Let’s start with an example showing the USGS topo around Mt Morgan (S) unshaded (from State of California):

Unshaded USGS Topo

Now here’s the same spot shaded (from ArcGIS map viewer):

Shaded USGS Topo

It’s much easier to get a sense of the topography at a glance with the shaded version. This led to one of my first cartographic decisions, how was I going to treat hillshading? I started by taking a look around the web for different treatments and wasn’t very happy with any of them. The shading shown above is pretty decent except for two things, the overall look of the map is dull and muted and there is no fine detail in the shading. It looks as if the shading were done with much lower resolution data than the contours would suggest.

Next, I turned to Google’s Terrain maps for the area around Mount Morgan (S) and Mount Tom:

Google Terrain Map

Note how metallic and unnatural this looks. The contrast makes the mountains pop way too much for my taste.

Next up, I took a look at TopOSM, a remarkable effort by Lars Ahlzen to provide topographic maps for the entire country (far more ambitious than me!) based on OpenStreetMap data:

TopOSM

I definitely appreciate the ability to turn off all map decorations besides the relief but, again, we get the metallic feel. With a bit more dynamic range it actually seems a little worse than the Google Terrain maps.

The real problem with this metallic looking shading is that it is no longer just a subtle visual cue, it’s become a first class element of the map. I’m after more of a light backdrop where I can overlay different layers of data.

One last example I found on the web was from ACME Mapper:

ACME Mapper

This one is pretty similar to the first shaded example from above but not quite as dull although it shares the problem of not having a lot of fine detail in the shading. What’s strange about this one though is that the ridges are actually casting shadows onto the plains below. It has the same effect as traditional hillshading in that you can more easily visualize it in 3-D but I found that it is both confusing and needlessly darkens the valley floors.

Creating the Hillshade

After my survey of what was out there I had a pretty clear vision in my head of what I was looking for. The first step was to get the actual data. I’m using the 1/3 arc-second (roughly 10m) USGS Seamless National Elevation Dataset (NED) which the state of California has conveniently cut into one-degree square blocks. I used wget to grab the files:

wget -r -np ftp://projects.atlas.ca.gov/pub/ned/38119/grid/grdn38w119_13/
... plus a bunch more

At about 500MB per square degree this took a few hours. Next I used GDAL to combine the files and generate the hillshading background image. The first step is to use GDAL to create a VRT that will make all of the separate one degree block files appear as one large dataset:

gdalbuildvrt dem.vrt grdn38w119_13/w001001.adf {more files}

I need the final output to be in the Closed Contour projection so I used GDAL to warp the data:

gdalwarp -t_srs "+proj=tmerc +lon_0=120w +k=0.9996 +ellps=WGS84" \
        -r cubic -tr 10 10 dem.vrt dem.tif

At this point I have a single TIFF file that has elevations for the entire area of interest. This file is 36515 x 55915 32-bit (that’s 2000+ megapixels!) floats representing elevation, weighing in around 8GB.

Next I used GDAL to create a visual hillshading representation:

gdaldem hillshade -s 1 -z 0.7 dem.tif shade.tif

I experimented with a variety of ‘-s’ and ‘-z’ arguments but couldn’t really get what I was looking for.  Here’s the output from the above:

Direct GDAL Output

Uh-oh, we’re back to that metallic feel. I was a bit stumped at this point. I knew that I could take an image like this into Photoshop and tweak the curves to get the effect I was looking for but I wasn’t too excited about asking Photoshop to open up my 8Gb image (I was on a 32-bit machine at the time). Besides, I wanted this to be reproducible in a script.

I poked around with GDAL a bit more and noted that the output from the hillshade program was an 8-bit grayscale image. If I could just remap those bytes into my own palette then I’d be set. I thought this would be pretty straightforward but a bunch of googling later and I still didn’t have a solution. Further digging revealed that GDAL had all the answers. First, I created a VRT from the hillshade output:

gdalbuildvrt remap.vrt shade.tif

Open up remap.vrt in an editor and you’ll see this:

...
   <VRTRasterBand dataType="Byte" band="1">
   <NoDataValue>0.00000000000000E+000</NoDataValue>
   <ColorInterp>Gray</ColorInterp>
...

It turns out you can manually create a custom color palette like so:

...
   <ColorInterp>Palette</ColorInterp>
   <ColorTable>
      <Entry c1="128" c2="128" c3="128" c4="255"/>
      ... 254 more color entries
   </ColorTable>
...

I wrote a small script (which I have unfortunately misplaced, although just ask if you’d like the generated color ramp that I used) that mapped the gray values in such a way that the darkest became 128 gray and rapidly curved up to white.  The last step was to convert the VRT to an actual TIFF, which was as simple as:

gdalwarp remap.vrt final_hillshade.tif

This what I ended up with:

Remapped GDAL Output

Ah, this is much better. Still a bit metallic when viewed in isolation, but once you start overlaying data it recedes into the backdrop and becomes just that 3-D topographic visual cue I was looking for. Here’s what this area of the map looks like in the latest version of the map:

Closed Contour Map

But Closed Contour is just a hobby for me and it’s unclear when I’ll ever have time to do these things. But I don’t want to let my desire for a cleaner, prettier, more featureful map stand in the way of releasing what I think is a pretty cool map.

So I’ve decided that it’s time I start telling people about the Closed Contour SPS map. Hopefully visitors will find it to be both useful and aesthetically pleasing. I’ve certainly enjoyed making it and look forward to continuing work on it in the future.

Mount Starr King from Liberty Cap

While we were making our way up Liberty Cap I realized that the coloring of the my map was too simple and somewhat misleading.  Here’s a snap of the Liberty Cap area from the v1 tiles (which are the published ones as I write this):

Liberty Cap (Closed Contour v1)

We left the trail around the ‘6000′ contour label and headed up the gully between Mount Broderick and Liberty Cap.  We then curled around and went up the broad gully on the north side of the dome.  So what’s the problem with the map?  It shows most of the route up the north side as being in the forest when in fact it was easy traveling over mostly rock.  In coloring the land-cover I was ignoring the density of the vegetation.  I spent some time last night working on some styling for this data and ended up with this:

Liberty Cap (Prototype v2)

This is a better representation of reality and is superior for trip planning (i.e. not going through dense forest if possible).  As you can see there are other styling elements added as well.

The same problem arose on the Mount Watkins hike.  Here’s the current map:

Mount Watkins (Closed Contour v1)

It shows the Snow Creek Trail (the steepest trail out of the valley) climbing through green, which could indicate shade while in reality this route is mostly exposed to the sun.  Also, the cross country travel is pretty straightforward once leaving the trail on the east side of Snow Creek.  Take a look at the v2 prototype map:

Mount Watkins (Prototype v2)

Unfortunately I’ve only applied these changes to the Yosemite land cover data at this time.  It’ll take a little bit of work and a re-import of the vegetation data to do it across the whole Sierra Nevada, but I feel like it’s a big improvement in the usability of the map.

I’ll close this post with a few other photos from the weekend.  First, the top of Nevada Falls:

Top of Nevada Falls

Because of the late spring snow and cool temperatures the rivers are just now reaching their peak flow.  Next, a couple from the Mount Watkins hike:

Half Dome from Mount Watkins

Clouds Rest from Mount Watkins

I think the Snow Creek Trail has the best views of Half Dome in the valley.  And the summit of Mount Watkins has the best views of Clouds Rest.  Two highly recommended hikes.

• Tiles for the outer zoom levels look like crap right now.  I’ve got a somewhat cleaner look and feel for those now, just need to generate.
• I’ve cleaned up my trails data a little bit and got Yosemite and Sequoia/Kings Canyon data in there.
• I’ve added some populated places to the map.
• Added more data to the DEM so hillshading will be better, especially zoomed out.

Anyway, I hope to have my new tileset up and published by the end of the long weekend.  Update: new tiles are up.