGCM2.3.2a

GISS_AOM

NASA Goddard Institute for Space Studies, C4x3

GISS_MODEL_E_H

NASA Goddard Institute for Space Studies, ModelE20/HYCOM

GISS_MODEL_E_R

NASA Goddard Institute for Space Studies, ModelE20/Russell

NCAR_CCSM3_0

National Center for Atmospheric Research,
CCSM3.0

NCAR_PCM1

National Center for Atmospheric Research, PCM1

GFDL_CM2_0

NOAA Geophysical Fluid Dynamics Laboratory, CM2.0 Model

GFDL_CM2_1

NOAA Geophysical Fluid Dynamics Laboratory, CM2.1 Model




Figure S1
. Graphical representation of the
spatially structured partitioning method used
to calibrate the models and avoid spatial autocorrelation.
This approach divides the data
into multiple strata with the origin at the median center point, which is measure
d as the
central tendency that minimize
s the Euclidean Distance
of

all observations. A measure of
direction to the median center point is added to each observation. This angular measure is
used to divide the localities into slices. Each slice is used in turn to test the model while
the rest of
the slices are used to train the model







Figure
S2
.

AUC validation values for all the 13 species known to be reservoirs of
henipavirus.







Figure
S3
.

Top:
Cynopterus brachyotis

current modeled distribution. Bottom: future
projected di
stribution





Figure
S4
.

Top:
Eidolon helvum

current modeled distribution. Bottom: future
projected distribution.





Figure
S5
.

Top:
Eonycteris spelaea

current modeled distribution. Bottom: future
projected distribution.





Figure
S6
.

Top:
Hipposideros larvatus

current modeled distribution. Bottom: future
projected distribution.




Figure
S7
.

Top:
Pteropus alecto

current modeled distribution. Bottom: future
projected distribution.



Figure
S8
.

Top:
Pteropus conspicillatus

current modeled

distribution. Bottom: future
projected distribution.






Figure
S9
.

Top:
Pteropus giganteus

current modeled distribution. Bottom: future
projected distribution.






Figure
S10
.

Top:
Pteropus hypomelanus

current modeled distribution. Bottom:
future p
rojected distribution.




Figure
S11
.

Top:
Pteropus poliocephalus

current modeled distribution. Bottom:
future projected distribution.





Figure
S12
.

Top:
Pteropus scapulatus

current modeled distribution. Bottom: future
projected distribution.






F
igure
S13
.

Top:
Pteropus tonganus

current modeled distribution. Bottom: future
projected distribution.




Figure
S14
.

Top:
Pteropus vampyrus

current modeled distribution. Bottom: future
projected distribution.





Figure
S15
.

Top:
Scotopilus kuhlii

current modeled distribution. Bottom: future
projected distribution.




Figure
S16
.
Cynopterus brachyotis

expansion/contracion/stable maps



Figure
S17
.
Eidolon helvum

expansion/contraction/stable maps.


Figure
S18
.
Eonycteris spelaea

expansion/contraction/stable maps.




Figure
S19
.
Hipposideros larvatus

expansion/contraction/stable maps.


Figure
S20
.
Pteropus conspicillatus

expansion/contraction/stable maps.


Figure
S21
.
Pteropus alecto

expansion/contraction/stable maps.



























Figure
S22
.
Pteropus giganteus

expansion/contraction/stable maps.

Figure S23.
Pteropus hypomelanus

expansion/contraction/stable maps



F
igure
S24
.
Pteropus poliocephalus

expansion/contraction/stable maps.


F
igure
S25
.
Pteropus scapulat
us

expansion/contraction/stable maps.


F
igure
S26
.
Pteropus tonganus

expansion/contraction/stable maps.



F
igure
S27
.
Pteropus vampyrus

expansion/contraction/stable maps.




F
igure
S28
.
Scotophilus kuhlii

expansion/contraction/stable maps.





Figure
S29. Top: Species richness map for current distribution based on the sum of 1
species.. Bottom: Species richness map for midcentury (2050’s) A2 scenario based
on 50% GCM agrement.