Frank Paul

Department of Geography, University of Zurich

Using Sentinel-2 MSI for

global-scale glacier mapping

The target: Glaciers

Alean

Paul

Transforming a satellite image into ...

... a glacier inventory with attributes

Area changes in : 1985-2000-2014

1 km

What we need and what we have

• Freely available multispectral satellite data covering glaciers

• acquired at the end of the ablation season (without clouds and seasonal snow)

• a band in the SWIR (1.5 mm) for automated classification of clean ice

• sufficient spatial resolution (10-30 m) to identify the target and repetition

• experts performing the processing and applying corrections scene-by-scene

• Satellites and DEMs being applied

• Landsat (MSS/TM: archive & change assessment, ETM+/OLI: ongoing)

• ASTER (end of life), SPOT (expensive), S2a/b (forthcoming)

• SRTM/GDEM-2 (widely used, artefacts), GIMP (Greenland), WorldDEM (future)

• A first version of a globally complete glacier inventory (RGI)

• required: quality improvement, temporal consistency, change assessment

• mapping guidelines for the community, RGI/GLIMS integration

• long-term strategy for an operational service (fast changes, personnel costs)

Spectral properties and glacier mapping

JPL

Paul

Ratio

TM5

TM3

glacier

debris

The Randolph Glacier Inventory (RGI)

IPCC (2014)

Pfeffer et al. (2014)

The RGI is a merged product

Interpretation errors and seasonal snow

Comparing S-2 MSI with Landsat OLI

ESA 2012

L8-OLI S2-MSI win

Coverage 180 km 100 km O

Revisit 16 10 / 5 M

SWIR band yes yes O/M

Resolution 30 m 10/20 m M

Operational yes no O

Orthorectific. GLS2000 ? ?

12-bit bands yes yes O/M

• Key improvements

• spatial resolution (debris,

mapping at 10 m, velocity)

• revisit period (seasonal snow, in

some regions >20 years waiting)

• 12-bit: better discrimination of

snow facies but ratio less good

Griesglacier 1992: 30 m Landsat TM (B3)

1 km

VAW/ETHZ

Griesglacier 1992: 10 m SPOT Pan

1 km

VAW/ETHZ

Fusing 30 m TM with 10 m SPOT (IHS)

1 km

Aerial photography 2012

1 km

Flow velocity and DEM accuracy

Mean annaual flow velocity

Heid & Kääb (2012)

Feature tracking (3 yr difference)

100 km tiles: 9 S-2 instead of 1 TM?

290 km Sentinel 2

100 km tiles

180 km Landsat

The LASER band: snow grain size mapping

Maximum change of spectral reflectance at 1065 nm,

which is where LASERs of ICESat and ALS are working

Snow grain size dependence

A related band would be nice to have in the future

ALSX

Höfle et al. (2007)

Hintereisferner: The unexplored near infrared

TM4

Summary

• Current global scale glacier mapping relies on Landsat 8 OLI

and some ASTER / ETM+ (still 30 m but huge TM archive)

• Sentinel 2 will help closing the acquisition gap (temporal) and

getting more cloud-free coverage without seasonal snow

• The higher spatial resolution of S-2 MSI compared to L8-OLI

will improve the quality of the outlines (clean ice) as well as

identification of debris covered ice

• Also creation of glacier velocity fields from feature tacking

will be considerably improved by the 10 m resolution

• The DEM used for orthorectification needs to be better than

SRTM &GDEM2 (and publicly available => drainage divides)

• The 100 km tiling seems suboptimal for glacier mapping

The Randolph Glacier Inventory (RGI)

• 170,000 glaciers worldwide covering 730,000 km2

• Volume of about 170,000 km3 or 41 cm SLE

• Numbers will change through RGI updates

IPCC (2013)