ArticlePDF Available

Abstract and Figures

Spatial patterns of land use change due to urbanization and its impact on the landscape are the subject of ongoing research. Urban growth scenario simulation is a powerful tool for exploring these impacts and empowering planners to make informed decisions. We present FUTURES (FUTure Urban – Regional Environment Simulation) – a patch-based, stochastic, multi-level land change modeling framework as a case showing how what was once a closed and inaccessible model benefited from integration with open source GIS.We will describe our motivation for releasing this project as open source and the advantages of integrating it with GRASS GIS, a free, libre and open source GIS and research platform for the geospatial domain. GRASS GIS provides efficient libraries for FUTURES model development as well as standard GIS tools and graphical user interface for model users. Releasing FUTURES as a GRASS GIS add-on simplifies the distribution of FUTURES across all main operating systems and ensures the maintainability of our project in the future. We will describe FUTURES integration into GRASS GIS and demonstrate its usage on a case study in Asheville, North Carolina. The developed dataset and tutorial for this case study enable researchers to experiment with the model, explore its potential or even modify the model for their applications.
Content may be subject to copyright.
A. Petrasovaa,b,
, V. Petrasa,b , D. Van Berkela, B. A. Harmona,d , H. Mitasovaa,b, R. K. Meentemeyera,c
aCenter for Geospatial Analytics, North Carolina State University, USA -
bDepartment of Marine, Earth, and Atmospheric Sciences, North Carolina State University, USA - (vpetras, akratoc, hmitaso)
cDepartment of Forestry and Environmental Resources, North Carolina State University, USA -
dDepartment of Landscape Architecture, North Carolina State University, USA -
Commission VII, SpS10 - FOSS4G: FOSS4G Session (coorganized with OSGeo)
KEY WORDS: GRASS GIS, FUTURES, urbanization, land change, open science, simulation
Spatial patterns of land use change due to urbanization and its impact on the landscape are the subject of ongoing research. Urban
growth scenario simulation is a powerful tool for exploring these impacts and empowering planners to make informed decisions. We
present FUTURES (FUTure Urban – Regional Environment Simulation) – a patch-based, stochastic, multi-level land change modeling
framework as a case showing how what was once a closed and inaccessible model benefited from integration with open source GIS. We
will describe our motivation for releasing this project as open source and the advantages of integrating it with GRASS GIS, a free, libre
and open source GIS and research platform for the geospatial domain. GRASS GIS provides efficient libraries for FUTURES model
development as well as standard GIS tools and graphical user interface for model users. Releasing FUTURES as a GRASS GIS add-on
simplifies the distribution of FUTURES across all main operating systems and ensures the maintainability of our project in the future.
We will describe FUTURES integration into GRASS GIS and demonstrate its usage on a case study in Asheville, North Carolina. The
developed dataset and tutorial for this case study enable researchers to experiment with the model, explore its potential or even modify
the model for their applications.
Population growth in cities worldwide drives changes in land use
often negatively impacting the environments in which people live
and undermining the resilience of local ecosystems. The need
to understand the trade-offs urban planners are facing gave rise
to a number of different land change simulation models, which
proved to be powerful tools for exploring alternative scenarios
and their impacts on various aspects of human-environmental sys-
tems (Chaudhuri and Clarke, 2013, Verburg et al., 2002, Sohl
et al., 2007, Waddell, 2002). Despite the influence of the spa-
tial structure and connectivity of urbanizing landscapes on bio-
diversity, water quality, or flood risks (Alberti, 2005), most ur-
ban growth models are based on cell-level conversions and have
not focused on generating realistic spatial structures across scales
(Jantz and Goetz, 2005). To bridge the gap between cell- and
object-based representation, we developed FUTURES (FUTure
Urban-Regional Environment Simulation), a patch-based, multi-
level modeling framework for simulating the emergence of land-
scape spatial structure in urbanizing regions (Meentemeyer et al.,
2013). The FUTURES model was successfully applied in sev-
eral cases including a study of land development dynamics in the
rapidly expanding metropolitan region of Charlotte, North Car-
olina (Meentemeyer et al., 2013) and an analysis of the impacts
of urbanization on natural resources under different conservation
strategies (Dorning et al., 2015). Most recently, FUTURES was
coupled with ecosystem services models to examine the impacts
of projected urbanization and urban pattern on several ecosystem
services and their trade-offs (Shoemaker, 2016, Pickard et al., in
In order to study the complex interactions between human and
natural systems, interdisciplinary researchers are coupling exist-
ing simulation models. Land change modeling plays often a cru-
Corresponding author
cial role in these coupled models. Previous case studies with FU-
TURES have demonstrated that the model can be applied to a
wide range of cases with different study systems and aims. The
initial implementation of model, however, was a prototype that
was not ready to be shared with scientific community. The model
accumulated too much “technical depth” (Easterbrook, 2014) dur-
ing its initial development, making it difficult to add new fea-
tures or run the simulation at larger scales. In order to continue
adding new capabilities to FUTURES and to promote its usage
both inside and outside of the land use community, we decided to
revise the implementation of the FUTURES model and develop
a new version which would be (a) more efficient and scalable,
(b) as easy to use as possible for a wider audience and (c) fully
open source and maintainable in the long run. To achieve these
goals we decided that instead of keeping FUTURES as a stan-
dalone application, we would take advantage of existing geospa-
tial software and integrate FUTURES into open source GRASS
GIS (Neteler and Mitasova, 2008). By using GRASS GIS’ effi-
cient geospatial libraries we can develop better and higher-level
code. Providing open source software to the scientific commu-
nity entails more than just releasing the actual code – documen-
tation, tutorials, installation instructions, binaries and support are
also needed and require considerable effort. Without this effort,
models cannot be practically used by other researchers. By us-
ing GRASS GIS’ existing infrastructure we could focus on de-
veloping the actual materials instead of managing our own server
In this article we present a new version of the FUTURES urban
growth model that is available as the r.futures module set from the
GRASS GIS add-on repository. This new version of FUTURES
streamlines data processing, provides opportunities to study ur-
banization on mega-regional scales, and allows for more repro-
ducible research in the land change community. We demonstrate
this new version of FUTURES with a case study of the Asheville
metropolitan area in North Carolina, USA.
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XLI-B7, 2016
XXIII ISPRS Congress, 12–19 July 2016, Prague, Czech Republic
This contribution has been peer-reviewed.
FUTure Urban-Regional Environment Simulation is a stochastic,
patched-based model for projecting landscape patterns of urban
growth (Meentemeyer et al., 2013). FUTURES has a modular
structure consisting of 3 main submodels: DEMAND, POTEN-
TIAL and PGA (patch-growing algorithm), see Figure 1. Land
conversion is driven by projected population demand computed
by the DEMAND submodel, and is spatially defined by a prob-
ability surface derived by the POTENTIAL submodel from mul-
tiple environmental and socio-economic predictors. The popu-
lation demand and the effects of land change drivers can vary in
space by subregions, such as jurisdictional units, allowing projec-
tions across heterogeneous landscape. FUTURES main strength
lies in realistically modeling the spatial structure of urban change
by growing patches that are parameterized by size and compact-
ness and calibrated using historical data. For a detailed explana-
tion of FUTURES’ components, please refer to Meentemeyer et
al. (2013).
Land cover maps
through time
population data
where to build?
how much land?
patch size & shape?
Figure 1: Simplified schema of FUTURES conceptual model
with inputs and outputs in gray and submodels in yellow
The original implementation of FUTURES consisted mainly of
the patch-growing algorithm, a standalone program written in a
mixture of C and C++. The PGA program itself utilized ineffi-
cient algorithms and required raster data in ASCII format as input
leading to very slow initialization. The DEMAND submodel was
computed in a spreadsheet and POTENTIAL coefficents were de-
rived using R statistical software. No official implementation
of these submodels existed so each researcher developed a dif-
ferent workflow. This made it difficult for peers to verify each
other’s work. Several scripts for calibrating patch characteris-
tics derived with FRAGSTATS (McGarigal et al., 2012) existed,
however these tools were written in an unnecessarily low-level
language for a specific case using the author’s directory layout.
When revising the original implementation of FUTURES we iden-
tified several issues which needed to be addressed. First, it is im-
portant to follow best practices for scientific computing (Wilson
et al., 2014) including use of a versioning system, writing doc-
umentation and testing. We also wanted to minimize tasks that
had been done manually in order to make the process more ef-
ficient and avoid errors that are often difficult to detect. When
automating tasks we had to compromise between the flexibility
and simplicity of the workflow. We also focused on making FU-
TURES scalable so that it can run large scale applications at a
relatively fine spatial resolution. Finally, we designed FUTURES
to be more user-friendly and easy to test so that anyone can con-
fidently apply it to their research.
GRASS GIS has had a long history as a platform for scientific
models (Chemin et al., 2015). As an open source GIS used by
researchers worldwide and one of the founding projects of OS-
Geo, GRASS GIS provides a stable environment for the devel-
opment of scientific models for studying problems from various
domains including geomorphology, hydrology, planetary science,
landscape ecology, hazard mapping, archaeology, renewable en-
ergy and transportation. Thanks to the numerous scientist and
developers who have been involved, GRASS GIS today provides
a large spectrum of geospatial modules ranging from basic GIS
functionality to highly specialized models. Most of the special-
ized tools are not part of standard GRASS GIS installations, but
are easily accessible from the add-on repository.
There were multiple reasons for our decision to integrate FU-
TURES into GRASS GIS as an add-on. Some of these reasons
were specific to FUTURES, but others apply to any spatial, sci-
entific model. Integrating a model into GIS gives both users and
developers a wide array of standard geospatial tools that simplify
the implementation of a model, and streamline pre- and post-
processing and visualization. GRASS GIS provides model devel-
opers a raster library for highly efficient data reading and writ-
ing. This means that FUTURES no longer has to read in ASCII
files, significantly reducing time needed for initialization. Fur-
thermore, raster data from FUTURES simulations are efficiently
compressed. Despite ever increasing disk space, it is still quite
important to reduce the file size, especially for stochastic spatio-
temporal simulations, which typically generate huge datasets. In
order to achieve the best speed performance, most GRASS GIS
functionality is implemented in C. Because of this, we could eas-
ily integrate FUTURES code, written in a mix of C and C++,
without major rewriting. For portability reasons we later decided
to use the C99 standard. While C and C++ are the preferred lan-
guages for computationally expensive algorithms, GRASS GIS
also supports Python as the primary scripting language. This
is crucial because FUTURES had many steps of data prepara-
tion that we were easily able to automate using Python scripting.
Model developers can appreciate GRASS GIS’ automatic gener-
ation of command line interfaces, Python interfaces and graph-
ical user interfaces (GUI). Simply by defining options in C or
Python modules we can call the same module from a GUI dia-
log, a Python script or a Bash script. A graphical interface makes
FUTURES easy to use, especially for users on the Windows plat-
form. A Python or Bash interface, however, is needed for more
advanced applications such as running FUTURES in parallel on a
high performance computer. GRASS GIS provides infrastructure
for publishing and distributing models to users on all major plat-
forms. Models and tools in GRASS GIS’s Add-on repository1
can be easily browsed and installed with their documentation, re-
lieving researchers of the burden of maintaining such infrastruc-
3.1 Implementation
We implemented FUTURES as a set of GRASS GIS modules
starting with a common prefix r.futures:
r.futures.demand extrapolates the area of developed land from
population trends and projections.
r.futures.devpressure computes the development pressure pre-
r.futures.potential models the development probability sur-
face through multi-level logistic regression.
r.futures.calib calibrates patch sizes and shapes.
r.futures.pga simulates urban development using the patch
growing algorithm.
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XLI-B7, 2016
XXIII ISPRS Congress, 12–19 July 2016, Prague, Czech Republic
This contribution has been peer-reviewed.
In addition, we implemented the add-on r.sample.category needed
for the workflow. Since its functionality is not specific to FU-
TURES, we kept it separate. All of these add-ons can be conve-
niently installed from GRASS GIS using the GUI or command
line2. Each individual add-on has a manual page accessible both
online and offline. Figure 2 shows FUTURES workflow and the
inputs needed for each tool. In the following sections we de-
scribe the developed tools, their functionality and implementation
Figure 2: Diagram of FUTURES workflow showing how are
r.futures modules (yellow boxes) chained and what are their
input data (grey boxes). As indicated by the light yellow box,
module r.futures.calib calls r.futures.pga.
3.1.1 r.futures.demand Based on historical land development
and population growth, the DEMAND submodel (implemented
as r.futures.demand) projects the rate of per capita land consump-
tion for each year of the simulation and each subregion. This
Python module uses GRASS GIS Python Scripting Library and
the NumPy, SciPy and matplotlib libraries for scientific comput-
ing to approximate the relation between population and land con-
sumption with a statistical model described by a linear, logarith-
mic or exponential curve. For example, a logarithmic relation
means that a growing population requires less developed land per
person over time. With enough data points, the module can select
the best curve for each subregion based on residuals. The primary
outputs are plain text files with tab-separated values representing
the number of cells to be converted to developed land each year
for each subregion. The module plots the resulting curves and
projected points for each subregion (Figure 3) so that the results
can be visually inspected. The module r.futures.demand provides
a fast way to estimate the land demand given a large number of
subregions with diverse population trends and thus allows us to
quickly explore different population scenarios.
3.1.2 r.futures.devpressure Development pressure is one of
the most important predictors of where development is likely to
happen. For each cell it is computed as a distance decay func-
tion of neighboring developed cells (Meentemeyer et al., 2013).
Compared to the tool previously used for computing development
pressure, the new Python module r.futures.devpressure provides a
faster and more efficient implementation by taking advantage of
the existing GRASS GIS module r.mfilter written in C for moving
window analysis with custom designed matrix filters. By precom-
puting the matrix of distances we avoid repeated distance compu-
tations resulting in faster processing. Because the new implemen-
tation is less memory intensive it can be used for larger regions
than the previous tool.
3.1.3 r.futures.potential uses multilevel logistic regression to
model development suitability based on environmental, infras-
tructural, and socio-economic predictors such as distance to roads
or topographic slope. We randomly sample these predictors and
2g.extension r.futures
developed cells
37021, RMSE: 1188.867
y= 2.505 + 0. 257 ln( x1.382)
Figure 3: An example of r.futures.demand output plot showing
the logarithmic relation between population and land
consumption for the county with ID 37021. Observed data are
showed as blue dots and predicted data as circles.
the observed change from undeveloped to developed cells to es-
timate the coefficients of the multilevel logistic regression. The
core of this module is a script in the R language (R Develop-
ment Core Team, 2008), which uses the package lme4 (Bates et
al., 2015) for fitting generalized linear mixed-effects models and
the package MuMIn (Barto, 2015) for automatic model selection.
The output file is a plain text file with tab-separated regression
coefficients. This script is wrapped in Python for more seamless
processing and chaining of modules. The coupling between R,
Python and GRASS GIS is intentionally very loose to make the
workflow possible in the Windows environment where some of
the other, more elegant, options such as rpy23are complicated to
use. We performed stratified sampling of observed new develop-
ment and predictors using GRASS GIS add-on r.sample.category.
Although we developed this add-on for urban growth modeling
with FUTURES, its application is much broader. In order to en-
courage its use in other applications we made it a general module
rather than making it part of the r.futures tool set.
3.1.4 r.futures.pga is the main engine of FUTURES – it sim-
ulates urban growth using inputs from the DEMAND and PO-
TENTIAL submodels. The patch growing algorithm (PGA) sto-
chastically allocates a seed for new development across the de-
velopment suitability surface, challenges the seed by comparing
it with a random number between 0 and 1, and then grows a
discrete patch from the seed if it survives (Meentemeyer et al.,
2013). This process repeats until the number of converted cells
specified by DEMAND is met. The development pressure predic-
tor and then the development suitability values are updated based
on the newly developed cells. (The development suitability is
computed internally from predictors and regression coefficients
supplied by POTENTIAL.) We kept the original patch growing
algorithm, but significantly improved its implementation to make
it faster, more memory efficient and simpler to use. We replaced a
custom, undocumented configuration file with a standard module
interface usable from GUI or the command line, and restructured
the input and output parameters and their names so that they are
easy for users to understand. We used efficient GRASS GIS li-
braries for reading and writing raster data, which minimized the
3rpy2 is a Python package for using R from Python (http://rpy2.
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XLI-B7, 2016
XXIII ISPRS Congress, 12–19 July 2016, Prague, Czech Republic
This contribution has been peer-reviewed.
time needed to initialize the simulation. FUTURES now reads
rasters in GRASS’s native format instead of ASCII files. This
decreased the time needed for model initialization from several
minutes to several seconds for a region with tens of millions of
cells. Furthermore, we replaced the static allocation of internal
structures with dynamic allocation and reduced the overall mem-
ory requirements so that FUTURES could run on large regions
with tens or hundreds of counties as well as smaller areas like our
case study. Finally, through the use of appropriate programming
techniques, such as binary search, we significantly increased the
speed of the algorithm.
3.1.5 r.futures.calib We developed a dedicated Python mod-
ule for calibrating patch sizes and shapes that runs the module
r.futures.pga with different combinations of patch parameters and
outputs a table with scores for each combination of patch param-
eters. The simulation is run multiple times for each combination
to account for the stochasticity of the model. To speed up the
calibration process r.futures.calib can take advantage of multiple
computer cores.
To demonstrate how the new FUTURES framework can be used
to simulate urban growth, we present a case study for Asheville
metropolitan area located in the Blue Ridge Mountains in the
west of North Carolina, USA. The region consists of five counties
with total area of 6,271 km2and around 477,000 people based on
2014 population estimates. It is characterized by rapid popula-
tion growth around Asheville, the largest city of the region. New
development is constrained by the steep mountainous terrain and
large national and state parks. We simulate urban growth from
2012 to 2030 using publicly available data, including the USGS’s
National Land Cover Database (NLCD) (Homer et al., 2015, Fry
et al., 2011, Homer et al., 2007, Fry et al., 2009), past estimates
and future projections of county populations (NCOSBM, 2015),
boundaries and roads provided by the United States Census Bu-
reau’s database (TIGER) and a digital elevation model from the
National Elevation Dataset (NED) distributed by the USGS.
4.1 Approach
There are several steps required to run the FUTURES simulation:
Preprocess the data.
Estimate per capita land consumption controlling the total
area of converted land.
Derive the development suitability statistical model to con-
trol where the new development happens.
Calibrate patch size and shape.
Run the urban growth simulation.
4.1.1 Data preparation The core input data for urban growth
modeling with FUTURES is a timeseries of land cover maps,
which can be derived by various methods from satellite imagery.
In this study we used the 2001, 2006 and 2011 NLCD Land
Cover products and 1992/2001 Retrofit Land Cover Change prod-
uct to derive a 30-meter binary representation of developed areas.
We excluded national and state parks, water bodies and wetlands
from further analysis. We used NLCD products that are avail-
able for the contiguous USA so that this study and its workflow
would be easier to reproduce and apply to other study areas. We
obtained population statistics from the North Carolina Office of
State Budget and Management, which are based on 2000 and
2010 censuses and include past as well as future estimates of pop-
ulation per county for each year up to 2035. Data for the 5 coun-
ties studied were extracted and formatted as a comma-separated
values (CSV) file.
Figure 4: 2011 land cover (Homer et al., 2015) and protected
areas (Anderson and Sheldon, 2011) in the Asheville
metropolitan area in the west of North Carolina, USA. Inset A is
used in Figure 5.
4.1.2 DEMAND We derived the relation between population
and land consumption from the series of binary rasters of de-
veloped areas and population statistics to model how much land
will be developed each year of the simulation. Using the mod-
ule r.futures.demand we explored different curve fitting methods
and derived the per capita land consumption from period 1992
through 2011, which was characterized by population growth with
decreasing demand for land per person over time. We expect sim-
ilarly low rates of per capita land consumption in the following
years because development is restricted by the mountainous ter-
rain and large protected areas. Based on RMSE and visual inspec-
tion of the plots created by r.futures.demand we selected either
linear or logarithmic relations for each county, where the function
coefficients were automatically determined using linear regres-
sion and non-linear least squares optimization in r.futures.demand
(Figure 3).
4.1.3 POTENTIAL We used multilevel logistic regression to
predict where new development happens based on environmental,
infrastructural and socio-economic site suitability predictors. Us-
ing r.sample.category we sampled predictors on 8000 randomly
selected locations and estimated the model coefficients using the
R package lme4 integrated into the module r.futures.potential.
The sample points were stratified by the response variable where
new sites developed since 1992 have a value of 1 and sites that
are undeveloped in 2011 have a value of 0. We included counties
as the group level indicator in the multilevel model to account
for differences across jurisdictional boundaries. From the initial
list of hypothesized predictors (slope, distance to water, protected
areas, interchanges, travel time to cities and towns, forest occur-
rence and road density) we identified a set of predictors (Table
1) resulting in a model with the lowest AIC (Akaike information
criterion) score. We verified the robustness of the selected pre-
dictors by repeating the random sampling and model selection
process multiple times. In addition to these predictors, we in-
cluded also development pressure, a special, dynamic predictor
that is updated during the simulation based on new simulated de-
velopment to enable positive feedback. We computed the initial
development pressure raster with r.futures.devpressure; its subse-
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XLI-B7, 2016
XXIII ISPRS Congress, 12–19 July 2016, Prague, Czech Republic
This contribution has been peer-reviewed.
undeveloped projected development
developed until 2011 protected areas and water roads 5 km
a) sprawl b) status quo c) inll
AshevilleAshevilleAshevilleAsheville AshevilleAsheville
Figure 5: Results of three realizations of multiple stochastic runs with different scenarios. Depending on the scenario, simulated
development is more diffuse (a) or more compact (c).
quent updates are performed in memory during the simulation.
Predictors Estimate*Std. Error
Intercept (varies by county) -2.593 0.269
Development pressure 0.058 0.005
Road density 0.118 0.007
Percentage of forest -0.013 0.002
Distance to protected areas -0.140 0.039
Distance to water bodies -0.148 0.022
* all P-values <0.001
Table 1: List of selected predictors and estimated coefficients for
site suitability model
4.1.4 Patch calibration Prior to running the urban growth
simulation implemented in r.futures.pga we calibrated the input
patch compactness and size to match the simulated patterns with
the observed patterns from 1992 to 2011. Since calibration is
a time consuming process, we ran the module r.futures.calib for
Buncombe county and applied the results to the rest of our study
region. We choose Buncombe County which includes the city
of Asheville because it is where most new development will oc-
cur. For each combination of patch parameters we compared the
patch characteristics averaged from 20 runs of the urban growth
simulation with the known patches. Based on the score we se-
lected patch parameters resulting in high compactness which is
expected for mountainous regions.
4.1.5 Urban growth simulation Having collected all neces-
sary input data, we ran r.futures.pga with a 1 year time step until
2035 for the entire study region at 30 m resolution. To account for
different future policies regarding new development, we explored
scenarios altering the site suitability to encourage infill or sprawl
by changing incentive power parameter of r.futures.pga. This
value transforms the probability pa cell is developed to pxwhere
x= 1 represents status quo, higher values of xresult in infill and
lower values in sprawl. In addition to the status quo we simulated
scenarios with xequals 0.25, 0.5, 2 and 4. We repeated each
scenario 50 times to account for the model’s stochastic behavior.
4.2 Results
The resulting development patterns of three realizations of the
random runs are visible in Figure 5 for the status quo, infill sce-
nario (x= 4) and sprawl scenario (x= 0.25). The simulated
patches realistically mimic the current patches of development
in shape and size and are mostly, but not exclusively adjacent to
roads as expected. Furthermore, we post-processed the results to
study how different urban growth policies influence the loss of
forest and agricultural land in the Asheville area (Figure 6) by
averaging the loss of both land use categories over the 50 runs. In
all scenarios, forest is more affected by future development than
farmland. The extreme case of urban sprawl results in twice as
much forest as farmland being developed. Status quo scenario
leads to the smallest difference between the areas converted to
forest and farmland. Interestingly, infill scenario develops the
forested area in a similar way as sprawl does, which is not surpris-
ing considering developed areas are largely surrounded by forest
0.25 0.5 1 24
16 forest farmland
inllsprawl status quo
Converted area in km2
Figure 6: Area in km2of converted land from forest (green) and
farmland (yellow) to urban differs for urban sprawl and infill
scenarios. Numbers 0.25 to 4represent the exponent xwhich
transforms development probability pto px.
Table 2 shows the computational resources necessary for run-
ning this case study and compares the time and memory require-
ments with the original implementation of FUTURES. Note that
our study area is fairly small (12 million cells) and when applied
to larger regions with more projected development, the expected
speed gain is even more significant as we changed the complexity
of one of the core algorithms from linear to logarithmic. Because
the individual stochastic runs are independent of each other this
simulation is an “embarrassingly parallel” problem (Herlihy and
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XLI-B7, 2016
XXIII ISPRS Congress, 12–19 July 2016, Prague, Czech Republic
This contribution has been peer-reviewed.
Shavit, 2012) in which the computation can easily be distributed
across multiple computer cores.
FUTURES version memory 1 run all runs (250)
original 1.7 GB 60 s 4 h 10 min
r.futures 0.86 GB 19 s 1 h 20 min
Table 2: Time and memory needed to run the simulations with
the old version of FUTURES and the new r.futures implemented
in GRASS GIS on a laptop with 64-bit Ubuntu 14.04 LTS, Intel
Core i7-4760HQ @2.10GHz using 1 CPU and running on
external hard drive.
The input data and instructions to run the model are available as
part of material developed for the US-IALE 2016 Annual Meet-
ing workshop on FUTURES4.
The new FUTURES framework is split into independent GRASS
GIS modules so that the modeling workflow is flexible and ex-
tendable. By using standardized inputs and outputs (raster lay-
ers and CSV files) and described interface we allow FUTURES’
users to replace DEMAND and POTENTIAL implementations
by their own tools, which may be better suited to the character-
istics and datasets available for their study systems. We ran all
previous studies on county level at 30 m resolution. FUTURES,
however, can be applied to larger or smaller scales as long as
there is data available and the patch characteristics are properly
calibrated. Future research will explore nested scales in order to
address the different scales of the population data and the spatial
drivers of land change.
We presented a new, open source version of the FUTURES ur-
ban growth model that is integrated into GRASS GIS, opening
new possibilities for environmental scientists and urban planners
to project and understand the impacts of urbanization at relevant
ecological and decision-making scales. Integration into GRASS
GIS allowed us to make FUTURES more efficient, simple to use
and transparent. With documented code running on all platforms,
FUTURES can now be easily tested and applied to study sites at
local to megaregional scales. We illustrated how FUTURES can
be used in a small case study of the Asheville metropolitan area.
We also provided the instructions and data needed to reproduce
this study as a step towards more reproducible research in land
change science.
We would like to thank Monica Dorning and Douglas Shoemaker
for discussing with us the original model implementation, and
Brian Pickard and Georgina Sanchez for testing new FUTURES
Alberti, M., 2005. The effects of urban patterns on ecosystem
function. International regional science review 28(2), pp. 168–
Anderson, M. and Sheldon, A. O., 2011. Conservation Status
of Fish, Wildlife, and Natural Habitats in the Northeast Land-
scape: Implementation of the Northeast Monitoring Framework.
Technical report, The Nature Conservancy, Eastern Conservation
Barto, K., 2015. MuMIn: Multi-Model Inference. R package
version 1.15.1.
Bates, D., Maechler, M., Bolker, B. and Walker, S., 2015. lme4:
Linear mixed-effects models using Eigen and S4. R package ver-
sion 1.1-9.
Chaudhuri, G. and Clarke, K. C., 2013. The sleuth land use
change model: A review. International Journal of Environmental
Resources Research 1(1), pp. 88–104.
Chemin, Y., Petras, V., Petrasova, A., Landa, M., Gebbert, S.,
Zambelli, P., Neteler, M., L ¨
owe, P. and Di Leo, M., 2015.
GRASS GIS: a peer-reviewed scientific platform and future re-
search repository. In: Geophysical Research Abstracts, Vol. 17,
p. 8314.
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.
Easterbrook, S. M., 2014. Open code for open science? Nature
Geoscience 7(11), pp. 779–781.
Fry, J., Coan, M., Homer, C., Meyer, D. and Wickham, J., 2009.
Completion of the National Land Cover Database (NLCD) 1992–
2001 Land Cover Change Retrofit Product. Technical report, U.S.
Geological Survey Open-File Report 20081379.
Fry, J., Xian, G., Jin, S., Dewitz, J., Homer, C., Yang, L.,
Barnes, C., Herold, N. and Wickham, J., 2011. Completion of
the 2006 National Land Cover Database for the Conterminous
United States. PE&RS 77(9), pp. 858–864.
Herlihy, M. and Shavit, N., 2012. The Art of Multiprocessor Pro-
gramming, Revised Reprint. Elsevier.
Homer, C., Dewitz, J., Fry, J., Coan, M., Hossain, N., Larson, C.,
Herold, N., McKerrow, A., VanDriel, J. and Wickham, J., 2007.
Completion of the 2001 National Land Cover Database for the
Conterminous United States. Photogrammetric Engineering and
Remote Sensing 73(4), pp. 337–341.
Homer, C., Dewitz, J., Yang, L., Jin, S., Danielson, P., Xian,
G., Coulston, J., Herold, N., Wickham, J., and Megown, K.,
2015. Completion of the 2011 National Land Cover Database
for the conterminous United States – Representing a decade of
land cover change information. Photogrammetric Engineering
and Remote Sensing 81(5), pp. 345–354.
Jantz, C. A. and Goetz, S. J., 2005. Analysis of scale dependen-
cies in an urban land-use-change model. International Journal of
Geographical Information Science 19(2), pp. 217–241.
McGarigal, K., Cushman, S. and Ene, E., 2012. FRAGSTATS v4:
spatial pattern analysis program for categorical and continuous
maps. Computer software program produced by the authors at
the University of Massachusetts, Amherst.
Meentemeyer, R. K., Tang, W., Dorning, M. A., Vogler, J. B.,
Cunniffe, N. J. and Shoemaker, D. A., 2013. FUTURES: Multi-
level Simulations of Emerging UrbanRural Landscape Structure
Using a Stochastic Patch-Growing Algorithm. Annals of the As-
sociation of American Geographers 103(4), pp. 785–807.
NCOSBM, 2015. County/State Population Projections.
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XLI-B7, 2016
XXIII ISPRS Congress, 12–19 July 2016, Prague, Czech Republic
This contribution has been peer-reviewed.
Neteler, M. and Mitasova, H., 2008. Open Source GIS: A GRASS
GIS Approach. Vol. 773, Springer, New York.
Pickard, B. R., Van Berkel, D., Petrasova, A. and Meentemeyer,
R. K., in prep. Future patterns of urbanization reveal trade-offs
among ecosystem services.
R Development Core Team, 2008. R: A Language and Envi-
ronment for Statistical Computing. R Foundation for Statistical
Computing, Vienna, Austria. ISBN 3-900051-07-0.
Shoemaker, D. A., 2016. The role of spatial heterogeneity
and urban pattern in modulating ecosystem services. Doctoral
dissertation, North Carolina State University. Retrieved from
Sohl, T. L., Sayler, K. L., Drummond, M. A. and Loveland, T. R.,
2007. The FORE-SCE model: a practical approach for projecting
land cover change using scenario-based modeling. Journal of
Land Use Science 2(2), pp. 103–126.
Verburg, P. H., Soepboer, W., Veldkamp, A., Limpiada, R., Espal-
don, V. and Mastura, S. S., 2002. Modeling the spatial dynamics
of regional land use: the CLUE-S model. Environmental man-
agement 30(3), pp. 391–405.
Waddell, P., 2002. UrbanSim: Modeling urban development for
land use, transportation, and environmental planning. Journal of
the American Planning Association 68(3), pp. 297–314.
Wilson, G., Aruliah, D., Brown, C. T., Hong, N. P. C., Davis, M.,
Guy, R. T., Haddock, S. H., Huff, K. D., Mitchell, I. M., Plumb-
ley, M. D. et al., 2014. Best practices for scientific computing.
PLoS Biol 12(1), pp. e1001745.
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XLI-B7, 2016
XXIII ISPRS Congress, 12–19 July 2016, Prague, Czech Republic
This contribution has been peer-reviewed.
... The FUTURES model was developed by Meentemeyer et al. in 2013 [62] and published as an open-source add-on of GRASS GIS software ( [63]. The FUTURES model integrates a multilevel regression model to determine the probabilities of urban development in sub-regions and a patchbased stochastic growth algorithm to simulate future landscape patterns in urbanizing regions [15,63]. ...
... [63]. The FUTURES model integrates a multilevel regression model to determine the probabilities of urban development in sub-regions and a patchbased stochastic growth algorithm to simulate future landscape patterns in urbanizing regions [15,63]. FUTURES allows researchers to modify the initial data or control parameters to construct different urban growth scenarios [62]. ...
... The FUTURES model was developed by Meentemeyer et al. in 2013 [62] and published as an open-source add-on of GRASS GIS software ( [63]. The FUTURES model integrates a multilevel regression model to determine the probabilities of urban development in sub-regions and a patch-based stochastic growth algorithm to simulate future landscape patterns in urbanizing regions [15,63]. ...
Full-text available
Over the last few decades, rapid urban expansion has spread over a great deal of arable and ecological land, leading to severe social and environmental issues. Although different urban growth scenarios cause varying types of urban forms to emerge, there is currently a lack of empirical studies and other research on these different forms. Therefore, it is important for decision-makers to have an improved understanding of the relationships between arable land and ecological land under different urban form conditions in order to implement sustainable urban development policies. This study utilized a patch-based, multilevel stochastic urban growth model to simulate Shenzhen’s urban growth until 2035. To determine the impacts of urban forms and population density on land use, we established five scenarios to simulate urban expansion and land-use changes at the sub-regional scale. The results revealed the trade-off relationships that emerge when altering the urban forms or population density, which shows that no single policy can conserve arable land and ecological land simultaneously. The results also revealed that sub-regions have distinct responses to alternative urban form scenarios compared with an entire region. Decision-makers and planners should consider the urban form in order to optimize development projects that fit local conditions and achieve more sustainable development.
... These maps are computed with GrassGis Software [7], and the pressure map is computed with devpressure function provided by the r.futures package [10]. The r.futures package consists of eight functions: r.futures.pga, ...
... For measuring the performance of MLLR, we use the implementation of the multi-level logistic function from FUTURES package in GrassGis [10] to measure the allocation accuracy. We use the maps listed in Table 1. ...
Conference Paper
Full-text available
Land use change is a global phenomenon that impacts directly to the urban growth and it should be addressed from different disciplines to minimize the potential negative effects of urbanization predicting the spatial urban growth. The urban growth dynamics might be very complicated and difficult to model, nevertheless it is necessary to understand the causes of the dynamics and the dynamics itself to build precise computational models that help to detect problems generated by urban land use change. For that reason, we propose the study of the land use suitability sub-model used by several models to make land use spatial predictions. This sub-model is implemented as a logistic regression based on linear correlations. The problem is that this model is limited to capture a variety of nonlinear relations among variables for prediction and classification purposes. We propose to use an alternative based on Deep Feedforward Networks able to deal with this problem. In Mexico, the urban growth will increase considerably the number of cities during the next decade where the Mexican population will be concentrated. That means that the generation and study of existing spatio-temporal computational frameworks for studying the Mexican urban growth is very relevant. Therefore we present an initial contribution comparing Deep Feedforward Networks with a Multi-Level Linear Logistic Regression as land suitability models applied to Mexican land use classification. We show that basic deep feedforward models outperform in allocation accuracy to linear logistic regression, and also minimizes the parameters tuned by trial and error.
... FUTURES has been validated [24], tested [27], and used for a variety of applications across a variety of study areas [13,25,27,28]. The model is available as an open-source GRASS GIS integration [29]. FUTURES was designed to explore the possible future locations, quantities, and spatial patterns of developed areas through the integration of three sub-models (DEMAND, PO-TENTIAL, and PGA), with a particular focus on the leapfrogging dynamic development process [24]. ...
Full-text available
Rapid urbanization results in farmland loss, habitat fragmentation, biodiversity decrease, and greenhouse gas emissions. Land-use policies and planning as administrative means are used to guide sustainable urban development and to balance the location of urban expansion and agricultural activities. To better understand the future implications of a variety of land-use policies, we used a FUTURES model scenario analysis to analyze the potential future patterns of urban areas and the loss and fragmentation of farmland and natural resources at the local level for Xi’an. We tested representative indicators of sustainable urbanization according to Plan 2014–2020. We found that scenarios representing the integration of several policies showed both synergetic spatial patterns and conflicting outcomes. The simulated land-use patterns of urban growth resulting from the combination of policies, were the most likely to support progress toward a livable compact city and natural resources’ conservation. These findings underscore the importance of simulation modeling and scenario analyses to quantify and visualize the results from policies and planning to support sustainable urbanization. Specifically, they show the value in simulation modeling for integrating information across scales, i.e., combining macro-level land-use policies with local-level spatial heterogeneity in socio-ecological settings, for identifying actionable planning solutions. Hence, these research results provide scientific support for land-use policy revision and implementation in Xi’an, as well as a reference point for other urbanizing cities in China.
... For the city development, urban growth scenarios and land use changes were derived from the study and associated datasets of Hackbarth (2021). This study compared the CLUE-S model and the FUTURES system, which are both operational within Grass GIS, and opted to rely on the latter given the active support on the one hand, and the possibilities for scenario development (Petrasova et al., 2016). It requires at least two LULC maps (Meentemeyer et al., 2013), which can be directly utilized by QGIS. ...
Full-text available
Predicting how a planned city will develop and expand after its construction, and which resources, such as energy, the city will need over time is only possible if one can rely on similar examples and reliable models. Given the existing spatial plans for the design of the new capital city of Indonesia, there is a need to develop and compare city development scenarios–in spatial expansion, population size, resource, energy and food requirements. A combination of various geospatial data approaches can address this knowledge and assessment gap. This article investigates spatial expansion, forest encroachment and sustainable energy infrastructure requirements using open access geodata and models. The hypothesis is that the constitution of the new capital city of Indonesia can rely on existing energy infrastructures but may also need to rely on additional resources. The research approach was to collect and integrate different types of geospatial data related to land use, terrain characteristics and population growth assumptions and connect these to both urban growth models and predictions and energy. This relied on land use change methodologies and urban growth models to simulate and predict spatial effects, with ca particular focus on the expansion of energy requirements. The choice to focus on energy requirements additionally required a comparison of different kinds of energy sources, such as solar and wind energy. The conclusion is that all design and expansion scenarios indicate a possible spatial conflict between locating sustainable energy production facilities with maintaining ecologically sensitive areas. A possible solution is to make use of existing mining infrastructures to enhance sustainable energy production and to make use of dual land and water solar energy systems.
... This involves setting of input variables for PGA such as patch size and compactness parameters. The calibration process was conducted to match observed urban growth patterns to those simulated by the model, including the sizes and shapes of new development [24]. Calibration requires the development binary raster at the beginning and end of the reference period to derive the patch sizes and compactness. ...
Full-text available
This study attempts to analyze and simulate urban growth pattern of Colombo city in Sri Lanka which is a dynamic and rapid urbanizing region. The spatiotemporal urban growth patterns during 1997–2019 were first analyzed by comparing Land Cover (LC) maps for time intervals between 1997–2008 and 2008–2019 using intensity and growth pattern analysis. Urban lands in Colombo have grown in a faster rate during 1997–2008 as compared to 2008–2019 period. The prominent spatial expansion pattern during 1997–2008 is outlying, as opposed to edge expansion which is predominant during 2008–2019. These major urban expansion patterns were modeled to predict the future urban structure of Colombo in 2030 using FUTURES (FUTure Urban-Regional Environment Simulation) model. FUTURES is a patch-based, multilevel modeling framework for simulating the emergence of landscape spatial structure in urbanizing regions. Simulated result generated from the model reveals substantial agreement with real ground urban changes showing a kappa value of 0.78. The model allows to predict three different scenarios, namely Business as Usual, Infill Growth and Sprawl showing over 100 km2 increase in urban lands by 2030. Predicted urban structure was then compared with proposed development plan. With certain limitations arising from available data, the model is effective in predicting possible urban scenarios and providing valuable inputs to support better decision making for sustainable development of Colombo city. The results demonstrated in this study would be useful in modelling urban growth in other cities and further validate the efficacy of the proposed workflow.
... In this simulation, uncorrelated and significant predictors (p b 0.05) were used to fit a multilevel model at the county level (Table 4). We computed all simulations with the open source GRASS GIS add-on FUTURES toolset (Petrasova et al., 2016). ...
Urban growth and climate change together complicate planning efforts meant to adapt to increasingly scarce water supplies. Several studies have independently examined the impacts of urban planning and climate change on water demand, but little attention has been given to their combined impact. Here we forecast urban water demand using a Geographically Weighted Regression model informed by socio-economic, environmental and landscape pattern metrics. The purpose of our study is to evaluate how future scenarios of population densities and climate warming will jointly affect water demand across two rapidly growing U.S. states (NC and SC). Our forecasts indicate that regional water demand by 2065 will increase by 37%–383% relative to the baseline in 2010, across all scenarios of change. Our results show future water demand will increase under rising temperatures, but could be ameliorated by policies that promote higher density development and urban infill. These water-efficient land use policies showed a 5% regional reduction in water demand and up to 25% reduction locally for counties with the highest expected population growth by 2065. For rural counties experiencing depopulation, the land use policies we considered are insufficient to significantly reduce water demand. For expanding communities seeking to increase their adaptive capacity to changing socio-environmental conditions, our framework can assist in developing sustainable solutions.
... The FUTURES framework is unique in that it combines a field-based with an object-based representation of land change, geared at providing a tool for analyzing the spatial structure of development of peri-urban landscapes. It was validated for the metropolitan region of Charlotte, North Carolina (Meentemeyer et al., 2013), and has been successfully applied in a set of case studies located in North Carolina (Dorning, Koch, Shoemaker, & Meentemeyer, 2015;Petrasova et al., 2016;Pickard, van Berkel, Petrasova, & Meentemeyer, 2017). This CA component enables the representation of environmental factors, their spatial heterogeneity, and their effect on development patterns. ...
Full-text available
Population growth and unrestricted development policies are driving low-density urbanization and fragmentation of peri-urban landscapes across North America. While private individuals own most undeveloped land, little is known about how their decision-making processes shape landscape-scale patterns of urbanization over time. We introduce a hybrid agent-based modeling (ABM)-cellular automata (CA) modeling approach, developed for analyzing dynamic feedbacks between landowners' decisions to sell their land for development, and resulting patterns of landscape fragmentation. Our modeling approach builds on existing conceptual frameworks in land systems modeling by integrating an ABM into an established grid-based land-change model-FUTURES. The decision-making process within the ABM involves landowner agents whose decision to sell their land to developers is a function of heterogeneous preferences and peer-influences (i.e., spatial neighborhood relationships). Simulating landowners' decision to sell allows an operational link between the ABM and the CA module. To test our hybrid ABM-CA approach, we used empirical data for a rapidly growing region in North Carolina for parameterization. We conducted a sensitivity analysis focusing on the two most relevant parameters-spatial actor distribution and peer-influence intensity-and evaluated the dynamic behavior of the model simulations. The simulation results indicate different peer-influence intensities lead to variable landscape fragmentation patterns, suggesting patterns of spatial interaction among landowners indirectly affect landscape-scale patterns of urbanization and the fragmentation of undeveloped forest and farmland.
The advancing global urbanization puts great pressure on the society and ecosystem, especially in developing countries. Reasonable land‐use policies adapted to local conditions are the key issues to prevent disordered urban expansion. Multi‐scenario simulation provides a new perspective for differentiated policy formulation based on regional heterogeneity. Regarding the Shaanxi Province, a rapid urbanized area with high spatial heterogeneity in western China, as the study case, we tracked its past land‐use changes and predicted the characteristics of urban expansion using the Future Urban‐Regional Environment Simulation model. We found that (1) during the past 35 years, built‐up land evidently increased accompanied by the loss of cropland, grassland, and unused land. (2) The urban expansion mainly will occur in the areas with flat terrain under three scenarios. (3) The urban expansion will transit into the “requiring land from mountains” pattern in the future, which means that it will gradually spread to areas with higher slopes. (4) Differentiated and optimal development patterns are proposed for different subregions by accounting the loss of ecosystem service in the procedure of urban expansion. This research can help local governments formulate differentiated future macro control of territorial spatial planning to optimize subregional land development.
Due to human-induced climate change, traditionally water-rich environments can no longer depend on stable patterns of freshwater supply. Climate scientists advise water managers to expect warmer temperatures, changes in the distribution and intensity of precipitation, and more prolonged droughts. Together with global climate forcings, rapidly expanding cities and suburban communities are altering the spatial and temporal availability of freshwater resources. Constantly changing environments urge for integrated land- and water-use planning efforts to help inform water-efficient development patterns. This dissertation contributes to emerging research which increasingly recognizes the spatial configuration and patterns of developed land use as major factors affecting water demand. Framed around three scientific research studies, this dissertation is the first modeling effort of its kind to examine how different spatial patterns of development are likely to affect water demand and supply under a scientifically plausible spectrum of future climate conditions. The first study examines the functional linkage between water use and spatial patterns of development, while accounting for other well studied socio-economic and environmental factors. Results consistently demonstrate that clusters of compact patterns of development show potential for more efficient use of water. The main products of the first study are empirically-derived coefficients of estimated water demand; these coefficients are then used in the second study to forecast how future land and climate changes are likely to impact local and regional water demand. Lastly, the third study examines the spatial distribution and severity of future water stress conditions associated with the combined and individual effects of urban growth, water demand, and climate change. While results show global climate forcings as the dominant driver of future water demand and supply, regional land use policies that promote urban infill and higher density development have the potential to reduce future water demand and water stress. These spatially-explicit policies would be particularly effective in areas with high projected population growth. Changes in flow regime associated with future land and climate conditions are also likely to affect the ecological integrity of aquatic ecosystems, which in turn can negatively impact human well-being. Supporting more environmentally sensitive regulations is critical to ensure the future availability of water. The methodological framework described here provides a platform to scientifically evaluate land-use policies and regulations that support a more efficient use of freshwater resources under future conditions of environmental change.
Full-text available
In this study, factors that cause drainage problems in 23 Nisan Park and Mevlana Park have been investigated in the scope of landscape design and landscape engineering. To propose regulations to protect the natural topography and natural drainage lines, evaluations and observations were conducted, and these data were digitized. Water accumulation analyses were performed using field observations, which were used to create a water accumulation analysis map that was compared with water collection area maps created using Arc Hydro extension in ArcGIS. The results show that surface waters accumulate in areas such as pedestrian ways and playgrounds due to deficiencies in landscape design. Surface water quantities were calculated using the rational method. To prevent structural damage and excessive water entry, deep drainage and concrete or grass-covered parabolic surface drainage systems are recommended. Additional suggestions are provided for other studies to be conducted in this regard.
Conference Paper
Full-text available
Geographical Information System (GIS) is known for its capacity to spatially enhance the management of natural resources. While being often used as an analytical tool, it also represents a collaborative scientific platform to develop new algorithms. Thus, it is critical that GIS software as well as the algorithms are open and accessible to anybody [18]. We present how GRASS GIS, a free and open source GIS, is used by many scientists to implement and perform geoprocessing tasks. We will show how integrating scientific algorithms into GRASS GIS helps to preserve reproducibility of scientific results over time [15]. Moreover, subsequent improvements are tracked in the source code version control system and are immediately available to the public. GRASS GIS therefore acts as a repository of scientific peer-reviewed code, algorithm library, and knowledge hub for future generation of scientists. In the field of hydrology, with the various types of actual evapotranspiration (ET) models being developed in the last 20 years, it becomes necessary to inter-compare methods. Most of already published ETa models comparisons address few number of models, and small to medium areas [3, 6, 7, 22, 23]. With the large amount of remote sensing data covering the Earth, and the daily information available for the past ten years (i.e. Aqua/Terra-MODIS) for each pixel location, it becomes paramount to have a more complete comparison, in space and time. To address this new experimental requirement, a distributed computing framework was designed, and created [3, 4]. The design architecture was built from original satellite datasets to various levels of processing until reaching the requirement of various ETa models input dataset. Each input product is computed once and reused in all ETa models requiring such input. This permits standardization of inputs as much as possible to zero-in variations of models to the models internals/specificities. All of the ET models are available in the new GRASS GIS version 7 as imagery modules and replicability is complete for future research. A set of modules for multiscale analysis of landscape structure was added in 1992 by [1], who developed the r.le model similar to FRAGSTATS ([10]). The modules were gradually improved to become in 2006. Further development continued, with a significant speed up [9] and new interactive user interface. The development of spatial interpolation module started in 1988 [11] and continued by introduction of new interpolation methods and finally full integration into GRASS GIS version 4 [13]. Since then it was improved several times [8]. The module is an important part of GRASS GIS and is taught at geospatial modeling courses, for example at North Carolina State University [14]. GRASS GIS entails several modules that constitute the result of active research on natural hazard. The r.sim.water simulation model [12] for overland flow under rainfall excess conditions was integrated into the Emergency Routing Decision Planning system as a WPS [17]. It was also utilized by [16] and is now part of Tangible Landscape, a tangible GIS system, which also incorporated the r.damflood, a dam break inundation simulation [2]. The wildfire simulation toolset, originally developed by [24], implementing Rothermel’s model [21], available through the GRASS GIS modules r.ros and r.spread, is object of active research. It has been extensively tested and recently adapted to European fuel types ([5, 19, 20]). References [1] Baker, W.L., Cai, Y., 1992. The r.le programs for multiscale analysis of landscape structure using the GRASS geographical information system. Landscape Ecology, 7(4):291-302. [2] Cannata M. and Marzocchi R., 2012. Two-dimensional dam break flooding simulation: a GIS embedded approach. - Natural Hazards 61(3):1143-1159. [3] Chemin, Y.H., 2012. A Distributed Benchmarking Framework for Actual ET Models. In Evapotranspiration - Remote Sensing and Modeling, Intech (Eds). [4] Chemin, Y. H. , 2014. Remote Sensing Raster Programming, 3rd Ed., Lulu (Eds). [5] Di Leo, M., de Rigo, D., Rodriguez-Aseretto, D., Bosco, C., Petroliagkis, T., Camia, A., San-Miguel-Ayanz, J., 2013. Dynamic data driven ensemble for wildfire behaviour assessment: A case study. IFIP Advances in Information and Communication Technology, vol. 413, pp. 11-22, 2013, ISSN:1868-4238. Special issue: "Environmental Software Systems. Fostering sharing information". [6] García, M., Villagarcía, L., Contreras, S., Domingo, F. & Puigdefábregas, J. (2007). Comparison of three operative models for estimating the surface water deficit using aster reflective and thermal data, Sensors 7(6): 860–883. [7] Gao, Y. & Long, D. ,2008. Intercomparison of remote sensing-based models for estimation of evapotranspiration and accuracy assessment based on swat, Hydrological Processes 22: 4850–4869. [8] GRASS GIS Trac, changelog for, 2015. [9] GRASS GIS Trac, changelog for, 2015. [10] McGarigal, K., and B. J. Marks. 1995. FRAGSTATS: spatial pattern analysis program for quantifying landscape structure. USDA For. Serv. Gen. Tech. Rep. PNW-351 [11] Mitas, L., and Mitasova H., 1988, General variational approach to the approximation problem, Computers and Mathematics with Applications, v.16, p. 983-992. [12] Mitas, L., and Mitasova, H., 1998, Distributed soil erosion simulation for effective erosion prevention. Water Resources Research, 34(3), 505-516. [13] Mitasova, H. and Mitas, L., 1993: Interpolation by Regularized Spline with Tension: I. Theory and Implementation, Mathematical Geology, 25, 641-655. [14] North Carolina State University, Geospatial Modeling Course, GIS/MEA582, 2015. [15] Petras, V., Gebbert, S., 2014. Testing framework for GRASS GIS: ensuring reproducibility of scientific geospatial computing. Poster presented at: AGU Fall Meeting, December 15-19, 2014, San Francisco, USA. [16] Petrasova, A., Harmon, B., Petras, V., Mitasova, H., 2014. GIS-based environmental modeling with tangible interaction and dynamic visualization. In: Ames, D.P., Quinn, N.W.T., Rizzoli, A.E. (Eds.), Proceedings of the 7th International Congress on Environmental Modelling and Software, June 15-19, San Diego, California, USA. ISBN: 978-88-9035-744-2 [17] Raghavan, v., Choosumrong, S., Yoshida, D., Vinayaraj, P., 2014. Deploying Dynamic Routing Service for Emergency Scenarios using pgRouting, GRASS and ZOO. In Proc. of FOSS4G Europe, Jacobs University, Bremen, Germany, July 15-17, 2014. [18] Rocchini, D., Neteler, M. ,2012. Let the four freedoms paradigm apply to ecology. Trends in Ecology & Evolution, 27: 310–311. [19] Rodriguez-Aseretto, D., de Rigo, D., Di Leo, M., Cortés, A., and San-Miguel-Ayanz, J., 2013. A data-driven model for large wildfire behaviour prediction in europe. Procedia Computer Science, vol. 18, pp. 1861-1870. [20] de Rigo, D., Rodriguez-Aseretto, D., Bosco, C., Di Leo, M., and San-Miguel-Ayanz, J., 2013. An architecture for adaptive robust modelling of wildfire behaviour under deep uncertainty. in Environmental Software Systems. Fostering Information Sharing, ser. IFIP Advances in Information and Communication Technology, J. Hˇrebíˇ cek, G. Schimak, M. Kubásek, and A. Rizzoli, Eds. Springer Berlin Heidelberg, 2013, vol. 413, pp. 367-380. [21] Rothermel, R. C., 1983. How to predict the spread and intensity of forest and range fires. US Forest Service, Gen. Tech. Rep. INT-143. Ogden, Utah. [22] Suleiman, A., Al-Bakri, J., Duqqah, M. & Crago, R. ,2008. Intercomparison ofevapotranspiration estimates at the different ecological zones in jordan, Journal of Hydrometeorology 9(5): 903–919. [23] Timmermans, W. J., Kustas, W. P., Anderson, M. C. & French, A. N. ,2007. An intercomparison of the surface energy balance algorithm for land (sebal) and the two-source energy balance (tseb) modeling schemes, Remote Sensing of Environment 108(4): 369 – 384. [24] Xu, Jianping, 1994. Simulating the spread of wildfires using a geographic information system and remote sensing. Ph. D. Dissertation, Rutgers University, New Brunswick, New Jersey.
Technical Report
Full-text available
Description Fit linear and generalized linear mixed-effects models. The models and their components are represented using S4 classes and methods. The core computational algorithms are implemented using the 'Eigen' C++ library for numerical linear algebra and 'RcppEigen' ``glue''.
Full-text available
Land that is of great value for conservation can also be highly suitable for human use, resulting in competition between urban development and the protection of natural resources. To assess the effectiveness of proposed regional land conservation strategies in the context of rapid urbanization, we measured the impacts of simulated development patterns on two distinct conservation goals: protecting priority natural resources and limiting landscape fragmentation. Using a stochastic, patch-based land change model (FUTURES) we projected urbanization in the North Carolina Piedmont according to status quo trends and several conservation-planning strategies, including constraints on the spatial distribution of development, encouraging infill, and increasing development density. This approach allows simulation of population-driven land consumption without excluding the possibility of development, even in areas of high conservation value. We found that if current trends continue, new development will consume 11% of priority resource lands, 21% of forested land, and 14% of farmlands regionally by 2032. We also found that no single conservation strategy was optimal for achieving both conservation goals. For example, strategies that excluded development from priority areas caused increased fragmentation of forests and farmlands, while infill strategies increased loss of priority resources proximal to urban areas. Exploration of these land change scenarios not only confirmed that a failure to act is likely to result in irreconcilable losses to a conservation network, but that all conservation plans are not equivalent in effect, highlighting the importance of analyzing tradeoffs between alternative conservation planning approaches.
Full-text available
Scientists spend an increasing amount of time building and using software. However, most scientists are never taught how to do this efficiently. As a result, many are unaware of tools and practices that would allow them to write more reliable and maintainable code with less effort. We describe a set of best practices for scientific software development that have solid foundations in research and experience, and that improve scientists’ productivity and the reliability of their software.
Tools for performing model selection and model averaging. Automated model selection through subsetting the maximum model, with optional constraints for model inclusion. Model parameter and prediction averaging based on model weights derived from information criteria (AICc and alike) or custom model weighting schemes. [Please do not request the full text - it is an R package. The up-to-date manual is available from CRAN].
Open source software is often seen as a path to reproducibility in computational science. In practice there are many obstacles, even when the code is freely available, but open source policies should at least lead to better quality code.
We present a multilevel modeling framework for simulating the emergence of landscape spatial structure in urbanizing regions using a combination of field-based and object-based representations of land change. The FUTure Urban-Regional Environment Simulation (FUTURES) produces regional projections of landscape patterns using coupled submodels that integrate nonstationary drivers of land change: per capita demand, site suitability, and the spatial structure of conversion events. Patches of land change events are simulated as discrete spatial objects using a stochastic region-growing algorithm that aggregates cell-level transitions based on empirical estimation of parameters that control the size, shape, and dispersion of patch growth. At each time step, newly constructed patches reciprocally influence further growth, which agglomerates over time to produce patterns of urban form and landscape fragmentation. Multilevel structure in each submodel allows drivers of land change to vary in space (e.g., by jurisdiction), rather than assuming spatial stationarity across a heterogeneous region. We applied FUTURES to simulate land development dynamics in the rapidly expanding metropolitan region of Charlotte, North Carolina, between 1996 and 2030, and evaluated spatial variation in model outcomes along an urban–rural continuum, including assessments of cell- and patch-based correctness and error. Simulation experiments reveal that changes in per capita land consumption and parameters controlling the distribution of development affect the emergent spatial structure of forests and farmlands with unique and sometimes counterintuitive outcomes.