Mapping the future
Converting storylines to maps
Nasser Olwero
Stanford, March 15, 2013
Objectives
• Intro to scenarios and tool
• Other tools out there? Tier?
• Can this work for you?
Why are
things
where
they are?
Scenarios
Good
Bad
Action
Scenarios
• They are not necessarily predictions
• Scenarios
– Narratives of alternative environments in which
today’s decisions may be played out (Adam Gordon –
futuresavvy.net)
– alternative pathways into the future
– Plausible future
– “What if” analysis
Land cover change modeling
• Quantity of change
– How much is changing?
– Stakeholder storylines
• Transition probabilities
– Statistical methods – regression, Bayesian probabilities….
– Artificial intelligence – artificial neural networks
– Stakeholder estimates
• Factors – spatial and environmental characteristics
• Decision rules
– Drivers, actors influence
– Eg development can occur only in areas less than 35% slope
• Constraints
• Transition procedures
Scenario Tools
•
•
•
•
•
•
•
•
•
Metronamica
PoleStar
IMAGE
WaterGAP
AIM
T21
GLOBIOM
Mirage
CLUE-S
• GTAP/MAGNET
• LandSHIFT
• International Futures
Model
• IDRISI Land Change
Modeler
• Marxan
• Dinamica
• GEOMOD
Scenario Tools
Demand
Allocation
Quantity of change
Where does change occur?
Land Change Modeler
• Uses two land cover maps
• Three steps
– Change analysis
• Compares two maps to identify where changes occur
– Transition potential
• Uses neural networks, logistic regression or machine
learning
– Change prediction
• Uses historical rates of change and transition potential
to create predicted changes
• Uses constraints and incentives
CLUE-S
• Conversion of LandUse and its Effects
• Has a demand module [policy team] and allocation
module [GIS team]
– Demand module provides annual area of each cover type
demanded
• Uses logistic regression to explain relationship between
spatial configuration and factors
• Creates probability map for each landcover for each
year
• Decision rules are created which specify what
transitions can happen e.g. from agriculture to urban
• Allocate pixels based on probability and the decision
rules and compare to the demand.
GEOMOD
• Simulate land cover change based on landscape
characteristics
• GEOMOD1 uses intuitive perspective from observation
• GEOMOD2 based on statistical analysis of landscape
structure
• Assumes land use change rates are known
• Calculates coefficient of relative suitability
• Uses a time loop with a time step of 1 year
• Simulation starts from the most suitable and grows using
adjacency and has to wait before it becomes seed.
• Randomly jumps to a new location every 10th time step
• Quantity calculated by interpolation
• User supplies number of cells at end of run
Dinamica
• Cellular Automata model
• Calculates spatial transition probabilities
based on static (eg elevation) and dynamic (eg
population) variables using logistic regression
• Calculates transition rates and saturation
value to determine amount of change (rate *
number of cells)
• To perform transition, cells with high
probability are ranked and selected randomly
• Uses expander and patcher functions
What are the storylines?
Landcover Types
Change
Rules
increase
along roads, in poor soils, on hilltops,
difficult to cultivate areas, in and
around cfrs & lfrs,
increase
along roads, in poor soils, on hilltops,
difficult to cultivate areas, in and
around cfrs & lfrs,
Tropical high forest increase
in and around cfrs and lfrs, not in nps
Degraded forest
decrease
in and around cfrs and lfrs, not in nps
Woodland
increase
outside pas
Broadleaved tree
plantation
Coniferous
plantation
How do you move from story to map?
Landcover Types Change
Rules
Broadleaved tree
plantation
increase
along roads, in poor soils, on
hilltops, difficult to cultivate areas,
in and around cfrs & lfrs,
Coniferous
plantation
increase
along roads, in poor soils, on
hilltops, difficult to cultivate areas,
in and around cfrs & lfrs,
Tropical high
forest
increase
in and around cfrs and lfrs, not in
nps
Degraded forest
decrease
in and around cfrs and lfrs, not in
nps
Woodland
increase
outside pas
?
Land cover transition
(Swetnam et al 2010)
Land cover transition
built
Original
Forest
1
Grassland
1
7
3
Scenario
Forest
Agriculture
Grassland
2
Agriculture
Built
Objectives: Cover as Objective
• Land cover change is driven by an objective
and the objective determines the decision rule
• Objective can be complimentary or conflicting
• Single objective modeling is simpler than
multi-objective
• Examples of objectives are: agricultural
development, conservation, urbanization etc
• In this model, the cover type is used as a proxy
for the objective. Cover types that increase
represent some objectives
Factors
• Factors are criteria that increases or reduces the
suitability of a parcel for a specific objective
• Proximity attributes that determine where
change occurs
–
–
–
–
Roads [transportation, ]
Rivers [transportation, proximity to water]
Slope [access, ag suitability]
Cities [market, population pressure]
• Rules combined added to attributes creates a
suitability layer
Multi Criteria Evaluation
Slope
Dist to Roads
Elevation
< 35%
x1 (std 0-1)
< 5km
> 1000m
X2 (std 0-1)
X3 (std 0-1)
w2
w3
Weights
Factors
Decision Rules
Suitability
Threshold
w1
Assigned by AHP
=
(x1w1
+
x2w2
+
x2w2) * Constraints
pixels with suitability values above
threshold are converted
User sets goal of conversion quantities
Addition of likelihood matrix and Overrides
Agriculture
Urban
Proximity
Proximity
distance
0
1
7
2
-30%
0
0
0
Grassland
0
0
3
1
-40%
0
0
0
Agriculture
0
0
0
0
50
1
10
2
Urban
0
0
0
0
10
1
5
1
Location
Quantity
Priority
Grassland
/Forest
Change
Forest
LOSS
GAIN
Transition Matrix
Constraints
• Limit the alternatives
• Create exceptions [‘no go’]
• Constraints can be simple (specific areas cannot
be affected) or more complex (eg minimum area
required for large scale agriculture)
• Constraints have varied degrees of effect
– 0 – no change
– 1 – no effect on change
• Multiple constraint layers combined by taking the
minimum value
• An example is protected areas
Scenario tool schematic
Transition
matrix
Current
landcover
Patch
size
Suitability
Factors
Constraints
Rules
Override
Allocation
Proximity
Scenario
Scenario tool schematic
Quantity
What goes first?
Transition
rate
Priority
Allocation
Override
Scenario
Allocation
While n > 0
Go to next
lower suitability
(n = n-1)
Suitable
cells > 0?
Suitability
Val (n) = 0 – 100
Start from n = 100
No
Cells grouped
To minimum
Patch size and selected
randomly
Yes
Suitable
cells <
goal?
No
Group and
select
Yes
Convert cells
to Ag
No
In order of
priority
No
Goal
met?
Yes
Go to
next cover
Preparing Suitability layers
Rules
Likelihood(+)
Factors(+)
weighted
Proximity (+)
Constraints (x)
Aggregate transition probability/suitability
Tool interface
Important
• Currently tool cannot handle large datasets,
limit to under 1M pixels [InVEST 3!]
• Minimize number of transitions, select most
important transitions to use
• Factors must have complete coverage, select
factors very carefully
• Iterative process
Limitations/Issues
• Accuracy depends on stakeholders
• Model grows cover, doesn’t shrink
• Model assumes a cover type either increases
or decreases but not both
• Assumes a single step transition
• Stakeholder given values are best for near
future
Virungas Example
BAU
Market
Current
Green
Virungas Example
What next?
•
•
•
•
•
•
Time step
Remote stakeholder
Validation
Alternative sources of probability
Selection of allocation algorithm
Demonstration
Download
• https://wwfcsp.org/natcap/scenario/ScenarioDistrib
ution.zip