Relations in OSM data could be Multipolygon, which represent different kind of land-uses such as lakes, rivers, forest and so on. The question is: there are not tags inside relations which can be used to identify land-uses. But, obviously, the OSM tiles can tell different land-uses. How do they do it?
In these cases, the landuse tags are probably on the outer way.
The decision tree for which tags to use (relation, outer ring, inner ring) in which cases (inner tags same/different from outer tags, different from relation tags?) can be hard to understand, and are different for each data user.
@Ldp: thank you for your kind help. I agree with you. It is hard to find the land-uses based on tags.
When we talk about spatial data quality, positional and attribute accuracy, completeness and consistency are three key concepts. Now OSM data has covered most of the world, which means the completeness is good. And when OSM data is used to analyze urban morphology, the accuracy is also good enough. However, the data consistency is not so good because the lack of standard when collecting OSM data. As Ldp said, it’s hard for users to extract the corresponding information. I think it worth thinking for OSM project.