search

ConservationInternational / trends.earth

trends.earth - measure land change
http://trends.earth
GNU General Public License v2.0
107 stars 45 forks source link

Nesting of land cover legend for imported local custom data #643

Closed gdaldegan closed 1 year ago

gdaldegan commented 2 years ago

I don't think Trends.Earth should have the rule below, given it is quite common to have distinct land cover classes between the initial and final layers. nesting

azvoleff commented 2 years ago

Can you tell me more what scenario we'd need this for @gdaldegan? My thought would be that if you have two different legends, then you shouldn't be calculating change among the two layers - I'm not sure how you'd know whether any "change" is due to actual change on the land, a change in class name between the two legends, or a change in class definition between the two legends.

gdaldegan commented 2 years ago

I'm trying to run Land Cover Change sub-indicator using imported custom LC layers for a small subset of the Colombian Amazon region. Importing LC 2000, the original data doesn't have the class 13 lc_2000

Class 13 is a non-forest natural vegetation, which in my interpretation likely represents forest pixels that were cleared at some point early to mid the 15y time-series and than abandoned, so began a natural regeneration succession - LC Legend: lc_classes

This scenario is very likely to happen in recently occupied natural landscapes.

When importing LC 2015, which now does have class 13, I match it to Grassland for a lack of another standard class. lc_2015

Then trying to run LC Change, I get that message lc_nesting

Thus I don't fully understand the rule, given I'm matching the original classes to the 7 standard UNCCD classes: lc_imported

azvoleff commented 2 years ago

Ah, ok. I see what you mean. Can you try saving the full legend from the 2015 data to a file, and then loading it from file when importing the 2000 data? This should then add that class to the legend even though it isn't in the file.

Also once #626 is merged, the best way to do this would be to first set a custom legend, and then import the land cover data. This would allow matching that legend directly and avoid the issue of different legends across years.

azvoleff commented 1 year ago

No longer an issue with new LC tool