Hi Sam, Welcome to the guild.
This is how I do it:
https://www.cartographersguild.com/s...l=1#post296662
How smooth your DEM will be depends on how many bands you use. But you cant make it really smooth without a lot of blending.
Hi all,
I have created a topographic map from scratch using a number of layers of colours in a gradient. I am looking to convert this to a height map to then create a DEM. I have converted the layers to grayscale (0-255) but now I need to blend those layers to create a smooth height map for conversion. Any ideas how?
I initially drew the map in Illustrator but also have photoshop available.
Cheers!
Sam
Hi Sam, Welcome to the guild.
This is how I do it:
https://www.cartographersguild.com/s...l=1#post296662
How smooth your DEM will be depends on how many bands you use. But you cant make it really smooth without a lot of blending.
Thanks so much on both fronts! Been meaning to join for a while.
Have you found any good ways to blend the contours? I have got to the same last step as you and often end with something looking fuzzy rather than blended, using Gausian Blur. Also it would be great to have more control over the actual gradients. I'm sure there is a way to do it, I've never been great at photoshop
Cheers again
The general method that I have been using is to make the approximate landscape using the method that I shown which gives a very smooth undulating DEM with no detail in it and then take real earth DEM data from the Space Shuttle missions from NASA (the SRTM data) and flatten that so it has loads of detail and no undulating height information in it. Then I add on the flattened detailed SRTM data on top of my own smooth undulating contour data. Of course you have to pick the resolution so that they match and the SRTM data is something like 1 pixel per km which is pretty coarse. You can butt many SRTM DEMs together to get it even more coarse but getting finer data is a challenge. There are other data sources than NASAs SRTM but it can be hard to get the right area of earth and NASA data is public domain and usable. I like to grab the Himalayas around Pakistan area for some nice peaky mountains. The other climates / terrains are a bit more numerous and less of an issue.
Otherwise your left with adding random noise like Perlin noise to the terrain. You can do that and then put it into Wilbur and it will make it better through the use of its algorithms which take into account the erosion but that takes some skill to work the app. Waldronate on here is best placed to answer any tricky questions about that app. But we have a large number of people who use it and have posted about it with lots of his answers to guide you. Otherwise for a larger monetary costs there are World Machine or Terragen and so on which do similar erosion / terrain modelling. In my opinion tho, nobodys - including my own app for this - does terrain modelling well enough to look as good as real earth data. Its a very hard problem to solve. Probably possible with enough compute power but that has been beyond most people until quite recently. Perhaps someone will step up again with the newer powerful graphics cards processing to do a better one now. AI could certainly be trained to generate the terrain trained on real earth DEM data which would also be very cool. I have not seen that done yet although a Microsoft researcher whos name is just beyond my memory did make a plug in for World Machine that did something very much like that and it was most impressive.
The problem of increasing the resolution of a low-resolution image is a hard one. The system has to hallucinate new information based on the constraints of the existing information and those hallucinations have to be plausible to someone who's not taking the same drugs. There are many ways to hallucinate the new data (various types of analysis to just add high-frequency noise like Hugues Hoppe did with the whole-US data set, stitching and plopping real-world data like Howard Zhou did in his IEEE paper, making a GAN trained on lots of similar things to make new data, physically-based erosion sims, and so on), but they all end up with characteristic issues. The major issue is that there is a whole lot of missing data that needs to be generated and it's hard to make it plausible beyond the first glimpse. It's even harder if you want very good fidelity to the original image.
Why did I start by describing ways to increase resolution of images when you asked about smoothing data? Because the contours comprise a low-detail and often ambiguous approximation of your surface. Just interpolating the contours using something like a thin-plate spline interpolator will get you the effect of vacuum-pulling a rubber sheet down over a layer cake: it won't look very good. Interpolating is going to smoothly connect the data that you have, and that's not fun for the human eye. To get an idea of what happens, consider a stream running downhill. A stream is constrained to follow the path of steepest descent, meaning that it will always cross contours at right angles. However, what that stream does in between contours isn't constrained at all by the contours. A simple interpolation between contours will give you very smooth (likely straight) stream paths, which look unnatural. You'll get a tiny amount of dynamics from the angles, but you'll get a really straight stream segment if the contour lines are far apart: in the real world, streams are likely to wiggle a lot between those wide contours.
Using just Photoshop and Illustrator really limits what you can do. Depending on your data, you might be able to introduce some additional contours by expanding and contracting a selection based on each contour level and then filling in between with an appropriate color if you have enough colors left over. However, you still won't have the morphological aspects of the stream network unless you manually tweak the contours. Depending on the scale of your map, that may not matter much, or it may be absolutely critical to being plausible. If keeping those contours as close as possible to what you already have is important, manual might be the best way to go forward (especially with the tool limitations).
Me, I'm lazy. I rarely need exactly what's there. I'm willing to take some garbage to avoid a lot of work. That's why I wrote a piece of software to do some of that stuff. It's not Photoshop, though, and it only runs on Windows. https://www.cartographersguild.com/s...ad.php?t=33087 shows how I abused a patient guild member's contour map into something 3Dish that's plausibly close to what they had. However, it's a hallucination of what the original map showed, not anything that would match up against a real-world comparison.
I'll stop ranting now. Interpolating is hard work.