Prominence Update 2023


In late 2022, I re-ran my global prominence analysis using 30m (one arcsecond) data from the Copernicus GLO30 digital elevation model, which was collected between 2011 and 2015. Like the previous 90m SRTM data, this is a surface model, meaning that it includes trees, buildings, and other things that are on top of the land. Coverage is nearly global, except for an area covering Armenia and Azerbaijan. I was able to fill those areas using the ALOS World 3D DEM, which is also 30m. It was collected between 2006 and 2011.

My previous analysis used 10m data in the US outside of Alaska, so I preserved those results and merged them into the GLO30 results elsewhere. That means the artifacts along the Canadian border are still present. However, other artifacts are fixed: spurious peaks along the Mexico/Guatemala border, and small areas of missing results in eastern Quebec and the coast of Norway.

Since 2017, I have improved the code to use data in floating point meters, rather than integer feet as before. This avoids losing a little bit of precision. I've also optimized the programs somewhat, especially the program that merges divide trees. Merging is now done in parallel in multiple threads, removing the need to merge in multiple steps. Even all of Eurasia was merged in a single run, and took about 30 minutes with 8 threads.

A visualization of the new results is here. (The previous results from 2017 are now here).

A zipped CSV file with the results is here. This is a list of every calculated peak with at least 100 feet of prominence, sorted by decreasing prominence. The elevation and prominence values are in floating point meters. (The previous CSV from 2017 is still here).

Peak counts

A summary of the number of peaks found by the previous analysis ("90m data") and the current one ("30m data") is below. Central Ameraica and South America are not strictly comparable between the two analyses, as the boundaries changed. In the new analysis, the border between Central and South America is the Panama Canal, which eliminated some of the artifacts in the area in the previous analysis. The USA outside Alaska is unaffected since the results were identical in the two analyses.

Sampling tends to underestimate summit elevations; a finer sampling underestimates less. The new analysis found 24% more P300s. The effect is less in areas that had some 30m data in the previous analysis, such as Canada, and greater in areas with the worst data in the previous analysis, such as Antarctica.


Below is a list of peaks with at least 1500m prominence that differ between this 30m analysis and the previous 90m analysis. Most of these ultras are already well-known, which indicates the importance of comparing DEMs with surveyed elevations, especially when using a surface model that includes treetops. DEMs tend to underestimate summit elevations since they average over an area. Entries in bold are the most plausible changes and should be verified against better local data like a topo map.


Lalla Khedidja, not Ras Timedouine.


Dent de Crolles, not Chamechaude

Aletschhorn, not Finsteraarhorn

Adamello, not Cima Presanella

Torstein, not Hoher Dachstein

Negoiu, not Varful Moldoveanu

Titov Vrh, not Korab

Not Sarektjåhkkå P1495

Not Ellmauer Halt P1484


28.1031,89.5478 P1500 China/Bhutan

38.9986,70.8786 P1527, not Petra Pervogo Range High Point

29.1981,96.9689 P1683 not Peak 6327, China

15.6181,121.3680 P1594 not Mingan Mountains High Point, Philippines

Kohe Shakawr, not Udren Zom, Pakistan

33.7503,74.4667 P1919 not Sanset, India

Not Shah Dhar P1488

Not Kataklik Kangri P1457

Not Rangrik Rang P354

Not Panch Chuli P1397

Not Gora Cherskogo P1470

Not Gora Oval'naya Zimina P1351

Not Jethi Bahurani P1421

Not Gaurishankar P1400

Not Melungtse P1426

Not Burnag Kangri P1387

Not Kaimuk Kangri P1439

Not Mazar Shan P1408

South America

Veronica, Peru not Nevado Sahuasiray

-46.6150,-72.9358 P1543 not Peak 2300, Chile

Not Wellington Island High Point, Chile P1472

Not Aprada-tepuí, Venezuela P1384

Not Auyan-tepuí, Venezuela P1495

Not Monte Roraima, Venezuela P1174

Not Nevado Champará, Peru P1477

North America

A smattering of alternative ultras, not supported by summit elevations in Canadian topo maps. Some examples:

Ambition Mountain not Scud Peak, British Columbia

Peak 2454, British Columbia not Lehua Mountain

62.1647,-148.7317 P1763 not Sovereign Mountain, Alaska

Prominence less than 1500m:

Not Sharktooth Mountain, BC P1479

Not Gataga Peak, BC P1472

Not Mount Macdonald, Yukon Territory P1498

Not Dunn Peak, BC P1485

Not Eglinton-Sam Ford Peak, Nunavut P1491


71.7156,-52.6092 P1565 not Solo Snow Dome


Potential new ultras:

Mount Bouvier P1663

Mount Sanderson P1533

Flat Top P1502

-67.4497,-67.0600 P1956

-61.0906,-54.7272 P1786 looks spurious

Not Mount Morning P1232

Not Dome Argus P1344

Known, but not in 90m analysis

These peaks were not ultras in the 90m analysis, but are already known to be ultras in Peakbagger:

Akbaba Tepesi, Turkey; Kiyamaki Dagh, Iran; Lavar, Iran; Denband, Iran; Kuhsefid, Iran; B21, Pakistan; Kolahoi, India; Khrebet Baldyrgannyg High Point, Russia; Kennedy Peak, Myanmar; Tongshanjiabu, China/Bhutan; Peak 6447, China; Kotaklik Shan, China; Geladandong, China; Kone Kangri, China; Baojie Ling, China; Qingliang Feng, China; Doro Dindi, Indonesia; Soputan, Indonesia; Mount Shungol, Papua New Guinea; Cerro Castillejos, Argentina; Cerro de Incahuasi, Chile/Argentina; Volcán Lonquimay, Chile; Volcán Puntiagudo, Chile; Sierra de Sangra, Argentina; Nordkronen, Greenland; Takkerne, Greenland; Frænkel Land High Point, Greenland; Mountain 2, Greenland; Berzelius Bjerg, Greenland; Lauper Bjerg, Greenland; Agdlerulik, Greenland; Peak 1830, Nunavut; Zirbitzkogel, Austria; Pelister, North Macedonia; Psili Koryfi, Greece; Mount Isto, Alaska; Pik Sovetskoy Konstitutsii, Kyrgyzstan; Monar, Iran; Djangspitze, Kyrgyzstan


In 2022, researchers at the University of Bristol released FABDEM, a version of the Copernicus GLO30 DEM with trees and buildings removed, in an attempt to make the surface model into a bare-earth terrain model. They did this by training a machine learning model on GLO30 data and other publicly available data like forest and building coverage maps, and then using it to estimate the difference between GLO30 elevations and LIDAR elevations. Their paper is here and their data is available here.

Although it's advertised as global, there are several gaps in coverage:

  • No coverage south of 60 degrees south latitude, which presumably isn't a big deal considering the lack of tree and building coverage there.

  • No coverage north of 80 degrees north latitude. This is a problem for prominence because it would introduce artificial peaks near the boundary.

  • The same gap over Armenia and Azerbaijan as GLO30.

  • A small number of tiles in northern latitudes have a single column of invalid values (NaNs).

I filled the Armenia/Azerbaijan hole with ALOS data, the same as in my GLO30 analysis. North of 80 degrees I copied over GLO30 data. I didn't bother analyzing Antarctica. Also, since I considered the data somewhat experimental, I did not splice in the 10m results for the U.S. There are many fewer peaks found in the U.S. with 30m data versus 10m data.

A visualization of the results is here.

A zipped CSV file with the results is here.

Below is a table comparing the number of peaks with at least 300 feet of prominence in GLO30 vs. FABDEM, in regions where the analysis is comparable. (Antarctica is not present in FABDEM, and North America is not comparable because I didn't splice in the 10m data for the U.S. in the FABDEM analysis)