[mapserver-dev] Scales and raster maps

Fri Aug 9 01:51:22 PDT 2013

Hey Frank,

On 9 August 2013 10:31, Frank Broniewski <brfr at metrico.lu> wrote:
> Thomas,
>
> I am very sorry for not being clear. This must be a language issue on my
> side.
>
> I am not so sure on what you mean with ground unit ambiguity. I thought the
> ground units for a map request are determined by the CRS? If I request a map
> in WGS84, ground units are DD, a request for the Luxembourg National
> reference system (EPSG:2169) would be in meter and for the dreaded Global
> Mercator also. And there is already the UNITS keyword which specifies the
> units of the map coordinates.
>
> So if my mapserver receives a request for a map in EPSG:2169 with the extent
> 80000,80000,90000,90000 and a width and height of 500x500 pixels I can tell
> that my resolution (meters per pixel) is 20. Or in other words, I'm packing
> 10km in 500 pixels. For me, from a cartographic standpoint, it is easy to
> generate styling rules from requests like this one.
The ambiguity I'm referring to is that if you are supporting multiple
clients, you are also possibly supporting multiple SRSs and this means
multiple ground units. i.e. there's an ambiguity when you've defined
your map resolutions in meters per pixel, and you receive and
epsg:4326 request which is expressed in degrees per pixel (even worse
in this case: the mapping from degrees per pixel to meters per pixel
is going to be latitude dependant).
As Edward pointed out, we can go back to look at layer->units or
map->units to determine which ground unit is being used to define the
layer min/max resolutions. This is already a first ambiguity.
Secondly, the layer->units and map->units are more or less used to
define the ground units of the data sources, i.e. might be completely
different than what is needed to determine min/max layer resolution
limits you are talking about (which need to be consistant throughout
layers).

tl;dr : using resolution instead of scale is preferable for digital
maps in the vast majority of cases. however there are a number of
corner cases that need to be addressed.

another point: if you factor out dpi completely and only rely on
pixels per unit, how do you differentiate a retina request from a
72dpi request? As you're going to serve the same image, you're going
to end up with either too small or too large features depending on the
device.

--
thomas

>
> Problems arise when a client software comes into play that doesn't have the
> same concept for scales/dimensions as Mapserver does, like my QGIS example
> from the last post. Then the client gets maybe a wrong map delivered and
> this can be avoided when we had a scale independent mechanism for
> determining style based "breakpoints".
>
> I hope this is a bit more clear,
>
> Frank
>
>
>
>
> Am 2013-08-09 10:07, schrieb thomas bonfort:
>
>> Frank,
>> While I again agree with all you have said, you have not responded as
>> to how to treat the "ground unit" ambiguity. If I'm not mistaken, you
>> are proposing to replace the dpi ambiguity with a ground unit one.
>>
>> --
>> thomas
>>
>> On 9 August 2013 10:02, Frank Broniewski <brfr at metrico.lu> wrote:
>>>
>>> Hi Thomas,
>>>
>>> Am 2013-08-09 09:11, schrieb thomas bonfort:
>>>
>>>> Frank,
>>>> While I agree with you that the concept of "scale" is not suited for
>>>> digital maps, I'm not seeing it go away as it is the the only metric
>>>> that gives you a tangible sense of "what should I be displaying when
>>>> I'm zoomed in at a given level". more inline...
>>>
>>>
>>>
>>> my point is probably a bit academic and I really don't want to replace
>>> the
>>> (existing) scale systems in our web mapping applications. Problem is that
>>> the scale that you are using for the determination of what you want to
>>> display is totally wrong. When you deliver a map with a certain pixel
>>> resolution (e.g. 800x800 pixels) it's scale varies greatly depending on
>>> the
>>> output device. OGC defines a pixel with 0.28^2 mm, the Iphone has a pixel
>>> size of 0.0779^2 mm. So the "real" dimension of map is totally different
>>> depending on the device and therefore the scale.
>>> It's maybe the word scale that bothers me here, because it is not *the*
>>> map
>>> scale we refer to when we talk about maps, but it is a ratio of real
>>> world
>>> dimensions against a internal setting.
>>>
>>>>
>>>> snip ...
>>>
>>>
>>>
>>>
>>>> 1. Openlayers allows you to work with scale *and* resolution, which is
>>>> even more confusing.
>>>
>>>
>>> I know, it wasn't probably the best example taken but I think this is
>>> something to consider. A two way approach ...
>>>
>>>
>>>> 2. While pixels per ground unit is great for tiling because "ground
>>>> unit" is strictly defined by your grid definition, this does not fit
>>>> with a mapserver where you want to define scale-based ruled rendering,
>>>> but you have no idea what the ground unit will be (i.e. depending on
>>>> the request, it can be meters, degrees, feet, miles, nautical
>>>> miles....)
>>>
>>>
>>>
>>> Again, your scale is based on wrong assumptions because you do not know
>>> the
>>> output mediums resolution at all. Leaving the DPI/PPI issues aside, you
>>> simply cannot know what dimension your map will span on the output
>>> medium.
>>> Scales need to be precise. If a user requests a map at a scale of e.g.
>>> 1:20.000 the scale needs to be the same across all output mediums. With
>>> paper maps (and PDF) it's easy because you know the dimensions. But with
>>> raster maps it's virtually impossible to know.
>>> And I think it is more precise, better easier, to define a style for a
>>> pixel
>>> with 5mx5m resolution than for a somewhat vague 1:20.000 scale (numbers
>>> are
>>> not related).
>>>
>>> Other problems arise when a client software doesn't have the same
>>> assumption
>>> about the DPI setting like Mapserver has (or you have defined in the
>>> Mapfile). I found this post on the QGIS hub [1] where the map layout is
>>> broken because QGIS does have another view on the clients DPI as the OGC
>>> has. Because then 1:20.000 isn't 1:20.000 anymore and the wrong map gets
>>> delivered to the client.
>>>
>>>
>>>>>
>>>>> Would this be possible to implement along the DPI/scale setting? I
>>>>> think
>>>>> this would be handy to have and would be also more future proof than
>>>>> the
>>>>> DPI/scale setting. Since this measurement is not based on an arbitrary
>>>>> estimated value but is based on real output dimensions that is the
>>>>> pixel.
>>>>> Well, most of the time I mean ;-)
>>>>
>>>>
>>>>
>>>> What are you proposing concretely ?
>>>
>>>
>>>
>>> What I would like to -have- see as an addition is another way of defining
>>> my
>>> cartography rules. Something like the resolutions from Mapproxy (to leave
>>> out OpenLayers for this time, even if Mapproxy isn't a cartography tool
>>> by
>>> itself) . I'd find it really helpful to make my styles dependent on pixel
>>> to
>>> ground units ratio than scales. May it be feet per pixel, meters, km or
>>> whatever. To speak in Mapfile terms I'd like to have an addition to the
>>> MIN-
>>> and MAXSCALEDENOMINATOR parameter. Something like a MIN- and
>>> MAXRESOLUTIONDENOMITATOR. The resolution itself is always easily
>>> calculable
>>> because you know the map's extent and the pixel output size. Having this
>>> mechanism available would decouple Mapservers output from any assumptions
>>> about the client.
>>>
>>> Just to repeat it: I don't think it is feasible to remove the scale based
>>> mechanism from Mapserver, what I think would be helpful is a more device
>>> neutral styling determination mechanism as an addition.
>>>
>>>
>>> thanks,
>>>
>>> Frank
>>>
>>>
>>> [1] http://hub.qgis.org/issues/6430
>>>
>>>
>>>>
>>>> regards,
>>>> thomas
>>>>
>>>>>
>>>>>
>>>>> Many thanks for reading,
>>>>>
>>>>> Frank
>>>>>
>>>>>
>>>>> [1]
>>>>>
>>>>>
>>>>> http://osgeo-org.1560.x6.nabble.com/MapProxy-Seeding-in-a-certain-scale-tp5071082p5071464.html
>>>>>
>>>>> --
>>>>> Frank BRONIEWSKI
>>>>>
>>>>> METRICO s.à r.l.
>>>>> géomètres
>>>>> technologies d'information géographique
>>>>> rue des Romains 36
>>>>> L-5433 NIEDERDONVEN
>>>>>
>>>>> tél.: +352 26 74 94 - 28
>>>>> fax.: +352 26 74 94 99
>>>>> http://www.metrico.lu
>>>>> _______________________________________________
>>>>> mapserver-dev mailing list
>>>>> mapserver-dev at lists.osgeo.org
>>>>> http://lists.osgeo.org/mailman/listinfo/mapserver-dev
>>>>
>>>>
>>>>
>>>>
>>>
>>>
>>> --
>>> Frank BRONIEWSKI
>>>
>>> METRICO s.à r.l.
>>> géomètres
>>> technologies d'information géographique
>>> rue des Romains 36
>>> L-5433 NIEDERDONVEN
>>>
>>> tél.: +352 26 74 94 - 28
>>> fax.: +352 26 74 94 99
>>> http://www.metrico.lu
>>
>>
>>
>
>
> --
> Frank BRONIEWSKI
>
> METRICO s.à r.l.
> géomètres
> technologies d'information géographique
> rue des Romains 36
> L-5433 NIEDERDONVEN
>
> tél.: +352 26 74 94 - 28
> fax.: +352 26 74 94 99
> http://www.metrico.lu