304 R.W. Sutherst et al. Agriculture, Ecosystems and Environment 82 2000 303–319
Vulnerability = Sensitivity 1 − Adaptive management where ‘Sensitivity’ is the impacts in the absence of
adaptation and ‘Adaptive management’ the response options as determined by the sustainability and ro-
bustness of available and putative management tools Sutherst et al., 1998.
The process described above has proved to be very appropriate for use on pests, diseases and weeds.
Sutherst 2000 described a conceptual framework for studying the effects of invasive species under
global change. This framework incorporates ap- proaches from the IPCC, and the quarantine pest risk
International Plant Protection Convention, IPPC and plant pathology communities. Sutherst et al.
2000 applied the IPCC approach in an assessment of the vulnerability to climate change of differ-
ent horticultural industries and growing regions in Australia.
The research issues facing global change research- ers, policymakers and managers of pests were
reviewed by Sutherst et al. 1996a. In summary, there is less support available for climate change research
than is allocated to the solution of problems of im- mediate concern. This is despite the initial need for
global change research to understand current prob- lems better. Such research will provide a true baseline
as well as an understanding with which to enhance risk assessments and the design of both current and
future management options. Other research issues are that risk assessments often involve multiple in-
teracting drivers of change, which degrade the value of isolated assessments. Furthermore, outputs relate
mostly to the policy-level, which means regional, national or global scales involving industries or re-
gions. Policy questions focus on defining impacts and vulnerability, via a combination of regionalindustry
scale sensitivity analysis and evaluation of adaptation options. Results of risk assessments have to be trans-
lated into socio-economic measures in order to be meaningful for policymakers. These demands are not
matched by available resources and therefore need a degree of pragmatism if they are to be addressed in
the foreseeable future.
The scientific community involved in biological research needs some common data sets, analytical
tools and languages to facilitate collaboration. The uncertainty of future global change scenarios and
our inability to conduct large-scale experiments on impacts of global change have forced reliance on
models to explore possible impacts. Global research networks are also needed to gain leverage through
collaboration and information sharing. The Global Change and Terrestrial Ecosystems Program GCTE
of the International Geosphere Biosphere Program IGBP has a number of crop and soil networks
Ingram et al., 1999 and is developing a pest and disease network Sutherst et al., 1996b; Scherm et al.,
1999. The crop modelling community has already made progress in developing international collabora-
tive research groups, which run model comparison exercises in a push to improve the models for use
in global change research Landau et al., 1998; In- gram et al., 1999. The partnership has been assisted
by the previous collaborative network built around the use of the DSSAT model http:agrss.sherman.
hawaii.edudssatindex.htm. In contrast the pest re- search community faces different challenges with its
large number of pest communities that vary greatly in composition in different regions and on different
crops, making it less appropriate to focus collabora- tion on particular species. Thus GCTE has focussed
effort initially on developing generic approaches to the assessment of risks under global change Sutherst
et al., 1996a; Scherm et al., 1999, while recognising the need to increase effort in experimentation and
monitoring activities.
Two modelling tools are discussed here, namely CLIMEX Sutherst et al., 1999, a tool for modelling
species’ responses to climate, and DYMEX May- wald et al., 1997, 1999, a generic and modular mod-
elling toolkit designed for biologists. Both CLIMEX and DYMEX offer the advantage that the models or
modules that they create can be readily shared with other users to avoid the need for wasteful repetition of
effort.
2. A biological toolkit for global change
The starting point for the design of any toolkit to address global change issues is to find out what sort
of questions the end-users of the products of research ask. Policymakers often do not know what technical
questions to ask, but they do want to define the likely winners and losers on both regional and industry scales
R.W. Sutherst et al. Agriculture, Ecosystems and Environment 82 2000 303–319 305
and therefore need answers in socio-economic terms. They also need to know the time-scale on which the
impacts are likely to be felt and the urgency with which adaptation measures need to be implemented.
Likely costs of adaptation are also important consid- erations. At the individual level, farmers want to know
whether the enhanced greenhouse effect is going to affect their local area and, if so, when they should im-
plement adaptation measures and how much they will need to spend. Thus the emphasis is on directions of
change and approximate indications of the scale of the likely problem, preferably with a probability attached
to projections.
Given the range of demands, the variable nature of the available data, and the logistics involved in a
detailed simulation modelling analysis, a hierarchy of data and tools is necessary Sutherst et al., 1996a. The
data range from expert opinion through qualitative ob- servations to rigorous, quantitative data on biological
processes or states. Software tools range from expert systems through pattern-matching tools for matching
climate or vegetation types to very simplified sim- ulation models like CLIMEX or process-based i.e.
mechanistic population and community models.
Population or ecosystem dynamics models pro- vide the most comprehensive method for quantifying
likely future adaptation options but they require an order of magnitude more data and effort than more
pragmatic approaches. Their use will therefore be re- stricted to key problems with sufficient resources to
collect the necessary data. For most other problems, there is a role for the less sophisticated approaches
outlined above. Our proposed hierarchical ‘climate impacts toolkit’ Sutherst et al., 1996a is designed
to meet the need for a range of analyses, includ- ing climate-matching and process-based modelling.
CLIMEX http:www.ento.csiro.auresearchpestmg- mtclimexclimex.htm and DYMEX http:www.
ento.csiro.auresearchpestmgmtdymexdymexfr.htm are core features of our toolkit and some of their key
features are described below.
The application of these two software tools in a ‘Modelling Workshop’ environment is illustrated
using the results of the GCTE Global Change Work- shop on modelling global change impacts on pests in
Bogor in 1998 Sutherst et al., 1999. Other exam- ples are used to illustrate particular points where
necessary. 2.1. CLIMEX modelling of species’ responses to
climate CLIMEX is a simplified dynamic model that is
used to infer species’ responses to climate from obser- vations of the geographical distribution and seasonal
abundance. It is appropriate for the vast majority of species, for which comprehensive lifecycle data are
not available. CLIMEX has been used for a number of years in global change research and has provided a
regional perspective of problems. It has enabled many analyses that would otherwise have been impossible
through lack of adequate data to build population mod- els Worner, 1991; Sutherst et al., 1995; Baker et al.,
1996.
The attributes and applications of the CLIMEX modelling software have been extensively documented
elsewhere Sutherst and Maywald, 1985, 1999; Sutherst et al., 1995, 1999; Sutherst, 1998. It uses
a combination of ‘growth’ and ‘stress’ indices to describe responses over the full range of climatic
conditions that occur in different seasons in different places around the world. Experience with CLIMEX
has shown how little biologists know about the role of climate in limiting species’ distributions. Sutherst et al.
1995 and Kriticos and Randall 2000 compared the different ways in which various climate-matching
approaches matched climates. Rather than repeat a description of the model here, the opportunity is taken
to highlight or elaborate on a number of features and issues related to CLIMEX and climate-matching that
are not widely appreciated.
1 It is often not possible to characterise the climatic requirements of a target species, such as
when targeting a collection or release site for a little known biological control agent. CLIMEX provides
a climate-matching facility — ‘Match Climates’ — that compares meteorological parameters from dif-
ferent locations directly. It also allows the user to select which meteorological variable or period of the
year to include when comparing data from different locations. This allows the user to explore a range of
hypotheses on the factors limiting a species’ geo- graphical distribution when it is not feasible to model
its growth and stress responses using the ‘Compare LocationsYears’ model. For example, it is possible to
compare a selection of geographical locations using similarities of minimum winter temperatures with one
306 R.W. Sutherst et al. Agriculture, Ecosystems and Environment 82 2000 303–319
Fig. 1. CLIMEX ‘Match Climates’ index for Mt Tamborine in Queensland, Australia with locations in New South Wales: a winter minimum temperature, b winter maximum temperature.
using maximum winter temperatures, which would have a different physiological effect Fig. 1. In the
example, a species that can persist at Mt Tamborine in southern Queensland in Australia is projected to be
able to colonise quite different regions depending on whether it is limited by minimum temperatures or by
a lack of thermal accumulation during winter.
Caution is needed when comparing climates be- cause the CLIMEX ‘Match Climates’ match index
does not currently indicate the direction of differences between the target and matched locations. For exam-
ple, the winter minimum temperatures of a matched location Tibooburra, NSW may be colder than the
target location Mt Tamborine but have the same match index 0.68 as a warmer location Port Kem-
bla, NSW as shown in Fig. 2. It is therefore necessary to support the analyses with direct examination of the
meteorological data in the context of the particular organism being investigated. CLIMEX provides both
graphical and tabular displays to enable the user to do this. The same need for supportive examination
of the raw meteorological data arises when match- ing climates close to the edges of the geographical
distribution of a species.
2 CLIMEX provides ‘Compare Locations’ or ‘Compare Years’ functions to model the stressful
mechanisms by which climate limits the geographical distribution of species and the suitable conditions that
enable population growth in the favourable season. These functions are suitable for analyses of risks from
pests. Fig. 3a shows the estimated limiting effect of heat stress on the geographical distribution of the
rice-field rat Rattus argentiventer in Asia. Fig. 3b and c shows the CLIMEX summary index — the annual
ecoclimatic index EI — of distribution and abun- dance in relation to climate for the rat in Asia. The EI
integrates a measure of population growth, the annual growth index GI
A
, with the sum of the limiting fac- tors called ‘stresses’. The analyses were done under
current climate and a future climate change scenario with a 2
◦
C increase in temperature Sutherst et al., 1999. A global climate grid of interpolated data
IPCC, 1999 was used to create the present results. As concluded by Worner 1991 and Atzeni et al.
1994, a temperature and moisture-based ‘growth index’ model, as contained in the CLIMEX ‘Compare
LocationsYears’ options, is often as useful as a popu- lation model for estimating the likely extent of spread
of a species for policy purposes. An example of a comprehensive analysis using CLIMEX was given
by Yonow and Sutherst 1998. Nix and Fitzpatrick 1969, Fitzpatrick and Nix 1970 and Stephens et al.
1989 obtained useful estimates of crop yields on a regional basis using such ‘growth index’ models. The
R.W. Sutherst et al. Agriculture, Ecosystems and Environment 82 2000 303–319 307
Fig. 2. Comparison of winter minimum temperatures of Mt Tamborine — with those of two locations, Tibooburra · · · and Port Kembla - - - in New South Wales, Australia, each of which has the same value 0.68 for the CLIMEX match index with Mt Tamborine.
results from CLIMEX can be scaled in some cases to reflect economic costs by relating the GI or EI
to recorded amounts of pest damage Sutherst et al., 2000.
3 CLIMEX is a climate- rather than weather-driven modelling program that is designed to provide in-
sights into species’ requirements for climate, as ex- pressed by their geographical distribution, seasonal
phenology and relative abundance. It is based on the premise that it is possible to define climates that are
conducive to the generation of particular weather patterns, which directly affect populations on a short
time-scale. Thus, for example, some plant pathogens require a very specific, short period of leaf wetness
that is only described precisely by local weather data with a resolution as short as a few hours. CLIMEX
is not designed to predict actual events in a crop us- ing real-time weather data, although it is being used
successfully to deliver weekly forecasts of pest and pathogen development based on thermal accumula-
tion in turf in USA C. Sann, pers. comm.. Rather, CLIMEX indicates, in a more strategic manner on
a landscape scale, the average weekly interpolated from monthly average data with the smoothing that
this involves climatic conditions that are conducive to the generation of the weather required to produce
particular temperature or moisture conditions. It does this by defining the climatic responses of the species,
which allow it to exist in its geographical distribu- tion. This resolution is more appropriate for policy
and planning questions about issues such as risk as- sessment in integrated pest management IPM and
quarantine, biological control or climate variability and long-term climate change. It is also more suited
to use with data on geographical distributions.
4 CLIMEX exploits a meteorological database with much higher temporal resolution than any
other climate-matching program for both its climate- matching Match Climates and its modelling Com-
pare LocationsYears functions Sutherst et al., 1995. For example, BIOCLIM users have usually used 7–32
seasonal data points to describe the 30-year aver- age climate of a given location Booth, 1996. The
Compare LocationsYears functions in CLIMEX use weekly, temporally interpolated data from averages
of five different variables: maximum and minimum temperatures, 9 a.m. and 3 p.m. RH, and rainfall.
This amounts effectively to 260 data points per lo- cation for the fitting process. Climates are matched
Match Climates using monthly values, with options to weight different variables or to mask out parts of
the year. Such high resolution partly explains why
308 R.W. Sutherst et al. Agriculture, Ecosystems and Environment 82 2000 303–319
Fig. 3. a CLIMEX heat stress index HS for the rice-field rat R. argentiventer in Asia under current climate, and the CLIMEX ecoclimatic index under b current climate, and c +2
◦
C.
R.W. Sutherst et al. Agriculture, Ecosystems and Environment 82 2000 303–319 309
Fig. 4. CLIMEX growth index for the predatory mite C. peregrinus in Brownsville: a current rainfall pattern, b rainfall 2 months earlier, and c rainfall 2 months later.
CLIMEX is so robust, because it puts huge, although hidden, demands on the parameter-fitting procedure.
Another advantage that it confers is the flexibility to address any climate, including novel climates that
may emerge under climate change. In particular, one of the most significant changes in climate that is usu-
ally overlooked is the possible change in the seasonal timing of rainfall. A shift forwards or backwards can
have very substantial effects on species by disrupting required synchronisation of suitable temperatures and
rainfall. For example, the response of a predatory mite species Coccidoxenoides peregrinus Timber-
lake that has an observed peak in population growth in autumn in Brownsville USA is quite different to
that when the year’s rainfall is either brought forward or delayed by two months Fig. 4a–c, as indicated
by the CLIMEX weekly growth index GI
W
. 5 CLIMEX parameters reflect the observed ge-
ographical distribution of a species and so relate to the sum of all the genetic variation contained in that
species, i.e. the distribution is made up of an aggre- gate of biotypes, which may vary to different degrees
in their response to climate. Hence it is not always safe to assume that a small sample of the whole pop-
ulation — as usually occurs in biological control — will be fully representative of the species. This caveat
applies to all approaches based on inferring species’ climatic limits.
6 Estimation of CLIMEX model parameter values requires information on the geographical distribution
in relation to each of the climatic limiting factors. As stated by Sutherst and Maywald 1985, this is best ob-
tained in heterogeneous environments. In some cases the current distribution does not offer an opportunity to
estimate the parameter value from field data. Such situ- ations occur when a species occurs only in a temperate
area such as Europe for example, where no part of the continent experiences a combination of high tempera-
tures and soil moisture that is usual in summer rainfall areas in the tropics. Similarly, coasts create abrupt bar-
riers to species’ distributions and prevent inference of limiting conditions in that direction. In these cases, the
limiting effect of the variable concerned remains un- defined and it is necessary to make this fact explicit to
the audience. In a similar vein, species’ distributions are dynamic but usually only average ranges are docu-
mented, so care is needed when interpreting such data.
7 CLIMEX, and all other climate-matching ap- proaches, currently rely on average meteorological
data to infer the conditions that limit species’ distri- butions. This carries the assumption that the variances
in different places are the same, which is not true. In particular, the phenomenon of intermittent, but very
severe freezes in southern parts of North America and Asia are known to limit the geographical distribution
of species that have low dispersal rates Krakauer,
310 R.W. Sutherst et al. Agriculture, Ecosystems and Environment 82 2000 303–319
1968; Sutherst et al., 1996c; Booth, 1999. Thus, there is a great need for providers of meteorological
data sets to include percentiles and extremes in the long-term average data sets currently available.
When designing management strategies to adapt to global change, it is often not feasible to use CLIMEX
because it does not describe species’ lifecycles, thus providing no mechanism with which to link the effects
of interventions. This task requires a process-based modelling approach, so we now describe one such
modelling paradigm that has been applied in global change research on pests, the DYMEX modelling
package.
2.2. DYMEX generic and modular modelling toolkit for biologists
The sustainability of available technologies and their robustness under global change need to be as-
sessed in order to estimate the likelihood that they will be available for use as management options
under global change Sutherst et al., 1998. Quanti- tative estimates of effectiveness and profitability of
those technologies are best obtained using mecha- nistic models. Process-based or mechanistic models
are powerful tools for analysing pest, disease and weed problems. They enable risk assessments and the
design of integrated adaptation strategies to be under- taken at scales ranging from fields to regions. They
also provide the capability to link host and pest com- ponents, thereby enabling the integration of impacts
on crops and on industries. The global community needs common process-based modelling tools and
languages to facilitate collaboration. To be effective, such tools need to be transparent, easy to use and
preferably self-documenting.
Models are uniquely suited to a number of tasks such as the following:
1. Clarifying mechanisms involved in global change responses, including interactions between compo-
nents of the environment e.g. crops and pests or between different drivers of global change e.g.
temperature, moisture and CO
2
. 2. Estimating the sensitivity of a production system
to perturbation in situations where an inferential, climate-based approach i.e. CLIMEX is consid-
ered to be inadequate. 3. Designing and quantifying management options
in terms of physical responses of the species or production system to interventions, and economic
costs and benefits. 4. Providing a means of quantitative risk assessment
that can be applied in different geographical regions and over different time-scales.
A generic modelling package for use in global change research ideally will include at least the fol-
lowing features if it is to contribute in the environment in which global change research has to be done.
1. A structure that provides the capability to assem-
ble modules of objects and processes into different combinations to create descriptions of the pop-
ulation dynamics of different taxa. Genericness and modularity enable the creation of a library
of reusable modules that describe physical and biological processes or ‘attributes’, which are
commonly needed to estimate impacts and man- agement options Reynolds and Acock, 1997. For
example, we need a library of modules describing global change drivers and their interactions, to-
gether with derived variables such as soil moisture.
2. Descriptions of biological processes and attributes, such as development, survival, fecundity and
longevity, which can be associated with different lifecycle stages of plants or animals.
3. A library of commonly used functions, which save time and duplication of effort.
4. A geographical modelling platform to enable mod- els to be run at multiple locations and the results
to be mapped in order to expedite the develop- ment and validation of models. Our experience
has demonstrated that there are substantial oppor- tunities to accelerate progress in the development
of process-based population models by running the models with meteorological data from a wide
geographical area and mapping the results. Results can then be compared with field data on the distri-
bution, seasonal phenology and abundance of the species. This process exploits the expanded range
of spatial and temporal variation that is available in different environments compared to the usual
practice of using time-series data from a single site as a validation data set Sutherst, 1998. This
approach is a core feature of the CLIMEX model and is readily applicable to process-based models
that are driven by meteorological data. The models
R.W. Sutherst et al. Agriculture, Ecosystems and Environment 82 2000 303–319 311
can then be used on a geographical scale that is relevant to policy formulation. Luo et al. 1995
used a model of rice blast within a geographical information system GIS to investigate regional
risks under climate change. Basher et al. 1998 illustrate the use of a model of cattle tick popula-
tions and yield losses linked with a GIS to estimate impacts of global change in Australia.
DYMEX is a generic and modular modelling pack- age Maywald et al., 1997, 1999 that addresses the
need to build population models rapidly without the need for a computer programmer. Modules are linked
to provide descriptions of the target system, such as crop, pest and natural enemy, in order to facilitate
the analysis of the system’s behaviour under differ- ent scenarios. DYMEX provides a user-friendly tool
for biologists to build and use mechanistic models of species’ populations and management options. It
is an MS-Windows program, which allows modules to be assembled interactively using icons and dialogs
and it has in-built data formatting, graphics and table generating facilities Maywald et al., 1997, 1999. It
was made possible by the advent of object-oriented programming languages. Models consist of a group
of linked ‘modules’ which are either built using the ‘Builder’ or introduced into the model from a library
of modules that is provided with DYMEX. The com- pleted model is run using a ‘Simulator’, which cur-
rently allows the user to conduct sensitivity analyses and optimisations of timing or number of interven-
tions for example, with associated costings. DYMEX currently lacks the desirable CLIMEX geographical
platform but that is under development at the time of publication.
An illustration of a simplified, but multi-component DYMEX model of a rice-brown plant hopper —
rice-field rat system is shown in Fig. 5. It was pro- duced during a global change training workshop in
Indonesia Sutherst et al., 1999. GMD files are gen- erated by the DYMEX ‘Builder’ and read by the
‘Simulator’. An extract of a DYMEX ‘generic model description’ GMD file illustrating the hidden syn-
tax that describes a function in the rice-field rat mortality module is shown in Table 1. The ‘Builder’
also creates a self-documenting model description in text format, as shown for the same module in Table 2.
A view of the DYMEX simulator dialogue box used to adjust parameter values is shown in Fig. 5.
Table 1 An extract from the RatHopper ‘GMD’ file describing a mortality
function in the rice-field rat’s lifecycle, illustrating the hidden DYMEX syntax used internally to describe models
“Begin Process” “Number” . . .
. . . . . .
“Begin Action”
“Begin Comment” A constant background mortality rate increases in a
quadratic manner above a threshold stress level “End Comment”
“Variable” “Stress” “Function Type” “Quadratic above Threshold”
“Begin Parameter” “Name” “Background Mortality” “”
“Value” “0.0” “0.01” “0.001” “Begin Comment”
Daily “background” mortality rate in mature rats from predators, disease, etc., in the absence of stress due to
food shortage “End Comment”
“End Parameter” “Begin Parameter”
“Name” “Stress Mortality: Threshold” “Value” “0.0” “0.4” “0.25”
“Begin Comment” The value of stress at which mortality of the mature
rats starts to increase above the background level “End Comment”
“End Parameter” “Begin Parameter”
“Name” “Stress Mortality: Multiplier” “Value” “0.1” “0.3” “0.2”
“Begin Comment” The coefficient that determines how rapidly mortality
increases with increasing stress above the stress threshold
“End Comment” “End Parameter”
“End Action” . . .
. . . . . .
“ End Process
”
3. A case study