Mapping landcover where it is not clear where farmland is

I am mapping around here: OpenStreetMap All that landcover is my work. Take a look at the satellite imagery. The problem is that for some parcels (they are divided by tree rows) I can not determine whether they are used for crops or they are abandoned and left for weed to grow.

I want the landcover here to look rich, like in European cities. Examples: Moscú, Riga, Londres.

Any suggestion of what to do for these parecels?

I figured one method is to compare different satellite photos. That will show more change in the farmland but this does not leave all cases clear.

1 Like

You can look if your government has open data on farmlands which you can use for cross reference (mind the license). In the netherlands, we have a digital farmland map that we are allowed to use.

1 Like

Hi, Farmland is not covered in growing crops all the time. Many farms follow a crop rotation method of production, including land being left fallow, (fallow is a farming technique in which arable land is left without sowing for one or more vegetative cycles). Crop rotation also includes planting grass for animals, often having animals grazing there. After this fertilization, the field is again used for crop growing. Even if there are weeds growing in a field it’s still land used for farming, farmland.

My point is looks can be deceiving, farmland is farmland no matter what it looks like. Also, OSM should reflect ground truth not merely something that looks “rich”.

My suggestion would be to go out and have a look.


Of course. In this case, the point of mapping such ground truth is that it look pretty. Why would I care about a patch of trees in the middle of a farm, if not to make my map pretty?

Excellent. This is useful information.

My general impression is that you’ve been too generous with natural=scrub, but this is OSM and everything can be refined. A lot of this looks like tall herb vegetation rather than scrub (you can see some informal paths close to residential areas) and may be fallow land as @BCNorwich suggests. Unfortunately we don’t have a good tag for tall herb vegetation on OSM, even though it is common in both natural and intensively farmed landscapes.

The area with lots of tracks close together looks like a new housing subdivision, so landuse=construction + construction=residential might be more appropriate.

Another way you could refine this would be to look at Sentinel imagery. It has less resolution, but is collected more frequently and has multi-colour bands which assist in interpreting vegetation. Unfortunately this requires some significant disk space, familiarity with a GIS program such as QGIS, and familiarity with how to find the imagery.


Currently Im seeking tutorials on supervised image classification with qgis. However, im not sure about how to map areas with overlapping vegetation. For instance a pach of woodland within a grassland. Should I use relations, or just map one on top of the other for easier maintenance? Many mappers are unfamiliar with relations and might end up breaking them while editing

What tagging do you suggest instead?

Note that first I tagged the whole municipio (roughly translated to county) as scrub, then carve holes for more specific uses.

Large areas look more a mix of forest, grass and farmland rather than scrub. Especially if you take the neighboring buildings as comparison for the crowns of the vegetation.

Also, I would personally cut out major roads and all settlements (tagged with landuse) from the scrub. That also prevents these massive multipolygons.

Indeed they are. The massive multipolygons are first approximation. I replace it incrementally with detailed mapping. See around here: OpenStreetMap

Great. Also, it is in general not needed to create a multipolygon for an area that can be drawn by a closed way like this one: Relation: 16177445 | OpenStreetMap . Only when where are holes in a landuse or it consists out of to many node (2000 I thought), a multipolygon is needed. This reduced complexity and makes editing easier. I generally start by splitting large areas based on roads or other large features and then cut it smaller and smaller

This is to share ways between adjacent areas. Nearly all ways in my landcover cited above are shared by 2 multipolygons.

See Mapping landcover with multipolygons
for the discussion on that.

What people think about my method of choice is none of my concern. It works and works very well. That is what I care about. I do not feel like reading tens of screens worth of messages on a non-problem.

You’re the one who asked for suggestions, what did you expect!

Read the OP and you will see it is unrelated to whether to use multipolygons.