OpenStreetMap has many tags, inherited from a natural language with blurred meaning and definitions, depending on each mapper's understanding of associated natural language term. Also, many tags are representing more than one property of an object, such as, say, type of flora, populating particular area and presence of management of this area. (Good example is
In an ideal case, any classification system should have only "atomic" properties instead of "molecular", where several real properties are linked. (Good example is the recently introduced scheme with
leaf_cycle= - independent properties instead of linked ones.)
One of the extremely widespread tags with both bad features is
It belongs to
natural=* class, and it gives people an idea, that only natural habitats should be tagged with it. Therefore, we also have
landuse=forest, which means the same kind of habitat, but more related to man-made objects.
Actually, it creates the really huge problem. First, let's try to explain what exactly we should map with it.
Both natural wood and any kind of planted forest are areas of vegetation, dominated by trees. But there is no clear definition (in OSM), what does "dominated" mean.
natural=wood should, supposedly, tells us, that area of vegetation forms "natural habitat". But there is no clear explanation, what does "natural" mean.
landuse=forest should, supposedly, tell us, that area of vegetation is managed, or planted, or forms artificial habitat, or trees there are non-native. There is no clear explanation, is there any difference between
landuse=plant_nursery - difference is only implied, because nursery should be only used to plant young trees for sale or for planting it somewhere else for forestry management purpose.
So many variants, so many assumptions and lots of guessing.
Currently, in British OSM community, there is a process of
habitat=* tagging scheme discussion going on. Concepts and definitions there are mainly based on Joint Nature Conservation Committee works and publications, including Handbook for Phase 1 Habitat Survey. I like this approach, but there is a problem: ecologists usually have different goals for their projects than OSM community has. They can use plain (single dimension) classifications with all existing variants, represented by compound classes.
This approach has positive features: it's impossible to classify any object with non-existent set of properties, and it's also impossible to classify anything incompletely, because each class has full set of required properties and its values. But in OSM it doesn't make any sense: mappers are non-professionals, and often they can't evaluate all required features to use compound classes. That's why multi-dimensional classification with no mandatory properties makes much more sense in OSM.
Multi-dimensional classification means that you can have any set of independently determined properties. For example, tagging scheme can have
material= properties. It means, we can tag only color, only material or both. We also can add third property (say,
shape) to scheme completely independently. Single-dimension scheme will require total revision in this case, because adding another property will require making completely new classes.
Like, originally we had classes: "red_steel", "green_steel", "red_wood", "green_wood", Adding shape to single dimension classification will make it look like: "red_steel_round", "green_steel_round", "red_wood_round", "green_wood_round", "red_steel_square", "green_steel_square", "red_wood_square", "green_wood_square".
Looks extensive, right? So, multi-dimensional classification with independent properties is definitely more suitable for OSM. Then, lets try to establish basic properties for different types of tree vegetation, that will allow us to map almost everything we want without implications and assumptions.
First, we need fundamental tag to show, that "here are trees". Trees in general, nothing more. I believe, that
natural=wood can probably work for it. Its meaning is currently blurred enough to use it like very broad thing.
What to map with this tag? Answer is simple: any area, where trees grow, regardless of anything. Yes, if trees are sparse enough to see them independently, you can try mapping them as independent trees with
natural=tree. But even if you can, it doesn't mean you have to.
Ecologists and forestry management using canopy coverage as a criterion for classifying any area as "forest". But I think, for our purposes we can have certain scale of percentage ranges, for example:
- 0..25% of coverage - single trees, not recommended to tag as wood, recommended to tag single trees
- 25..50% of coverage - sparse wood
- 50..100% of coverage - wood
These values (single, sparse, normal) can be assigned to density key.
It's pretty easy to use these ranges, because if canopies covering a bit less than a half of territory, it's sparse, if more - normal wood.
Origin of wooded vegetation area can be tagged (if we know it), and it's pretty obvious, which values to use for this property:
But it only a kind of historical reference information, because 100 years old plantation of native trees is currently unlikely looks differently from completely natural forest, even in aspect of rows.
Therefore, we need something for current habitat state. It's only about how it looks and how similar is it to reference natural habitat. Based on ecological classifications and practical experience, we can define values for this key as:
- natural (which means similar to natural)
- semi-natural (which means, this area has significant traces of planting, cleanup, removal of dead branches, or, otherwise, being artificially planted, it has traces of succession, growth of new young trees and scrub)
- artificial (which means, it looks completely or almost completely different from natural habitat typical for this area).
These definitions are just a framework, but it explains the idea.
Management is another obvious property, describing the situation, but it could be tricky to find right values for this key, because management can consist of very different works. Therefore, I suggest using general tag like
managed=yes in case if it's managed and making separate scheme (considering namespace usage) for types of management, such as cleanup, fire protection, pest control and so on.
Obviously, these framework keys and values are only an example, because correct tagging scheme should be discussed and tested on different real cases. Also, definitions for values should be given in clear manner to avoid any contradictions, broadening of usage and so on. But as a framework, it gives good example of classification, based on independent "atomic" properties.
For sure, it doesn't solve certain problems completely. Like, it doesn't help to filter out tree groups within city limit (which should be done using spatial functions of SQL extensions). But it is potentially able to solve all controversy of real situations, like proper and simple description of difference between city park vegetation and wild forest vegetation.