Creating Urbanization Scenarios with the FUTURES Model
Vaclav (Vashek) Petras, Anna Petrasova, Georgina M. Sanchez, Derek Van Berkel & Ross K. Meentemeyer
NCGIS 2019 Winston-Salem
Feb 27 - Mar 1, 2019
Motivation
Long-term planning as people are moving
to a from coast
to suburbs or city centers
…
Assessing impact of these changes before they happen
No Room for a Black Box
Can we understand the behavior of the model?
Can we make sure it is working as described?
There are numerous challenges for the urban growth and land change models,
one of them is that the algorithms are often black boxes.
Black box here means that the model provides little explanatory
insight into the influence of the independent variables
in the prediction process.
Also many published models do not provide their software implementation,
so all possible problems are hidden and the models algorithms
cannot be adjusted when applied to different study system.
FUTURES
FUT ure U rban-R egional
E nvironment S imulation (Meentemeyer et al., 2013)
urban growth model
patch-based
stochastic
accounts for location, quantity, and pattern of change
positive feedbacks (new development attracts more development)
allows spatial non-stationarity
FUTURES, an urban growth model
developed originally by Meentemeyer et al.
FUTURES is a stochastic model, where new development occurs in discrete patches,
it accounts for ...
FUTURES, A Simplified View
turning green cells into orange cells
-1: undeveloped, 0: initial development, 1: developed in the first year, …
Modeling framework
This is a basic schema of FUTURES, where the modeling framework is
based on 3 components: POTENTIAL submodel providing the information
where will urbanization likely happen, the DEMAND specifies how much
land will be developed and the third component PGA (meaning Patch Growing Algorithm)
is the actual engine of FUTURES, growing the patches of calibrated size and shape.
Potential Submodel
multilevel logistic regression for development suitability
accounts for variation among subregions (for example policies
in different counties)
inputs are uncorrelated predictors (distance to roads and development, slope, ...)
surface: potential, orange: developed areas, green: undeveloped areas
Demand Submodel
estimates the rate of per capita land consumption for
each subregion
extrapolates between historical changes in population and land conversion
inputs are historical landuse, population data, population projection
Patch Growing Algorithm (PGA)
stochastic algorithm
converts land in discrete patches
inputs are patch characteristics (distribution of patch sizes and compactness)
derived from historical data
FUTURES Prototype
Private/proprietary
Only the core of the model formalized in code
Poorly documented code with many hardcoded constants
User interface: configuration file and C code editing
The original paper went through the classic peer-review process and was published in a scientific journal.
FUTURES Prototype
Calibration data, tools and documentation distributed in a password-protected ZIP files by email.
Open Source FUTURES
To pay the technical debt and to go beyond experimental prototype we needed to make FUTURES:
more efficient and scalable
as easy to use as possible for a wider audience
open source, integrated into a larger modeling project, and maintainable in the long run
⇒ new FUTURES GRASS GIS add-on:
a set of modules called r.futures
Open science
Image credit:
Open Science Logo v2 ,
CC BY-SA 3.0 Greg Emmerich
The original implementation of FUTURES was a research prototype
which couldn't be effectively shared with the land change community
because it was difficult to run it and didn't scale very well.
So we decided to go beyond this prototype. We made the model
simpler to run so that all our colleagues and also wider audience
can use it in their research without extensive training.
With the goal to study large-scale urbanization in high detail
we made the model more efficient and parallelized.
We don't share just binaries but we made the model
fully open source and implemented it in GRASS GIS, a stable, powerful
geospatial platform to ensure the model is available and maintained
and can be used by the community even if the original
authors cannot support it for lack of funding for example.
Why GRASS GIS?
Advantages for model developers (and all tool/plugin developers) :
modular architecture: modules in C, C++, and Python
all needed GIS functions at hand
efficient I/O libraries (several further improvements in GRASS GIS since the decision was made)
ability to process large datasets
automatically generated CLI and GUI
infrastructure for online manual pages
daily compiled binaries from C/C++ for Windows(thanks to M. Landa, FCE CTU in Prague)
code in common add-on repository partially maintained by community and core developers
Why GRASS GIS?
Advantages for model users:
multiplatform
graphical user interface
scriptable (Bash, Python, R, …)
easy installation from GUI or command line: g.extension r.futures
more tools available for further analysis and visualization
Why GRASS GIS?
Advantages for model users:
spatio-temporal analysis and visualization
Example: Animation tool
r.futures
Information flow diagram for the set of modules implementing FUTURES
Additionally, r.futures.parallelpga can be used instead of r.futures.pga .
GUI
Graphical User Interface
CLI
Command Line Interface
r.futures.pga -s subregions=counties developed=urban_2011 \
output=final demand=demand.csv discount_factor=0.1 compactness_mean=0.1 \
predictors=road_dens_perc,forest_smooth_perc,dist_to_water_km,dist_to_protected_km \
devpot_params=potential.csv development_pressure=devpressure_0_5 \
n_dev_neighbourhood=30 gamma=0.5 patch_sizes=patches.txt num_neighbors=4 output=final
TUI
Tangible User Interface
Getting Started: Data
Import or link data into a GRASS GIS Spatial Database
Includes reprojection into the same SRS
Unify resolutions and extents (g.region, r.resamp.stats, … )
Landuse classes draped over topography (3D view in GRASS GIS)
Getting Started: Calibration
Potential submodel [calibrated using difference between two years in the past]
Predictors (distance to water, slope, travel time to city center, …)
Development pressure (Dynamically modified during the simulation)
Patch calibration [calibrated using difference between two years in the past]
Size and shape of patches of new development
Demand submodel [calibrated using all available past years]
Equation to relate new development and population growth (past and projected)
Distance to forest edge computed using r.grow.distance
Getting Started: Scenarios
stimulus is spatially variable increase potential for development (e.g. zoning)
constrain_weight is spatially variable limits to development (e.g. city park)
incentive_power influences infill and sprawl (e.g. government policy)
change inputs for predictors (e.g. new road) and population growth
left: infill, middle: status quo, right: sprawl
References
Meentemeyer, R. K., Tang, W., Dorning, M. A., Vogler, J. B.,
Cunniffe, N. J. and Shoemaker, D. A., 2013.
FUTURES: Multilevel
Simulations of Emerging UrbanRural Landscape Structure
Using a Stochastic Patch-Growing Algorithm . Annals of the Association
of American Geographers 103(4), pp. 785–807.
Dorning, M. A., Koch, J., Shoemaker, D. A. and Meentemeyer,
R. K., 2015.
Simulating urbanization scenarios reveals tradeoffs
between conservation planning strategies . Landscape and Urban
Planning 136, pp. 28–39.
Pickard, B. R., Van Berkel, D., Petrasova, A. and Meentemeyer,
R. K., 2017.
Future patterns of urbanization reveal trade-offs
among ecosystem services . Landscape Ecology Volume 32, Issue 3, pp 617-634
Petrasova, A., Petras, V., Van Berkel, D., Harmon, B. A., Mitasova, H., and Meentemeyer, R. K., 2016.
Open Source Approach to Urban Growth Simulation .
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLI-B7, 953-959.
If you can't get to any of these, we can send them to you!
Tutorials
Highlights
realistic spatial pattern [Pickard 2017]
modular (different submodels and data can be plugged in)
transparent (open source)
integrated with analytical tools (in GRASS GIS)
available (open source including its dependencies)
Pickard, B., Gray, J., and Meentemeyer, R.K., 2017.
Comparing quantity, allocation and configuration accuracy of multiple land change models.
Land 6.3: 52.
FUTURES stands out because it can model realistic spatial patterns,
is modular and makes the modeling transparent thanks
to its new open source implementation.