12 replies to this topic

[Steve B]

Hi--Am trying to digitize data from old ethnographic maps in a GIS using ArcMap. One of the less complex examples is Algeria which is attached. These are maps of tribes, clans, ethnicities...very fuzzy data.

The problem is that there are often labels (without polygons) of a variety of sizes and covering small and large areas. Any recommendations how how to digitize this info?

Options I've considered--
--Add a point feature where the middle of the label is
--Estimate the polygon that the label would cover
--Enter it into the GIS as annotation (but then you can't do queries on it nor apply domain rules)

Thanks,

Steve

Attached Files

•  algeria_tribes_1971.jpg   112.89K   62 downloads

[Hans van der Maarel]

Hi Steve, and welcome to Cartotalk.

My 2 cents:
1: Point feature. Not a good idea imho. By the spread/spacing of the label you can sort of convey the spread of whatever it is that you're mapping. Condense that back to a point and you lose all that information.

2: Polygon feature. Not good either. As you mention, the boundaries are fuzzy, by digitizing a polygon feature you're creating very distinct boundaries, which don't accurately represent reality. You may know that those cold hard lines are different in reality, but your dataset might end up being used by other people who don't know that, and consequently start using it the wrong way.

3: Annotation. Not a fan of that either. GIS packages seem to handle these in different way and it's a problem to exchange them between different GIS packages.

I think that this is a prime example of a map feature that cannot be properly represented in GIS. It's a fuzzy thing, and GIS is not really meant for that.

Hope I've made myself a bit clear.

[François Goulet]

I'd hate to have to do that one... not to mention that the legend specifies that the labels on the map are "Selected tribes" which seems to indicate that there's more... With polygons, you could enclose unnamed tribes into some other territory.

If I absolutely had to do it, I'd go with points, even thought as Hans mentioned, you'd lose the approximate geographical extent shown by the label... You could add a field with an attribute for the text size which will give you an idea of the importance of the tribe as well as a field with orientation that could give you a general, a bad - but general - approximation of the size of the territory. Note ideal, far from it, but at least, you'll won't place the Touareg in the North.

If you're digitizing polygons, I'd take a generic shape, rectangle or ellipse and have fate (even though we know it not really true) that the cartographer expanded it's labels to cover most of the territory (we can't estimate otherwise)...

Other than that, I'd be really confuse about how to do it... I'm not sure my options are optimal given the circumstances, but they're the only way of doing it I can think of...

Good luck and keep us informed! As a cartographer and historian, I'd love to find a good way to represent those so-so-so unclear boundaries I found sometimes on historical maps...

Attached Files

•  _algeria_tribes_1971.jpg   161.36K   29 downloads

[Steve B]

Hi Steve, and welcome to Cartotalk.

My 2 cents:
1: Point feature. Not a good idea imho. By the spread/spacing of the label you can sort of convey the spread of whatever it is that you're mapping. Condense that back to a point and you lose all that information.

2: Polygon feature. Not good either. As you mention, the boundaries are fuzzy, by digitizing a polygon feature you're creating very distinct boundaries, which don't accurately represent reality. You may know that those cold hard lines are different in reality, but your dataset might end up being used by other people who don't know that, and consequently start using it the wrong way.

3: Annotation. Not a fan of that either. GIS packages seem to handle these in different way and it's a problem to exchange them between different GIS packages.

I think that this is a prime example of a map feature that cannot be properly represented in GIS. It's a fuzzy thing, and GIS is not really meant for that.

Hope I've made myself a bit clear.

Good to be here, Han. Thanks for helping me explore these options which, I agree, point to the limitations of a GIS. It's ironic that GIS doesn't do fuzzy data well, but with the exception of perhaps man-made objects, it's all fuzzy! Will let you know what we end up doing.

Best,

Steve

[Steve B]

I'd hate to have to do that one... not to mention that the legend specifies that the labels on the map are "Selected tribes" which seems to indicate that there's more... With polygons, you could enclose unnamed tribes into some other territory.

If I absolutely had to do it, I'd go with points, even thought as Hans mentioned, you'd lose the approximate geographical extent shown by the label... You could add a field with an attribute for the text size which will give you an idea of the importance of the tribe as well as a field with orientation that could give you a general, a bad - but general - approximation of the size of the territory. Note ideal, far from it, but at least, you'll won't place the Touareg in the North.

If you're digitizing polygons, I'd take a generic shape, rectangle or ellipse and have fate (even though we know it not really true) that the cartographer expanded it's labels to cover most of the territory (we can't estimate otherwise)...

Other than that, I'd be really confuse about how to do it... I'm not sure my options are optimal given the circumstances, but they're the only way of doing it I can think of...

Good luck and keep us informed! As a cartographer and historian, I'd love to find a good way to represent those so-so-so unclear boundaries I found sometimes on historical maps...

François--Its good to know I'm not imagining this problem! I like the ellipse example you attached...at least it is spatial rather then a numerical indication of font size and orientation (not to mention amount of curvature, too!). I'll be running these thoughts by my colleagues here and will let you know what we end up doing. The objective of the map is to somehow convey: Who lives here?
--Steve

[Charles Syrett]

The map you posted (as your source material) looks clear and communicates effectively. Given the stated purpose of your map, why use GIS at all? If I was asked to do this project, I'd look for more samples of the kind of map you posted, and then create my own in a similar fashion.

Charles Syrett
Map Graphics
http://www.mapgraphics.com

The objective of the map is to somehow convey: Who lives here?
--Steve

[David Medeiros]

This is, to me, a clear example of the limitations of GIS for straight cartography… it doesn’t handle non-discrete data very well.

If it were me I’d be doing this map in Illustrator, or at the very least I’d be shipping the base layers out of the GIS for post production to Illy and doing the area labels there.

[David Medeiros]

--Enter it into the GIS as annotation (but then you can't do queries on it nor apply domain rules)

Should have added in my reply above that I don't really think one should be doing queries on data like this. If you estimate its extent you certainly can do a query but to what purpose? Estimated data of this type is just too vague to be used to answer spatial questions with any degree of accuracy. Aside from the inaccuracies of your estimation of extent you have to factor in that you don’t know anything about the original authors’ estimations.

[Steve B]

This is, to me, a clear example of the limitations of GIS for straight cartography… it doesn’t handle non-discrete data very well.

If it were me I’d be doing this map in Illustrator, or at the very least I’d be shipping the base layers out of the GIS for post production to Illy and doing the area labels there.

I should have explained that the ultimate output of this project is a web mapping application that will consolidate ethno-religious-tribal-language data on a world-wide scale from multiple sources. So a GIS, ultimately, needs to be a part of the solution. Otherwise, I'd definitely go with Illustrator!
Best regards.

[eli]

I work with indigenous data all the time, and it is often challenging and complex - which makes it all the more rewarding (imho).

I'll point you to an example from the BC Government of languages, which is a fuzzy representation of data. There's another example from the Musuem of Anthropology, which is a little bit more GIS-oriented - but notice those dangly lines! Lovely.

What I would recommend is finding other sources for the area - the one from the BC Gov't uses two main sources of data and then blends them. For example, I randomly chose a tribe from your map, Tuareg, and searched for it on Wikipedia. Low and behold, there's a map. Probably not every one will be that easy, but I sincerely doubt the map you shared is the only map of indigenous nations in Algeria.

Please do keep us updated - it's really interesting stuff! Good luck.

[natcase]

I like the approach National Geographic did on its Jamestown map a couple years ago (or was it last year?): generalized, soft-edged amoeboids. At some level, maps of non-nation-state territories ought to be more of a raster than a polygon-based system. Really, it would be pretty cool to do a raster-based political map, based on the degree of allegiance the inhabitants have to their nation-state. Saturated colors in Kent and Indiana, more faded in Tibet and North-west Frontier Province... some hard boundaries, some blurred and faded boundaries.

[eli]

This topic is old, but maybe this'll help someone in the future:

I came accross this map and thought it might help you with the line work - if in fact that's the way you chose to go.

http://www.globalhum...om/africahr.jpg

(Also, I was curious about the Jamestown map that Nat mentioned. The one I found was an interactive version here.)

[frax]

When doing queries etc, one has to keep in mind that the ethnicities also overlap and co-exist, so it wouldn't be so easy to display in one single raster. One would have to use one column for each group and have the raster value be the percentage of inhabitants in each pixel...

For practical purposes, I would try to digitize rough polygons where the labels should be. Those polygons should never be displayed.