Mapping and modeling species distributions
Department of Botany and Zoology, Masaryk University
Bi9661 Selected issues in Ecology, Autumn 2013
Borja Jiménez-Alfaro, PhD
Introduction:
ASSESSING
SPECIES DISTRIBUTIONS
“attempts to explain why species and higher taxa are
distributed as they are, and why the diversity and taxonomic
composition of the biota vary from one region to another”
Biogeography
ASSESSING SPECIES DISTRIBUTIONS
Philip Sclater
(1829-1913)
is the science that attempts to
document and understand spatial
patterns of biodiversity.
Reconstruct the origins, dispersal, and extinctions of taxa
Primarily focused on evolution, dispersal and vicariance
Historical biogeography
Primarily focused on present distributions, species responses to
biotic environment and interactions with other organisms
Ecological biogeography
Combines historical and ecological biogeography, investigating the
relationships between communities (abundance, distribution, and
diversity of species) and abiotic conditions
Paleoecology
ASSESSING SPECIES DISTRIBUTIONS
Work on the protection and restoration of natural environments
Conservation biogeography
Special issue
Diversity and distributions
13 (3) 2006
ASSESSING SPECIES DISTRIBUTIONS
Special feature:
The application of
predictive modeling of
species distribution to
biodiversity conservation
Computational power (computers)
Geographic Information Systems
Geostatistics
Recent tools in biogeography
MAPPING MODELING
ASSESSING SPECIES DISTRIBUTIONS
2002
ASSESSING SPECIES DISTRIBUTIONS
2010
ASSESSING SPECIES DISTRIBUTIONS
2011
ASSESSING SPECIES DISTRIBUTIONS
Predicting species occurrences and estimating ranges
Modeling ecological spatial responses
Reconstructing past distributions
Biogeography of genetic and physiological data
Assessing responses to climate changes
Establishing diversity patterns (endemicity, richness)
and much more…
Species distribution models (ecological niche models)
are used for:
ASSESSING SPECIES DISTRIBUTIONS
ASSESSING SPECIES DISTRIBUTIONS
ASSESSING SPECIES DISTRIBUTIONS
ASSESSING SPECIES DISTRIBUTIONS
Three main steps:
1. compile spatial data associated with the target element
and environmental data for the area of interest
2. build a statistical model based on the association of the
element to environmental variables at sites of known
occurrence
3. Map the model via GIS across the area of interest
ASSESSING SPECIES DISTRIBUTIONS
Environmental
Data
Spatial
Model
Range Prediction
Specialist
Review
Species-
Environment
Relationship
Localities
Environmental DataSpecialist’s
Knowledge
D
E
D
U
C
T
I
V
E
I
N
D
U
C
T
I
V
E
Distribution Modeling
www.naturserve.org
ASSESSING SPECIES DISTRIBUTIONS
Franklin
2009
ASSESSING SPECIES DISTRIBUTIONS
Model building process (from Guisan and Zimmerman 2000)
ASSESSING SPECIES DISTRIBUTIONS
Distribution modeling (per se) is EASY
Just some technical skills are required
Anyone can compute it with user-friendly software
Applying distribution modeling is more TRICKY
You need a good purpose to do it
(research question, conservation goal)
You must know how to do it properly
ASSESSING SPECIES DISTRIBUTIONS
About this course:
Part 1 – Mapping
Dealing with occurrence data
Environmental variables
Spatial terms and PRACTICE with GIS
Part 2 – Modeling
Background theory (niche concept)
Modeling methods
Maximum Entropy and PRACTICE with MaxEnt
Part 3 – Mapping and Modeling
Model implementation and evaluation
Applications and future challenges
Using your OWN DATA and GROUP PRACTICE
Part 1:
MAPPING
OCCURRENCE DATA
A record for one species/organism/community in one locality
Presences (and absences) are the MAIN dependent variable for
Species Distribution Modeling (but there are more)
What is occurrence data?
OCCURRENCE DATA
Guisan & Zimmerman (2000)
Two main options:
Where to obtain occurrence data?
OCCURRENCE DATA
2. Using existing data (e.g. biodiversity databases)
Pro: huge amount of data around the world
Con: uncertainties on sampling, accuracy, etc.
1. Desing your own field sampling
Pro: you have control on your data
Con: many times you have not time or money
Probabilistic design is required to quantify the species responses
along gradients, in order to consider the edges of
environmental distribution
1. Sampling design
OCCURRENCE DATA
Geographic distribution of a species
OCCURRENCE DATA
Distribution in the environmental space
OCCURRENCE DATA
Distribution in the environmental space is different!
OCCURRENCE DATA
There is no universal design for all questions, but…
Even better can be mixed designs, e.g random stratified sampling
Simple random design is used for relatively homogeneous spaces
(when the probabilities of occurrences are equal) but it is not
a good option if you have to sample organisms which are rare
or disjuntly distributed
Regular, systematic, clustered or stratified designs are prefered to
sample occurrence data if the organism is clearly influenced by
geographial, environmental or topotraphical gradients
OCCURRENCE DATA
You “should” avoid bias survey sampling
-> Geographic bias: along roads, near the cities,…
-> Taxonomic bias: wrong identification of species
OCCURRENCE DATA
E.g.: identification of bias in biological collections of Lupinus hspanicus
(Parra-Quijano et al.): diferent geographic cover
Genetic bank
Both
Herbarium / literature
You “should” avoid purposive sampling
-> non-probabilistic, based on aprioristic knowledge
-> usually produces undersampling of the study subject
OCCURRENCE DATA
E.g.: Comparison of sampling survey desings for predicting lichen species in USA
(Edwards et al. 2006)
A visual example of designed versus purposive sampling
(vegetation plots in Picos de Europa, Spain)
Designed (systematic)
(N = 80)
Purposive (biased)
(N = 100)
OCCURRENCE DATA
More important that the number of observations is the degree to which the
range of the environmental space occupied by the species are covered in the
sample (= COMPLETENESS) and the frequency of events (records of species
presences) from the sample (= PREVALENCE)
OCCURRENCE DATA
How many samples?
Or, a better question,
What is the mínimum sample size for my study?
For SDMs, there are some rules:
→ A mínimum of 50 observations can be fine
→ 20-40 times as many observations as predictors
→ For rare species and some algorithms, 20 occurrences can be enough…!
OCCURRENCE DATA
Wisz et al. 2008. Effect of sample
size on the performance of species
distribution models.
Diversity and Distributions 14: 763
Very few samples can be valid for rare organisms
… but it depends on the method
OCCURRENCE DATA
In summary, the quality of our data for modeling
distributons will depend on many factors:
-> EXTENT of the study area and ACCURACY of occurences
-> The ECOLOGY of the species
-> How we sample the ENVIRONMENTAL SPACE
-> How many PRESENCES and ABSENCES are sampled
-> The PREDICTORS and the modeling METHOD
OCCURRENCE DATA
2. Using existing data (e.g. biodiversity databases)
OCCURRENCE DATA
-> SDMs are mainly used to map unkwon species
distributions
-> Species mapping has however a long history using known
distributions from many different sources
Main types of sources:
- Grid-based atlases (compilation of information)
- Natural history collections (museums, botanic gardens)
- Surveys (conservation, vegetation or faunistic surveys)
Grid-based Atlases
Pro: Cover large territories and represent distribution ranges well
Con: Coarse grain (10 km, 50 km) and small spatial acuracy
ATLAS FLORAE EUROPAEAE
50km x 50 km
2559 species
(20% of European flora)
OCCURRENCE DATA
Natural history collections
Pro: Large amount of data for all the world
Con: Low spatial resolution and high uncertainty
OCCURRENCE DATA
Biodiversity Surveys
Pro: Spatial accuracy is heterogeneous, although can be good
Con: Generally biased or purposive samping
Czech national
Phytosociological
Database
OCCURRENCE DATA
OCCURRENCE DATA
There is some
sampling design
behind this?
PROBABLY NOT
From
Franklin 2009
Is the data valid?
PROBABLY YES
Problems associated with biodiversity databases
OCCURRENCE DATA
1. Low spatial accuracy: location and coordinates
(if existing) are generally imprecise
2. Unknown sampling design: generally biased or
purposive, but in general not reported
How this affects our data:
- Incomplete distributions (bias)
- Undersampling
- Pseudo-replication
- Spatial autocorrelation of samples
- Low spatial accuracy of the analyses
Next week we will
go back to the
spatial issues
How to solve these limitations?
OCCURRENCE DATA
Georeferenciating: it takes time but it allow us to measure
spatial uncertainty
Resampling: to have some control of the data (e.g.
analyzing subsets separately)
Adaptative sampling: resampling after a first assessment
Evaluating bias: using spatial information
Measuring spatial autocorrelation
Georeferencing
OCCURRENCE DATA
The main challenge of biological collections is the assignment of geographic
coordinates to millions of historical records (Baker & al., 1998)
Spatial autocorrelation
OCCURRENCE DATA
Oversampling of aras produces pseudo-replication and further
overfiting of the models.
What to do?
Sampling (or resampling) according
to spatial criteria
Assessing spatial autocorrelation
(e.g. Moran’s I) after modeling
b)
Part 1:
MAPPING
ENVIRONMENTAL VARIABLES
ENVIRONMENTAL VARIABLES
In the ecological space, what factors are important
for the distribution of species?
At the macro-distributional scale, ultimate controlling factors have to do with
energy requirements of species.
Energy requirements are, in turn, determined by physiology and morphology
http://ruig.grid.unep.ch/?p=95
World Potential
Evapotranspiration
ENVIRONMENTAL VARIABLES
And… what is the primary source of energy for
the Earth?
ENVIRONMENTAL VARIABLES
Solar energy brings
Light
(quantity and quality)
Heat
ENVIRONMENTAL VARIABLES
What factors affect solar radiation and temperature?
Topography:
- Elevation
- Slopes
- Exposure
Latitude
ENVIRONMENTAL VARIABLES
which is also affected by latitude and topography
But the factory of primary production (vegetation)
... also needs WATER
ENVIRONMENTAL VARIABLES
There are also functional manifestations of the
interplay of all these factors
evapotranspiration
productivity
ENVIRONMENTAL VARIABLES
In sum
NOTE: Energy and water income is dynamic in time. For some
questions regarding the eco-geographic distribution of species,
the time dimension is crucial
There are distal factors that determine directly or indirectly the
distribution of all species (at broad spatial scales):
-> Amount of light
-> Amount of heat
-> Amount of water
-> Topography
ENVIRONMENTAL VARIABLES
When modelling individual species, more proximal variables
become relevant:
-> Soil types
-> Evapotranspiration
-> Primary productivity
-> Light quality
-> Number of frost days
NOTE: A necessary field of research is needed to achieve a better
understanding of the inclusion of different types of variables in the
modelling process, as well as the effect of redundancy on model quality.
Implications
ENVIRONMENTAL VARIABLES
E.g.: interactions, parasitisms…….
But EACH organism has its own requirements
In statistical terms, there are two main variables:
QUANTITATIVE
elevation, temperature, precipitation, etc.
QUALITATIVE (CATEGORICAL)
Soil type, land cover, vegetation
->They will be used differently in the modeling process
Types of environmental variables
ENVIRONMENTAL VARIABLES
In practice, we should distinguish the most appropiate variables for
our study case, and especially the SCALE
Types of environmental variables
Broad scale studies:
More focused on DIRECT variables, mostly climatic:
Temperature, precipitacion, solar radiation, evapotranspiration
Local scale studies:
More focused on INDIRECT variables, mostly topogaphic:
Elavation, slope aspect, exposition, topograhical índices, etc.
ENVIRONMENTAL VARIABLES
Conceptual model of
relationships between
resources, direct and
indirect variables, and their
influence on plant
performance (from Guisan
& Zimmerman 2000)
ENVIRONMENTAL VARIABLES
ENVIRONMENTAL VARIABLES
WORLDCLIM
Averaged from long-term (30yr) series of Temp and Prec
BIOCLIM: “Bioclimatic variables are derived from the monthly temperature
and rainfall values in order to generate more biologically meaningful
variables. These are often used in ecological niche modeling (e.g., BIOCLIM,
GARP). The bioclimatic variables represent annual trends (e.g., mean annual
temperature, annual precipitation) seasonality (e.g., annual range in
temperature and precipitation) and extreme or limiting environmental factors
(e.g., temperature of the coldest and warmest month, and precipitation of the
wet and dry quarters). A quarter is a period of three months (1/4 of the year)”
ENVIRONMENTAL VARIABLES
WORLDCLIM (www.worldclim.org)
BIO1 = Annual Mean Temperature
BIO2 = Mean Diurnal Range (Mean of monthly (max temp - min temp))
BIO3 = Isothermality (P2/P7) (* 100)
BIO4 = T Seasonality (standard deviation *100)
BIO5 = Max T of Warmest Month
BIO6 = Min T of Coldest Month
BIO7 = T Annual Range (P5-P6)
BIO8 = Mean T of Wettest Quarter
BIO9 = Mean T of Driest Quarter
BIO10 = Mean T of Warmest Quarter
BIO11 = Mean T of Coldest Quarter
BIO12 = Annual Prec
BIO13 = Prec of Wettest Month
BIO14 = Prec of Driest Month
BIO15 = Prec Seasonality
BIO16 = Prec of Wettest Quarter
BIO17 = Prec of Driest Quarter
BIO18 = Prec of Warmest Quarter
BIO19 = Prec of Coldest Quarter
ENVIRONMENTAL VARIABLES
National data at higher resolution (e.g. Spain, 200m x 200m)
ENVIRONMENTAL VARIABLES
TOPOGRAHY
Potential solar radiation (from the Digital Elevation Model)
Indirect variable reflecting heat acumulation
ENVIRONMENTAL VARIABLES
TOPOGRAHY
Topographic Position Index (from the Digital Elevation Model)
Indirect variable reflecting moisture and wind exposure
ENVIRONMENTAL VARIABLES
MORE VARIABLES (AT DIFFERENT SCALES)
ENVIRONMENTAL VARIABLES
MORE VARIABLES (AT DIFFERENT SCALES)
ENVIRONMENTAL VARIABLES
MORE VARIABLES (AT DIFFERENT SCALES)
A map of land use in Europe. Yellow: cropland and arable, light green:
grassland and pasture, dark green: forest, light brown: tundra or bogs,
unshaded areas: other (including towns and cities).
ENVIRONMENTAL VARIABLES
Last inter-glacial (LIG; ~120,000 - 140,000 years BP)
Mid-Holocene (~6000 BP)
Paleoclimatic models
ENVIRONMENTAL VARIABLES
Last glacial maximum (LGM; ~21,000 years BP)
Paleoclimatic models
ENVIRONMENTAL VARIABLES
Climate future projections
ASSESSING SPECIES DISTRIBUTIONS
NEXT WEEK
Part 1 – Mapping
Dealing with occurrence data
Environmental variables
Spatial issues and PRACTICE with GIS (October 18)
Part 2 – Modeling
Background theory (niche concept)
Modeling methods
Maximum Entropy and PRACTICE with MaxEnt
Part 3 – Mapping and Modeling
Model implementation and evaluation
Applications and future challenges
Using your own data and GROUP PRACTICE