I am looking for advice of how to relate multiple objects in space. I have a set of N objects. Each of the N objects are surrounded by varying object types in the environment which are in the range A to Z, that is, 26 different types of objects. Furthermore, these surrounding objects in the environment can occur many times (e.g. A can occur 10 times whereas B may only occur once). Because its 2D space I can calculate the distance from each object in the set N to each surrounding object in the environment. Is there a way I can associate or find the best combination of surrounding objects which I can use to characterize the possible occurrence of objects in set N?
For each of the N objects, you might use some "select by location" command to select all objects within a fixed distance threshold, than add a new field to attribute an id to this configuration. In this way, you will have 26 configurations for each distance treshold you have fixed.
I am not sure this was what you wanted!
Thanks Daniele. Thats definitely part of it. I'm also looking for a way to find patterns for each of these distances, not from a pairwise comparison, but with many other features. For instance, it could be that 4 objects are always clustered near to the feature of interest. Using pairwise comparison you lose information on how surrounding features in the environment make the best combination of features to characterize an object of interest. Individually a set of objects of the same type may not seem to clusters around a feature of interest. However, if you add another related set and combine them they may now make sense with respect to clustering around an object of interest. I've seen paper focus on colocation of two objects, and I'm sure there are others extending this to N features of interest, but I'm possibly not searching the right way and need some guidance herein. I'm not sure if this makes sense but I'll be all to happy to explain things once again from another point of view.
Many thanks
yeah, very interesting issue.
I was wondering if it ould be possible to use standard cluster analysis, using some distance variables.
also: there might be particular spatial autocorrelation techniques looking for correlated combinations of objects' variables.
I thought about using standard cluster analysis but then you may lose the unique contribution of each of the surrounding variables types and combinations of these in characterizing various objects of 'N' type. With respect to distance, definitely this is a major factor. Interestingly, I didn't think about the autocorrelation idea which makes a great deal of sense. I'm still looking for relevant literature but I it really is an interesting issue. Thanks for the ideas Daniele
- Badges
- Science topic
- Similar topics
- Geoscience
- Geoinformatics