Hello everybody,
does anyone know an easy/possible function/way/method/package to cluster spatial objects (polygons in my case, a grid of 500mx500m cells) with an equal number of objects per cluster using QGIS or R?
I'm looking for something like ArcGIS's Grouping analysis tool (http://pro.arcgis.com/en/pro-app/tool-reference/spatial-statistics/how-grouping-analysis-works.htm).
I'm attaching an example of what I mean but, in this example, the number of cells per cluster is not the same....
Regards
Hi Maurizio.
Interesting question, I think that results might depend on the dimension of your data. If you have just coordinates then it is easier (check e.g. https://www.r-bloggers.com/spatial-clustering-with-equal-sizes/) for how to do this. If you have a rich data and you want them to be clustered and spatially constrained then the solution might be slightly harder to get. I would try DBSCAN algorithm or to weight the distance matrix of attributes by the distance matrix of spatial distances (coordinates) and cluster this.
I would not bet on QGIS itself, but there are clustering plugins available for general and spatial clustering (ClusterPoints etc.), but you might implement the python function (check Clusterpy and Scipy Point Clustering plugins and https://stackoverflow.com/questions/23588329/clustering-with-specific-sized-groups).
You can also take a look at research from Spatial Data Mining and Visual Analytics Laband their REDCAP software (http://www.spatialdatamining.org/software/redcap). They have really interesting publications and software about spatial data mining including clustering.
Best
Lukas
Hi Lukas and thank you for your reply.
Yes, I just want a spatial clustering of coordinates to develope something like this paper (http://www.nrcresearchpress.com/doi/abs/10.1139/cjfr-2016-0013#.WfrdyLynFBR).
I read the page you shared and it is not easy :)
So I think that the best thing to do is to implement a loop where:
0) Calculate the number of cells (i.e. objects) per cluster (n)
1) begin with a target cell
2) Calculate the spatial distance from each cell to the selected
3) order the distances ascending
4) select the n cells to attribute its cluster
5) remove the selected cells from the dataset
6) restart from point 1)
It will be quite long but it should work...
Cheers
Hi Maurizio!
Here's a link to an implemented example with equal sized clusters in R:
https://www.r-bloggers.com/spatial-clustering-with-equal-sizes/
Here's a an example of a modified K-means clustering algorithm for obtaining equal sized clusters: https://stackoverflow.com/questions/5452576/k-means-algorithm-variation-with-equal-cluster-size
... and the link to a paper explaining the R-package containing the K-means clustering algorithm in depth:
https://www.jstatsoft.org/article/view/v072i05/v72i05.pdf
Hope this helped!
Best regards,
Karolina
Dear Maurizio,
Another solution could be using a fuzzy clustering approach (e. g. cluster::fanny). Fuzzy clustering provides you with a membership degree of all clusters. This is accessible as the matrix membership in the fanny-object. You can sort the membership vectors for all clusters and cut them at the required count.
Best wishes,
Oliver
- Badges
- Science topic
- Similar topics
- Biological Science
- Bioecology
- Systems Ecology
- Landscape Ecology
- Connectivity