Get documents about "Computer Network Performance Modelling Types Approaches - DOC
W
Description
Computer Network Performance Modelling Types Approaches document sample
Document Sample
Bioclimatic Discussion Report Version updated 24/06/2005
BIOCLIMATIC MODELS - Discussion
Participants: Miguel Araujo, Michel Bakkenes, Pam Berry, Isabelle Chuine, Matt
Fitzpatrick, Pru Foster (Rapporteur), Theresa Mau-Crimmins, Camille Parmesan, Sarah
Shafer, Bob Sutherst (Chair), Martin Sykes, Wilfried Thuiller
The motivation for the Bioclimatic modelling group is the need to restore the confidence of
international policy-makers and other users of bioclimatic model simulations in the results of
bioclimatic analyses of risks to species posed by global change. Conservation agencies
dealing with rare species involving minimal amounts of observational data also need answers
to their questions on the risks to such species from global change.
Terminology – The description ‘Distribution Models’ was preferred by the group to
‘Bioclimatic Models.’ The models include both statistical and process-based tools, which can
simulate plant functional types, biomes, species and genotypes as desired with the necessary
caveats.
The main challenges for the group were expressed as (i) incorporating temporal scale in
model simulations, and (ii) moving to more process-based modelling approaches. There was
interest in ‘Hybrid’ approaches involving statistical models informed by process-based
approaches.
Topics of interest in priority order
1. Model Comparisons
Current or recently active groups or networks identified:
NCEAS how to extract niche information from geographical distributions, comparing
different modelling approaches, data quality (Peterson (also GARP), Morris), but no
focus on global change.
Conservation International (CI) – Montpellier statistical model comparison using
South Africa taxa.
1
Bioclimatic Discussion Report Version updated 24/06/2005
POPNET (How to incorporate population biology data into bioclimatic models)
(York, UK) popnet.ac.uk (contact - ), Report on January 2005 meeting is in
preparation.
Model Validation
There is a need to clarify the definition of what constitutes good model performance –
goodness of fit or predictive ability involving target taxa in defined regions. Different types
of models may be appropriate in different situations.
There are opportunities to use resources from the following communities to facilitate model
validation using independent data:
a. Invasive species distributions in different continents as test cases (there was
strong support to link with GISP), including parameter estimation tests using both
native and introduced ranges. Model prediction errors provide the opportunity for
new insights into factors limiting species distributions and detection of non-
equilibrium distributions.
b. Reciprocal transplants – lots of insect examples
c. Historical time-space datasets
d. Consensus forecasts (probability density functions) based on techniques used in
other fields, including GCM simulations.
e. Selection of models based on objective selection criteria including the conditions
under which such models outperform (Biosecurity, WTO. IPPC demand both
transparency and reliability with proven performance)
Proposed Collaboration Mechanisms
Network
Activity 1 – Focus on model development: pattern versus process, validation
processes, spatial scales; data quality; exploration of other classes of models
(e.g., DGVMs) and their potential; identify important processes in available
models.
2
Bioclimatic Discussion Report Version updated 24/06/2005
Activity 2 – Focus on benchmarking that incorporates / collates information from all
groups; and provides user guidance. The aim is to provide access for a variety
of users, including policy makers and natural resource managers, to metrics
and processes for assessment of model performance, demonstrations, model
limitations and scope. For a benchmarking exercise to work, thoroughly pre-
tested instructions are necessary for both developing model simulations and
data. There may be opportunities to use past and existing model comparison
projects as examples, such as the Paleoclimate Modelling Intercomparison
Project, Phase II (PMIP2; http://www-lsce.cea.fr/pmip2), and the GCTE crop
model comparisons
(http://mwnta.nmw.ac.uk/gctefocus3/activities/a31info.htm)
There is perceived to be a need for a resource (paper/report) that targets users,
including policy makers and conservation and natural resource organizations,
to inform them of limits to the uses of bioclimatic models, the scope for model
improvements, and guidelines on how, when and where to apply them. This
reports would go beyond the reports in the literature (e.g., Pearson & Dawson,
2003, Global Ecology & Biogeography, 12:361-371).
How do we apply bioclimatic models to biodiversity questions involving many
species?
a. Can we achieve reliable automation of model parameter estimation
with either statistical or process-based models that have been shown to
give reliable predictive ability? Early results suggest that it is possible.
b. Alternative approaches that deserve consideration are to model
selective, representative species or functional groups with similar
geographical ranges and extrapolate the results to the other species.
Resources required include (i) ongoing Workshops and (ii) a central Platform (data
sets, sharing agreements, rules of access, criteria and processes for objective model
validation using independent and ‘blind’ datasets, a computer portal or other collation
vehicle) for model comparison and communication to policy makers, model users, etc.
3
Bioclimatic Discussion Report Version updated 24/06/2005
2. What controls species ranges?
a. A ‘commonsense’ approach to estimating species parameters based on basic ecological
principles is needed to guide model selection. Does the projected model make sense?
b. Clarify whether the target species is limited by climatic or non-climatic factors. A list
of items to check before assuming that climate is limiting (e.g. Sutherst 2003 Journal of
Biogeography, 30:805–816) could be developed further. This will require the
development or refinement of a number of different detection techniques including tests
for internal consistency in model parameterisation processes, identifying abrupt edges to
ranges, etc.
1. Host availability may limit a species and so the climatic limit of the host needs to
be included in an analysis of a predator.
2. Patterns of replacement in space, coincidence of species range changes with
climatic change, checkerboard/mosaic patterns.
c. Link species to climate through specific processes, e.g. physiological tolerances –
survival, thermoregulation.
d. Species distribution models are often used to make projections of species range shift
based on process-based understandings of how species respond to climate change (e.g.,
mechanistic models) or on measured relationships between observed species’ ranges and
climate (e.g., statistical models). However, population responses to changing climate are
likely to vary spatially due to local adaptation. Spatial variation in genetic diversity might
also result in varying levels of adaptive capacity to climate change.
e. Define populations densities and demographic processes along environmental gradients
1. Define shape of species boundaries for representative target groups – can we
detect consistent patterns linked to taxonomic or functional groups and to specific
a regions or biome types?
2. Shape and steepness of gradients at borders (Fortin et al. OIKOS 108: 70/17,
2005)
Proposed Collaboration Mechanisms
A computer portal or other collated database of species attributes, distributions,
physiological data, life history traits and functional traits for understanding species
ranges would be extremely useful. Examples include Grime et al.’s (1989) book on
comparative ecology, Natural Biodiversity Network (nbn.org.uk); Prentice et al.
4
Bioclimatic Discussion Report Version updated 24/06/2005
tree dataset in preparation, Thompson et al., 1999, Atlas of Relations between
Climatic Parameters and Distributions of Important Trees and Shrubs in North
American (http://pubs.usgs.gov/pp/p1650-a/).
o Such a portal could usefully include species lists with citations (e.g.
CABI), inaccessible literature sources, and grey literature.
Ongoing workshops to design and implement access to distribution data or
incorporate data into a mirror DIVERSITAS site.
A network of modellers and experimentalists interested in participating in a large-
scale experiment to improve our knowledge of species responses to climate along
range edges (use ‘seed’ money from DIVERSITAS to organise this group)
Write a project proposal for submission under EC FP7 (7-year projects, more
than 50 partners, probably more than 20 million euros) and maybe US NSF.
Species distribution models would provide the hypotheses to be tested (projected
range shifts and limits of tolerance) with field observations (animals?) and
experiments (plants?). Observations would include genetic and behavioural data
collected in selected parts of species ranges; experiments would seek to measure
physiological responses of species.
3. Model Extensions
a. Can process-based models be used to inform and so improve the performance of statistical
models?
There are a variety of physical processes that are important in determining the distribution of
species (e.g., species interactions, disturbance regimes) that are not simulated by statistical
distribution models. It may be possible, however, to use information about these physical
processes as simulated by physically-based process models (e.g., LPJ, MC1) to inform
statistical model simulations of species distributions. A number of examples were discussed,
including:
Example: Incorporating vegetation response to atmospheric CO2.
Statistical models have been used to simulate future distributions of plant taxa under
future climate conditions, but have not been able to incorporate the potential
distribution changes resulting from vegetation response to changes in atmospheric
CO2 concentrations. A number of mechanistic models, however, do simulate the
5
Bioclimatic Discussion Report Version updated 24/06/2005
effects of changes in atmospheric CO2 on vegetation. A process model could be used
to inform a statistical model by first simulating future plant taxa distributions using a
statistical model, simulating the same taxa (or plant functional types, etc.) using a
process model that includes the CO2 effect, and then using the process model
simulations to constrain or expand as appropriate the statistical model distributions to
include the effect of CO2 in the final distribution.
Example: Effects of fire regimes.
A statistical model may simulate species to occur where the fire regime (i.e., the
frequency and intensity of fires) would exclude the species. A number of process
models simulate the effects of fire on species, particularly on plant taxa and plant
functional types. Simulations done using physically-based process model could
identify regions where statistical model simulations would need to be constrained for
particular species to account for the effects of fire. For example, if a statistical model
simulated a tree species to occur in a particular location where a process model that
simulated the effects of fire on vegetation indicated no trees could occur, then it might
be appropriate to constrain the statistical model simulation using this information.
Example: Species interactions limiting distributions.
Species interactions may affect the distributions of species and define certain range
limits. Some process models simulate species interactions (e.g., gap models,
Bugmann, 2001, Climatic Change 51:259-305) and simulations from these models
could be used to determine where a species simulated by a statistical model would not
occur as a result of species interactions. The distributions simulated by statistical
models could then be constrained accordingly.
The above are only examples of ways that process model simulations could inform statistical
model simulations. More thought would need to be given to determining what specific
processes, models, and spatial and temporal scales would be appropriate and how best to use
the information from process models to constrain distributions simulated by statistical models.
There are also many other processes that are calculated by process models that potentially
could be used to constrain or expand statistical model simulations, such as processes
associated with photosynthesis (e.g., net primary productivity) and soil hydrology.
6
Bioclimatic Discussion Report Version updated 24/06/2005
b. Can statistical models be used to inform and so improve the performance of process
models?
Process models often use bioclimatic variables that are known to (or assumed to) limit species
distributions to determine the initial presence or absence of a species (taxa, functional types,
etc.). Statistical models may be able to be used to relatively quickly simulate species
distributions using a large number of bioclimatic variables to identify those variables that
define species distributions but that are not currently used by process models. If these
variables were determined to represent (or be proxies for) physiological or other limits on the
species in question then process models could be adjusted to include these variables.
c. Can process-based models be extended by including extra physical and biological
variables?
Process models are becoming increasingly complex but there are still a number of processes
important in determining species distributions that are either not currently simulated by
process models or for which simulations could be improved by additional information,
including processes related to:
nitrogen deposition/availability effects
predation, herbivory and synergy involving taxa in different trophic levels
species interactions including competition and synergy within trophic levels
species assemblage rules
species physiological limits for use as parameters in models
phenology
Climate
Modern global climate data are available from the Climatic Research Unit (Univ. of
East Anglia) and the Tyndall Centre for Climate Change Research on 0.5o and 10-
minute grids including annual time series with monthly averages (1901-2000) and
long-term averages (1961-1990 30-year mean).
(http://www.cru.uea.ac.uk/cru/data/hrg.htm)
The Intergovernmental Panel on Climate Change Assessment Report 4 (IPCC AR4)
future climate scenario data are available via the Program for Climate Model
7
Bioclimatic Discussion Report Version updated 24/06/2005
Diagnosis and Intercomparison (PCMDI) website (http://www-
pcmdi.llnl.gov/ipcc/about_ipcc.php).
Time series of daily data for the past 10 years are available at NCDC
(http://lwf.ncdc.noaa.gov/oa/ncdc.html) for use in calculating measures of climate
variability and extremes.
Proprietary software (http://grads.iges.org/grads/grads.html) is available to read and
manipulate the data CRU.
It will facilitate model comparisons if agreement can be reached on using the same
predictive climatic variables.
CO2
Plant responses to atmospheric CO2 involve interactions with moisture and nutrients, with
significant observed variation in the extent of down-regulation in photosynthesis with
duration of plant exposure. This variability in plant response makes simplification of the
description of responses that are compatible with distribution modelling problematical and in
need of interdisciplinary consideration. Any analysis of species competition will need to take
account of concurrent habitat alternation or biome shifts in response to CO2 enrichment.
Paleo-evidence suggests that the CO2 effect is as large as the climatic effects for some plants.
Herbivores are also sensitive to changes in the nutritional quality of their host plants and
forage resulting from changes in CO2.
Incorporation of the effects of nitrogen deposition on species will also rely on the definition
of suitable species response functions for process-based models. Incremental steps to
incorporation of CO2 into models are recommended, including post-processing of model
outputs.
Nitrogen
Incorporation of the effects of nitrogen deposition on species will also rely on the definition
of suitable species response functions for process-based models. Similar caveats apply to
nitrogen as those listed above for CO2. Global datasets of nitrogen availability and deposition
as well as future scenarios of change in the nitrogen budget are required.
8
Bioclimatic Discussion Report Version updated 24/06/2005
Atmospheric Pollution and Reduced Solar Radiation
Incorporation of the effects of regionally reduced solar radiation, resulting from atmospheric
pollution, on species will also rely on the definition of suitable species response functions for
process-based models. Similar caveats apply to those listed above for CO2. Global datasets
and scenarios are required.
Disturbance
Disturbance was the third dimension of Grimes’ C_S_R triangular ordination model (the
others being Stress and Productivity). Incorporation of the effects of regional scale
disturbance from fire or changes in land use or cover, on species will also rely on the
definition of species response functions for each disturbance process, in process-based
models. Similar caveats apply to those above for CO2. Global datasets and scenarios are
available for presence / absence effects.
Species Interactions
a. Need to take Within-trophic and Multi-trophic scales into account.
b. Plant competition is feasible at the functional group level (tree/shrub/grass) in
terms of estimating habitat change.
c. Gap/Patch models are particularly appropriate for this.
d. Simulation of invertebrate competition has been shown to be feasible in process-
based models if the functions can be parameterised.
Migration / Dispersal
Migration / Dispersal was acknowledged but not discuss
9
