What strategies, criteria, or rules do you use for selecting coordinate systems for
(I humbly offer my reply to a related question about watershed analysis as an example of the considerations involved in (b).)
What are the pitfalls to watch out for?
Links to Web sites you find particularly helpful in this regard are welcome.
I discovered FlexProjector yesterday. It interactively lets one explore various world projections, tweak their parameters, and even invent new projections while displaying results on screen immediately, complete with Tissot Indicatrix (though I don't know yet if they're approximated or accurate).
Flex Projector is open source (GPL2) and cross platform (linux, mac, windows), written by Bernhard Jenny, Institute of Cartography ETH Zurich. Truly a wonderful gift to the geospatial professions and community.
a. It is good practice to store a version of the data in the projection in which it was captured. Re-projection can be a lossy process, and it is important to have the original. I have a preference for storing in WGS84 for the sake of simplicity.
b. Depending on your software you may need to re-project into a metre based projections such as UTM. Support for native geodetic is being added (PostGIS, SQL server). For raster analysis it can be preferable to keep the data in the format it was supplied to avoid data loss through interpolation.
c. Web mapping systems have standardised on spherical mercator, and to overlay tiles they need to be in this projection. For overlaying vector on google maps you use WGS84. If you are not targeting web mapping there will usually be a local projection that is standard.
For detailed, beautiful and well presented information on nearly all aspects of map projections at world scales I cannot praise Carlos Furuti's Map Projection pages enough.
The great amount of information there can be daunting and scares people off sometimes, so here are two noteworthy starting pages to get you hooked: Assessing and Measuring Distortion using Tissot Indicatix. All the circles in the following image comprise the same area and distance on the ground. (Matt Perry has a short gdal/ogr script for generating tissot shapes here). Secondly see the Summary of projections table with thumbnails.
Edited May 2021 to update urls. Sadly Carlos Furuti's reference pages are now only available via web.archive.org (but how wonderful we have that!)
Our strategy is to use the same for a, b and c. Reprojection is costly.
(We store, analyze and display data in the projection most used by our users - our national UTM zone - or one of them actually. This being Norway, any Mercator projection is distorting much. Still, the WGS84/Google projections EPSG:4326/EPSG:900913 are the other relevant ones we have to deal with. YMMV)
Without going into a lot of details (which can be obtained from some of the resources mentioned in the earlier answers) the standard cartographic answer is "it depends on your use and your audience." All projections distort at least one of the following: shape, area, or direction. For instance, if your analysis requires acccurate measurements of area, you need to use a projection that preserves area. Your choice will also depend on the geographic area of interest: is it large (the entire earth, an entire continent) or relatively small (a city, county, or even a region in some cases). UTM and State Plane coordinate systems (based on projections) may work well in these areas, but they are useless across an entire continent. Your audience is also important because they may have a particular projection that they are accustomed to seeing (even if they don't know it), or a particular standard that they area expecting. One thing for sure though, even if you store your dataset in geographic coordinates, I wouldn't recommend ever using GC for an actual cartographic representation unless you absolutely had to -- shape, area, and direction are ALL distored, and besides, the resulting map is likely to be ugly.
As with Relet and Jonatr we use a local coordinate system if at all possible, which in our case is Albers. Our reasoning:
All coordinate systems are arbitrary. By a coordinate system, we simply mean an alphanumeric system by which the positions of geographic objects can be be unambiguously described. By arbitrary, we mean there's no magic to how these systems were developed...they are completely contrived. Their sole purpose is to enable users to unambiguously describe an object's position in geographic space. It doesn't matter what system you use: spherical (Global Reference System) or planar (Albers, Lambert, UTM, State Plane, etc, etc...), as long as you can place objects in the correct location.
In summary:
- a coordinate system allows you to describe an object's geographic position,
- the object's position is important,
- how the object's position is described (i.e. coordinates) is not important.
The only real issue for us is that if data from a variety of sources are to be integrated, they must all reside in the same coordinate plane. Because of the theoretical limitations of UTM:
- UTM zones are limited to 3° on either side of the central meridian,
- mapping beyond the 3° limitation results in unacceptable aerial distortion
- the Yukon spans 4 UTM zones...
...and the simple fact that our users wish to be able to:
- integrate a variety of spatial data,
- create maps that span multiple UTM zones
...but don't give a hoot about the actual coordinates, we have abandoned UTM for Albers.
Instrumental in setting this standard was "Standard for the Use of Map Projections in British Columbia for Resource, Cultural and Heritage Inventories Ministry of Environment, Lands and Parks and Ministry of Forests for the Digital Data Working Group Resources Inventory Committee" (link) (how's that for an unwieldy title? :)
I responded to a not unrelated question on QGIS forum earlier:
It would depend on which CRS you are using. All coordinate systems warp reality in one way or another. they either warp the graphic depiction so that measurement is localized and more accurate for a "smallish" area, or they warp the measurement so that we like what we see. As in the US48 albers is a common "picture" of the US with a nice curve on canadian border, and Maine bigger than Texas, curve nicely centered somewhere in the midwest. However there is a relatively small area (in the center) where measurements can be made accurately. UTM and Stateplane are made for the US to localize and allow accurate measurement. Yes the geodetic measurement will always be different than the object length unless you build the object with 3d measures. I am not familiar with qgis measurement methods but you would want to use a geodetic measurment as the accurate one. It will come closest (understand that the coordinate system you are using utilizes an approximation of the ellipse of the earth where you are, and that could have quite a bit of + or - built in) to being the same measurement as the real world. Hope this helps Also don't forget that lat long is not a coordinate system, it is an angular system (thus deg units)
...
when I calculate the length of a polyline with the field-calculator, the value is different from the one measured with the measurement-tool. I think that this is a problem with the geodetic differenze between the real length, and the flat projected lenth.
My question now is: Wich of the two values is the exact one?
Using the ESPG website you should be able to locate a local system that will give both graphic quality and measurement usability.Spatial reference
Futher discussion on lat,lon should follow now...Lat Lon wiki
I have given the example that you take a basketball and hold a dowel (stick) pointing straight to the center (of the basketball). when you imagine the stick crossing through the center of the earth and through the intersection of equator/greenwich mean median that would be lat 0, lon 0. Moving the dowel to the east or west increases the angle from 0,0 either by positive numbers or in the case of the western hemisphere by - or negative numbers -100 lon, 52 lat. this is 100 degrees measured from the center of the earth west of greenwich mean median, and 52 degrees north of the equator (median). Take a basketball and a rod and do it yourself. It is really helpful to understand lat lon.
