Saturday, November 21, 2020

What's new in version 3.1?

Optimized Wall layout for polygon-based models

For 3.1, I changed the way walls are set up if the models was created via a polygon. This means that the raster contains NoData (NaN) cells. Cells containing a elevation were each simply expressed as a quad,  with a top and bottom and possibly with one or more walls. This simple rasterization method worked but lead to a gazillion "corner quads" (i.e. quads with 2 or 3 walls). And while these corners are typically nearly imperceptible in the final 3D print, I found that it was very hard on the motion system of my 3D printer to rattle along all those tiny teeth and it also took longer to print.

Old method

My solution is to "snip off" half of any quad that has 2 adjacent walls and replace them with a single diagonal wall:

New method


Note that I did not optimize quads with 3 adjacent wall b/c it would have to either randomly choose one of the two possible diagonals or do something drastic like removing this quad or replace it with some sort of "nose".

Here's an example of a non-trivial case created with this optimized wall strategy:


Added the MERIT 90m DEM source

It's only 90m resolution but is more precise, so maybe it's useful for those printing river networks and other flat areas (like Iowa :)

MERIT DEM a high accuracy global DEM at 3 arc second resolution (~90 m at the equator) produced by eliminating major error components from existing DEMs (NASA SRTM3 DEM, JAXA AW3D DEM, Viewfinder Panoramas DEM). MERIT DEM separates absolute bias, stripe noise, speckle noise and tree height bias using multiple satellite datasets and filtering techniques. After the error removal, land areas mapped with 2 m or better vertical accuracy were increased from 39% to 58%. Significant improvements were found in flat regions where height errors larger than topography variability, and landscapes such as river networks and hill-valley structures became clearly represented.

Another potential benefit is that MERIT is not defined for offshore locations, making it convenient for clean models of coastal areas, where you may not want a "flat ocean"

Using the AW3D30 DEM:
Using the MERIT DEM:

Friday, October 9, 2020

TouchTerrain 3.0: New GUI, place search, kml polygon file upload

What's new in Version 3? 

New GUI: With version 3.0 (October 2020), TouchTerrain has gotten a serious facelift. Gone is the utilitarian ca. 1995 aesthetic, replaced by a Bootcamp 4 based GUI. This is my first ever dip into using bootcamp and there are probably still some details to be worked out. 

GUI responsiveness: The Bootcamp GUI implements what it calls responsiveness which means that (to an extent) the app will try to dynamically"compact" itself when used with small screens (tablets, smartphones). Note that this should work on a smartphone but will look a bit weird. Hopefully most users will have a reasonably wide display when using the app!

GUI size adjustments: You will notice that there are 3 main parts ("cards" in Bootstrap speak) to the right of the map. Clicking on their titles (e.g. on Area Selection Box:) will grow or shrink them. By default, only the Terrain Settings and the 3D Printer Options parts are "open", the Area Selection Box (which hides the lat/long coordinates of the box corners) is closed. Opening all three together will make the green Export Button on the bottom to be shoved out of sight, so you'll have to close a part or scroll down to click on it.

Manual box coordinate entry: In version 3, the four coordinate fields inside Area Selection Box are now active. Changing any of the numbers and hitting Enter will immediately resize the red box accordingly.

"Oversampling" warning: If you configure things in a way that TouchTerrain will interpolate the DEM into a higher (more detailed) resolution that what the DEM source can provide, the Current DEM resolution value will turn yellow. Ex: The source DEM resolution is 30 m but you're requesting 20 m. Using 20 m will still "work" but it's counterproductive b/c it won't magically create more detail and your files will be needlessly larger than if you had configured it to resample at say 32 m instead.

Place search bar: As is standard in many other Google Map applications, you can now search for a place by typing a search term into the box and pick one of the five suggestions displayed.  Your map view will fly to the result you picked. Be aware that it will NOT move the red selection box to your new view, so don't forget to hit the blue Re-center box button when needed!

Use a Polygon from a kml file: If you have a kml file that contains a polygon or a polyline (e.g. the outline of a state) you can upload it to the app and use it as boundaries for the 3D model you download. 

To upload the kml file, click on Browse, select a kml file from you local drive and hit Open. Note that is has to be kml file, not a kmz file! Kmz is just the zipped version of kml. To use a kmz, unzip the kmz file, which will give you a file called doc.kml, just rename that and then upload it. You could of course also use Google Earth to digitize a polygon, save it into a kml file and upload it to the app.

After the upload your map view will jump to the area and will see your polygon in yellow inside the red box. Adjust the other settings and then hit Export. You will see a note (Using X points from kml file as polygon) and you model (and geotiff file) will be clipped (masked) by the polygon.

Note that this polygon will be cleared if you move the red box or if you use functionality that requires the app to reload, such as changing the DEM source or the hillshade settings. This is needed, b/c the polygon isn't inlined into the URL (like all the other settings). But, it should be pretty quick to re-upload the kml file again! Finally, be aware that if your file contains multiple polygons or polylines, only the first will be taken.

Other changes:

  • URL: Depending on you system you may no longer see the full URL in the browser. However, we now give you the full URL at the end of the export (as simple browser text), so you can still use it to "give" your model to somebody else.
  • GPX path lines:  On the standalone version of TouchTerrain, you can now drape path lines from gpx files over your model.
  • Map with jupyter notebook: The jupyter notebook standalone version can now show you an interactive Google Map in a notebook cell (using the geemap module). It also lets you digitize a rectangle, circle or polygon onto that map and use it as your model area.

Wednesday, May 13, 2020

touchterrain 2.5: New Hillshading, more DEM sources, lower_leq setting and new folder structure (May 13)

The May 13 release of touchterrain (v. 2.5) contains a whole bunch of new and changed features:

New Hillshading

  • Hillshading creates a overlay map that shows a pseudo-3D relief of the underlying DEM. This is useful as a pre-visualization of how your 3D print will roughly look and helps users to select interesting areas to print.
  • Hillshading (in its simplest form) assumes that a sun is positioned above the terrain at certain angles:
    • Azimuth: Defines the direction (compass heading) from which the Sun shines to create the hillshade relief. Note that South will typically result in a flipped relief perception!
    • Elevation: Defines the vertical Sun angle, i.e. how high the Sun is above the horizon. Use it to enhance detail for different types of terrain.
  • So basically we have a 2 angles, a horizontal (0 - 360, 0 = North) and vertical (0 - 90, 90 = perfectly above) that define the position of the light source used to create the impression of a 3D relief by creating greyscale values from 0 (dark) to 255 (super bright). Note that this does not create shadows, only the effect of specular highlights, which is a very strong perceptional cues for creating a 3D effect.
  • The standard hillshade parameter values are: azimuth of 315° (NNE) and 45° for elevation, which are good all-around settings.
  • However, it can be fun (and rewarding) to change these angles, which is what the new version allows you to do (in a limited fashion)!
  • For example, for best results the Sun should come at an angle roughly perpendicular to ridges in your terrain. Here's a comparison for different azimuth values (elevation angle is 60) (source):

  • 270, 315, 0 and 45 all bring out different aspects of the same data, e.g. 45 emphasizes NE-SW oriented features. Also, to me, some values seem to make my brain reverse the topography i.e. ridges look like valleys and vice versa. 
  • Looking at the elevation value, values lower than 45, create darker and darker overlays but also more and more emphasize low relief terrain features, such as drainage channels (but at the cost of blowing out high relief features).
  • For touchterrain 2.5 I have attempted to implement a simple way to experiment with these two hillshade angles without overwhelming and confusing the user:
  • Sun direction (new) has a few direction presets shown as words (West, North-West, etc.). I only sweep the northern directions in 45 steps b/c I'm pretty sure that using southern direction would result in relief confusion. But, just to force the effect I include South as well.
  • I renamed the Elevation parameter to Sun Angle (b/c I want to avoid confusion with terrain elevation) and again offer just a few presets: Normal (45), Steep (55) and several progressively flatter settings to go below 45 all the way to 5
  • Finally there's a gamma value via a direct number input. This gamma correction factor is used to somewhat compensate that low sun angles tend to create hillshades that look too dark.  Gamma values > 1 increase the "lightness" so as the sun angle becomes lower, this gamma will increase to compensate. Note that the specific gamma values are my "opinion" only! If you want to tweak the gamma after selecting a sun angle preset, just change the number and hit enter. Feel free to experiment, but be aware that your tweak will be overwritten next time you choose another sun angle preset!
  • Be aware that changing any of the hillshade parameters will require a reload of the map, b/c I need to give Earth Engine these new values, get back a new hillshade overlay calculated for the new values and then display it. So it might take a split second, depending on your internet.
  • Note that I have tried to preserve all other input values you might have chosen before doing such a reload. This includes the type of Google Map map in the background and the transparency. However, this is not fully tested, so please let me know if values for some fields you changed earlier are getting reset.
  • If you're curious, go and change the Google Map to Terrain (Map -> Terrain), which also has hillshading backed in, and lower your transparency to mix them both.
  • Transparency (inverse of opacity) is still given in percent but I made the slider much, much smoother (1% steps). I also removed the direct numeric input b/c I feel having a precise number value is not how transparency is used, user should simple move the slider until they are happy.  Changing Transparency will not trigger a reload, b/c it's set via your browser (client) not via the server.
  • Finally, there are a bunch of new query parameters in the URL that store the new hillshade settings and transparency: transp= 0 -100,  hsazi= -360 - 360 (hs azimuth), hselev=  0 - 90 (hs elevation) and gamma= 0 - 99, so if you need values that cannot be generated in the GUI, just change those in the URL and hit enter. Note that none of these parameters matter when processing the terrain into 3D models, they are purely for a better visual experience.
  • So ... got nuts and play with these values!  Set the transparency to 0 (full left) and set the Sun angle to extremely flat. I think this looks very interesting (or at least artistic :)

More DEM sources

  • I have added a few new sources and added a link to their meta data (via earth engine):
    • USGS/NED: 10 m, continental USA only. link
    • ALOS DSM: Global: 30 m, worldwide, but has some small "holes". link
    • USGS/SRTMGL1_003: 30 m, "worldwide", but not very far north (lower quality and older than ALOS!). link
    • USGS/GMTED2010: ~230 m, truly worldwide. link
    • GTOPO30: 30 Arc-Second, 1000 m, 1996, worldwide. link
    • CryoSat-2 Antarctica: 1000 m, antarctica only. link
    • NOAA/NGDC/ETOPO1: 2000 m, worldwide, with bathymetry. link
  • For any non-US area ALOS should be your preferred choice as it's much newer and of better quality. However, it misses data in a few tiny areas ("holes")
  • Using the Antarctica data is sadly a bit silly at the moment b/c Google maps are way too distorted.

lower_leq setting

  • Thanks to Iden Craven a manual setting was added to lower any elevation values that are lower than a threshold (say <= 0 m, i.e. anything not land) by a millimeter value when printed out.
  • Example "lower_leq":[0, 5] as a manual setting will lower offshore areas by 5 mm.
  • This can emphasize shorelines while still printing area below sealevel. (ignore_leq would completely omit these areas when printed)

 New folder structure

  • I decided to create a proper module for most of the touchterrain code, which required a new folder structure:
    • standalone was remove and the two "main" files, and its jupyter notebook version, TouchTerrain_standalone_jupyter_notebook.ipynb (recommended) are now at the root level.
    • All other code (the common and server folders) was moved into a new folder called touchterrain
    • Before running the stand alone version, a module called touchterrain needs to be build and installed in your system's site-packages folder. This is done via, best by using pip: pip install . <-dot!
    • The jupyter notebook now explains the installation and can even run it inside a cell as a shell command!
    • auxiliary files (examples, etc.) were mover to a new folder called stuff

Tuesday, March 31, 2020

Adapted to Earth Engine API changes (Mar. 31)

  • Google announced a while back that they are changing the JS and Python API to be supported by it's Cloud API
  • For me, that change broke several things, both in Python and in JS:
    • I was suddenly unable to d/l any geotiffs from EE to make models from, which took time to figure out.
    • Even then, none of my hillshade overlays showed up, again, that took some time to fix.
  • Impact: 
    • There was an outage of a couple of days that prevented any d/l. Several users got in contact with me; if you've been affected - I'm sorry!
    • But, nobody complained about the lack of hillshade overlays :)
    • Google has apparently decided to lower the maximum size that can be downloaded. It should not affect TouchTerrain, b/c I only request appropriately downsampled geotiffs from them anyway and our server also caps how many pixels you can process on it, but I have not tested this well. It's possible that users will run into the Google quota if they request very large models.
    • But, that pretty much kills the "download at source resolution" option I had and so I took it out. It's still available in standalone to make models at the exact same resolution as a local raster (geotiff). 
    • It is still possible to "export" large rasters into something running in the Google Cloud but I have no intentions to run our server in the Google cloud. At least not for the foreseeable future.
  • Silver lining: After going back to digging around the Earth Engine API I noticed a whole bunch of new terrain and terrain related datasets!  My next plan is to add some of these to Touchterrain.

Exporting Terrain models with real world coordinates

Exporting Terrain models with real world coordinates

  • Some people would like to import TouchTerrain model files (STLs or OBJs) into software for CAD or 3D Modelling. 
  • At my University (Iowa State), an architecture student imported terrain into Rhino, manually mach a building he designed to its correct place and CNC'd the entire thing!
  • Another request asked for the ability to import a TouchTerrain STL into Blender in order to make changes to it there. Specifically, to combine it with a GPS track in order to make the track visible when the 3D terrain is printed out (more about this later, provided this turns out to have worked ...): Update: this is now an official touchterrain option (thanks to KohlhardtC!), see the gpx option in the github ReadMe.

  • As a result, I decided to add an option to set the xyz coordinates of the vertices inside the STL/OBJ files as something in a global context. (Typically, the coordinates are in mm and the origin is the center of your buildplate ...)
  • Using unprojected lat/long coordinates is out of the question for many reasons (distortions, height, sphere vs plane, etc.), so the only real option is using UTM coordinates, which TouchTerrain already requests from EarthEngine. This makes the x/y coordinates be large numbers, something in 100,000s or even millions, with z (elevation) always in meters. 
  • These UTM x/y coordinates are distances from the intersection of the equator with Greenwich. Here's a web app that show those numbers for the (nearly) entire globe.
  • In theory, terrain models with these numbers, when imported into a 3D modelling system, should show up in the correct place and should also be correct with regard to other, UTM encoded, objects.
  • Sadly, the only system I remotely know how to use is Blender, for which these standard UTM coordinates did NOT work (at least for me)!

  • However, after looking into BlenderGIS, I realized that I could trick Blender in positioning the STL model in the correct place, if I first imported the geotiff of the terrain (which is always given in the zip folder you download). The trick seems to be to still use meters as units but set the origin (0,0,0) at the center of the geotiff.  For this to work, you need to use the "centered" option (see below).
  • I've documented my Blender adventure here (very detailed, illustrated instruction pdf and a zip with example files used). It only deals with draping gpx lines on the topography (which was added as an official option to standalone in version 3) but it will also show you how to center a touchterrain model for use with BlenderGIS. So I imagine this method should also work for placing other types of geometry (like buildings) onto the model using Blender.

  • Using ArcGIS Pro, I was able to import a OBJ terrain mesh from TouchTerrain (with the "UTM" option, see below) into a local 3D Scene and it does overlap nicely with its geotiff. I used the Import 3D Files (3D Analyst) function.
  • To show the fit, I've set the build in terrain to flat. The "shadow" over which the imported OBJ seems to be floating is the geotiff.
TouchTerrain 3D model of Sheep Mountain, imported as OBJ into ArcGIS Pro, local 3D Scene

  • So what now? Well, if you want to try using a TouchTerrain Model in the context of a 3D Modelling software, all you need to do is to set the manual option:
    • "use_geo_coords":"centered" will set the UTM origin to the center of the full tile
    • "use_geo_coords":"UTM" will use the official UTM x/y coordinates
  • Again, this is highly experimental and only tested in Blender (with "centered" and using BlenderGIS to import the geotiff first!) and ArcGIS Pro. If it works for you in another 3D software, please let me know! (If it fails, also let me know, maybe I can make it work ...)