17 replies to this topic

[Dany]

Hello everyone

It's my first post. Any comment on the attached hillshades are welcome.

My aim was to create a decorative map of the Alps, with no text. I used free data for the continent (srtm) and the sea (gebco), and followed Tom Patterson's advice in his section on 'resolution bumping' (http://www.shadedrel...om/dem/dem.html).

There are still two issues I am not able to address: (1) eliminate the "terraced" colour gradient in flat areas (look at the southwest of the Alps, on the Italian side); and (2) make the coarse scale shading sharper (as it was done by downgrading the DEM's resolution, angles are rounded up).

Thanks for your views

Attached Files

•  AlpsLowRes.jpg   425.27K   343 downloads
•  AlpsZoomLowRes2.jpg   505.11K   229 downloads

[CentralAmericaExpe]

...I read quickly through Tom's very detailed topographic rendering article. Lots and lots of detail there.

In my mind, I think that more info is needed on your precise procedures that got you to where you are now...

Something seems missing in those 2 topography renders; it appears that the raw DEM is showing through too much (the hillshaded/modeled raster seems not to be very present: but I think there is no modeled hillshade derived from the raw DEM?). Are you using Photoshop lighting on the DEM? What SRTM product are you using (3 arcsecond?) ? How did you change the DEM resolution precisely (Photoshop interpolation? Another resampling software?).

I recall years ago trying to use Photoshop to light up a DEM, and I felt (then) that it was a very labor-intensive exercise when compared with other GIS (commercial and open-source) raster software. However, I have to admit that I know may way around Photoshop fairly well, but I'm no "expert" Photoshop wonk

Cheers,
Derek-

[Dany]

Thanks a lot for your feedback. I agree 'something seems to be missing', but what???
Regarding procedure, my first step was to create three hillshades of the DEM with ArcMap:

(1) with the 'raw' 3 arcsecond data (good for small valleys)
(2) with the raw dem downgraded to a 4 minute grid (representing major alpine structures)
(3) with the raw dem downgraded to an intermediate resolution, broadly around 30 arcseconds (emphasizing intermediate size valleys)

Then, with Photoshop:
(1) I blured out the two coarser scale hillshades, after resampling them at 3 arcseconds, to make the large pixels invisible.
(2) I overlayed the DEM (coloured by a self-made gradient map) and the hillshade (opacity 65%)
(3) Through a layer mask 'Brightness/Contrast' combined with the two coarse scale hillshades, I darkened the shaded side of the intermediate and major alpine structures
(4) Through a layer mask 'Brightness/Contrast' combined with the raw DEM, I darkened the plains and brightened the peaks

All resolution/resampling were done with Menu: Image/Change Image size (Bicubic interpolation)

Had I not used those downgraded hillshades to emphasize larger structures, the image would have looked like that attached here.
Hope this helps find what's missing...

Thanks
Dany

...I read quickly through Tom's very detailed topographic rendering article. Lots and lots of detail there.

In my mind, I think that more info is needed on your precise procedures that got you to where you are now...

Something seems missing in those 2 topography renders; it appears that the raw DEM is showing through too much (the hillshaded/modeled raster seems not to be very present: but I think there is no modeled hillshade derived from the raw DEM?). Are you using Photoshop lighting on the DEM? What SRTM product are you using (3 arcsecond?) ? How did you change the DEM resolution precisely (Photoshop interpolation? Another resampling software?).

I recall years ago trying to use Photoshop to light up a DEM, and I felt (then) that it was a very labor-intensive exercise when compared with other GIS (commercial and open-source) raster software. However, I have to admit that I know may way around Photoshop fairly well, but I'm no "expert" Photoshop wonk

Cheers,
Derek-

Attached Files

•  AlpsWithoutCoarseScaleHllshades2.jpg   900.99K   74 downloads

[David Medeiros]

To me it seems like your gradient colors and color breaks are overpowering any sense of the real landscape. They seem to be replicating the raw elevation file in color with too much contrast between the low, mid and high elevation colors.

The terraced color gradient in flat areas is probably the result of an un-blended gradient and/or even breaks in your gradient ramp. The color changes at that elevation regardless of whether or not it fits in with the elevation change so you have a color break at say 50m when everything to either side is a gradual slope up or down. You want level terrain to have a more gradual change in color. Try more blending of the colors so there are no hard color changes and set your breaks so that lower elevations have more elevation between the breaks which should further soften their color transitions. Play with where each color breaks to move the bands of color up or down the terrain and open or close some bands up. Hard to describe in writing.

For the coarse texture you might try skipping the Arc re sampling and use just photoshop to adjust the variuous layers with some of the blur tools. The surface blur is very good at diminishing the fine shadows and detail and emphasizing larger relief shapes. Use this with various layers and higher detail shading to get the right mix. Once finished apply a very small Gaussian blur over all layers to soften and blend the overall appearance of the terrain.

[CentralAmericaExpe]

I like several of David's suggestions, especially focusing on simplifying color gradient at gentle slopes.

I always keep terrain modeling simple (the fewer the steps, the easier to comprehend subtle changes when blending). IMO, I would simply try David's interval shading suggestion using just the 3 arcesecond hillshade and DEM, and the transparency tool in ArcMAP. Then, bring in your resampled coarse GRID (using your line of resampling methodology) using the same color ramping you used to colorize your 3 arcsecond hillshade, and try blending these with the DEM; etc...

What keeps sticking in my craw is that Photoshop blurring. I get a bad feeling about that step.

David: I like your approach to mapping in Overland Journal (i'm a subscriber); keep up the good work.

D-

On edit: I just about forgot about the stepping gradient anomaly in some DEMs (getting old I guess!): if you ever encounter stepping in any DEM product, try working with that DEM and it's associated hillshade(s) in a 3D rendering software (3D ArcScene; 3D Studio VIZ; AutoCAD; etc...) oriented pure top view (i.e. not oblique). The 3D rendering algorithm will smooth out the stepping anomaly. BTW, how did you deal with the myriad holes in SRTM data inherent in the mountainous regions that you are focusing on (are you using V 2.1; was it fairly clean ?) ?

[David Medeiros]

Thanks Derek. Glad to see another OJ subscriber on the carto forum. I'm anxiously awaiting the Gear issue!

I agree completely on keeping the process streamlined. I do a lot of relief modeling (mostly OJ) and its very much a kitchen chemistry process. If I had to handle multiple relief layers and a complex process it would take way longer than it does. Keeping to one layer if possible really helps. Here I guess its ok because it sounds like the relief is the main attraction in the map. Still, I bet it would be fine working with only one relief layer and applying the correct resampling and post processing.

Blur is an effect you have to be very careful with in terrain modeling. There are several blur tools in PS and they have dramatically different effects. The surface blur is not a blur as much as it is a simplification effect (similar to how you might use the simplify command in Illy to smooth an overtly detailed street or river line). It drops out the fine details and smooths down the major feature lines for ridges and valleys, creating a somewhat faceted look to things (depending on the variable settings). Up close it also applies a blur but when you zoom back out you will see that the edges remain fairly sharp and well defined. So much so in fact that after using the Surface Blur I usually also have to apply a Gaussian Blur to blend the relief. I've found that in high quality print publication of shaded relief, if you don't blur it a little it looks too sharp, if that makes sense. This is especially true when using high res DEMs like the 30m SRTM for small scale maps. The detail is too fine and too sharp and when seen at a normal viewing distance will look unnatural. The blur introduces a sense of distance that your eye expects when looking down from space over a portion of the earth and it smooths over data errors and artifacts very nicely.

[CentralAmericaExpe]

...if you don't blur it a little it looks too sharp, if that makes sense. This is especially true when using high res DEMs like the 30m SRTM for small scale maps...

Absolutely!

I always select the DEM grain size that fits the ultimate production scale. Its quite straight forward to resample a 10m DEM to 20m; a 90m to 250m or 500m (GTOPO30 is superb for small-scale presentations)...

Cheers,
D-

[David Medeiros]

...if you don't blur it a little it looks too sharp, if that makes sense. This is especially true when using high res DEMs like the 30m SRTM for small scale maps...

Absolutely!

I always select the DEM grain size that fits the ultimate production scale. Its quite straight forward to resample a 10m DEM to 20m; a 90m to 250m or 500m (GTOPO30 is superb for small-scale presentations)...

Cheers,
D-

Yeah you should certainly resample where appropriate. I have better luck with resampling the SRTM data than using GTOPO30. The GTOPO30 data just never looks right to me (at the regional scales I typically work in), it seems to round over the terrain where a resampling of the SRTM gives you sharp terrain with less detail between larger features.

My point on the blur applies equally to resampled data. Even at the correct meter to pixel size for your scale the surface of the relief image can look too sharp and plastic. It's surely a personal preference thing but to my eye I can usually tell when relief has been added to a map with no post processing and it always looks raw.

[Dany]

Thanks for your very constructive explanations! Further to these, I did three main things:

- I smoothened transitions between green, yellow and red, but created an abrupt break between red and white (around 3000m, where perrenial snow starts), to mimic snow cover
- I applied 'surface blur' to the hillshade to produce two images (a minor blur to depict medium valleys, and a major one for larger structures), which I combined with a 'brightness/contrast' mask to darken shaded hillslopes. As you said, this blur keeps crests sharp...
- Finally I used a gaussian blur of 0.5 pixel on the final flattened image.

The terraced DEM was caused by an export from ArcMap in 8bit, rather than 16bit... So sorry for this basic mistake.
Regarding Derek's question on SRTM holes, I cannot recall seeing any in the raw data. Guess the new versions have improved.

I have attached the new map, which I am happy with, but if you have still suggestions, I'd be glad to hear from you.

All good wishes
Dan

Hello everyone

It's my first post. Any comment on the attached hillshades are welcome.

My aim was to create a decorative map of the Alps, with no text. I used free data for the continent (srtm) and the sea (gebco), and followed Tom Patterson's advice in his section on 'resolution bumping' (http://www.shadedrel...om/dem/dem.html).

There are still two issues I am not able to address: (1) eliminate the "terraced" colour gradient in flat areas (look at the southwest of the Alps, on the Italian side); and (2) make the coarse scale shading sharper (as it was done by downgrading the DEM's resolution, angles are rounded up).

Thanks for your views

Attached Files

•  AlpsZoom50percent.jpg   490.26K   71 downloads
•  AlpsLowRes3.jpg   474.9K   69 downloads

[Lui]

Hi Dany!

You have used an interesting hypsometric color scale.

The main problem in small scale hillshading nowadays is the high detail of DEMs. This is evident also in your larger hillshading image. The image is some kind noisy. To produce a good looking hillshade image we have to genaralize DEM before we perform analytical hillshading. This can be tricky but using segmentation accompanied with small feature suppression can produce a better DEM for hillshading.

Well I just hijacked you first hillshading image of Alps and perform some noise (detail) suppression filtering and line (ridge) enhancment. I have also performed a small height driven sharpening of detail to inflate the aerial perspective a little bit. I hope that you are not angry?

Lui

Attached Files

•  Alps.jpg   988.76K   126 downloads

[Nick H]

Hi Dan, please bear in mind that I am not an expert (and there are real, world-class experts who post here).

The land masses are seriously distorted. Perhaps it would have been a good idea to project the DEMs before doing anything else (to UTM, may be).

These things are subjective, but to me something is wrong with the elevation colours. Using black and deep greys for the lower elevations doesn't work and produces solid dark areas, like those in the Po Valley for example. Perhaps the only blacks and greys on a colour-shaded relief map should be those used for terrain-shading? In the high mountains the shading and elevation colours work together quite well, but this area is only a small part of the whole.

Regards, Nick.

[Dany]

Lui, I am amazed by what you've done, especially with this downgraded image!

Personally I am fascinated by the fractal nature of river networks, so I want the map to express these nested levels of valleys in the Alps. The challenge, as you stressed, is to limit the noisy impression given by small features. In principle, this map would be printed on a one by one metre poster, and the reader would ideally contemplate the large mountain structures from a distance, without being bothered by the details, which he would then discover as he approaches the wall. I am not a cartographer, but I presume that depicting relief over a wide span of scales is a typical issue.

With respect to your map, I want to keep your improvements, but without discarding the smallest details, if that's possible. Anyway, first I'll try reproducing your transformations. Did you use tools such as 'Sharpen', 'Sharpen Edges', and 'Median filter' in Photoshop?

I am attaching a small tile of the DEM (in 8 bit to limit size), if you or anyone else want to play with it.

Thanks for your contribution!
Dany

Hi Dany!

You have used an interesting hypsometric color scale.

The main problem in small scale hillshading nowadays is the high detail of DEMs. This is evident also in your larger hillshading image. The image is some kind noisy. To produce a good looking hillshade image we have to genaralize DEM before we perform analytical hillshading. This can be tricky but using segmentation accompanied with small feature suppression can produce a better DEM for hillshading.

Well I just hijacked you first hillshading image of Alps and perform some noise (detail) suppression filtering and line (ridge) enhancment. I have also performed a small height driven sharpening of detail to inflate the aerial perspective a little bit. I hope that you are not angry?

Lui

Attached Files

•  Alps_DEM_tile2.jpg   854.87K   56 downloads

[EOSGIS]

I will try to give another tip about terrain modelling i use in my own maps where there is a huge difference in heights as this one or others with valleys and mointains.
my technique is just to treat the DEM in a GIS or a program that accepts DEM manipuñation with a formula (ie:Surfer) and apply a formul that makes highest elevatins more high and lower elevatins as they are. Sometimes I include in the formula the slope. Using this method, i am trying to replicate human perception, in the sense that we ussualy try to percieve great heights differences for mountains, but areas within them with a low slope are just "plains" for us, even if they are in the top.
So... i cant give an exact formula, because i play with the Dem untill i have what i want. It dependes a lot on thhe zone, scale, ... but usually i use the slope as an exponential multiplying factor for the heigh, since exp() is a function that increases rapidly from low values to the highest ones.
its clear that the resulting DEM is useless for scientific use since the values have been modified, but i am just searching a good shaded relief with hipsometry for map making and human perception.
Another program that makes something similar interactivelly , is terrainbender, and other generalization programs you can find in the samw page. Scuse me for not posting the url for the program, but i'm currently out of office.
The technique i describe makes that plain areas maintain somewhat plain, and mountains get. bigger. The best results are when applied to a generalized dem and then overlying the original dem in photoshop to give details in the shaded relief. Some what a bumpmapping.
hope i helped.
best regards

[Kathi]

The land masses are seriously distorted. Perhaps it would have been a good idea to project the DEMs before doing anything else (to UTM, may be).

These things are subjective, but to me something is wrong with the elevation colours. Using black and deep greys for the lower elevations doesn't work and produces solid dark areas, like those in the Po Valley for example. Perhaps the only blacks and greys on a colour-shaded relief map should be those used for terrain-shading? In the high mountains the shading and elevation colours work together quite well, but this area is only a small part of the whole.

I agree with Nick on both his points. The distortion from using WGS-data unprojected is just ugly, so projecting before the export from ArcMap (or at least changing the Projection of your map view) would make it look better. You could try the Swiss projection (CH1903) or one of the Europeans.

The colours you used in the lower valleys are too dark. They take away too much attention from the hillshade and make the lowlands look like abyssal holes (especially the Po Plain). IMHO a hillshade should be a rather subtle colouring, maybe even a greyscale to just show which areas are "sun"-lit and which are in the shadow. If you must add colouring to show elevation (which is a second information and not directly related to hillshading), then use subtle colours as well. To emphasize the fractal nature of river systems it might work better to actually show the rivers themselves as lines.

[Dany]

Hi Nick and Kathi

My eyes got used to the default projection I guess, but I agree, it doesn't look right. Your idea, Kathy, of displaying the river network is also great. I have to find a suitable colour ramp now. Actually my choice for blacking out low elevations, which I understand is arguable, is to shed more light on the alpine structure.. My point is that the map is not intented as a backrgound for any other information or text, but is a 'stand-alone' picture. However, I still reckon that the colour ramp could be improved.

I am grateful for your useful comments.
All good wishes
Dany