Mapping the Loss of Mt. Kenya's Glaciers

geosciences

Article Mapping the Loss of Mt. Kenya’s Glaciers: An Example of the Challenges of Satellite Monitoring of Very Small Glaciers

Rainer Prinz 1,* ID , Armin Heller 2, Martin Ladner 2,3 ID , Lindsey I. Nicholson 4 and Georg Kaser 4 1 Department of Geography and Regional Science, University of Graz, Heinrichstrasse 36, 8010 Graz, Austria 2 Institute of Geography, University of Innsbruck, Innrain 52, 6020 Innsbruck, Austria; [email protected] (A.H.); [email protected] (M.L.) 3 Austrian Alpine Club, Olympiastrasse 37, 6020 Innsbruck, Austria 4 Institute of Atmospheric and Cryospheric Sciences, University of Innsbruck, Innrain 52, 6020 Innsbruck, Austria; [email protected] (L.I.N.); [email protected] (G.K.) * Correspondence: [email protected]; Tel.: +43-316-380-8296

Received: 31 March 2018; Accepted: 5 May 2018; Published: 11 May 2018

Abstract: Since the last complete glacier mapping of Mt. Kenya in 2004, strong glacier retreat and glacier disintegration have been reported. Here, we compile and present a new glacier inventory of Mt. Kenya to document recent glacier change. Glacier area and mass changes were derived from an orthophoto and digital elevation model extracted from Pléiades tri-stereo satellite images. We additionally explore the feasibility of using freely available imagery (Sentinel-2) and an alternative elevation model (TanDEM-X-DEM) for monitoring very small glaciers in complex terrain, but both proved to be inappropriate; Sentinel-2 because of its too coarse horizontal resolution compared to the very small glaciers, and TanDEM-X-DEM because of errors in the steep summit area of Mt. Kenya. During 2004–2016, the total glacier area on Mt. Kenya decreased by 121.0 × 103 m2 (44%). The largest glacier (Lewis) lost 62.8 × 103 m2 (46%) of its area and 1.35 × 103 m3 (57%) of its volume during the same period. The mass loss of Lewis Glacier has been accelerating since 2010 due to glacier disintegration, which has led to the emergence of a rock outcrop splitting the glacier in two parts. If the current retreat rates prevail, Mt. Kenya’s glaciers will be extinct before 2030, implying the cessation of the longest glacier monitoring record of the tropics.

Keywords: glacier monitoring; glacier inventory; satellite remote sensing; Pléiades satellite images; Sentinel-2; TanDEM-X; DEM; Mount Kenya; tropical glacier

1. Introduction Glaciers act as low pass ﬁlters of climate variability [1], and are, therefore, key indicators of climate change [2]. For this reason, glacier monitoring is implemented in the Global Climate/Terrestrial Observing System (GCOS/GTOS) via a Global Hierarchical Observing Strategy (GHOST) [3]. Following the tiers deﬁned in this strategy, long-term observations of glaciers currently demonstrate historically unprecedented global glacier area and mass losses [4], which pose challenges to existing monitoring networks, such as glacier disintegration, increasing debris cover, and complete extinction of glaciers [5]. To keep pace with the rapid ongoing changes, glacier monitoring at appropriate spatial and temporal scales becomes increasingly important [6,7]. Enhanced Earth observing systems facilitate glacier monitoring in most remote areas [8], however, the variations of very small glaciers (<0.1 km2) in complex mountainous terrain remain a challenge for detection [9]. The glaciers on Mt. Kenya (0◦90 S, 37◦180 E) are among the best studied tropical glaciers. Glacier changes have been reported since the late 19th century, including (sub) decadal mappings

Geosciences 2018, 8, 174; doi:10.3390/geosciences8050174 www.mdpi.com/journal/geosciences Geosciences 2018, 8, 174 2 of 14

Geosciences 2018, 8, x FOR PEER REVIEW 2 of 14 from 1934 onwards, and surface mass balance measurements from 1979 to 1996, and from 2010 to 2014 [10–The12] (andglaciers references on Mt. Kenya therein). (0°9′ S, The 37°18 last′ E) glacierare among inventory the best studied on Mt. tropical Kenya glaciers. was carried Glacier out in changes have been reported since the late 19th century, including (sub) decadal mappings from 1934 2004 [13], showing that between 1899 and 2004, the total area of 18 glaciers decreased from 1.64 km2 to onwards, and surface mass balance measurements from 1979 to 1996, and from 2010 to 2014 [10–12] 2 − 0.27 km(and (references84%), while therein). 8 glaciers The last vanishedglacier inventory completely on Mt. [Kenya14]. The was most carried recent out in mapping 2004 [13], ofshowing the largest glacierthat (Lewis, between Figure 1899 and1) was 2004, performed the total area in of 2010 18 glaciers [ 15,16 decreased], which from conﬁrmed 1.64 km² the to 0.27 strong km² retreat (−84%), rates, indicatingwhile that8 glaciers between vanished 1934 completely and 2010 the[14]. glacier The most lost recent 90% (79%) mapping of its of volumethe largest (area) glacier [12 ].(Lewis, A climate sensitivityFigure study1) was based performed on energy in 2010 balance [15,16], modellingwhich confirmed found the that strong glacier retreat recession rates, indicating since the latethat 19th centurybetween can be 1934 primarily and 2010 attributed the glacier lost to atmospheric 90% (79%) of dryingits volume resulting (area) [12]. in reducedA climate cloudiness, sensitivity study snowfall, and albedobased on [17 energy]. Similar balance studies modelling proved found the that climate glacier proxy recession potential since the of tropicallate 19th glacierscentury can [18 be–20 ] as theyprimarily capture climate attributed signals to atmospheric from the tropicaldrying result mid-troposphere,ing in reduced cloudiness, where our snowfall, understanding and albedo is scarce and controversial[17]. Similar studies [21,22 proved]. Although the climate Mt. Kenya’s proxy potential glaciers of have tropical limited glaciers socioeconomic [18–20] as they relevance capture [23], climate signals from the tropical mid-troposphere, where our understanding is scarce and they have the longest record of glacier monitoring in the tropics. Recent reports of substantial glacier controversial [21,22]. Although Mt. Kenya’s glaciers have limited socioeconomic relevance [23], they changeshave [ 24the,25 longest] demand record an of update glacier of monitoring Mt. Kenya’s in the glaciers tropics. inventory, Recent reports offering of substantial unique witnessing glacier of the impendingchanges [24,25] deglaciation demand an of update an entire of Mt. massif. Kenya’s glaciers inventory, offering unique witnessing of Directthe impending glacier observationsdeglaciation of on an Mt.entire Kenya massif. ceased in 2014 [24], leaving remote sensing as the only option forDirect a continuous glacier observations monitoring on strategy. Mt. Kenya In ceased this study, in 2014 we [24], use leaving very high remote resolution sensing (0.5as the m) only satellite imagesoption from for Pl aé iadescontinuous tri-stereo monitoring scenes strategy. (acquired In th inis 2016), study, whichwe use are very processed high resolution to an (0.5 orthophoto m) and tosatellite a digital images elevation from Pléiades model tri-stereo (DEM) ofscenes the summit(acquired area.in 2016), From which the are orthophoto, processed weto an derive a neworthophoto glacier inventoryand to a digital of Mt. elevation Kenya model and quantify(DEM) of the glacier summit area area. changes From the by orthophoto, comparison we to the derive a new glacier inventory of Mt. Kenya and quantify glacier area changes by comparison to the previous inventory, which was derived from the last complete glacier mapping in 2004. For quantifying previous inventory, which was derived from the last complete glacier mapping in 2004. For glacier volume changes, we subtract Lewis Glacier DEMs from 2016 and 2010. Additionally, to explore quantifying glacier volume changes, we subtract Lewis Glacier DEMs from 2016 and 2010. the capabilityAdditionally, of available to explore remote the capability sensing of available data to monitor remote sensing the changes data to ofmonitor Mt. Kenya the changes glaciers, of Mt. we test the PlKenyaéiades glaciers, orthophoto we test and the DEM Pléiades against orthophoto freely availableand DEM products,against freely like available Sentinel-2 products, satellite like images and theSentinel-2 TanDEM-X-DEM. satellite images and the TanDEM-X-DEM.

Figure 1. Lewis Glacier above Lewis Tarn seen from its late 19th century lateral moraine on 01 March Figure 1. Lewis Glacier above Lewis Tarn seen from its late 19th century lateral moraine on 2011. Note the steep walls of the two highest peaks of the massif, Batian (5199 m) and Nelion (5188 01 March 2011. Note the steep walls of the two highest peaks of the massif, Batian (5199 m) and m), which rise to the upper left corner. Photo credit R. Prinz. Nelion (5188 m), which rise to the upper left corner. Photo credit R. Prinz. 2. Materials and Methods 2. Materials and Methods 2.1. Pléiades Tri-Stereo DEM and Orthophoto 2.1. Pléiades Tri-Stereo DEM and Orthophoto The Pléiades system (operated by the Centre National d’Études Spatiales) consists of two Theidentical Pléiades optical system satellites (operated Pléiades by 1A the and Centre 1B, positioned National with d’É atudes 180° phase Spatiales) shift in consists a sun-synchronous of two identical optical satellites Pléiades 1A and 1B, positioned with a 180◦ phase shift in a sun-synchronous orbit at 694 km altitude enabling a daily acquisition of any point on Earth. During a single ﬂy Geosciences 2018, 8, x FOR PEER REVIEW 3 of 14 orbit at 694 km altitude enabling a daily acquisition of any point on Earth. During a single fly in longitudinal-scanGeosciences 2018, 8, 174 mode (along-track), image triplets (tri-stereo) can be taken in a few seconds in3 ofthe 14 direction of forward, backward, and near perpendicular (nadir). Access to the Pléiades scenes was granted via a mapping project of the Austrian Alpine Club in longitudinal-scan mode (along-track), image triplets (tri-stereo) can be taken in a few seconds in the (http://www.gis.tirol/AV.MAP/). For a new edition of the trekking map of Mt. Kenya six Pléiades tri- direction of forward, backward, and near perpendicular (nadir). stereo images (Table 1) were purchased from Airbus Defense and Space, and serve as terrain and Access to the Pléiades scenes was granted via a mapping project of the Austrian Alpine Club land surface layers. The total map domain covers around 1050 km². In this study, we examine only a (http://www.gis.tirol/AV.MAP/). For a new edition of the trekking map of Mt. Kenya six Pléiades subdomain covering the summit area of Mt. Kenya (6.3 km²), representing the same domain as the tri-stereo images (Table1) were purchased from Airbus Defense and Space, and serve as terrain and map of 2004 [13]. The Pléiades acquisition time at Mt. Kenya is around 8:00 UTC (11:00 local time, land surface layers. The total map domain covers around 1050 km2. In this study, we examine only a Table 1). This timing fortunately avoids problems associated with large shadows, as the solar subdomain covering the summit area of Mt. Kenya (6.3 km2), representing the same domain as the elevation is high at this time, and also issues of clouds, which tend to form from noon onwards map of 2004 [13]. The Pléiades acquisition time at Mt. Kenya is around 8:00 UTC (11:00 local time, following the pronounced diurnal weather cycle in the tropics [17]. The acquisition date at the end of Table1). This timing fortunately avoids problems associated with large shadows, as the solar elevation the dry season, is ideal for glacier monitoring as it is generally associated with the minimum extent is high at this time, and also issues of clouds, which tend to form from noon onwards following of seasonal snow cover. the pronounced diurnal weather cycle in the tropics [17]. The acquisition date at the end of the dry season, is ideal for glacier monitoring as it is generally associated with the minimum extent of seasonal Table 1. Details of the Pléiades tri-stereo primary products used in this study. snow cover. Global Track Track Solar Solar Time Satellite ImagingTable 1. DateDetails of the PléiadesIncidence tri-stereo primaryIncidence products Incidence used in thisAzimuth study. Elevation (UTC) (°) Across (°) Along (°) (°) (°) Time Global Track Incidence Track Incidence Solar Solar Satellite Imaging Date PHR 1B 21 December 2015(UTC) 7:58:54Incidence (◦5.62) Across (◦) 5.36 Along 1.69(◦) Azimuth 138.3 (◦) Elevation 58.0 (◦) PHRPHR 1A 1B 21 22 December February 2015 2016 7:58:54 8:02:30 5.62 3.03 5.36−1.91 1.692.36 138.3130.0 58.058.0 PHRPHR 1B 1A 22 4 FebruaryFebruary 2016 2016 8:02:30 8:02:45 3.03 2.94 −1.91−2.85 2.36−0.76 130.0124.0 58.058.2 PHRPHR 1A 1B 174 February February 2016 2016 8:02:45 8:02:14 2.94 4.99 −2.85 −3.63 −0.76−3.43 124.0116.3 58.258.5 PHR 1A 17 February 2016 8:02:14 4.99 −3.63 −3.43 116.3 58.5 PHRPHR 1A 1A 17 17 February February 2016 2016 8:02:26 8:02:26 4.07 4.07 −3.40−3.40 2.242.24 116.0116.0 59.959.9 PHRPHR 1B 1B 23 23 February February 2016 2016 8:06:20 8:06:20 12.55 12.55 −11.83−11.83 −4.29−4.29 112.6112.6 61.961.9

To generate the terrain model, PlPléiadeséiades tri-stereo images for the entire domain were processed to a DEM (PLE) with ESRI ArcGIS Pro (version 2.01, Environmental System Systemss Research Research Institute, Institute, Redlands, CA, USA), according according to to the the scheme scheme shown shown in in Figure Figure 22,, usingusing nono groundground controlcontrol points.points. Mosaicking overlapping scenes scenes reduced reduced cloud cover. Point Point clouds resulting from the stereo model builder were filtered filtered for outliers (no noise) to improve the terrain model, which was used for orthorectificationorthorectification ofof the the panchromatic panchromatic and and multispectral multispectral images. images. Applying Applying pansharpening, pansharpening, the images the imageswere finally were combinedfinally combined to high geometricto high geometri resolutionc resolution images. Thus,images. three Thus, orthophotos three orthophotos (forward, (forward,backward, backward, and nadir) and with nadir) no cloudwith no cover cloud in co thever summit in the summit domain domain and a digitaland a surfacedigital surface model model(DSM) (DSM) at 1 m at resolution 1 m resolution (UTM (UTM WGS84, WGS84, elevations elevations referring referring to mean to mean sea level)sea level) are are available available for forour our purpose purpose (Table (Table2). Note,2). Note, because because in thein the summit summi domaint domain of of Mt. Mt. Kenya Kenya plants plants do do not not grow grow tallertaller than tussock, and there are just a few small huts, th thereere is no difference between the surface elevation and the terrain elevation. Hence, we use the terms DSM and DEM synonymously.

Figure 2. Step-by-step processing of the digital surf surfaceace model (DSM) and the orthophoto with ESRI ArcGIS Pro 2.01. * calculation step individually per scene. ESRI provides different matching algorithms for DEM construction, of which we tested two: the semi-global matching (SGM) algorithm [26] and the extended terrain matching (ETM) tool, Geosciences 2018, 8, x FOR PEER REVIEW 4 of 14

Geosciences 2018, 8, 174 4 of 14 ESRI provides different matching algorithms for DEM construction, of which we tested two: the semi-global matching (SGM) algorithm [26] and the extended terrain matching (ETM) tool, a feature- abased feature-based stereo matching stereo matching algorithm algorithm based on based the Harris on the feature Harris featurepoint detector point detector [27]. SGM [27]. derives SGM derives much muchdenser denser point pointclouds, clouds, but presumably, but presumably, due dueto the to the distortion distortion of ofthe the steep steep terrain, terrain, the algorithmalgorithm introducedintroduced artefactsartefacts toto thethe DEMDEM matching,matching, whichwhich causedcaused invalidinvalid results.results. ETM provided much better resultsresults forfor cartographiccartographic purposes than SGM, and was ﬁnallyfinally used in this study. FigureFigure3 3 shows shows the the differencesdifferences andand illustratesillustrates thethe defectivedefective resultsresults ofof thethe SGMSGM algorithm.algorithm.

(a) (b)

Figure 3.3.Contour Contour lines lines (20 (20 m) m) derived derived from from PLE withPLEdifferent with different algorithms: algorithms: (a) SGM (semi-global(a) SGM (semi-global matching), andmatching), (b) ETM and (extended (b) ETM terrain (extended matching), terrain which matching), better resolves which thebetter features resolves of the the mountain features summit. of the Themountain minimum summit. (maximum) The minimum elevations (maximum) using SGM elevations and ETM using algorithm SGM areand 3168 ETM m algorithm (7457 m) andare 42043168 m (5156(7457 m),m) and respectively. 4204 m (5156 The m), altitude respectively. reference The for altitude the highest reference peak (Batian)for the highest is 5199 peak m. (Batian) is 5199 m. 2.2. The Survey of 2004 2.2. The Survey of 2004 In 2004, the last complete glacier inventory of Mt. Kenya was compiled in a map (scale 1:5000) basedIn on 2004, aerial the photogrammetry last complete glacier [13]. Asinventory the original of Mt. data Kenya is unavailable was compiled (S. Hastenrath in a map (scale pers. comm.)1:5000) thisbased map on aerial was digitized photogrammetry to derive [13]. the glacierAs the original extents data from is 2004, unavailable and to serve(S. Hastenrath as a reference pers. comm.) for the comparisonthis map was with digitized Pléiades to and derive Sentinel the glacier imagery. extents As the from map 2004, coordinates and to followserve as the a virtualreference projection for the establishedcomparison duringwith Pléiades an early and expedition Sentinel toimagery. Mt. Kenya As the [28 map], georeferencing coordinates follow to the the WGS84 virtual UTM projection datum wasestablished necessary. during To achievean early this,expedition we used to Mt. 10 groundKenya [28], control georeferencing points (GCPs), to the well WGS84 distributed UTM datum over thewas domain.necessary. Four To achieve GCPs were this, surveyedwe used 10 during ground an control earlier points study (GCPs), [15] around well distributed Lewis Glacier over and the complementeddomain. Four byGCPs field were measurements surveyed during surveyed an withearlier a Trimble study Receiver[15] around R1 (real-timeLewis Glacier kinematic and accuracycomplemented 0.7–1.0 by m), field concurrently measurements to the surveyed Pléiades acquisitionwith a Trimble [29]. AllReceiver GCPs R1 can (real-time be identified kinematic in the fieldaccuracy as well 0.7–1.0 as in m), the concurrently map, because to they the representPléiades acquisition prominent [29]. peaks All or GCPs boulders, can be or theidentified color marks, in the paintedfield as well for theas in 2004 the airbornemap, because survey, they which represen aret still prominent visible inpeaks the or field boulders, and depicted or the color in the marks, map. Thepainted digitized for the map 2004 was airborne transformed survey, into which the WGS84 are still UTM visible datum in the using field aand spline depicted transformation, in the map. which The performeddigitized map well was in similartransformed studies into [30 ,the31]. WGS84 The transformation UTM datum isusing exact a atspline the GCPs, transformation, and no warping which wasperformed detected. well in similar studies [30,31]. The transformation is exact at the GCPs, and no warping was detected.In the next step, the map contour lines were digitized semi-automatically using the ArcScan extensionIn the from next ESRI’sstep, the ArcMap map contour 10.6, and lines corrected were digitized with the semi-automatically topology rules defined using in the ArcGis ArcScan Pro 2.1.1extension to avoid from topologicalESRI’s ArcMap errors. 10.6, Additionally, and corrected 90with elevation the topology points rules digitized defined manually in ArcGis from Pro 2.1.1 the mapto avoid complemented topological theerrors. elevation Additionally, information 90 elevation from the points contour digitized lines. Finally,manually using from ArcMap’s the map topo2rastercomplemented tool, the based elevation on the information Anudem algorithm from the [ 32co],ntour all elevation lines. Finally, information using ArcMap’s was interpolated topo2raster to a hydrologicallytool, based on consistent the Anudem DEM (ROS)algorithm with a[32], cell sizeall elevation of 2 m (Table information2). was interpolated to a hydrologically consistent DEM (ROS) with a cell size of 2 m (Table 2).

Geosciences 2018, 8, 174 5 of 14

2.3. Additional Topographic Data

2.3.1. Sentinel-2 Scene The Sentinel-2 mission (European Space Agency) consists of two polar orbiting satellites with a phase shift of 180◦. The mission provides multispectral images (horizontal resolution for different bands 10/20/60 m) for earth exploring research. Due to a larger number of bands than the Pléiades instrument, it is possible to derive a normalized difference snow and ice index (NDSI) enabling a semi-automatic classiﬁcation of ice [9]. We used a Sentinel-2A scene (acquired on 15 March 2016, cloud free over the summit region, downloaded from https://scihub.copernicus.eu/) to explore the semi-automatic classiﬁcation of ice as an alternative method of glacier delineation from Pléiades scenes.

2.3.2. TanDEM-X-DEM The TanDEM-X-DEM (TDX) [33] is a global digital elevation model at ~12 m horizontal resolution generated by applying RADAR interferometry from space. TDX access for scientiﬁc use was granted by the DLR (German Aerospace Center, TerraSAR-X/TanDEM-X mission) through its call for global data access in 2017. As its horizontal resolution falls between the DEM of the Shuttle Radar Topography Mission (SRTM) or the global DEM of Advanced Spaceborn Thermal Emission and Reﬂection Radiometer (ASTER GDEM) (~30 m) and–usually expensive–very high resolution products, TDX is valuable for various studies depending on high resolution terrain recording [34–36]. TDX constraints are the signal penetration into snow and ice and the time period of acquisition ranging over multiple years. For this study, we focused on the terrain representation of TDX in the same domain as the PLE to focus on the potential of this freely-available TDX product as a tool for monitoring very small glaciers, such as those remaining on Mt. Kenya.

Table 2. Overview of the digital elevation models (DEMs) used in this study.

DEM Acquisition Method Cell Size Acquisition Date PLE optical 1 m 23 February 2016 TDX X-band RADAR 12.32 m 2010–2014 ROS optical, airborne 2 m 1 September 2004

2.4. Glaciological Analyses Glacier extents were digitized manually from the 2004 inventory [13], and the 2016 Pléiades orthophoto. Additionally, the NDSI was derived from the Sentinel-2 scene, applying the QGIS semi-automatic classification plugin, to map the glaciers using different thresholds to separate ice from snow and to perform the mapping in shadow areas. Glacier area changes were determined from all glaciers, while glacier volume and mass changes are derived for Lewis Glacier only from differencing the co-registered DEMs [37] of 2010 [12] and 2016 (PLE). The DEM of 2010 was surveyed by differential Global Navigation Satellite System, by crossing the glacier several times, ground controlled by a subset (those surrounding Lewis Glacier) of the same GCPs used for georeferencing the map of 2004 (Section 2.2), and processed to a DEM at 5 m horizontal resolution. The volume change is derived as the difference of the surface height between the two Lewis Glacier DEMs multiplied by the glacier extent of 2010. To convert volume into mass, a constant ice density of 900 kg/m3 was assumed. This is justified as the equilibrium line altitude of Lewis has been permanently above the glacier’s highest point since the 1990s, causing the loss of the firn area, which leaves a glacier stratigraphy of only ice, with an occasional shallow snow cover of a few centimeters during the rainy seasons [12]. Geosciences 2018, 8, 174 6 of 14 Geosciences 2018, 8, x FOR PEER REVIEW 6 of 14

3. Results

3.1. Pleiades Orthophoto The PlPléiadeséiades orthophoto and the georeferenced ma mapp of 2004 are the primary sources for glacier delineation in this study. With With a a simple simple test, test, we we investigate investigate if if both both products products overlap overlap accurately. accurately. Assuming that tarn level fluctuationsfluctuations of deeper tarnstarns have little impact on the general shape of the shore line and on the surface area, we consider them as rather stable features for confirming confirming minimal horizontal shifts between both products. We We compared the shorelines of 6 tarns around Mt. Kenya summit between the PlPléiadeséiades orthophoto and the georeferencedgeoreferenced map of 2004. The difference of the tarn areas areas between between 2004 2004 and and 2016 2016 are are below below 4% 4% (Figure (Figure 4),4 ),illustrating illustrating good good alignment alignment between between the theproducts, products, although although the shorelines the shorelines from from 2004 2004 contai containn a generalization a generalization to the to map the map scale scale of 1:5000. of 1:5000. An Anexception exception is the is shallow the shallow and andboggy boggy Kami Kami Tarn, Tarn, where where temporal temporal surface surface area areachanges changes due to due tarn to level tarn levelfluctuations fluctuations are very are very high. high. Nevertheless, Nevertheless, this this comparison comparison allows allows the the assessments assessments of of glacier glacier area change from the PlPléiadeséiades orthophoto andand thethe 20042004 mapmap toto bebe viewedviewed withwith confidence.confidence.

Figure 4. Comparison of the shorelines of two tarnstarns onon Mt.Mt. Kenya indicating the generally good alignment betweenbetween thethe Pl Pléiadeséiades orthophoto orthophoto and and the the map map of 2004,of 2004, but bu alsot also one one misalignment misalignment in the in case the ofcase shallow of shallow Kami Kami Tarn. Tarn.

3.2. DEM Comparison 3.2. DEM Comparison We analyzed three different DEMs (Table 2) and compared their horizontal alignment as a We analyzed three different DEMs (Table2) and compared their horizontal alignment as a relative relative quality control of the Pléiades processing. For altitude intercomparison, the TDX was quality control of the Pléiades processing. For altitude intercomparison, the TDX was corrected for corrected for geoid undulation, by reference to mean sea level. The co-registration of the DEMs failed, geoid undulation, by reference to mean sea level. The co-registration of the DEMs failed, because the because the identification of homologous points in the DEMs was not possible, due to the large identiﬁcation of homologous points in the DEMs was not possible, due to the large differences in cell differences in cell size in each DEM. Thus, our analysis relies upon relative representations of specific size in each DEM. Thus, our analysis relies upon relative representations of speciﬁc surface topography surface topography features, like shaded reliefs and catchment divides instead. features, like shaded reliefs and catchment divides instead. Table 3 shows the elevation ranges of the individual DEMs within the ROS domain, and Table3 shows the elevation ranges of the individual DEMs within the ROS domain, demonstrates the different representations of the lowest and highest pixels. Batian is the highest peak and demonstrates the different representations of the lowest and highest pixels. Batian is the of Mt. Kenya (5199 m), and was used as input elevation point in ROS, and is thus correctly highest peak of Mt. Kenya (5199 m), and was used as input elevation point in ROS, and is thus represented. TDX falls short of achieving this maximum summit elevation by 219 m and PLE by 43 correctly represented. TDX falls short of achieving this maximum summit elevation by 219 m and PLE m. The minimum elevations are better constrained, where just PLE is 4 m above TDX and ROS. by 43 m. The minimum elevations are better constrained, where just PLE is 4 m above TDX and ROS. Errors in steep terrain become visible when plotting the shaded reliefs of the DEMs (Figure 5). Errors in steep terrain become visible when plotting the shaded reliefs of the DEMs (Figure5). Although the timing of PLE acquisition on the late morning coincides with minimum cloud cover Although the timing of PLE acquisition on the late morning coincides with minimum cloud cover and high solar elevation (Table 1), errors due to shadows and distortion still occur in very steep and high solar elevation (Table1), errors due to shadows and distortion still occur in very steep terrain >~50°. TDX errors are flagged by the provided height error map, which is the standard deviation of the random error per pixel due to interferometric decorrelation and incoherence. In the

Geosciences 2018, 8, 174 7 of 14 terrain >~ 50◦. TDX errors are ﬂagged by the provided height error map, which is the standard Geosciences 2018, 8, x FOR PEER REVIEW 7 of 14 deviation of the random error per pixel due to interferometric decorrelation and incoherence. Indomain, the domain, this error this is error highest is highest over complex over complex terrain terraincharacterized characterized by large by vertical large vertical gradients, gradients, and is andfurther is further the reason the reason for missing for missing the highest the highest peaks peaks (Table (Table 3), if3 ),they if they are arebuilt built of steep of steep walls walls instead instead of ofgentle gentle slopes slopes (Figure (Figure 1).1 ).TDX TDX contains contains corrupted corrupted data data from from artefacts, which introduceintroduce completelycompletely different terrain properties. While While in in PLE the ET ETMM algorithm smoothly interpolates errors due to shading of steep walls, TDX falsely introduces different features in the summit region like rough terrain or an expansive ﬂatflat area (Figure5 5).). Hence,Hence, neitherneither PLEPLE nornor TDXTDX resultresult inin anan acceptableacceptable terrainterrain representation for the glacier areas, and TDX is invalid for most of the summit region, and must be rejected for for studies studies demanding demanding a aDEM DEM from from Mt. Mt. Kenya. Kenya. For the For less the complex less complex topography topography of the Lewis of the LewisGlacier Glacier catchment, catchment, both products both products reveal revealgood resu goodlts, results, with the with constraint the constraint that the that four the year four TDX year TDXacquisition acquisition period period averages averages over overthe glacier the glacier changes changes that thatoccurred occurred within within this thisperiod. period. Hence, Hence, for forsubsequent subsequent estimates estimates of glacier of glacier volume volume changes, changes, we weuse use PLE PLE in the in theLewis Lewis Glacier Glacier catchment catchment only. only.

Table 3. Elevation ranges of differentdifferent DEMs in the ROS domain.

ElevationElevation (m) (m) DEMDEM Cell Cell Size Size (m) (m) MinMin Max Max Mean MeanStd. Deviation Std. Deviation TDXTDX 12.3212.32 41994199 4980 4980 4560 4560 153 153 PLEPLE 1.001.00 42044204 5156 5156 4588 4588 179 179 ROSROS 2.002.00 41994199 5199 5199 4592 4592 183 183

Further, we explored the mainmain ridgeridge lineslines ofof thethe Mt.Mt. Kenya summit domain applying ESRI’s basin algorithm (without prescribed pour points) as a qualitative measure of confiningconfining hydrological catchments (Figure6 6).). WeWe includedincluded SRTMSRTM andand ASTERASTER DEMsDEMs forfor comparisoncomparison toto widely used datasets. TDX watersheds are rather ill defined,defined, and deviate from the other DEMs due to the large height error in steep terrain (Figure5 5).). SRTM,SRTM, ASTERASTER GDEM,GDEM, ROS,ROS, andand PLEPLE yieldyield veryvery similarsimilar results,results, althoughalthough the differences between their cell sizessizes areare large,large, i.e.,i.e., ~30~30 mm vs.vs. <2<2 m. This result indicates that TDX deficienciesdeficiencies in terrain representation areare notnot duedue toto itsits cellcell sizesize (~12(~12 m), which is coarser than PLE but finerfiner than SRTM or ASTER GDEM.GDEM.

(a) (b)

Figure 5. Comparison ofofthe the shaded shaded reliefs reliefs from from (a )(a PLE) PLE and and (b )(b TDX) TDX showing showing errors errors in steep in steep terrain terrain and glacierand glacier outlines outlines from from 2016 2016 in green. in green. PLE errorsPLE errors are smoothed are smoothed and restricted and restricted to steep to walls,steep walls, TDX errors TDX areerrors spread are spread over the over area the witharea complexwith complex terrain, terrain which, which is interpolated is interpolated falsely falsely asa as rough a rough surface surface or ﬂator flat area. area. Lewis Lewis (lower (lower right right corner) corner) is theis the only only glacier glacier not not affected affected by by DEM DEM interpolation interpolation errors. errors.

Geosciences 2018, 8, 174 8 of 14 Geosciences 2018, 8, x FOR PEER REVIEW 8 of 14

Figure 6. RepresentationRepresentation of basin borders derived from the individual DEMs. Contours are taken from the 2004 map (Rost2004). Additionally, Additionally, basin borders derived from widely used coarser global DEMs (cell sizesize 3030 m) m) SRTM SRTM (version (version 2) 2) and and the the ASTER ASTE (GDEMR (GDEM version version 2) are 2) shownare shown (both (both products products are freely are availablefreely available from the from U.S. the Geological U.S. Geological Survey Survey via EarthExplorer via EarthExplorer (https://earthexplorer.usgs.gov/ (https://earthexplorer.usgs.gov/).).

3.3. Glacier Changes

3.3.1. Changes of Glacier Area The high horizontal resolution of the PlPléiadeséiades orthophoto and expert knowledge proved to be essential toto distinguishdistinguish the the very very small small glaciers glaciers from from patches patches of seasonalof seasonal snow snow and and to delineate to delineate ice from ice debrisfrom debris or rock or rock in shadow in shadow areas. areas. The The Sentinel-2 Sentinel-2 scene scene was was found found to to be be invalid invalid for for glacierglacier mapping on Mt. Kenya, because the large errors introduced by its coarser horizontal resolution (20 m) cannot be compensated by findingfinding the best NDSI thresholds for glacier ice detection. However, this result was expected,expected, as as this this method method reaches reaches a limit a limit at glaciers at glaciers <0.1 km <0.12 [9 ].km² Therefore, [9]. Therefore, we confine we theconfine mapping the tomapping the manual to the analysis manual ofanalysis the Plé ofiades the orthophoto.Pléiades orthophoto. Throughout Throughout the mapping the mapping history ofhistory the glaciers of the onglaciers Mt. Kenya, on Mt. theKenya, glacier the area glacier was area defined was as defined the clean as icethe surfaceclean ice and surface a potential and a debris potential cover debris was nevercover was mapped. never Since mapped. most Since glacier most margins glacier are margins well defined are well by defined rock walls, by rock debris-covered walls, debris-covered areas are veryareas limited.are very limited. Table4 4shows shows the the glacier glacier area area of 2004of 2004 and and 2016, 2016, and theand changes the changes that occurred that occurred during during this period. this Glacierperiod. retreatGlacier is retreat substantial, is substantial, ranging betweenranging thebetween total lossthe oftotal Gregory loss of (reported Gregory earlier(reported in [ 15earlier]) and in a minor[15]) and area a minor loss of area 13% loss of Forel, of 13% awell-shaded of Forel, a well-shaded hanging glacier. hanging The glacier. largest The glacier largest (Lewis) glacier lost (Lewis) 46% of itslost surface 46% of area, its surface which area, is close which to the is close average to the loss average (44%) of loss all (44%) glaciers of onall theglaciers mountain. on the mountain. Figure7 7 demonstrates demonstrates thethe downwastingdownwasting of of LewisLewis GlacierGlacier (see(see FigureFigure1 1for for comparison). comparison). The The area area loss is not only due toto terminusterminus retreat, but ratherrather due to shrinkage from all sides, including the separation of the lower from the upper half of the glacier, now divided by a rock outcrop. The glacier surface accumulates sediments washed down from the debris surroundings, which enhance melt by lowering the the surface surface albedo albedo of of the the glacier glacier [17]. [17 ].Lewis Lewis Glacier Glacier area area change change can canbe subdivided be subdivided into intotwo twoperiods, periods, and results and results in −28.6 in − ×28.6 10³ m²× 10 (−321%,m2 (−−5.221%, × 10³− 5.2m² ×per10 year)3 m2 fromper year) 2004 fromto 2010 2004 and to in 2010 −34.1 and × in10³− m²34.1 (−32%,× 10 3−5.7m2 ×( −10³32%, m² −per5.7 year)× 10 from3 m2 per2010 year) to 2016, from respectively. 2010 to 2016, Rates respectively. of area change Rates slightly of area changeincreased slightly during increased the later during period, the but later are period, still wi butthin are bounds still within of the bounds steady of shrinkage the steady previously shrinkage previouslyobserved over observed recent overdecades recent on decadesLewis Glacier on Lewis [12,23]. Glacier [12,23].

Geosciences 2018, 8, 174 9 of 14 Geosciences 2018, 8, x FOR PEER REVIEW 9 of 14

Figure 7. Lewis Glacier seen on the Pléiades orthophoto (pansharpened, nadir) from 23 February 2016. Figure 7. Lewis Glacier seen on the Pléiades orthophoto (pansharpened, nadir) from 23 February 2016. Changes of Lewis Glacier 2004–2016 are depicted by the respective outlines. Former Gregory Glacier, Changes of Lewis Glacier 2004–2016 are depicted by the respective outlines. Former Gregory Glacier, which was still northerly connected toto LewisLewis inin 2004,2004, vanishedvanished inin 20112011 [[15].15].

Table 4.4. Changes of glacier area on Mt. Kenya 2004–2016. Error ranges were derived from horizontal accuracies reported in [[13]13] and this study. Glacier Area 2004 (10³ m²) Area 2016 (10³ m²) Difference (10³ m²) Difference (%) 3 2 3 2 3 2 GlacierLewis Area 136.1 2004 ± (10 2.2 m ) Area 73.3 2016 ± 3.4 (10 m ) Difference−62.8 ± 1.5 (10 m ) Difference−46 (%) LewisTyndall 136.1 51.7 ±± 1.32.2 38.0 73.3 ±± 2.43.4 −13.7−62.8 ± 0.4± 1.5 −26− 46 ± ± − ± − TyndallKrapf 14.951.7 ± 0.71.3 12.4 38.0 ± 1.42.4 −2.513.7 ± 0.1 0.4 −17 26 Krapf 14.9 ± 0.7 12.4 ± 1.4 −2.5 ± 0.1 −17 ForelForel 12.7 ±± 0.7 11.0 11.0 ±± 1.31.3 −1.7−1.7 ± 0.1± 0.1 −13− 13 CesarCesar 18.6 ±± 0.8 9.6 9.6 ±± 1.21.2 −9.0−9.0 ± 0.4± 0.4 −48− 48 DarwinDarwin 12.7 ±± 0.7 4.2 4.2 ±± 0.80.8 −8.5−8.5 ± 0.4± 0.4 −67− 67 HeimHeim 5.3 ± 0.40.4 3.0 3.0 ±± 0.70.7 −2.3−2.3 ± 0.1± 0.1 −43− 43 Northey 3.9 ± 0.4 1.1 ± 0.4 −2.8 ± 0.2 −72 Northey 3.9 ± 0.4 1.1 ± 0.4 −2.8 ± 0.2 −72 Diamond 3.1 ± 0.3 1.0 ± 0.4 −2.1 ± 0.2 −68 GregoryDiamond 15.6 3.1 ±± 0.30.7 1.0 ± 00.4 −2.1 −± 15.60.2 −68− 100 Gregorytotal 274.6 15.6 ±± 0.73.1 153.6 0 ± 5.0 −−121.015.6 ± 2.5 −100− 44 total 274.6 ± 3.1 153.6 ± 5.0 −121.0 ± 2.5 −44 3.3.2. Changes of Glacier Volume 3.3.2. Changes of Glacier Volume Deriving glacier volume changes from DEM differencing was not carried out for the whole summitDeriving domain, glacier as PLE volume is erroneous changes in from many DEM of the di glacierfferencing locations, was not i.e., carried small cirques out for below the whole steep wallssummit (Figure domain,5). However,as PLE is erroneous because the in many Lewis of Glacier the glacier catchment locations, is less i.e., influenced small cirques by steepbelow walls, steep PLEwalls quality (Figure is5). sufficient However, for because computing the Lewis volume Glacie changesr catchment between is less 2016 influenced and 2010, by steep referring walls, to PLE the previousquality is DEMsufficient of Lewis for computing Glacier [12 volume]. We did changes not calculate between volume 2016 and changes 2010, referring using TDX, to the because previous its acquisitionDEM of Lewis period Glacier of four [12]. years We (2010–2014)did not calculate makes volume comparisons changes to using the DEMs TDX, of because 2010 or its 2016 acquisition doubtful. periodPLE of wasfour co-registered years (2010–2014) to the 2010makes DEM, comparisons and resampled to the to DEMs the same of 2010 cell size or 2016 (5 m). doubtful. DEM differencing showedPLE a volumewas co-registered decrease of to 0.89 the× 2010106 m DEM,3 for Lewis and re Glaciersampled in the to periodthe same 2010–2016, cell size equivalent (5 m). DEM to a massdifferencing loss of 0.80showed× 10 a9 kg,volume or an decrease annual massof 0.89 balance × 106 ratem³ for of −Lewis1.47 kg/mGlacier2 (Table in the5 ).period Strong 2010–2016, negative annualequivalent mass to balancea mass loss rates of in 0.80 the × 1990s 109 kg, resulted or an annual from the mass loss balance of the firnrate bodyof −1.47 and kg/m² accompanying (Table 5). Strong negative annual mass balance rates in the 1990s resulted from the loss of the firn body and accompanying decrease of albedo [12]. The moderate mass loss between 2004 and 2010 is potentially

Geosciences 2018, 8, 174 10 of 14 decrease of albedo [12]. The moderate mass loss between 2004 and 2010 is potentially due to very dry conditions, which favor the more energetically demanding ablation process of sublimation over melt [17,38], while the increase of negative mass balance rates since 2010 is caused by decreasing albedo, due to sediment accumulation and increased longwave heating from the surroundings and the recently developed rock outcrops (Figure7).

Table 5. Volume change and mass balance rate of Lewis Glacier 2004–2016. Values for the period 2004–2010 were taken from [12]. Error ranges were derived from error propagation using horizontal and vertical accuracies from [12,13] and this study.

Year Volume (106 m3) Area (106 m2) Period Mass Balance Rate (kg/m2/y) 2004 2.37 ± 0.49 0.136 ± 0.007 1993–2004 −2.22 ± 0.44 2010 1.90 ± 0.30 0.107 ± 0.001 2004–2010 −0.63 ± 0.77 2016 1.02 ± 0.34 0.073 ± 0.003 2010–2016 −1.47 ± 0.75

4. Discussion

4.1. Accuracy On the basis of the rational polynomial coefficients delivered with Pléiades primary data, the root mean square error of the panchromatic nadir orthophoto for the total domain is 8.5 m (circular error in the 90% confidence interval). The acquisition of well distributed GCPs over the total domain was not possible from a cost point of view, and from the acquisition effort. However, 10 recorded off-glacier GCPs in the summit domain (see Section 2.2) were used as control points for validation, yielding a mean absolute horizontal (vertical) error of 3.4 (5.3) m, which is sufficient for the mapping purpose in a scale of 1:25,000, and for delineating glacier extents (Table4). For DEM differencing and calculation of glacier volume changes, we followed the advice of earlier publications [8,39] for DEMs constructed without GCPs, and evaluated a vertical bias correction of PLE in the Lewis Glacier catchment. For this, we used the four off-glacier GCPs closely surrounding Lewis from an earlier field campaign [12], which yielded a vertical bias of 0.08 m and a mean absolute error of 1.55 m. Due to the low bias we did not correct PLE but used the error for uncertainty estimates of the volume change (Table5), by propagating errors from the individual surveys. As TDX deviates widely from the reference survey (ROS) in the summit domain, a quantitative vertical accuracy estimate was not deemed valuable. The horizontal alignment on flat slopes is acceptable, but for steep slopes in complex terrain, TDX is not an option for cartography and other applications requiring an accurate DEM.

4.2. The Glacier Inventory of 2016 Figure8 shows the location of the glaciers on Mt. Kenya. Glacier delineation demands high resolution images and expert knowledge, as some of the very small glaciers are located in narrow cirques surrounded by steep walls that cast the glaciers into persistent shadows (Krapf, Cesar, Forel, Heim) or are difficult to distinguish from seasonal snow cover (Northey). Additionally, to the Pléiades orthophoto, we used a Sentinel-2 scene to explore a semi-automatic classification of ice, but due to its coarser resolution and larger shadows, it was impossible to find thresholds for plausible glacier ice detection. Hence, Sentinel-2 is inappropriate for mapping the very small glaciers on Mt. Kenya, and thus, we confirm the interpretation that glaciers <0.1 km2 are beyond the limit of detection from platforms like Sentinel-2 or Landsat 8 [9]. This is of potentially wider significance, as receding glacier ice and glacier disintegration means that an increasing proportion of the remaining mountain glaciers of the world become very small, and monitoring these in the future will require the data to meet the same constraints as required presently to map the very small glaciers on Mt. Kenya [40]. Appropriate mapping techniques for very small glaciers may also have increasing significance for providing Geosciences 2018, 8, 174 11 of 14 accurate information for resolving socioeconomic conflicts over diminishing glacier resources in arid mountainousGeosciences 2018, 8 regions, x FOR PEER [41]. REVIEW 11 of 14

Figure 8. Overview of the glaciers of Mt. Kenya in their 2016 extent. Figure 8. Overview of the glaciers of Mt. Kenya in their 2016 extent.

A major increase of debris cover is not observed on Mt. Kenya’s glaciers, unlike in some other mountainA major regions increase where of glacier debris retreat cover is isassociat not observeded with increasing on Mt. propor Kenya’stions glaciers, of debris-covered unlike in someglacier other ice [42,43]. mountain Except regions minor where debris glacier deposits retreat from isrock associated fall on Tyndall with increasing and Lewis, proportions Mt. Kenya’s of debris-coveredglaciers remain glacierdebris-free. ice [42 Field,43]. observations Except minor in debris periglacial deposits environments from rock fall adjacent on Tyndall to glaciers and Lewis, or in Mt.paraglacial Kenya’s areas glaciers of former remain glacier debris-free. loss (Gregory) Field observations show some in geomorphological periglacial environments processes adjacent typical toof glacierssuch regimes, or in paraglacial but ongoing areas permafrost of former degradation glacier loss (Gregory) due to limited show ground some geomorphological snow cover and processesresulting typicalhigher radiation of such regimes, receipts [44] but is ongoing in line with permafrost a general degradation decay of the due cryosphere to limited in groundthe summit snow region. cover and resultingIn total, the higher glacier radiation area on receipts Mt. Kenya [44] decrea is in linesed with from a 274.6 general × 10³ decay m² to of 153.0 the cryosphere × 10³ m² (− in44%), the summitfrom 2004 region. to 2016. This is in line with data from nearby Kilimanjaro [45] and tropical South America × 3 2 × 3 2 [46], Inindicating total, the tropical glacier glacier area onretreat Mt. is Kenyastronger decreased than the global from 274.6mean. Absolute10 m annualto 153.0 retreat10 ratesm − (on44%), Mt. Kenya from 2004have tobeen 2016. rather This constant, is in line withbetween data 10 from × 10³ nearby m² and Kilimanjaro 13 × 10³ m² [45 per] and year tropical over the South last America50 years. [Corresponding46], indicating tropicalrelative glacierannual retreatretreat isra strongertes increased than thefrom global 0.8% mean.per year Absolute (1947–1963) annual to × 3 2 × 3 2 retreat3.2% per rates year on (1993–2004) Mt. Kenya [14], have and been currently rather constant, reach 3.8% between (2004–2016). 10 10 Globalm and estimates 13 10 form futureper yearglacier over mass the loss last 50suggest years. that Corresponding the preservation relative of glaciers annual retreatby ambitious rates increased measures from is possible 0.8% per in year the (1947–1963)long term [47]. to 3.2%However, per year if current (1993–2004) retreat [14 rates], and prevail, currently Mt. reachKenya’s 3.8% glaciers (2004–2016). will vanish Global before estimates 2030. forLinear future extrapolations glacier mass of loss the suggest rates of thatvolume the preservationchange of Lewis of glaciers will result by ambitious in the extinction measures of the is possible glacier inbefore the long2025, termleaving [47 a]. void However, in tropical if current glaciology retreat studies rates and prevail, their importance Mt. Kenya’s for glaciers understanding will vanish the beforeclimatology 2030. of Linear low latitudes extrapolations [48]. of the rates of volume change of Lewis will result in the extinction

Geosciences 2018, 8, 174 12 of 14 of the glacier before 2025, leaving a void in tropical glaciology studies and their importance for understanding the climatology of low latitudes [48].

5. Conclusions This study explores the practicability of DEMs from Pléiades (PLE) and TanDEM-X (TDX) for monitoring very small glaciers in complex terrain. PLE was generated without ground control for a different purpose (topographic mapping), and its use in a smaller domain is found to be sufficient for glacier delineating and calculation of glacier volume change. On Mt. Kenya, the benefits of new and highly appreciated techniques (high spatial and temporal resolution, free data access, well documented procedures) like glacier mapping from Sentinel-2 or surface height change detection from TDX, reach their limit of feature identification. Therefore, monitoring of very small glaciers demands products of highest spatial resolution (e.g., Pléiades images), counterbalanced by the acquisition costs of such products. The updated glacier inventory of Mt. Kenya demonstrates a persisting glacier retreat at constant pace for glacier area changes and increased mass loss rates, resulting in the projected complete deglaciation of the massif by the end of the next decade. The extinction of Mt. Kenya’s glaciers connotes the cessation of the longest time series of glacier monitoring in the tropics.

Author Contributions: R.P. developed the storyline, wrote the paper, and performed the glaciological analysis. M.L. and A.H. focused on the Pléiades analysis, the DEM generation and comparison, and the GIS methods. L.I.N. and G.K. contributed to the glaciological analysis and helped to refine the storyline and the English. Funding: This research was funded by the FFG (Austrian Research Promotion Agency) grant number 4948512. Additionally, the authors acknowledge the financial support by the University of Graz. Acknowledgments: We acknowledge the DLR (German Aerospace Center) for providing the TanDEM-X-DEM data (grant: DEM_GLAC1560) acquired by the TerraSAR-X/TanDEM-X satellite. Thanks are due to the reviewers for their valuable and constructive comments. Conflicts of Interest: The authors declare no conflict of interest.

References

1. Marzeion, B.; Champollion, N.; Haeberli, W.; Langley, K.; Leclercq, P.; Paul, F. Observation-Based Estimates of Global Glacier Mass Change and Its Contribution to Sea-Level Change. Surv. Geophys. 2017, 38, 105–130. [CrossRef][PubMed] 2. UNEP. WGMS Global Glacier Changes: Facts and Figures; UNEP: Nairobi, Kenya, 2008. 3. Haeberli, W.; Cihlar, J.; Barry, R.G. Glacier monitoring within the Global Climate Observing System. Ann. Glaciol. 2000, 31, 241–246. [CrossRef] 4. Zemp, M.; Frey, H.; Gärtner-Roer, I.; Nussbaumer, S.U.; Hoelzle, M.; Paul, F.; Haeberli, W.; Denzinger, F.; Ahlstrøm, A.P.; Anderson, B.; et al. Historically unprecedented global glacier decline in the early 21st century. J. Glaciol. 2015, 61, 745–762. [CrossRef] 5. Paul, F. Rapid disintegration of Alpine glaciers observed with satellite data. Geophys. Res. Lett. 2004, 31, L21402. [CrossRef] 6. Zemp, M.; Armstrong, R.; Gärtner-Roer, I.; Haeberli, W.; Hoelzle, M.; Kääb, A.; Kargel, J.S.; Khalsa, S.J.S.; Leonard, G.J.; Paul, F.; et al. Global glacier monitoring—A long-term task integrating in situ observations and remote sensing. In Global Land Ice Measurements from Space; Kargel, J.S., Leonard, G.J., Bishop, M.P., Kääb, A., Raup, B.H., Eds.; Springer Verlag: Berlin/Heidelberg, Germany, 2014; p. 876. 7. Paul, F.; Kääb, A.; Haeberli, W. Recent glacier changes in the Alps observed by satellite: Consequences for future monitoring strategies. Glob. Planet. Chang. 2007, 56, 111–122. [CrossRef] 8. Paul, F.; Bolch, T.; Kääb, A.; Nagler, T.; Nuth, C.; Scharrer, K.; Shepherd, A.; Strozzi, T.; Ticconi, F.; Bhambri, R.; et al. The glaciers climate change initiative: Methods for creating glacier area, elevation change and velocity products. Remote Sens. Environ. 2015, 162, 408–426. [CrossRef] 9. Paul, F.; Winsvold, S.; Kääb, A.; Nagler, T.; Schwaizer, G. Glacier Remote Sensing Using Sentinel-2. Part II: Mapping Glacier Extents and Surface Facies, and Comparison to Landsat 8. Remote Sens. 2016, 8, 575. [CrossRef] Geosciences 2018, 8, 174 13 of 14

10. Hastenrath, S. The Glaciers of Equatorial East Africa; D. Reidel Publishing Company: Dordrecht, The Netherlands; Boston, MA, USA; Lancaster, UK, 1984. 11. Hastenrath, S. Glaciological Studies on Mount Kenya; University of Wisconsin-Madison: Madison, WI, USA, 2005. 12. Prinz, R.; Fischer, A.; Nicholson, L.; Kaser, G. Seventy-six years of mean mass balance rates derived from recent and re-evaluated ice volume measurements on tropical Lewis Glacier, Mount Kenya. Geophys. Res. Lett. 2011, 38, L20502. [CrossRef] 13. Rostom, R.S.; Hastenrath, S. Variations of Mount Kenya’s glaciers 1993–2004. Erdkunde 2007, 61, 277–283. [CrossRef] 14. Hastenrath, S. The glaciers of Mount Kenya 1899–2004. Erdkunde 2005, 59, 120–125. [CrossRef] 15. Prinz, R.; Nicholson, L.; Kaser, G. Variations of the Lewis Glacier, Mount Kenya, 2004–2012. Erdkunde 2012, 66, 255–262. [CrossRef] 16. Prinz, R. Lewis Glacier, Mount Kenya, 2010 (1:2500). In Fluctuations of Glaciers 2005–2010, Volume X; Zemp, M., Frey, H., Gärtner-Roer, I., Nussbaumer, S.U., Hoelzle, M., Paul, F., Haeberli, W., Eds.; World Glacier Monitoring Service: Zürich, Switzerland, 2012; p. 66. 17. Prinz, R.; Nicholson, L.; Mölg, T.; Gurgiser, W.; Kaser, G. Climatic controls and climate proxy potential of Lewis Glacier, Mt. Kenya. Cryosphere 2016, 10, 133–148. [CrossRef] 18. Francou, B.; Vuille, M.; Wagnon, P.; Mendoza, J.; Sicart, J.E. Tropical climate change recorded by a glacier in the central Andes during the last decades of the twentieth century: Chacaltaya, Bolivia, 16◦ S. J. Geophys. Res. 2003, 108, 4154. [CrossRef] 19. Vuille, M.; Francou, B.; Wagnon, P.; Juen, I.; Kaser, G.; Mark, B.; Bradley, R.S. Climate change and tropical Andean glaciers: Past, present and future. Earth-Sci. Rev. 2008, 89, 79–96. [CrossRef] 20. Mölg, T.; Cullen, N.J.; Hardy, D.R.; Winkler, M.; Kaser, G. Quantifying climate change in the tropical midtroposphere over East Africa from glacier shrinkage on Kilimanjaro. J. Clim. 2009, 22, 4162–4181. [CrossRef] 21. Karl, T.R.; Hassol, S.J.; Miller, C.D.; Murray, W.L. Temperature Trends in the Lower Atmosphere: Steps for Understanding and Reconciling Differences; General Books LLC: Washington, DC, USA, 2006. 22. Pepin, N.C.; Lundquist, J.D. Temperature trends at high elevations: Patterns across the globe. Geophys. Res. Lett. 2008, 35, 1–6. [CrossRef] 23. Mölg, T.; Cullen, N.J.; Hardy, D.R.; Kaser, G.; Nicholson, L.; Prinz, R.; Winkler, M. East African glacier loss and climate change: Corrections to the UNEP article “Africa without ice and snow”. Environ. Dev. 2013, 6, 1–6. [CrossRef] 24. WGMS. Global Glacier Change Bulletin No. 2 (2014–2015); Zemp, M., Nussbaumer, S.U., Gärtner-Roer, I., Huber, J., Machguth, H., Paul, F., Hoelzle, M., Eds.; ICSU(WDS)/IUGG(IACS)/UNEP/UNESCO/WMO; World Glacier Monitoring Service: Zurich, Switzerland, 2017. 25. Nicholson, L. Photographs of Lewis Glacier Change 2003–2014. Available online: http://lindseynicholson.o rg/2014/10/photographs-of-lewis-glacier-change-2003-2014/ (accessed on 20 March 2018). 26. Hirschmüller, H. Accurate and efﬁcient stereo processing by semi-global matching and mutual information. In Proceedings of the 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR’05), San Diego, CA, USA, 20–25 June 2005; Volume 2, pp. 807–814. 27. Harris, C.; Stephens, M. A combined corner and edge detector. Proc. Alvey Vis. Conf. 1988, 15, 147–151. 28. Charnley, F.E. Some observations on the glaciers of Mount Kenya. J. Glaciol. 1959, 3, 483–492. [CrossRef] 29. Ladner, M.; Heller, A.; Grillmayer, E. GNSS und Pléiades-Bilder in der Hochgebirgskartographie. AGIT J. Angew. Geoinform. 2017, 3, 13–23. 30. Heller, A. Die Alpenvereinskarte als Grundlage für Geographische Informationssysteme. Wiss. Alpenvereinshefte 2001, 34, 61–81. 31. Heller, A. Georeferenzierung von Alpenvereinskarten mit radialen Basisfunktionen. In Angewandte Geographische Informationsverarbeitung XIV; Strobl, J., Blaschke, T., Griesebener, G., Eds.; AGIT: Salzburg, Austria, 2002; pp. 162–171. 32. Hutchinson, M.F. A new procedure for gridding elevation and stream line data with automatic removal of spurious pits. J. Hydrol. 1989, 106, 211–232. [CrossRef] 33. Wessel, B. TanDEM-X Ground Segment DEM Products Speciﬁcation Document; Public Document TD-GS-PS-0021, Issue 3.1; EOC, DLR: Oberpfaffenhofen, Germany, 2016. Geosciences 2018, 8, 174 14 of 14

34. Rankl, M.; Braun, M. Glacier elevation and mass changes over the central Karakoram region estimated from TanDEM-X and SRTM/X-SAR digital elevation models. Ann. Glaciol. 2016, 57, 273–281. [CrossRef] 35. Rott, H.; Floricioiu, D.; Wuite, J.; Scheiblauer, S.; Nagler, T.; Kern, M. Mass changesof outletglaciers along theNordensjköld Coast, northern Antarctic Peninsula, based on TanDEM-X satellitemeasurements. Geophys. Res. Lett. 2014, 41, 8123–8129. [CrossRef] 36. Pipaud, I.; Loibl, D.; Lehmkuhl, F. Evaluation of TanDEM-X elevation data for geomorphological mapping and interpretation in high mountain environments—A case study from SE Tibet, China. Geomorphology 2015, 246, 232–254. [CrossRef] 37. Klug, C.; Bollmann, E.; Galos, S.P.; Nicholson, L.; Prinz, R.; Rieg, L.; Sailer, R.; Stötter, J.; Kaser, G. Geodetic reanalysis of annual glaciological mass balances (2001–2011) of Hintereisferner, Austria. Cryosphere 2018, 12, 833–849. [CrossRef] 38. Hastenrath, S.; Polzin, D.; Mutai, C. Diagnosing the droughts and floods in Equatorial East Africa during boreal autumn 2005–08. J. Clim. 2010, 23, 813–817. [CrossRef] 39. Berthier, E.; Vincent, C.; Magnússon, E.; Gunnlaugsson, Á.Þ.; Pitte, P.; Le Meur, E.; Masiokas, M.; Ruiz, L.; Pálsson, F.; Belart, J.M.C.; et al. Glacier topography and elevation changes derived from Pléiades sub-meter stereo images. Cryosphere 2014, 8, 2275–2291. [CrossRef] 40. Fischer, M.; Huss, M.; Barboux, C.; Hoelzle, M. The New Swiss Glacier Inventory SGI2010: Relevance of Using High-Resolution Source Data in Areas Dominated by Very Small Glaciers. Arct. Antarct. Alp. Res. 2014, 46, 933–945. [CrossRef] 41. Nicholson, L.; Marín, J.; Lopez, D.; Rabatel, A.; Bown, F.; Rivera, A. Glacier inventory of the upper Huasco valley, Norte Chico, Chile: Glacier characteristics, glacier change and comparison with central Chile. Ann. Glaciol. 2009, 50, 111–118. [CrossRef] 42. Stokes, C.R.; Popovnin, V.; Aleynikov, A.; Gurney, S.D.; Shahgedanova, M. Recent glacier retreat in the Caucasus Mountains, Russia, and associated increase in supraglacial debris cover and supra-/proglacial lake development. Ann. Glaciol. 2007, 46, 195–203. [CrossRef] 43. Shukla, A.; Gupta, R.P.; Arora, M.K. Estimation of debris cover and its temporal variation using optical satellite sensor data: A case study in Chenab basin, Himalaya. J. Glaciol. 2009, 55, 444–452. [CrossRef] 44. Grab, S.W.; Gatebe, C.K.; Kinyua, A.M. Ground thermal profiles from Mount Kenya, East Africa. Geogr. Ann. Ser. A Phys. Geogr. 2004, 86, 131–141. [CrossRef] 45. Cullen, N.J.; Sirguey, P.; Mölg, T.; Kaser, G.; Winkler, M.; Fitzsimons, S.J. A century of ice retreat on Kilimanjaro: The mapping reloaded. Cryosphere 2013, 7, 419–431. [CrossRef] 46. Rabatel, A.; Francou, B.; Soruco, A.; Gomez, J.; Cáceres, B.; Ceballos, J.L.; Basantes, R.; Vuille, M.; Sicart, J.E.; Huggel, C.; et al. Current state of glaciers in the tropical Andes: A multi-century perspective on glacier evolution and climate change. Cryosphere 2013, 7, 81–102. [CrossRef] 47. Marzeion, B.; Kaser, G.; Maussion, F.; Champollion, N. Mass loss commitment limits influence of climate change mitigation on glaciers. Nat. Clim. Chang. 2018. online first. [CrossRef] 48. Kaser, G. Glacier-climate interaction at low latitudes. J. Glaciol. 2001, 47, 195–204. [CrossRef]

© 2018 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).