DEM Comparisons: SRTM, ASTER, ALOS, Copernicus DEM Map Renderings

[Odainsaker 2,134]

I have done some test renderings to see what quality differences occur in SimCity 4 maps generated from 16-bit digital elevation maps (DEMs) downloaded from several different public sources.

This is an expansion deviating off an already lengthy reply post made in a previous thread, "How can I ensure I'm creating a valid 16 png for use with SC4Mapper?" by @Han Solo, but I thought it might be useful for any mapmakers to see these actual comparisons.

Many of us will already know the outlines of the expected results, but I don't recall many side-by-side comparisons and hopefully it will be interesting to see them objectively in one place.

 

 

DEM SOURCES

 

The following four DEM collections are now publicly available as digital surface maps (DSMs) at 1 arc second, or roughly 30-meter, interval postings in GeoTiff format:

NASA's Shuttle Radar Topography Mission (SRTM) Version 3.0 Global 1 arc second (SRTMGL1)

NASA and METI's Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) Global Digital Elevation Model (GDEM) Version 3 (ASTGTM) 

JAXA's Advanced Land Observing Satellite (ALOS) World 3D - 30m (AW3D30)

ESA's Copernicus Digital Elevation Model (COP-DEM) global 30m (GLO-30)

 

Individual national services may publicly offer higher resolution data for limited areas, such as the USGS's freely available National Elevation Dataset (NED) at 1/3 arc-second, or about 10 meters, for the United States, and 3 meters in smaller areas.  Other countries limit data for commercial licensing or even national security reasons, so you mileage will vary depending on area of interest.  JAXA offers purchasable ALOS 5-meter and even 2.5-meter DEMs, and ESA offers purchasable Copernicus DEMs and its original commercial AIRBUS WorldDEMs at 12-meters.  Depending on your project, 10-meter, 5-meter, and 3-meter DEMs can either be useful or overkill.

SRTM - SRTMGL1 is downloadable from the USGS Earth Explorer (EE).  Free registration is required via the EROS Registration System.  Within Earth Explorer you will have to filter to the desired dataset:  Data Sets --> NASA LDPAAC Collections --> NASA SRTM (SRTM 3) Collections --> NASA SRTM3 SRTMGL1.  (Alternatively, OpenTopography provides a direct portal, Shuttle Radar Topography Mission (SRTM GL1) Global 30m, without requiring registration.)

ASTER - ASTGTM is downloadable from the USGS Earth Explorer (EE).   Free registration is required via the EROS Registration System.  Within Earth Explorer you will have to filter to the desired dataset:  Data Sets --> NASA LDPAAC Collections --> ASTER Collections --> ASTER Global DEM V3.

ALOS - AW3D30 is downloadable from the JAXA EORC ALOS Global Digital Surface Model "ALOS World 3D - 30m (AW3D30)" page.  Free registration is required via the User Registration page.  (Alternatively, OpenTopography provides a direct portal, ALOS World 3D - 30m, without requiring registration.  OpenTopography does not include the auxiliary mask files.)

COP-DEM - GLO-30 is downloadable from ESA's Copernicus Space Component Data Access Planetary Data Access Catalogue (PANDA).  Free registration and then a digitally-signed public access license agreement are required via the Copernicus User's Personal Area: Service Project & Dataset Management Portal.  Within PANDA, you will have to filter to the desired dataset:  Filters --> select "PRISM Catalogue" --> select "Datasets" --> COP-DEM_GLO-30-DGED.  (Alternatively, OpenTopography offers a download portal, Copernicus GLO-30 Digital Elevation Model, without requiring registration.  OpenTopography does not include the auxiliary mask files.)

(Note that while the official download sites provide their DEMs as grid tile GeoTiffs keyed to simple whole number latitude and longitude grid corners, OpenTopography will crop results to a use-defined map selection area.  This lossy selection can make it difficult to merge or overlay different DEMs from different sources if you are not exactingly monitoring the selection coordinates so that all the involved maps align.  I usually find it much more convenient when using OpenTopography to instead manually enter selection coordinates that match the nearest simple whole number latitude and longitude grid corners.)

 

 

BATHYMETRY

 

None of the four DEM datasets listed above include bathymetry.  Worldwide available coverage for bathymetry is still very poor due to our limits in canvasing areas with sonar and lidar, and most freely available public DEMs are going to be at 15 arc second, or roughly 500-meter, interval postings.  Individual national services may offer better resolutions for limited areas, and you mileage will vary depending on the area of interest.  Because of the vast difference in resolutions, care must be taken in cropping, pasting, and aligning them into a land DEM from the above sources with the caveat that the transition from above sea level to below sea level will be grossly imprecise.  Thankfully, ALOS and COP-DEM include useful water masks, though GIMP's common graphics editing selection tools help as well.

The General Bathymetric Chart of the Oceans (GEBCO) offers freely available data most recently released for 2021 as the GEBCO_2021 Grid.

GEBCO - GEBCO_2021 Grid is downloadable from their portal, GEBCO Gridded Bathymetry Data Download.  Registration is not required.

Alternatively, OpenTopograpahy offers download portals for both the Global Bathymetry and Topography at 15 Arc Sec: SRTM15+ V2.1 and the Global Multi-Resolution Topography (GMRT) Data Synthesis, both of which similarly do not require registration.  The SRTM+ and GMRT bathymetry DEMs have shared information with GEBCO and at these resolutions they are not significantly different.  I have seen that GEBCO has filled DEM holes missing from SRTM+ data and that both have additional small water bodies not found in the GMRT data.

(Note that while the official land topography download sites provide their DEMs as grid tile GeoTiffs keyed to simple whole number latitude and longitude grid corners, GEBCO and OpenTopography will both crop their results to a user-defined map selection area.  This lossy selection can make it difficult to merge or overlay different DEMs from different sources if you are not exactingly monitoring the selection coordinates so that all the involved maps align.  I usually find it much more convenient when using both GEBCO and OpenTopography to instead manually enter selection coordinates that match the nearest simple whole number latitude and longitude grid corners.)

 

 

TOKYO REGION MAP

 

The map used here for test rendering is a scale-accurate section of the Greater Tokyo Metropolitan Area, generally spanning from the Tone River in the north to Hatsushima island in the south, and from the Fuji River in the west to the eastern shoreline of the Bōsō peninsula to the east.  The region is 122,880 m x 188,416 m, required a DEM image of 7,681 pixels x 11,777 pixels, and produced a SimCity 4 region map of 30 x 46 large city tiles.  The 16-bit DEM was processed with Prof. Peter Guth's MicroDEM 64-bit build 2016.4.28.199, edited with GIMP 2.10.18, and imported to SimCity 4's Regions folder with @wouanagaine's SC4 Mapper 2013 (v1.0) patched with the NTCore 4GB Patch.  The complete region view image was captured and exported with @JeanLucPicard2's SimCity4 Startup Manager (SUM) 3.0, also patched with the NTCore 4GB Patch.  SimCity 4 Rush Hour version 1.1.640.0 has been patched with the NTCore 4GB Patch and set to hardware rendering graphics mode.

MicroDEM is downloadable from the MicroDEM Home Page on the USNA.  Registration is not required.  The current standard version for "Complete MicroDEM install" is MicroDEM build 2020.8.20.1 and I have found that the functions of that build still work for SimCity 4 mapmaking purposes.  However, be aware that I have found that the currently offered updates with later builds break much of that functionality, and so I have NOT applied any updates suggested by the "Update program EXE" page nor any updates from within the program's updater.

GIMP is downloadable from the GIMP - GNU Image Manipuation Program homepage.  GIMP is currently now up to version 2.10.30.

SC4 Mapper 2013 (v1.0) is downloadable from SC4 Devotion's LEX.  Free registration is required

SimCity4 Startup Manager 3.0 is downloadable from SimCityKurier:  Modding-Tools.  Registration is not required.

NTCore 4GB Patch is downloadable from NTCore.  It also comes bundled with the Network Addon Mod.  Registration is not required.

 

The in-game images have been captured with custom terrain mods applied.  I have used the Gobias' Sudden Valley Terrain Mod (v1.0), option High Altitude, edited with @ilive's iLive Reader 0.9.3 to shift the snowline and treeline of the mod on the in-game Mount Fuji to roughly correspond to the snowline of the real-world Mount Fuji by changing the "TemperatureRangeFactor" exemplar value to 0.56 and the "Flora: elevation scale range" exemplar value to 3000.  The rock mod is @shokthrpy's SHK HD Rock Mods, option 03.  The beach mod is @_marsh_'s HD White Sands Beach Mod.  The water mod is Gobias' Gobias Water Pack 1 (v1.0), option GB Bodensee Water.  Although I have not planted trees in these images, the tree controller is @blunder's Cascadia Tree Controller With MMPS.  I do not claim that any of these mods or their combinations best represent the Tokyo area, only that they instead suited my particular taste at the time.

Sudden Valley Terrain Mod (v1.0) is downloadable from SC4 Devotion's LEX.  Free registration is required.  SimCity 4's hardware rendering graphics mode is required for use.

SHK HD Rock Mods is downloadable here from Simtropolis' STEX.  Registration is not required.  SimCity 4's hardware rendering graphics mode is required for use.

HD White Sands Beach Mod is downloadable here from Simtropolis' STEX.  Registration is not required.  SimCity 4's hardware rendering graphics mode is required for use.

Gobias Water Pack 1 (v1.0) is downloadable from SC4 Devotion's LEX.  Free registration is required.  SimCity 4's hardware rendering graphics mode is required for use.

Cascadia Tree Controller is downloadable here from Simtropolis' STEX.  Registration is not required.

iLive - Reader 0.9.3 is downloadable here from Simtropolis' STEX.  Registration for the STEX is not required.  Ilives Reader (v0.9.3) is also downloadable from SC4 Devotion's LEX.  Free registration for the LEX is required.  Alternatively, Dropbox download links for iLives Reader Version 1.5.4 can be found on the "iLives Reader 1.x Official support thread" at SC4 Devotion.

 

An extremely large region map of Tokyo is not necessarily the preferred map for these renderings, and it just happened to be the one I was working on.  However, it did have some fortunate features:  a high and smoothly symmetrical mountain, a broad plain, lowlands just at and even below sea level, a complex waterway system with a variety of shore and edge types, a major city with large and tall structures, and ongoing major landform changes in the form of costal land reclamation.  There will be lots of conditions to inspect.

 

 

With everything collected and identified so that others can recreate the experiment, let's finally see the rendered results.  I will show the complete regions and then various focal areas.  Each focal area will include renderings using all four public DEM sources, in order of when the DEM datasets were generally made publicly available.  I have set SimCity 4 to display at 1920 x 1080 resolution and taken screenshots at that resolution.  Images were converted to the lossy JPG format for upload to Imgur at a reasonable size.  Region images were reduced 25% in size for display, as free Imgur will automatically resize large images down to a maximum of 5,000 pixels width.  The Simtropolis Forums will resize embedded large images to maintain forum layout, so you may have to expand the images to see some details.

 

FULL REGION VIEW

SRTM - SRTMGL1

Note the striping visible as green and yellowish bands in the lowest/flattest elevations.  This is likely from a known flaw in SRTM data traced to uncompensated oscillation in the space shuttle's boom holding the detector and is most noticeable on relatively flat topography, which here just happened to be a contrasting contour color transition.  Selective blur filtering in GIMP can probably remove them.

ASTER - ASTGTM

Note that while the shorelines and rivers look very clean, there is very little of the lowest elevations shown by the default SC4 Mapper color coding as a yellowish white.  ASTER has been noted as having comparatively high deviation from direct elevation measurements at elevations below 7 m.

ALOS - AW3D30

COP-DEM GLO-30

 

(Oddly, in all four maps the bathymetry of Lake Kasumigaura at the bottom corner is misaligned, as is this bathymetric tile in general throughout though to lesser degrees in other parts of the map.  The tile comes from GEBCO, though the issue also exists in tiles of this area from other sources, most of whom use shared data.)

 

 

MOUNT FUJI

SRTM - SRTMGL1

Note that SRTM was originally first released as 90-meter DEMs, which have since been reprocessed with later techniques to create 30-meter DEMs.

ASTER - ASTGTM

Note the pockmarks around the darker contour color bands of Mount Fuji.  ASTER data is generally noisy, the noise apparently coming from its earlier stereoscopic imaging process that created a high spread of variations in its elevation data even on identical, contiguous slopes.  That noise shows as a uniform spread of small depressions and lumps.  Selective blur filtering in GIMP can reduce them.

ALOS - AW3D30

Note the striping in on the snowy sides on the upper portions of Mount Fuji and which seem to run in orthagonal direction with the SimCity 4 city tile lines.  I do not know if they are caused specifically by the DEM or the terrain mod, but they do not appear with the other DEMs.

COP-DEM - GLO-30

Note that with much of the noise and striping gone, crevasses running down Mount Fuji's smoother snowy slopes are now strongly defined.

 

(Frustratingly, a visible city tile seam appears across the middle of Mount Fuji on the most prominent facing side in all four images.  I think the source of this seam comes from the extreme elevation difference between where one tile ends and the other begins, even between contiguous DEM pixels, on extreme slopes.  I have seen it in other high, snowy mountains.  A solution is to change the position of city tile seams by altering the config.bmp so that they fall on less extreme slopes.  It can also be masked by darker terrain colors, forests of dark trees, and even direct editing of the edges with God Mode terrain shaping tools.)

 

MOUNT FUJI SLOPE AND SUMMIT

SRTM - SRTMGL1

ASTER - ASTGTM

ALOS - AW3D30

COP-DEM GLO-30

 

 

BASE OF MOUNT FUJI

SRTM - SRTMGL1

ASTER - ASTGTM

Note again the pockmarked or pitted appearance of numerous shallow holes and lumps created by noise in the DEM.

ALOS - AW3D30

COP-DEM GLO-30

 

 

CENTRAL TOKYO

SRTM - SRTMGL1

Note the relatively rough outlines of the coastal edges.

ASTER - ASTGTM

Note again how SC4 Mapper's default lowest elevation colors are lost as ASTER doesn't well capture these lowest elevations.  This is unfortunate, for historical urban design patterns in old Edo and early Tokyo can be divided between the "High City" of the aristocratic mansions and estates centered around Edo Castle on the higher elevations, and the "Low City" of the commoners built on the lowest elevations, reclaimed lands, and canals along the waterfront and Sumida River.

ALOS - AW3D30

Note the cluttered river lengths.  ALOS does include a separate mask image as an auxiliary file to assist in editing water bodies and channel, but I have not done such manual editing here.  On the other hand, ALOS does include numerous islets, rocks, and shore structures missed by the water masking of other datasets.  At least for this area, ALOS has the most up-to-date data as can be seen in the reclaimed disposal islands of Tokyo Bay.

COP-DEM GLO-30

Note that the Copernicus DEM has the clearest waterways, and the DEM includes a water mask image as an auxiliary file as well as water transparency in the DEM itself.  Also note that the reclamation island in Tokyo Bay is not as extensive as that in ALOS, suggesting the data here may not be as recent.

 

 

IMPERIAL PALACE

SRTM - SRTMGL1

ASTER - ASTGTM

Note the shallow lumpiness resulting from both baked-in building filtering and noise.  I had found that in GIMP 2.10.18, a Selective Gaussian Blur with a pixel radius of 5 and a max. delta of 0.001 was effective at removing most of the lumps without erasing the visible details on the slopes of Mount Fuji.

ALOS - AW3D30

Note that major buildings and even urban street corridors are readily apparent.  Free ALOS DEMs were based on a commercial 5-meter DEM and 2.5-meter imagery that have been reduced for public release.  That high resolution, however, had cleanly picked up enough building surfaces that even the reduced DEMs yield identifiable building mesas in SimCity.  I do not yet know what level of filtering will be needed to eliminate them for SimCity, especially as such filtering would need to be strong enough to eliminate the building mesas yet also not so strong as to inadvertently erase the moat walls of the Imperial Palace.

COP-DEM GLO-30

Note that there is not as much human structure clutter as the ALOS DEM.   Where the ALOS satellite had a resolution of 2.5 meters, Copernicus DEM is based on data from the TanDEM-X and TerraSAR-X twin satellites, which produced a 12-meter resolution WorldDEM.  Moreover, while ALOS and ASTER use imagery to calculate visible surface heights from photometric displacements, Copernicus and SRTM were based on radar ground measurement, which to a limited extent penetrated through some of the sponginess of ground clutter like vegetation and human structures.  All the datasets then use some filtering and corrective offsets, with varying fidelity and effects.

 

 

TOKYO HARBOR

SRTM - SRTMGL1

ASTER - ASTGTM

ALOS - AW3D30

COP-DEM GLO-30

 

 

SHIMIZU

SRTM - SRTMGL1

ASTER - ASTGTM

ALOS - AW3D30

COP-DEM GLO-30

Note that Copernicus did not pick up or has masked away the breakwater.

 

 

TONE RIVER

SRTM - SRTMGL1

Note how with SRTM much of this area in SimCity becomes underwater with numerous tiny pools.  This is actually a lowland area, with parts of it devoted to farming or rice paddies, but not the extent shown here.

ASTER - ASTGTM

Note how ASTER generally does not utilize the lowest elevation color banding of SC4 Mapper.  It also lost all trace of the rice paddies, 

ALOS - AW3D30

Note that in ALOS the lowland rice paddies have, and some of their individual rice paddy separations are visible as stripped berms.

COP-DEM GLO-30

Note Copernicus's water masking has correctly revealed Lake Inba-numa and even the narrower Shintone River paralleling the much wider Tone River.

 

 

RICE PADDIES

SRTM - SRTMGL1

ASTER - ASTGTM

ALOS - AW3D30

COP-DEM GLO-30

 

 

CONCLUSIONS

SRTM - SRTMGL1 is based on much older collected data, primarily an 11-day space shuttle mission in 2000, so recent changes in topography may be missing.  Water body masking is very rough, with imprecise shorelines and missing islets.  SRTM was previously publicly released as 90-meter DEMs, later reprocessed and released as 30-meter DEMs.  I find them still relatively blurry compared to the latest options, and I have noticed large scale striping in flat areas.  Interestingly, I recall drunkapple/dobdriver once posted that he used SRTM DEMS in his maps and he described issues with water edges and shorelines.  As his maps largely were all uploaded before SRTM 30-meter DEMs were publicly released starting in 2014, he may have been using the 90-meter DEMs.

ASTER - ASTGTM has more recent data than SRTM and better water body masking as it has been collected and updated since 1999 and was the first worldwide dataset made publicly available at 30-meter resolution.  A separate water body mask can be downloaded.  Potentially, ASTER can have finer details than SRTM, however, ASTER data is exceedingly noisy and will require significant user filtering.  I have found in GIMP 2.10.18 that a Selective Gaussian Blur with a blur radius of 5.00 and with a max. delta of 0.001 has so far yielded greatly improved results for my own purposes.  Compared to the other datasets, ASTER reportedly has significantly height variability issues compared to GPS ground measurements for elevations below 7 meters above sea level, and this is quickly seen in SC4 Mapper as the lowest land elevation colors are noticeably less applied.  I have also found major islands in the DEM shifted to the wrong locations, or even missing entirely.  Lastly, MicroDEM has had the most errors for me with certain downloaded ASTER DEMs, either showing blank maps or wrongly offset landforms, which I have managed to correct by first opening and then resaving the downloaded affected GeoTiffs in QGIS.

ALOS - AW3D30 is generally more recent still and, being derived from a 5-meter DEM and 2.5-meter imagery, has better height accuracy and water body masking than both SRTM and ASTER.  Some river channels are hard to distinguish as they reach sea level, but a separate water mask is provided to assist in editing.  Human structures such as large buildings are particularly prominent, especially in urban areas with large clusters of tall buildings, and will require significant user filtering, though I do not know yet what filtering options would be best.  Reportedly, the finer resolution skews elevation data broadly adjusted for tree canopies to be somewhat lower than the other comparable datasets, as the finer resolution is actually picking up more open areas between trees.  I have also noticed striping on some extreme slopes.  Japan's ALOS satellite may have been damaged by meteoroids and has been rendered dead in space, so it's dataset spanning from 2006 to 2011 will not be further updated with new imagery and some areas may have lesser coverage.

COP-DEM GLO-30 is the most recently released (2020), being derived from the 12-meter commercial AIRBUS WorldDEM collected and updated since 2011.  It reportedly has the current best height accuracy of the four DEMs listed here.  It's water masking is also very accurate and a separate water mask and even water transparency within the DEM are provided.  However, it doesn't capture as many islets or finer shoreline details as ALOS.  Being radar-based rather than radiometer stereo mapping, it definitely does not suffer from the overall noise of ASTER or much of the human structure clutter of ALOS.  I do admit I found the PANDA registration process the slowest and most cumbersome, actually requiring an account registration approval review that took me 24-hours.  However, once acquired the DEMs will require the least cleanup editing for use in SimCity.

 

Use the Copernicus DEM.  Actually, more specifically, I have been using Copernicus DEMs manually edited with additional shore and islet information from ALOS DEMs.  The previous DEMs were good for their time, and, when SimCity 4 was released, SRTM 90-meter at 1/3 the resolution shown here was actually the best publicly available option, but they are now becoming outdated as newer, higher resolutions DEMs collected by higher precision instruments and processing techniques have become available.

 

__________

 

Somehow, someday, I have to finish and upload this map.

[CorinaMarie 29,605]

Excellent and insightful post. Thanks!

[Odainsaker 2,134]

In the initial post I mentioned settling for the recent Copernicus DEM GLO-30 dataset as the best generally available set of global DEMs, but we also saw that it still showed clutter from ground structures, even if to a degree less than some of the other datasets.  Because in SimCity 4 we will build or plop our own buildings and trees, we want our DEM maps to be less of a digital surface model (DSM), which includes all surfaces such as buildings and tree canopies, and more of a digital terrain model, which should just be terrain elevations.  If your map is small, you can easily just edit the map by hand with GIMP tools and an eye on Google Maps as a guide, but this may not be practical for large maps or large numbers of maps.  I'm next going to examine two potential solutions to bring the Copernicus DEM GLO-30 dataset closer to being a accurate digital terrain model by filtering out the buildings and trees.

 

GIMP SELECTIVE GAUSSIAN BLUR FILTER

 

The first potential solution is to filter the set ourselves.  Short of elaborate GIS tools and cross-reference databases, our best available tool is GIMP, which has a variety of filters of which perhaps the most promising is the Selective Gaussian Blur filter.  Selective Gaussian Blur will blur a specified radius of pixels if the maximum difference in the values between those pixels is within a specified threshold.  This allows us some control as to where and to what degree blurring and smoothing occurs, and if we carefully select our parameters we can generally blur undesirable DEM spikes from buildings while preserving desirable DEM changes from terrain.  Trial and error is the most straightforward means to find he parameters for the Selective Gaussian Blur function, recognizing that it is still a grossly imprecise method as it cannot directly identify buildings or trees, but instead only general differences in pixel values across and area.

 

FOREST AND BUILDINGS REMOVED COPERNICUS DEM

 

The second potential solution I stumbled upon while doing trial and error testing of the first.  Researchers as the University of Bristol have actually applied GIS tools, machine learning, and and a variety of databases to the Copernicus DEM to more narrowly target urban structures and forest canopies to be removed in a way that brings the new DTM to more closely match known ground and lidar-measured elevations.  The result of their efforts is the Forest And Buildings removed Copernicus DEM (FABDEM), made available in December 2021.  Having much more sophisticated tools and data to work with than any SimCity fan is likely to possess, they have already done much of the hard work for us.  Details can be read in the paper "A 30 m global map of elevation with forests and buildings removed" on IOP.

FABDEM is downloadable from the University of Bristol FABDEM Dataset page.  Registration is not required.  The data are downloaded as 10° x 10° mosaic groups of 1° x 1° map tiles at 1 arc second grid spacing, so the download size will be large.  The auxiliary mask files that originally come bundled with Copernicus DEM GLO-30 tiles are not included.

 

 

I'm going to do another SimCity rendered map comparison again, this time of the original Copernicus DEM, a version of the Copernicus DEM edited with GIMP's Selective Gaussian Blur filter, and FABDEM.  The same Tokyo region map and SimCity terrain mods were used as in the initial post.  After trial and error, parameters for the Selective Gaussian Blur were narrowed down to Radius: 3 and Max Delta: 0.002 as having the best balance of effectiveness at removing undesirable buildings while not excessively removing desirable terrain.

Again, I will show the complete regions and then various focal areas.  I have set SimCity 4 to display at 1920 x 1080 resolution and taken screenshots at that resolution.  Images were converted to the lossy JPG format for upload to Imgur at a reasonable size.  Region images were reduced 25% in size for display, as free Imgur will automatically resize large images down to a maximum of 5,000 pixels width.  The Simtropolis Forums will resize embedded large images to maintain forum layout, so you may have to expand the images to see some details.

 

FULL REGION VIEW

COP-DEM GLO-30

SELECTIVE GAUSSIAN BLUR

FABDEM

Note how abstracted and almost cartoonish the low-lying areas have become.  It should be remembered that these are SC4 Mapper colors of elevation, while actually rendered SimCity maps should not have that effect.  Perhaps it does help emphasize that these are largely sedimentary plains and river drainage, and this DEM was developed specifically with an eye towards improved modeling of drainage and flooding.

 

 

MOUNT FUJI

COP-DEM GLO-30

Buildings and forests should not exist at the summit of Mount Fuji, so we should expect minimal to zero effects from any correction at the summit.  Mount Fuji does have forests at its base, so we definitely want to compare and see what changes have occurred there.

SELECTIVE GAUSSIAN BLUR

Note that shallower, fainter crevices at the base and periphery of the mountain begin to disappear, while the straight line of a power transmission line cutting through the forests at the lower left are largely unaffected.

FABDEM

The straight path of the power transmission lines has largely disappeared as the high canopy of the forest they cut through has been reduced.

 

 

MOUNT FUJI SLOPE AND SUMMIT

COP-DEM GLO-30

SELECTIVE GAUSSIAN BLUR

FABDEM

 

 

BASE OF MOUNT FUJI

COP-DEM GLO-30

SELECTIVE GAUSSIAN BLUR

FABDEM

The paper supporting this this DEM indicate they have further removed artifact pits still found in the Copernicus DEM, many of which no doubt are artifacts of forests and buildings.

 

 

CENTRAL TOKYO

COP-DEM GLO-30

Note that lumpiness clusters are especially visible around the Imperial Palace, Shinjuku, and Shibuya.  The lumps are the results of buildings being detected by satellite sensors and remaining as artifacts in the DEM.

SELECTIVE GAUSSIAN BLUR

FABDEM

Note that the process to remove the effects of building elevations in the DEM has resulted in terrain now showing as below sea level.  This actually is correct for this area of Tokyo, which can be at -1.0 m at low tide.  This does create a problem for SimCity 4, which does not model sea level in a way that would not submerge such low-lying elevation, and it will be necessary to further edit this area of the map to be above sea leave to be playable.

 

 

IMPERIAL PALACE

COP-DEM GLO-30

Note the lumpiness from artifacts of buildings in the original Copernicus DEM.  The starkest lump seems to be the Tokyo Midtown Hibiya tower and there is broad, pitted clustering around Tokyo Station.  Toranomon Hills and the Ark Hills Sengokuyama Mori Tower would both be in the area covered by this map, but must not have been built when the data for this DEM was collected or were otherwise not as exaggerated as the Hibiya tower.  Mount Atago is a hill at the middle of the southern edge of the map and which historically was one of the scenic high viewing points for panoramas of old Edo and old Tokyo.

SELECTIVE GAUSSIAN BLUR

Note how the Imperial Palace is losing its northern moat and definition, while the Tokyo Midtown Hibiya tower starkly remains.   This suggests that further smoothing using this method risks erasing the moats altogether while still not fully erasing the tower.

FABDEM

Note that the Imperial Palace moats are still recognizable, particularly along the northern section of the palace, while the Tokyo Midtown Hibiya tower as well all the towers around Tokyo Station have been significantly reduced in prominence.  Unfortunately, the historic Mount Atago, today somewhat hemmed in by taller buildings, may have been erased into faint obscurity.

 

 

TOKYO HARBOR

COP-DEM GLO-30

Note that the Minato Ward area also includes numerous skyscrapers and Tokyo Tower, though Tokyo Tower is not significantly picked out in the DEM.

SELECTIVE GAUSSIAN BLUR

Note that while a water selection mask was used, land areas on the periphery of that mask were affected by the smoothing such that they still inadvertently became submerged.

FABDEM

Note that while the researchers behind FABDEM also explicity barred areas approaching Copernicus DEM water masks from being affected by their elevation correcting process in order to preserve water body edges, that safeguard also doesn't seem to have fully held, as this map has lost some water edges, numerous islands and islets, and breakwaters.  In this map, the historical naval defense battery island Dai-Roku Daiba has completely vanished.  On the other hand, the large skyscrapers of Minato Ward have been effectively smoothed away.

 

 

SHIMIZU

COP-DEM GLO-30

Note that a section of the elevated Tomei Expressway is clearly traceable before it enters what would be the Okitsu Tunnel through the mountains at one end and the Satta Tunnel at the other end nearer the coast.  Further to the right in the image a smaller section of the Tokaido Shinkansen can be seen before it too enters a tunnel.  Also recall that the Copernicus DEM was already missing the breakwater leading into Shimizu Port.

SELECTIVE GAUSSIAN BLUR

FABDEM

 

 

TONE RIVER

COP-DEM GLO-30

SELECTIVE GAUSSIAN BLUR

FABDEM

Note that again large sections of the lowlands around the Tone River have become submerged.  This stretch of the Tone River does have high levees to prevent flooding of the surrounding farmlands, much of which are rice paddies.  Much of the submerged area was historically reclaimed land as major efforts began in the Edo period to control the course of the Tone River.

 

 

RICE PADDIES

COP-DEM GLO-30

SELECTIVE GAUSSIAN BLUR

FABDEM

 

 

CONCLUSIONS

GIMP's Selective Gaussian Blur offers us a method to input our own thresholds, but, because it is untargeted, the results are average and it is at the extremes that it fails.  It cannot fully erase large buildings without also erasing terrain features we might want to keep.  The erasure of the Imperial Palace or the shallower crevices of Mount Fuji are not acceptable to me, and while such features can be masked so that a stronger smoothing blur can be applied to the urban areas, there comes point where the effort at careful masking and filtering begs the question why not go with a dataset where that masking and filtering has already been done at a much more sophisticated level.

FABDEM does effectively remove buildings and forest canopies and creates map readily playable for SimCity without the need to use God Mod terrain editing tools to further level areas, but it introduces numerous below-sea-level issues for lowland areas for SimCity.  Thankfully, Copernicus DEM GLO-30 does come with auxiliary water masks and even transparency, so it is possible to carefully use these masks along with layers to restore coastal features or infill low-lying areas.  Broadly, I worry that FABDEM might have overzealously smoothed some areas, particularly urban areas where small water channels have been erased as the protection of the water mask was imperfect.  Fortunately, these are not irrepairable.

FABDEM, is...ahem...a fabulous recent upgrade of Copernicus DEM making it much more ready for SimCity.  It will take some additional work in addressing water body and lowland issues, but, especially in heavily built-up urban areas, it has already solved for SimCity mapmakers numerous issues with strategic DEM smoothing.