Buffel grass and climate change: a framework for projecting … · 2018-10-09 · Buffel grass and climate change: a framework for projecting invasive species distributions when data

ORIGINAL PAPER

Buffel grass and climate change: a framework for projectinginvasive species distributions when data are scarce

Tara G. Martin . Helen Murphy . Adam Liedloff .

Colette Thomas . Iadine Chades . Garry Cook .

Rod Fensham . John McIvor . Rieks D. van Klinken

Received: 25 October 2014 /Accepted: 15 July 2015 / Published online: 22 July 2015

� Springer International Publishing Switzerland 2015

Abstract Invasive species pose a substantial risk to

native biodiversity. As distributions of invasive

species shift in response to changes in climate so will

management priorities and investment. To develop

cost-effective invasive species management strategies

into the future it is necessary to understand how

species distributions are likely to change over time and

space. For most species however, few data are

available on their current distributions, let alone

projected future distributions. We demonstrate the

benefits of Bayesian Networks (BNs) for projecting

distributions of invasive species under various climate

futures, when empirical data are lacking. Using the

introduced pasture species, buffel grass (Cenchrus

ciliaris) in Australia as an example, we employ a

framework by which expert knowledge and available

empirical data are used to build a BN. The framework

models the susceptibility and suitability of the Aus-

tralian continent to buffel grass colonization using

three invasion requirements; the introduction of plant

propagules to a site, the establishment of new plants at

a site, and the persistence of established, reproducing

populations. Our results highlight the potential for

buffel grass management to become increasingly

important in the southern part of the continent,

whereas in the north conditions are projected to

become less suitable. With respect to biodiversity

impacts, our modelling suggests that the risk of buffelElectronic supplementary material The online version ofthis article (doi:10.1007/s10530-015-0945-9) contains supple-mentary material, which is available to authorized users.

T. G. Martin (&) � I. Chades � J. McIvor �R. D. van Klinken

CSIRO, Ecosciences Precinct, 41 Boggo Rd, Dutton Park,

QLD 4102, Australia

e-mail: [email protected]

H. Murphy

CSIRO, Maunds Road, Atherton, QLD 4883, Australia

A. Liedloff � G. CookCSIRO, 564 Vanderlin Drive, Darwin, NT 0821, Australia

C. Thomas

Catchment to Reef Research Group, TropWATER,

Australian Tropical Science and Innovation Precinct,

Building 145, James Cook University, Douglas,

QLD 4811, Australia

R. Fensham

Department of Environment and Heritage Protection,

Queensland Herbarium, The Queensland Government,

Mt Coot-tha, QLD, Australia

R. Fensham

School of Biological Sciences, The University of

Queensland, Brisbane, QLD, Australia

123

Biol Invasions (2015) 17:3197–3210

DOI 10.1007/s10530-015-0945-9

grass invasion within Australia’s National Reserve

System is likely to increase with climate change as a

result of the high number of reserves located in the

central and southern portion of the continent. In situ-

ations where data are limited, we find BNs to be a

flexible and inexpensive tool for incorporating exist-

ing process-understanding alongside bioclimatic and

edaphic variables for projecting future distributions of

species invasions.

Keywords BN � Bayesian belief network � Expertjudgement � Expert elicitation � Invasive species �Exotic pasture � Cenchrus ciliaris � Speciesdistribution models

Introduction

Invasion by exotic species has been identified as a

significant threat to biodiversity under climate change

(Hellmann et al. 2008; Mooney and Hobbs 2000;

Thomas et al. 2004). Whether passengers or drivers of

biodiversity change, dominance of invasive species can

affect ecosystem structure and function as well as

distribution of native species (MacDougall and Turk-

ington 2005). Under rapid climate change the distribu-

tions of some invasive species are likely to expand,

while others may contract. The relative risk posed by

particular invasive species is therefore likely to change.

Management issues associated with a species that is

currently invasive may be alleviated in the future.

Likewise, species which currently pose little threat may

become problematic as a result of increasing habitat

suitability with climate change (Walther et al. 2009).

Mitigation and adaptation strategies for dealing

effectively with invasive species under climate change

will depend on good predictions of the likely change in

habitat suitability and susceptibility to invasion under

different global change scenarios, and the subsequent

impacts given these shifts. The problem is that for

many invasive species the necessary data required to

build species distributionmodels are lacking. Often we

must turn to expert knowledge to fill this information

gap (Kuhnert et al. 2010;Martin et al. 2012a). Here, we

employ a modelling framework that encapsulates

expert knowledge and available empirical data on the

ecology and spread of an invasive species and use this

information to project the future habitat suitability and

susceptibility to invasion under climate change.

We illustrate the benefits of our framework by

considering the distribution of buffel grass, Cenchrus

ciliaris L. across Australia. Buffel grass is one of many

exotic grasses introduced widely to improve livestock

production and aid in soil stabilization followingmining

and other development. Native to parts of Africa, Asia

and the Middle East, buffel grass is now widely

distributed across the United States, Mexico and

Australia (Arriaga et al. 2004; Lawson et al. 2004; Tu

2002; vanDevender et al. 1997).Buffel grass is amongst

a suite of commercially valuable invasive species,

highly valued as a pasture species but widely unpopular

among those concerned with its threat to native

biodiversity (Friedel et al. 2011; Grechi et al. 2014).

The management of such species is contentious and

offers a compelling case study to examine the impact of

climate projections on its future colonization success.

At present, there is considerable uncertainty about

the relative influence of climate change on the degree

of threat posed by buffel grass on native biodiversity

and how this threat will vary across different regions

(Sutherst et al. 2007). Despite this uncertainty there is

substantial pressure on natural resource managers to

make management recommendations, particularly

where areas of high conservation and cultural value

such as Australia’s system of protected areas, called

the National Reserve System (NRS), are at risk

(Fig. 1). Buffel grass is considered the most threaten-

ing invasive plant within Australia’s World Heritage

listed Uluru-Kata Tjuta (Ayers Rock) National Park

(Anon 2010). Protecting the National Reserve System

from buffel grass invasion presents enormous chal-

lenges, not least due to its continued promotion as a

Fig. 1 Buffel grass invasion within Australia’s National

Reserve System, Ochre Pits, West MacDonald Ranges; a site

of high cultural and ecological value. Photo. T.G. Martin

3198 T. G. Martin et al.

123

pasture species, its capacity to spread beyond the farm

gate and the labour intensiveness of current control

methods (hand pulling).

Species distribution models (SDMs) are the pri-

mary tools for predicting environmental suitability for

species (Elith et al. 2006). By relating species

occurrences to spatial environmental data, SDMs fit

the realized environmental niche of species. Com-

bined with climate projections SDMs are used to

project species distributions in the future (Keith et al.

2008; Ponce-Reyes et al. 2012). Bayesian Networks

(BNs) are one of the many different types of models

that can be used to predict species distributions

(Murray et al. 2014; Smith et al. 2007; Van Klinken

et al. 2015) and they are increasingly being adopted in

ecology to project species distributions under future

climate conditions (Mantyka-Pringle et al. 2014;

Murray et al. 2012; van Klinken et al. 2009). BNs

are probabilistic models that represent conditional

dependencies between nodes in a directed acyclic

graph. The nodes represent variables that affect some

outcome of interest and the links represent interactions

between the nodes (Marcot et al. 2001). Data feeding

into BNs can be based on expert judgement through an

expert elicitation process and/or empirical or modelled

data about the relationships of interest. Expert elici-

tation represents a way of capturing knowledge and

informing management and policy when empirical

data are limited but it presents a number of interesting

challenges, namely the collection of robust and

accurate, unbiased information from one or more

experts. Robust methods for eliciting and using expert

knowledge to inform ecological models are gaining

prominence in ecology and conservation science

(Martin et al. 2012a).

Using a BN we assess the relative threat of buffel

grass colonization across Australia and its greatest

conservation asset, the National Reserve System

(NRS). Specifically we ask:

1. What is the current susceptibility and suitability of

the Australian continent to buffel grass

colonization?

2. How will projected climate change scenarios

influence suitability to buffel grass colonization

in the future?

3. What are the management implications of these

projected changes?

Buffel grass colonization in Australia

Buffel grass is thought to have first arrived in Australia

between 1870 and 1880 as an unintentional passenger

in Afghan camel harnesses (Humphreys 1967). By

1910, its value as a pasture species in arid landscapes

had been recognized and seed was being deliberately

introduced and its spread actively encouraged (Mar-

riott 1955). Buffel’s capacity to produce high yields,

tolerate drought and heavy grazing, and grow vigor-

ously after fire makes it highly valued by some

livestock producers in arid landscapes (Tu 2002).

However, these same traits, coupled with a capacity

for establishment in disturbed areas (McIvor 2003),

rapid growth, fast maturation, prolonged flowering/

fruiting, prolific seed production and high seed

dispersal (Franks 2002) also make it a successful

colonizer of non-target areas. Buffel grass can form

dense single-species stands and out-compete native

plant species, threatening native animal species

through displacement of native vegetation. Several

studies have highlighted its negative impact on

biodiversity within remnant vegetation, tropical for-

ests and woodlands of Queensland (Eyre et al. 2009;

Fairfax and Fensham 2000; Franks 2002; Jackson

2005) and in the arid landscapes of central Australia

(Clarke et al. 2005; Smyth et al. 2009). Buffel grass

can generate high fuel loads, and so alter fire regimes

by carrying frequent and more intense fires. Positive

feedback between fire and buffel grass increases the

impact of buffel grass colonization (Butler and Fairfax

2003). Buffel grass is widely spread across Australia’s

rangelands (Australia’s Virtual Herbarium 2005) and

has been identified as one of the key threats to

rangeland biodiversity (Marshall et al. 2012; Martin

et al. 2006). Previous modelling of buffel grass

distribution in Australia implemented in CLIMEX

(Sutherst and Maywald 1985) projected buffel grass

expansion over 60 % of the Australian continent under

current climatic conditions (Lawson et al. 2004).

Model and methods

We model the susceptibility and suitability of the

Australian landscape to buffel grass colonization

under climate change using a Bayesian Network.

The BN model is organized around three invasion

Buffel grass and climate change 3199

123

requirements—the introduction of plant propagules to

a site, the establishment of new plants at a site, and the

persistence of viable populations at a site (Smith et al.

2012; Table 1). The establishment and persistence

nodes combine to influence the suitability to invasion

and the introduction and suitability nodes combine to

influence a sites susceptibility to invasion (Fig. 2).

Introduction, establishment and persistence are influ-

enced by key environmental variables such as land-

scape (e.g., soil type, tree cover, fire frequency),

climate (rainfall, temperature, soil moisture), and

dispersal (distance to nearest population). The rela-

tionship between these key environmental variables

and the invasion requirements are defined by experts

and empirical data and are illustrated through an

influence diagram which forms the basis of the BN

(Fig. 2; see Supplementary Material S2 for a descrip-

tion of relationships within the BN).

To illustrate the predictions of the model spatially

we use GIS spatial layers to represent the key

environmental variables directly or as proxies (Sup-

plementary Material, Table S1). For example the

current known distribution of buffel grass as mapped

by the Australian Virtual Herbarium is used to

estimate the distance to an existing population

(Fig. 2).

Eliciting expert knowledge

We captured the current understanding of buffel grass

ecology, management and invasion through a review

of the literature, expert workshop and follow up

elicitations with small groups and individual experts.

After an initial review of existing empirical informa-

tion, a two day workshop was convened with experts

in buffel grass ecology and management (see

acknowledgements for list of participants). During

this workshop experts identified the key environmen-

tal variables which influence buffel grass invasion

requirements—introduction, establishment and per-

sistence and the relationship between these variables

(Fig. 2). Buffel grass ecology (morphology, maturity

and growth habit) are known to vary considerably

throughout Australia (Humphreys 1967). In this

analysis we capture this variability through the

experts’ collective experience working on and manag-

ing buffel grass throughout the country. The spatial

layers available for mapping the key environmental

Table 1 Definition of BN nodes and their states as shown in Fig. 2, for susceptibility, suitability and the three invasion requirements

Node Definition State

Low Moderate High

Susceptibility Risk of being colonized by buffel grass

within 10 years based on the

probability of being introduced

within a 10 year timeframe and

suitability of the site

Low risk of being

invaded by buffel

grass within a

10 year timeframe

Moderate risk of being

invaded by buffel

grass within a 10 year

timeframe

High risk of being

invaded by buffel

grass within a 10 year

timeframe

Suitability Ability of buffel grass to establish and

persist

Can support isolated

plants only

Can support scarce to

moderate densities

Can support moderate

to high buffel

densities—extensive

monocultures possible

Establishment Frequency and density of seedling

recruitment (assumes seeds

available)

Recruitment absent

or infrequently in

low densities

Recruitment in low

densities most years

or moderate densities

every 5–10 years

Recruitment moderate

every year or dense

every 5–10 years

Persistence Ability of established buffel to survive,

grow and reproduce

Poor—Adults fail to survive or small proportion of adult plants survive,

grow and reproduce

Good—Most adult plants survive, grow and reproduce

Introduction The arrival of seeds from known

sources within a 10 year time-frame

No—none or\1 % chance of introduction from dispersal or direct

planting

Yes—introduction via dispersal from plants within 100 km or seeded at

the site

3200 T. G. Martin et al.

123

variables directly or indirectly were also identified.

Using a combination of facilitated group discussion,

small breakout groups and feedback, we developed a

set of key environmental variables for consideration in

the BN (Fig. 2).

The influence diagram was developed live during

the workshop using Netica 4.16 (Norsys Software

Corporation 1998) allowing experts to visualize the

relationships they were expressing and facilitating

easy updating and modifications as the workshop

progressed. Over 2 days the initial structure of the BN

and respective states of the nodes were developed. The

relationships between the nodes were defined by

conditional probabilities and were developed after-

wards in consultation with small groups of experts. A

conditional probability measures the probability of an

event given that another event has occurred. The

conditional probabilities for the entire network are

stored in a conditional probability table (CPT).

Conditional probabilities that are elicited from experts

are called elicited probability tables (EPTs). The EPTs

were developed via the following process. First, a

facilitated group discussion about the combination of

variables in question was undertaken. This discussion

was then followed by an independent elicitation of the

conditional probabilities from each expert. The group

then reconvened and discussed these independent

judgements. Finally each expert was invited to re-

evaluate their respective response (if desired) based on

the group discussion. This method captures the

benefits of group judgement while at the same time

maintaining independence (Martin et al. 2012a). The

mean response from the individual experts was then

taken and entered in the BN. For large conditional

probability tables (CPTs), Cain’s (2001) CPT calcu-

lator was used to generate the full CPT table from the

EPTs. The CPT calculator works by reducing the

number of scenarios in a CPT to key anchoring points

which are then interpolated to complete the entire

table. For example in the first set of probabilities

elicited, all of the parent nodes were in positive states

(benefited buffel grass introduction, establishment and

persistence) and in the second set, all of the parent

nodes were in negative states (impeded buffel grass

introduction, establishment and persistence). The CTP

calculator then works through the scenarios such that

only one parent node is not in its most favourable state

(i.e., most favourable for introduction, establishment

or persistence in this case).

Climate scenarios

Using outputs from the CSIRO mk3.5 Global Circu-

lation Model downloaded from OzClim (www.csiro.

Soil moisture

Fire frequency

Temperaturewe�est quarter

Rainfall Temperature

Distance tosource

Soil type

EstablishmentIntroduc�on

Planted

Persistence

Suscep�bility Suitability

Soil pH

Buffel grass

Climate

Invasion requirements

Soil surface salinity

Soil qualityGrazing index

Environmental variables

Grazing intensity

Tree cover Plant compe��on

(1)

(5)

(4) (2) (3)

(11)

(6)(7)

(8)

(12)

(10)

(13)

(9)

(14)

(17)(16)(15)

(18)

Fig. 2 Conceptual model

of the key climate,

environmental and invasion

requirements driving

landscape susceptibility and

suitability to buffel grass

colonization (see

Supplementary Material S2

for a review of the

conceptual model). Each

link is numbered based on

the relationships explained

in S2

Buffel grass and climate change 3201

123

au/ozclim), three climate scenarios were examined.

First, we modelled the predicted landscape suscepti-

bility and landscape suitability of buffel grass across

Australia based on current climate, centred on 1990

(recording range 1976–2005). Second we examined

the landscape suitability of buffel grass in 2070

(recording range 2065–2075) using a medium impact

A1B emissions scenario, and a high impact A1FI

emissions scenario (IPCC 2000). Monthly climate

change grids were downloaded at 0.25� resolution for

maximum temperature, minimum temperature, rain-

fall and evaporation, by specifying the above scenarios

in OzClim. Spatial downscaling was carried out using

the ANUCLIM software (Houlder et al. 2000) and

incorporates three submodels; ESOCLIM, which

outputs raw climate variable grids, BIOCLIM (Busby

1986) which outputs grids of average bioclimatic

parameters, and GROCLIM which can output gridded

indices from simple growth models. The beta release

of ANUCLIM version 6.0 was used, which allows

climate change grids to be applied over the historical

1990 centred climate surfaces. Software (www.csiro.

au/products/OzConverter-Software.html) was written

to interpolate the raw 0.25� CSIRO grids to cover the

whole Australian land mass, and relate evaporation

change to the date range used in ANUCLIM 6 (Har-

wood and Williams 2009).

Connecting the BN to GIS

All spatial layers (Table S1) were converted into a

25 km2 national grid, generating a 146 row 9 179

column matrix of grid cells totalling 26,134 cells. This

spatial scale ofmodellingwas deemed appropriate given

the environmental GIS spatial layers available. To read

theGIS data into the BN, we developed code which took

as input a text file containing the GIS layers as a string of

26,134 values and fed it into the BN software (Netica,

version 4.16, http://www.norsys.com/). Output from

Neticawas then convertedback into a textfile and then to

a raster for projection using ArcMap toolkit. Spatial

analyst and R (version 2.9.2; http://www.r-project.org/)

were then used to calculate differences between the

current and 2070 medium and 2070 high projections.

Model sensitivity

The sensitivity of each of the three invasion require-

ments; introduction, establishment and persistence to

the environmental variables included in the BN were

tested using entropy reduction. Entropy reduction

measures the degree to which findings at any node can

influence the beliefs in another, given the findings

currently entered in BN network. The degree of entropy

reduction I, is the expected difference in information H

between nodeQwith q states and findings nodeFwith f

states (Marcot 2006) and is calculated as:

I ¼ HðQÞ � HðQjFÞ ¼X

q

Xf

Pðq; f Þ log2½Pðq; f Þ�PðqÞPðf Þ

ð1Þ

In general, entropy reduction calculates the degree

to which nodeQ influences the response node Fwithin

the BN. In our case, the higher the value of I the greater

the sensitivity of each of the buffel grass invasion

requirements (introduction, establishment and persis-

tence) and buffel grass suitability and susceptibility to

key environmental variables included in the BN

(Marcot 2006).

Results

Bayesian network

The BN captured the key relationships between the

invasion requirements and key environmental vari-

ables as defined by experts, literature and as dictated

by available GIS layers. Introduction of buffel was

defined as being influenced by the proximity of the site

to source populations (‘distance to source’ node) and

whether or not buffel grass had been deliberately

‘seeded’ at a site (Fig. 2, Table S1, Supplementary

Material S2). Establishment was influenced by avail-

able soil moisture, C4 photosynthetic pathway, soil

quality and competition from tree cover and other

plants which in turn are moderated by livestock

grazing and rainfall. Warm season growing grasses

(C4) are linked to specialized Kranz leaf anatomy that

particularly adapts grasses like buffel to hot climates.

Where buffel grows well, the growing period coin-

cides with summer rainfall. In contrast to northern

Australia, in southern parts, where buffel is currently

absent, most of the rain falls in winter when it is too

cold for buffel establishment. With increasing mean

winter temperatures in the future, this impediment

may disappear. To include the interaction between C4

3202 T. G. Martin et al.

123

plant growth and temperature and rainfall we used the

BIOCLIM layer mean temperature of wettest quarter

(Table S1). Soil quality was derived from three GIS

layers; soil surface salinity, soil type and soil pH

(Table S1, Supplementary Material S2). Persistence

was influenced by soil moisture, temperature and soil

quality. Finally, ‘suitability’ to buffel grass coloniza-

tion is determined by ‘establishment’ and ‘persis-

tence’, whereas ‘susceptibility’ to buffel grass

colonization is influenced by ‘introduction’ and ‘suit-

ability’ (Fig. 2).

Current susceptibility and suitability to buffel

grass colonization

Under current climate conditions, the probability of

high landscape susceptibility reflects regions which

are currently experiencing the highest pressure of

‘introduction’ from buffel grass (Fig. 3). In general,

central Australia, central Queensland (Qld) and

pockets of Western Australia (WA) are currently

expected to experience the highest colonization pres-

sure. Removing the ‘introduction’ requirement, the

model reveals regions which are projected to have

highest suitability (43–95 %) for buffel grass (Fig. 4).

In other words, if buffel were to be introduced via

dispersal or seeding, these areas are projected to be

most suitable for buffel establishment and persistence

under current climate conditions. Highly suitable areas

comprise the majority of WA, NT, and Qld excluding

themost northerly and coastal regions. Mid to northern

South Australia (SA) and New South Wales (NSW)

are also projected to be highly suitable. The only states

to contain no highly suitable habitat are Victoria and

Tasmania.

Future suitability to buffel grass colonization

Examining the suitability of the continent to buffel

grass establishment and persistence under a 2070

Fig. 3 Probability of high

susceptibility for buffel

grass under current climate

conditions. Locations of the

national reserve system are

shown in green

Buffel grass and climate change 3203

123

medium impact A1B emissions scenario, there is a

shift in high suitability southwards and a decreasing

suitability in the northern and north-western parts of

the continent. Central Australia and Queensland

remain buffel strongholds and southern parts of WA

and much of western NSW and South Australia

become highly suitable. Under the 2070 high impact

A1F1 emissions scenario (Fig. 5), the southward shift

in suitability becomes even more pronounced spread-

ing into pockets of south WA, SA, QLD, NSW and for

the first time northern Victoria. In contrast, the

suitability of central Australia declines for the first

time below 40 %.

Impact of buffel grass across the National Reserve

System

We examined the impact of projected buffel grass

suitability across the National Reserve System (NRS)

(Fig. 6). Suitability was divided into four classes

corresponding to low (\20 %), medium (20–39 %),

high (40–59 %) and very high ([60 %) probability of

suitability. Overall the risk to the NRS is projected to

increase with a greater proportion of reserves pro-

jected to be high to very highly suitable in 2070 as

compared to projections under current climate condi-

tions. As a result, the suitability of ecoregions (Anon.

2012) within the NRS to buffel grass invasion is also

projected to change in the future. Western and central

Australia’s deserts and xeric shrublands, also known

as hummock grasslands will continue to be a

stronghold for buffel grass establishment and persis-

tence in the future. Within the south and south-east,

mediterranean forests and woodlands, temperate

woodlands and south-eastern forests are all projected

to become more suitable. Only the northern tropical

savannas are projected to become less suitable in the

future.

Fig. 4 Probability of high

suitability for buffel grass

colonization under current

climate conditions

3204 T. G. Martin et al.

123

Change in buffel grass suitability

Areas of increasing and decreasing suitability for

buffel grass in the future are illustrated by subtracting

the 2070 high suitability projections from those under

the current climate scenario (Fig. 7). Shades of brown

show regions projected to increase in suitability for

buffel whereas regions shaded in blue reveal areas

which are projected to decrease in suitability. The

southward shift in suitability is largely driven by

changes in soil moisture and timing of rainfall

projected under the 2070 scenarios. Buffel grass’s,

C4 photosynthetic pathway means the timing of

rainfall for buffel grass establishment and persistence

is paramount. The increase in winter temperatures

projected under the 2070 high emission scenario

coinciding with sufficient rainfall would allow buffel

Fig. 5 Probability of high

suitability for buffel grass

colonization under 2070

high emissions climate

scenario

0

10

20

30

40

50

60

<20(low)

20 - 39(medium)

40 - 59(high)

60 - 100(very high)

Prop

or�

on o

f NRS

Projected suitability (%)

current

2070 medium

2070 high

Fig. 6 Proportion of the national reserve system projected to be

of low (\20 %), medium (20–39 %), high (40–59 %), and very

high ([60 %) suitability to buffel grass colonization under three

climate scenarios; current, 2070 medium emissions, 2070 high

emissions

Buffel grass and climate change 3205

123

grass to establish and grow further south, whereas

under current conditions, insufficient rainfall during

the warmer months prevents buffel establishment.

Model sensitivity

The sensitivity of the three invasion requirements to

the environmental variables reflects the expert judge-

ments involved in constructing the conditional prob-

ability tables. For the invasion requirement

‘introduction’, whether or not buffel grass was

‘seeded’ at a site was the most important feature

influencing this node (Fig. 8). Soil quality, and in

particular soil type, was the key determinant of

persistence. Soil quality and mean temperature of

wettest quarter and to a lesser extent tree cover and soil

moisture were the most important variables influenc-

ing establishment. Suitability was driven equally by

persistence and establishment, whereas susceptibility

was influenced more by suitability than introduction.

The invasion requirements were least sensitive to high

temperatures, fire frequency and grazing intensity.

Evidence suggests buffel grass is tolerant of very high

temperatures (Cox et al. 1988). The limited influence

of grazing on invasion success of buffel grass may be

due in part because grazing intensity was considered

low across all of the rangelands (Bastin, AcRIS

Management Committee 2008), but is also supported

by recent experimental data (Fensham et al. 2013)

which concluded that both grazing and fire do not

substantially promote invasion.

Discussion

As climates shift, the relative risk associated with

invasive species is likely to change. To best manage

these dynamic threats, an understanding of where

changes in projected species distributions are greatest

is required. For most invasive species we lack basic

Fig. 7 Change in projected

high suitability between the

current climate and 2070

high emissions scenario,

where shades of brown

reveal regions of increasing

suitability and shades of

blue decreasing suitability

3206 T. G. Martin et al.

123

information on current distributions and habitat suit-

ability, inhibiting the use of traditional species distri-

bution models. Our framework acknowledges this

information gap and provides a method for projecting

species distributions under different climate scenarios

when empirical data is not sufficient. Through the

elicitation and use of expert judgements within a BN

combined with available empirical data and spatial

data layers, we provide a set of projections of the

relative risk across the Australian continent and its

greatest biodiversity asset, the National Reserve

System, to colonization by buffel grass, a contentious

invasive species due to its pastoral value and negative

biodiversity impact.

Suitability to buffel grass colonization

and management implications

Our results highlight the potential for buffel grass

management to become increasingly important in the

southern part of the continent, whereas in the north the

threat of buffel grass is likely to lessen with climate

change. This is potentially good news for Australian

rangeland biodiversity. However our work also

suggests that overall the risk of buffel grass coloniza-

tion, establishment and persistence within the NRS is

likely to increase with climate change; a consequence

of the high number of reserves located in the central

and southern portion of the continent.

Many species distribution modelling approaches

have been proposed for projecting the distribution of

invasive species, the most common being bioclimate

niche modelling (Venette et al. 2010). The benefit of

using BNs over bioclimatic niche modelling is their

ability to capture process-understanding even in data

poor environments by drawing on both expert judg-

ments and empirical data. The key difference between

our BNmodelling approach with that of Lawson et al’s

(2004) CLIMEX projections of buffel grass distribu-

tion under current climatic conditions is our ability to

capture landscape scale processes. A key process

driving current buffel grass susceptibility was propag-

ule pressure in the landscape via seeding and distance

to source population. The inclusion of this process

alone, translated into notable variation between the

two modelling approaches. For example, our model

projected high suitability for buffel throughout an area

dominated by brigalow (Acacia harpophylla) which

0 0.05 0.1 0.15 0.2

Suitability

Introduc�on

0 0.05 0.1 0.15

Peristence

Establishment

0 0.05 0.1 0.15 0.2

Soil quality

Plant compe��on

Temp we�est qtr

Tree cover

Soil moisture

0 0.05 0.1 0.15 0.2

soil quality

Temperature

Soil Moisture

Susc

ep�b

ility

Suita

bilit

yEs

tabl

ishm

ent

0 0.8

Seeded

Distance to source

0.05 0.1 0.15 0.2 //

Pers

iste

nce

Intr

oduc

�on

Fig. 8 Sensitivity of each

of the invasion requirements

(introduction, establishment

and persistence) to key

environmental variables

included in the BN

Buffel grass and climate change 3207

123

extends from Northern Queensland to mid-northern

western New South Wales. In contrast, Lawson et al’s

model projected high suitability in the northern section

of the brigalow only with the southern region deemed

unsuitable as a result of unfavourable climatic condi-

tions. Known buffel distribution as mapped by the

Australian Virtual Herbarium reveal many records in

both the northern and southern portions of the

brigalow belt suggesting it is highly suitable for buffel

grass establishment and persistence, corroborating our

findings. Two other recent studies on buffel grass

invasion have found that landscape scale processes

can be as important in driving invasion, as climatic

factors (Butler et al. 2014; Fensham et al. 2013).

In future using BNs in conjunction with bioclimatic

models such as CLIMEX (Sutherst, Maywald 1985) is

likely to prove useful by capitalizing on its sophisti-

cated climate and edaphic sub-models as well as the

process-understanding that can be incorporated within

a BN (Murray et al. 2012, 2014; Smith et al. 2012; Van

Klinken et al. 2015).

Three major sources of error can influence the

validity of the projections from a Bayesian network

(Van Klinken et al. 2015). First, the model may be

correct, but the spatial data is wrong. While the

advantage of BNs is that they can utilize a variety of

data types, when using BNs to project species

distributions across land and seascapes one is con-

strained by the spatial data layers that are available. In

many cases it may be necessary to use a surrogate for

the variable in question. The manner in which the

spatial data are categorized will also influence the

analysis. For example, the ‘grazing intensity’ node

used here was based on a spatial layer that was rather

coarse and mapped the majority of the rangelands as

low grazed, despite our knowledge that there is

considerable variability in grazing pressure across

the rangelands. In another example, the node ‘distance

to source’ relied on the Australian Virtual Herbarium

records of known buffel grass locations. These data are

biased towards roads and easily accessible regions and

are by no means complete. We are aware for example,

that there are many occurrences of buffel grass in

south-west South Australia which did not appear in the

herbarium data. In future, the expected spatial uncer-

tainty within the spatial layers themselves could be

elicited from experts and used in a sensitivity analysis

to determine the impact of this uncertainty on

projections. The second source of error is that the

expert knowledge may be correct, but the model is

wrong. We assessed the validity of the model through

a review of the model outputs with the experts. All

experts agreed that the model accurately reflected their

understanding of buffel invasion dynamics. In cases,

where the model does not reflect the knowledge of the

experts a process of revision including examining the

model structure, definitions of nodes and the condi-

tional probabilities is undertaken. The final source of

error is that the expert knowledge used to construct the

model itself might be wrong. This is also called

structural uncertainty or conceptual uncertainty and

arises when a key process or variable is overlooked or

unknown and thus not included in the Bayesian

Network. In future, this could be tested through the

comparison of independently constructed Bayesian

networks and through field experiments to examine

key environmental variables associated with the

invasion process (Van Klinken et al. 2015).

BNs derived from expert judgements and available

spatial layers offer a cost-effective and timely alter-

native to models which rely solely on empirical data or

climate only information. In many environmental

management situations, we do not have the luxury of

long time frames. Often decisions must be made

quickly or else opportunities to control an invasive

species or recover an endangered species are lost

forever (Martin et al. 2012c). In these situations,

decisions based on models parameterized with expert

judgements may be better than delaying decisions

until empirical data is available (Martin et al. 2012a).

Our analysis is already being used to inform threat

management priorities for invasive plant management

across central Australia (Firn et al. 2013). Using our

projections, Firn et al. (2013) find that buffel grass

management is orders of magnitude more expensive

than the management of nine other plant species

invading central Australia. The high management

costs were driven by buffel grasses extensive projected

distribution and the lack of effective management

strategies to control its spread. Given buffel grasses

ability to transform entire landscapes (Marshall et al.

2012), this is sobering news and emphasizes the need

to focus protection on high value assets such as Uluru

and Kata Tjuta National Park until more cost-effective

management strategies are developed.

The development of cost-effective and efficient

management strategies which account for trade-offs in

production and biodiversity benefits will be a valuable

3208 T. G. Martin et al.

123

contribution towards managing the dynamic threat of

buffel grass invasion in the future (Friedel et al. 2011;

Grechi et al. 2014). Our results suggest that climate

change will not diminish the issue of how to manage

invasive species such as buffel grass, but rather

highlights the need to develop intelligent management

policy in response to projected shifts in the spatial

distribution of such species and subsequent threats to

national assets.

Acknowledgments This project was funded through the

Department of Environment, Water, Heritage and the Arts and

some initial findings of this research were published as a report

(Martin et al. 2012b). This project would not have been possible

without the generous contribution of time and expertise from

scientists and natural resource managers across the country:

Gary Bastin, Kerrie Bennison, John Clarkson, Keith Ferdinands,

Margaret Friedel, David Gobbett, Tony Grice, Ben Lawson,

Neil Macleod, Cam McDonald, Samantha Setterfield, Stephen

VanLeeuwen, Wayne Vogler, and DickWilliams. We thank our

colleagues involved in the wider National Reserve System

project: Michael Dunlop, Simon Ferrier, Tom Harwood, David

Hilbert, Alan House, Suzanne Prober, Anita Smyth and Kristen

Williams. Finally, our gratitude to Jennifer Firn, John Dwyer

and two anonymous reviewers for providing helpful comments

on this manuscript.

References

Anon (2010) Uluru-Kata Tjuta National Park management plan

2010–2010. Director of National Parks, Canberra

Anon (2012) Interum biogeographic regionalisation for Australia

(IBRA) version 7. Department of Sustainability, Environ-

ment, Water, Population and Communities, Canberra

Arriaga L, Castellanos AE, Moreno VE et al (2004) Potential

ecological distribution of alien invasive species and risk

assessment: a case study of buffel grass in arid regions of

Mexico. Conserv Biol 18:1504–1514

Australia’s Virtual Herbarium (2005) Herbaria records of

Cenchrus cililaris in Australia. Available at http://www.

anbg.gov.au/cgi-bin/avh.cgi. Accessed 13/04/2010

Bastin G, AcRIS Management Committee (2008) Rangelands

2008—taking the pulse, published on behalf of the AcRIS

Management Committee by the National Land and Water

Resources Audit Canberra

Busby JR (1986) Bioclimate prediction system (BIOCLIM)

user’s manual, version 2.0. Bureau of Flora and Fauna,

Canberra

Butler DW, Fairfax RJ (2003) Buffel grass and fire in a gidgee

and brigalow woodland: a case study from central

queensland. Ecol Manag Restor 4:120–125

Butler S, McAlpine C, Fensham R et al (2014) Climate and

exotic pasture area in landscape determines invasion of

forest fragments by two invasive grasses. J Appl Ecol

51:114–123

Cain J (2001) Planning improvement in natural resource man-

agement. Centre for Ecology & Hydrology, Wallingford

Clarke PJ, Latz PK, Albrecht DE (2005) Long-term changes in

semi-arid vegetation: invasion of an exotic perennial grass

has larger effects than rainfall variability. J Veg Sci

16:237–248

Cox JR,Martin-RMH, Ibarra-F FA et al (1988) The influence of

climate and soils on the distribution of four African grasses.

J Range Manag 41:127–139

Elith JH, Graham CP, Anderson R et al (2006) Novel methods

improve prediction of species’ distributions from occur-

rence data. Ecography 29:129–151

Eyre TJ, Wang J, Venz MF et al (2009) Buffel grass in

Queensland’s semi-arid woodlands: response to local and

landscape scale variables, and relationship with grass, forb

and reptile species. Rangel J 31:293–305

Fairfax RJ, Fensham RJ (2000) The effect of exotic pasture

development on floristic diversity in central Queensland,

Australia. Biol Conserv 94:11–21

Fensham RJ, Donald S, Dwyer JM (2013) Propagule pressure,

not fire or cattle grazing, promotes invasion of buffel grass

Cenchrus ciliaris. J Appl Ecol 50:138–146

Firn J, Martin TG, Walters B et al (2013) Priority threat man-

agement of invasive plants species in the Lake Eyre Basin.

CSIROandQueenslandUniversity of Technology, Australia

Franks A (2002) The ecological consequences of Buffel Grass

(Cenchrus ciliaris) establishment within remnant vegeta-

tion of Queensland. Pac Conserv Biol 8:99

Friedel MH, Grice AC, Marshall NA et al (2011) Reducing

contention amongst organisations dealing with commer-

cially valuable but invasive plants: the case of buffel grass.

Environ Sci Policy 14:1205–1218

Grechi I, Chades I, Buckley YM et al (2014) A decision

framework for management of conflicting production and

biodiversity goals for a commercially valuable invasive

species. Agric Syst 125:1–11

Harwood TD, Williams KJ (2009) CSIRO OzConverter: Edit

0.25� OzClim*.asc output files. http://csiro.au/products/

OzConverter-Software.html

Hellmann JJ, Byers JE, Bierwagen BG et al (2008) Five

potential consequences of climate change for invasive

species. Conserv Biol 22:534–543

Houlder D, Hutchinson MF, Nix HA et al (2000) ANUCLIM

Version 5.1. Centre for Resource and Environmental

Studies, Australian National University, Canberra. http://

cres.anu.edu.au/outputs/anuclim.html

Humphreys LR (1967) Buffel grass (Cenchrus ciliaris) in Aus-

tralia. Trop Grassl 1:123–134

IPCC (2000) Emissions scenarios. special report of the inter-

governmental panel on climate change. Cambridge

University Press, UK

Jackson J (2005) Is there a relationship between herbaceous

species richness and buffel grass (Cenchrus ciliaris)?

Austral Ecol 30:505–517

Keith DA, Akcakaya HR, Thuiller W et al (2008) Predicting

extinction risks under climate change: coupling stochastic

population models with dynamic bioclimatic habitat

models. Biol Lett 4:560–563

Kuhnert PM,Martin TG, Griffiths SP (2010) A guide to eliciting

and using expert knowledge in Bayesian ecological mod-

els. Ecol Lett 13:900–914

Buffel grass and climate change 3209

123

Lawson BE, Bryant MJ, Franks AJ (2004) Assessing the

potential distribution of buffel grass (Cenchrus ciliaris L.)

in Australia using a climate-soil model. Plant Prot Q

19:155–163

MacDougall AS, Turkington R (2005) Are invasive species the

drivers or passengers of change in degraded ecosystems?

Ecology 86:42–55

Mantyka-Pringle CS, Martin TG, Moffatt DB et al (2014)

Understanding and predicting the combined effects of cli-

mate change and land-use change on freshwater macroin-

vertebrates and fish. J Appl Ecol 51:572–581

Marcot BG (2006) Characterizing species at risk I: modeling

rare species under the Northwest Forest Plan. Ecology and

Society 11:10 online

Marcot BG, Holthausen RS, Raphael MG et al (2001) Using

Bayesian belief networks to evaluate fish and wildlife

population viability under land management alternatives

from an environmental impact statement. For Ecol Manag

153:29–42

Marriott JA (1955) Buffel grass. Address to the Queensland

Branch of the Australian Institute of Agricultural Science

summarised. Australian Institute of Agricultural Science

21:277

Marshall VM, Lewis MM, Ostendorf B (2012) Buffel grass

(Cenchrus ciliaris) as an invader and threat to biodiversity

in arid environments: a review. J Arid Environ 78:1–12

Martin TG, Campbell S, Grounds S (2006) Weeds of Australian

rangelands. Rangel J 28:3–26

Martin TG, BurgmanMA, Fidler F et al (2012a) Eliciting expert

knowledge in conservation science. Conserv Biol

26:29–38

Martin TG, Murphy H, Liedloff A (2012b) Invasive species and

climate change: a framework for predicting species dis-

tribution when data are scarce CSIRO Climate Adaptation

Flagship Working Paper No. 13G. http://www.csiro.au/en/

Organisation-Structure/Flagships/Climate-Adaptation-Flag

ship/CAF-working-papers.aspx

Martin TG, Nally S, Burbidge A et al (2012c) Acting fast helps

avoid extinction. Conserv Lett 5:274–280

McIvor JG (2003) Competition affects survival and growth of

buffel seedlings—is buffel grass a coloniser or an invader?

Trop Grassl 37:176–181

Mooney HA, Hobbs RJ (2000) Invasive species in a changing

world. Island Press, Washington

Murray JV, Stokes KE, van Klinken RD (2012) Predicting the

potential distribution of a riparian invasive plant: the

effects of changing climate, flood regimes and land-use

patterns. Glob Change Biol 18:1738–1753

Murray J, Berman D, van Klinken R (2014) Predictive mod-

elling to aid the regional-scale management of a vertebrate

pest. Biol Invasions 16:2403–2425

Norsys Software Corporation (1998) Netica applciation for

belief networks and influence diagrams: user’s guide.

Norsys Software Corporation, Vancouver

Ponce-Reyes R, Reynoso-Rosales V-H, Watson JEM et al

(2012) Vulnerability of cloud forest reserves in Mexico to

climate change. Nat Clim Change 2:448–452

Smith C, Howes AL, Price B et al (2007) Using a Bayesian

belief network to predict suitable habitat of an endangered

mammal—the Julia Creek dunnart (Sminthopsis douglasi).

Biol Conserv 139:333–347

Smith C, van Klinken RD, Seabrook L et al (2012) Estimating

the influence of landmanagement change on weed invasion

potential using expert knowledge. Divers Distrib

18:818–831

Smyth AK, Friedel M, O’Malley C (2009) The influence of

buffel grass (Cenchrus ciliaris) on biodiversity in an arid

Australian landscape. Aust Rangel J (in press)

Sutherst RW, Maywald GF (1985) A computerised system for

matching climates in ecology. Agric Ecosyst Environ

13:281–299

Sutherst RW, Baker RHA, Coakley SM et al (2007) Pests under

global change - meeting your future landlords. In: Canadell

JG, Pataki DE, Pitelka LF (eds) Terrestrial ecosystems in a

changing world. Springer, Berlin, pp 211–226

Thomas CD, Cameron A, Green RE et al (2004) Extinction risk

from climate change. Nature 427:145

Tu M (2002) Species management summary: element stew-

ardship abstract for Cenchrus ciliaris L. Accessed 25/10/

2014 http://www.imapinvasives.org/GIST/ESA/esapages/

cenccili.html

van Devender TR, Felger RS, Burquez AM (1997) Exotic plants

in the Sonoran Desert Region, Arizona and Sonora. In:

Kelly M, Wagner E, Warner P (eds) Proceedings of the

California exotic pest plant council symposium, pp 10–15

van Klinken RD, Lawson BE, Zalucki MP (2009) Predicting

invasions in Australia by a Neotropical shrub under climate

change: the challenge of novel climates and parameter

estimation. Glob Ecol Biogeogr 18:688–700

van Klinken R, Murray JV, Smith C (2015) Process-based Pest

RiskMapping using Bayesian Networks andGIS. Pest Risk

Modelling and Mapping for Invasive Alien Species. CAB

International, pp 171–188

Venette RC, Kriticos DJ, Magarey RD et al (2010) Pest risk

maps for invasive alien species: a roadmap for improve-

ment. Bioscience 60:349–362

Walther G-R, Roques A, Hulme PE et al (2009) Alien species in

a warmer world: risks and opportunities. Trends Ecol Evol

24:686–693

3210 T. G. Martin et al.

123