global 10 m landuse data

123 views
Skip to first unread message

Justin

unread,
Jun 24, 2021, 2:08:19 PM6/24/21
to datameet

Justin

unread,
Jun 24, 2021, 2:08:42 PM6/24/21
to datameet
On Thursday, June 24, 2021 at 2:08:19 PM UTC-4 Justin wrote:
g10m2020lulc.png

Sharad Lele

unread,
Jun 24, 2021, 10:23:20 PM6/24/21
to datameet
Hi Justin,
Thanks for this. I downloaded a tile, but could not find the legend information anywhere. Please advise.
Sharad

J M

unread,
Jun 24, 2021, 10:46:02 PM6/24/21
to data...@googlegroups.com

--
Datameet is a community of Data Science enthusiasts in India. Know more about us by visiting http://datameet.org
---
You received this message because you are subscribed to a topic in the Google Groups "datameet" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/datameet/w-Bz-XMqQUg/unsubscribe.
To unsubscribe from this group and all its topics, send an email to datameet+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/datameet/4fccab33-95a2-45a7-8def-dd0a0376c144n%40googlegroups.com.

Sharad Lele (शरच्चंद्र लेले)

unread,
Jun 25, 2021, 1:59:50 AM6/25/21
to data...@googlegroups.com, J M

Works! Thanks, Justin!

Sharad

Sharad Lele

unread,
Jun 25, 2021, 5:11:09 AM6/25/21
to datameet
Since several folks emailed me one-on-one, I am adding this clarification:

1. The link below only gives information on what the map is about, how it was made, and (importantly) what its accuracy levels are.

2. To actually download tiles for your area, you need to go to the downloader page, which for some reason does not seem to work from the above page. You need to go to:

3. You can also download the whole set of tiles at once (60GB) 

4. Upon perusing the documentation, and doing some ground truthing of my own for areas I have already mapped at high resolution, my observations are:
a) the 10-fold classification is generally too coarse for many applications. E.g., 'trees' includes horticultural plantations (coconut, areca, eucalyptus, rubber, coffee, etc) along with forest of all densities and types. 'cropland' includes fallows.
b) the documentation itself acknowledges that the accuracy for certain classes is low (~50%): grassland, and scrubland. Thus, most forest edges are classified as 'bare', when they are likely to be grassland or scrubland.
In other words, the spatial resolution may be good, but the class-resolution (and associated accuracy) is not good enough for many applications. Not a criticism of this effort, but something to keep in mind when using the product.

Again, a big thanks to Justin for bringing this source of data to our attention and hosting links on his website (https://revolutionarygis.wordpress.com/) which has tonnes of other goodies as well!

Best,
Sharad



Reply all
Reply to author
Forward
0 new messages