A hierarchical inventory of the world’s mountains for global comparative mountain science

The generation of this map of the world’s mountains consisted of five steps (Fig. 1): (i) the identification and hierarchisation of named mountain ranges and the recording of range-specific information; (ii) the manual digitization of the ranges’ general shape; (iii) the definition of mountainous terrain (and the inventory’s outer borders) using a DEM-based algorithm; (iv) the automatic refinement of the digitized and named ranges’ inner borders; and (v) the preparation of the final layers. The resulting products consist of a refined mountain definition (GMBA Definition v2.0), two versions of the inventory (GMBA Inventory v2.0_standard & GMBA Inventory v2.0_broad), and a set of tools to work with the inventories.

Step i: Identification and hierarchisation of mountain ranges

In a first step, we identified mountain ranges worldwide. To do so we adopted the mountain ranges identified in the GMBA Inventory v1.4^10,14 and searched existing resources in any languages for other named ranges not yet included. The ranges added could either be adjacent to, included in (child range or subrange) or including (parent range or mountain system) mountain ranges of the GMBA Inventory v1.4. The resources used for our searches included world atlases (e.g. The Times Comprehensive Atlas of the World¹⁹, Knaurs grosser Weltatlas²⁰, Pergamon World Atlas²¹); topographic maps (e.g. http://legacy.lib.utexas.edu/maps/imw/, http://legacy.lib.utexas.edu/maps/onc/, https://maps.lib.utexas.edu/maps/tpc/, www.topomap.co.nz, https://norgeskart.no, www.ign.es/iberpix/visor/); encyclopaedias (www.wikipedia.org; www.britannica.com); online gazetteers and reference sites (e.g. www.wikidata.org, www.geonames.org (GeoNames), www.mindat.org); mountain classification systems (e.g. the International Standardized Mountain Subdivision of the Alps or SOIUSA for the Alps²², Alpenvereinseinteilung der Ostalpen²³, Classification of the Himalaya²⁴, www.peakbagger.com/rangindx.aspx (PEMRACS), www.carpathian-research-network.eu/ogulist, http://www.sopsr.sk/symfony-bioregio/lkpcarporog, www.dinarskogorje.com, https://bivouac.com/, https://climbnz.org.nz/); and national or regional landscape, geomorphological, or physiographic maps and publications^{4,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42}. The full list of the consulted sources and references is available on GitHub at https://www.github.com/GMBA-biodiversity/Inventory (GMBA Mountain Inventory v2.0 References.pdf).

All identified mountain ranges were recorded in a Microsoft Access relational database (“Mountain database”, see below) and given a name, a unique 5-digit identifier (GMBA_V2_ID), and the corresponding Wikidata unique resource identifier (URI), when available. This URI gives access to a range’s name as well as to its Wikipedia page URL in all available languages and lists other identifiers for given mountain ranges in a variety of other repositories such as GeoNames or PEMRACS. The primary mountain range names were based on the resources used for range identification and were preferably recorded in English. Names used nationally, locally, as well as/or by indigenous people and local communities were extracted from Wikidata and recorded in a separate attribute field.

In the process of cataloguing, we attributed a parent range to each of the mapped mountain ranges. Information about parent ranges is included in PEMRACS, often also in Wikidata as a property that can be extracted though a SPARQL query, in the corresponding Wikipedia pages description, and in regional hierarchical mountain classifications that exist for the European Alps (SOIUSA), the Carpathians, and the Dinaric Alps. When no such information was available, we relied on other sources of information that we found either using a general web search (leading to specific papers, reports, or web pages on mountain ranges) or by consulting (online) topographical maps and atlases at different scales. The information about parent ranges was used to construct a hierarchy of up to 10 levels using a recursive SQL query (see Step v). The result of this step was a relational database with a hierarchy of mountain systems and (sub-) ranges (Fig. 1, “Mountain database”).

Step ii: Digitization of the mountain ranges

In a second step, we digitized all identified ‘childless’ mountain ranges (i.e. smallest mapping units, called ‘Basic’ as opposed to ‘Aggregated’ in the database) in one vector GIS layer. To do so, we used the Google Maps Terrain layers (Google, n.d.) as background and the WHYMAP named rivers layer⁴² as spatial reference since descriptions of mountain range areal extension is often given with reference to major rivers. The digitization, which was done in QGIS⁴³ using the WGS 84 / Pseudo-Mercator (EPSG 3857) coordinate reference system, consisted in the drawing of shapes (polygons) that roughly followed the core area of each mountain range. In general, the approximate shape and extent of the mountain ranges we digitized could be distinguished based on the terrain structure as represented by the shaded relief background that corresponded to the placement and orientation of the range’s name label on a topographical map, atlas or other resource. As the exact placement and orientation of mountain range labels in each specific source can be influenced by cartographic considerations (e.g. avoiding overlaps with other features), the final approximation of the mountain range was obtained by consulting a variety of sources for each mountain range. Occasionally, the mountain terrain’s geomorphological characteristics strongly hampered the accuracy of our visual identification of mountain subranges within larger systems. This was particularly the case in old, eroded massifs such as the Brazilian Highlands or the highlands of Madagascar, where individual mountain ranges are not separated by deep well-defined valleys and have a very complex topography. In these cases, we referred to available topographical descriptions of range extent and to the river layer (see above). Other complex regions included Borneo and the Angolan Highlands, whereas subranges in mountain systems such as the European Alps, the Himalayas, and the North American Cordillera were comparatively easy to map. Moreover, the density of currently available mountain toponymical information varied quite strongly between regions. Accordingly, regional variation in the size of the smallest mountain range map units can be considerable. The result of this step was a (manually) digitized vector layer of named mountain ranges shapes (Fig. 1, “Manual mountain shapes”).

Step iii: Definition of mountainous terrain

In a third step, we defined mountainous terrain (GMBA Definition v2.0). To distinguish mountainous from non-mountainous terrain, we developed a simple algorithm which we implemented in ArcMap 10.7.1⁴⁴. This algorithm is based on ruggedness (defined as highest minus lowest elevation in meter) within eight circular neighbourhood analysis windows (NAWs) of different sizes (from 1 pixel (≈ 250 m) to 20 (≈ 5 km) around each point, Fig. 2c) combined with empirically derived thresholds for each NAW (Fig. 2). The decision to use multiple NAW sizes was made because calculating ruggedness based on only a small or a large NAW comes at the risk of identifying the many local irregularities typically occurring in flat or rolling terrain as mountainous or of including extensive flat ‘skirts’ through the smoothing and generalization of large NAWs³. Accordingly, our approach ensures that any point in the landscape classified as mountainous showed some level of ruggedness not only at one but across scales. This also resulted in a smooth and homogeneous delineation of mountainous terrain, very suitable for our mapping purpose.

We used the median value of the 7.5 arc second GMTED2010 DEM⁴⁵ as our source map. To reduce the latitudinal distortion of the raster, and thus the shape and area of the NAWs, we divided the global DEM into three raster layers corresponding to three latitudinal zones (84° N to 30° N, 30° N to 30° S and 30° S to 56° S) excluding ice-covered Antarctica and projected the two high latitude zones to Lambert Azimuthal Equal Area and the equatorial zone to WGS 1984 Cylindrical Equal Area. We used these reprojected DEM layers to produce eight ruggedness layers, each using one of the eight NAWs.

To determine the threshold values of our algorithm, we selected 1000 random points within the area defined by the geometric intersection (Fig. 1b) of the three commonly applied mountain definitions, i.e. the definitions by UNEP-WCMC⁴⁶, GMBA¹⁵, and USGS³. These layers (referred to as K1, K2, and K3, respectively by Sayre and co-authors¹²) were obtained from the Global Mountain Explorer⁴⁷. We eliminated 80 clearly misclassified points (i.e., points that fell within lakes, oceans, or clearly flat areas according to the shaded relief map we used as a background) and used the remaining 920 to sample the eight ruggedness layers. For each of the 8 layers, we retained the lowest of the 920 ruggedness values as the threshold for the layer’s specific NAW (Fig. 2c). The eight threshold values were then used to reclassify each of the eight layers by attributing the value 1 to all cells with a ruggedness value higher than or equal to the corresponding threshold and the value 0 to all other cells. Finally, we performed a geometric intersection (see Fig. 1b) of the eight reclassified layers to derive the new mountain definition.

After these calculations, we reprojected the three raster layers to WGS84 and combined them through mosaic to new raster. To eliminate isolated cells and jagged borders, we then generalized the resulting raster map by passing a majority filter (3 × 3 pixels, majority threshold) three times. This layer corresponds to the GMBA Definition v2.0.

The resulting mountain definition (GMBA Definition v2.0) distinguishes itself from previous ones because of the empirically derived thresholds method used to develop it and the use of eight NAWs. In line with the previous GMBA definition, it relies entirely on the ruggedness values within NAWs. The GMBA Definition v2.0 was used to determine the outer delineation of this inventory’s mountainous terrain. As expected, it includes neither the wide ‘skirts’ of flat or undulating land around mountain ranges nor the topographical irregularities that are both typically included when other approaches are applied. It also successfully excludes extensive areas of rolling non-mountainous terrain such as the 52,000 km² Badain Jaran Desert sand dunes in China. However, this mountain definition is conservative and only includes the highest, most rugged cores of low mountain systems, as for example in the Central Uplands of Germany, and therefore excludes some lower hill areas still considered by some as mountains.

As a further step towards generalization, we considered that small (<100 km²) inner-mountain flat areas corresponding to valley floors, small depressions, and isolated high plateaus were part of the mountainous terrain. Additionally, to avoid self-intersecting polygons in the final product we also eliminated mountain ‘appendages’ consisting of isolated raster cells smaller than 2 km² and touching the main mountain area through one corner only. For the generalization process, we used a vectorized version of the mountain definition that we reconverted to a raster file for use in the creation of the GMBA Inventory v2.0_standard shapes in step iv. This simplification of the GMBA Definition v2.0 was deemed necessary to generate the cleanest possible range shapes. Despite this generalization, we consider that these shapes can be used for most comparative mountain studies. However, for very precise area calculations, the new inventory layer can be intersected with the GMBA Definition v2.0.

Step iv: Refinement of the ranges’ inner borders

To generate the final shape layer, we extended the hand-drawn polygons to the nearest surrounding river by allocating the value of the GMBA_V2_ID to all intermediate raster cells using the ArcMap tool ‘Cost Allocation’. For this, we intersected the simplified GMBA Definition v2.0 (output step iii) with a rasterized river layer from Hydrosheds². This resulted in a mask layer (Fig. 1, “Mask (GMBA)”) with value 1 for areas that are mountainous and not rivers and ‘NoData’ for non-mountain areas or rivers. We then combined this mask with the digitized vector layer of named ranges (Fig. 1 “Manual mountain shapes”, output step ii) and allocated to each cell the 5-digit GMBA_V2_ID value of the nearest digitized mountain range shape. In the allocation, rivers and non-mountain areas (value ‘NoData’ in the mask) acted as barriers to the cost allocation. This resulted in individual mountain ranges separated by rivers inside the overall GMBA Definition v2.0 area. As the river cells maintained a value ‘NoData’ during the cost allocation operation, we performed a second cost allocation with a mask consisting of the GMBA Definition v2.0 only, to fill these (river) cells with the nearest GMBA_V2_ID. Finally, we converted the resulting raster map to shapes representing the smallest mountain map units (‘Basic’ unit) of our inventory (Fig. 1, “Processed mountain shapes”).

Step v: Preparation of the final products

To create the final products, we first developed a recursive query to convert the parent-child relations recorded in the “Mountain database” (output step i) into unique hierarchical paths (see Fig. 3e), leading from the basic mapping unit up to the highest level of aggregation (Level 1, continents and oceans). We then combined the mountain range shapefile layer (“Processed mountain shapes”, output step iv) with an export query of the “Mountain database” as an attribute table containing the complete hierarchy for each mapped mountain range. This allowed us to construct all the higher parent ranges levels by dissolving according to individual levels in the hierarchy. The resulting layer representing mountain range shapes at the ten levels in the hierarchy were merged into one final ‘stacked’ shapefile entitled GMBA Inventory v2.0_standard containing all mountain range shapes at all (overlapping) hierarchical levels (Fig. 3).

To produce a mountain inventory that can be intersected with any of the three mountain definitions currently in use and available on the Global Mountain Explorer, we also applied the ‘cost allocation’ (step iv) to an additional, considerably broader mountain layer that we obtained in step (iii) by geometric union (Fig. 1c) of the UNEP-WCMC, GMBA, and USGS layers. A 5 km buffer was added to ensure that small, isolated mountain patches would be connected and thus facilitated the cost allocation procedure. We then processed the resulting layer (“Broad”) following the exact same steps (iv and v) as above to generate a second version of the raster map and a second version of the ‘stacked’ shapefile. This second version, entitled GMBA Inventory v2.0_broad, enables comparative mountain science based on a different definition of mountains than the one presented here but needs to be intersected with the chosen definition before use.

Source: Ecology - nature.com