DESCRIPTION
Module r.futures.calibration is part of FUTURES,
land change model.
It is used for calibrating certain input variables for patch growing
algorithm r.futures.pga, specifically patch size and compactness parameters.
The calibration process is conducted to match observed urban growth patterns
to those simulated by the model, including the sizes and shapes of new development.
The calibration is achieved by varying the values of the patch parameters,
comparing the distribution of simulated patch sizes to those observed
for the reference period, and choosing the values that provide the closest match.
For the details about calibration see below.
Patch size
As part of the calibration process, module r.futures.calibration
produces patch size distribution file specified in patch_sizes parameter,
which contains sizes (in cells) of all new patches observed
in the reference period. The format of this file is one patch size per line.
FUTURES uses this file to determine the size of the simulated patches.
Often the length of the reference time period does not match
the time period which we are trying to simulate.
We use the discount factor to alter the size of simulated patches
so that after the reference period they closely match the observed patterns.
During the simulation, this factor is multiplied by the patch sizes listed in the patch size file.
The values of discount factor can vary between 0 and 1,
for example value 0.6 was used by Meentemeyer et al. 2013.
Patch compactness
The shapes of patches simulated by FUTURES are governed
by the patch compactness parameter (Meentemeyer et al. 2013, Eq. 1).
This variable doesn't represent actual patch compactness, it is rather
an adjustable scaling factor that controls patch compactness
through a distance decay effect.
By specifying the mean and range of this parameter in module
r.futures.pga, we allow for variation in patch shape.
As the value of the parameter increases, patches become more compact.
Calibration is achieved by varying the values specified in compactness_mean
and compactness_range and comparing the distribution
of the simulated patch compactness (computed as
patch perimeter / (2 * sqrt(pi * area)))
to those observed for the reference period.
Meentemeyer et al. 2013 used mean 0.4 and range 0.08.
Calibration input and output
Calibration requires the development binary raster in the beginning
and end of the reference period (development_start and development_end)
to derive the patch sizes and compactness.
It is possible to set the minimum number of cells of a patch in patch_threshold
to ignore too small patches.
For each combination of values provided in compactness_mean,
compactness_range and discount_factor, it runs
module r.futures.pga which creates new development pattern.
From this new simulated development, patch characteristics are derived
and compared with the observed characteristics by histogram comparison
and an error (histogram distance) is computed.
Since r.futures.pga is a stochastic module, multiple
runs (specified in repeat) are recommended, the error is then averaged.
Calibration results are saved in a CSV file specified in calibration_results:
input_discount_factor,area_distance,input_compactness_mean,input_compactness_range,compactness_distance
0.1,1.01541178435,0.1,0.02,3.00000005937
0.2,1.26578803108,0.1,0.02,4.12442780529
0.3,1.17631210026,0.1,0.02,3.86904462396
0.4,2.31700278644,0.1,0.02,15.0569602795
0.5,1.08655152036,0.1,0.02,3.72484862687
0.6,2.97628078734,0.1,0.02,21.6358616001
0.7,3.61632549044,0.1,0.02,25.4492265706
0.8,2.72789958233,0.1,0.02,18.1083820007
0.9,2.45915297845,0.1,0.02,18.4500322711
0.1,1.05473877995,0.1,0.04,3.09321560218
...
Optimal values can be found by visual examination of the second and fifth columns.
Providing too many values in compactness_mean,
compactness_range and discount_factor results in very long computation.
Therefore it is recommended to run r.futures.calibration
on high-end computers, with more processes running in parallel using nprocs parameter.
Also, it can be run on smaller region, under the assumption
that patch sizes and shapes are close to being consistent across the entire study area.
For all other parameters not mentioned above, please refer to
r.futures.pga documentation.
EXAMPLES
SEE ALSO
FUTURES,
r.futures.pga,
r.futures.devpressure,
r.futures.demand,
r.futures.potential,
r.sample.category
REFERENCES
-
Meentemeyer, R. K., Tang, W., Dorning, M. A., Vogler, J. B., Cunniffe, N. J., & Shoemaker, D. A. (2013).
FUTURES: Multilevel Simulations of Emerging
Urban-Rural Landscape Structure Using a Stochastic Patch-Growing Algorithm.
Annals of the Association of American Geographers, 103(4), 785-807.
DOI: 10.1080/00045608.2012.707591
- Dorning, M. A., Koch, J., Shoemaker, D. A., & Meentemeyer, R. K. (2015).
Simulating urbanization scenarios reveals
tradeoffs between conservation planning strategies.
Landscape and Urban Planning, 136, 28-39.
DOI: 10.1016/j.landurbplan.2014.11.011
- Petrasova, A., Petras, V., Van Berkel, D., Harmon, B. A., Mitasova, H., & Meentemeyer, R. K. (2016).
Open Source Approach to Urban Growth Simulation.
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLI-B7, 953-959.
DOI: 10.5194/isprsarchives-XLI-B7-953-2016
AUTHOR
Anna Petrasova, NCSU GeoForAll
Last changed: $Date$