Problems with ZONING/zoning.R. I’ve rewritten the script which now should work as expected.
This is a simple fix – but deserves a new sub version number (9.5.3) because its a crucial update. Scripts have now been updated to reflect the loss of gWidgetstcltk from the CRAN repositories and its replacement gWidgets2tcltk. Thanks Naveen for reminding me to do this!
Link to new version APoLUS 9.5.3
We recently began work on the EU-funded INTRANCES project (Grant agreement ID: 886050). Under INTRANCES, an integrated scenario model including transport, urban land use and air pollution wil be developed based on the SIMLANDER and APoLUS modelling tools.
Developments, publications and links to download open access resources, like datasets and tools used by the project, will be reported on this site on the dedicated project web page.
I just noticed that I can no longer install the gWidgetstcltk package because it is no longer maintained in the CRAN repository.
Without this package APoLUS will give you an error because quite a few scripts contain the commands:
Fortunanately, this error can be fixed by installing gWidgets2tcltk instead and correcting the scripts that caused the problem to require gWidgets2tcltk instead. Off hand, this includes SWITCH/switchb.R, NHOOD/selectmatlist.R and probably one or two others. You will easily be able to tell, because the script will fail telling you that gWidgetstcltk could not be loaded..
I’ll fix this in the next version of APoLUS..
The user guide has also been updated. It is included in the download package for the new version and can also be accessed separately here.
A new version of SIMLANDER, version 1.0.6, is now also available.
I was recently invited to Deakin University, Melbourne, Australia, by Professor Brett Bryan to give a course on land use modelling in R using the SIMLANDER/APoLUS framework to the Land Use Futures modelling group.
Seemed to go down well, students were very engaged and got on well with the model. Fun had by all, including me.
One bright participant noticed a stupid error in the accessibility calculation. This can be summarized as follows:
1. I’d mistyped White’s et al’s (1997) accessibility equation. The correct equation is to be found here on the updated accessibility page of this website.
2. I was entirely wrong about the way to put the coefficient into the equation. It’s a single number, not a reclassified distance map. Of course, if you don’t want to use the accessibility equation, you can of course, reclassify the distance map. But not both!
The correct procedure is described as follows here
Over the coming days I will correct this error in the script in SIMLANDER and APOLUS, and post an update when I do this…..
Full details on setting the neighbourhood effect using the Automatic Rule Detection (ARD) procedure in SIMLANDER are described by Majid Shadman here
Some major revisions and improvements in version 9.5, including:
- Calculate land use demand using a tendency curve, instead of just inputting final demands
- Improvements to accessibility and suitability calibration
- introduction of map comparison, cross tabulation and goodness-of-fit statistics (see MAPCOMPARISON folder) to help model calibration.
To download the new version, go here
Towards automatic calibration of neighbourhood influence in cellular automata land-use models
One of the hardest things to deal with in CA land use models is calibrating the neighbourhood effect. The shape of the distance decay curve (nrules) and the size of the cell neigbourhood (nsize) have a strong influence on the behaviour of land use simulations. In general, steep decay curves lead land use patches to clump together, while shallower curves give more dispersed effects, (see a note on calibrating the neighbourhood effect). The problem is not so much identifying the pattern you want to replicate, but knowing what settings to give to nsize and nrules. Usually, this is done simply by experimenting, i.e. running simulations with one set of rules, comparing the results, and going back to modify the rules. But not only is this really tedious and time consuming, it’s also not clear how you know when to stop. If you find a rules set that seems to fit, should you just go with it, or carry on experimenting until you find the best one? And is there really a single “best” set of rules our there? or will a range of different combinations of nsize and nrules give you the result you want?
A new paper, entitled “Towards automatic calibration of neighbourhood influence in cellular automata land-use models” published last week in Computers, Environment and Urban Systems, led by Majid Shadman Roodposhti, orginally of the University of Tasmania, now at Swinburne University of Technology (both Australia), presents a way to automate the process by employing a structured procedure to experiment with many different rules combinations and test the results against the reference maps using standard goodness-of-fit testing measures. The results are then ranked so the user can explore the values which give the best results. All of the code is R, and presented in supplementary material together with the paper, which is open access and can be downloaded for free from here.
simlandeR was cited by Simon Moulds, in his PhD thesis.
Moulds, S. (2016). Toward integrated modelling systems to assess vulnerability of water resources under environmental change (Doctoral dissertation, Imperial College London).
The thesis is available here:
Thanks Simon, I hope you found simlandeR interesting.