Landslide Detection for Rapid Mapping Using Sentinel-2

Alexander Ariza
Jan 25, 2021 · 5 min read

A guest blog post by Alexander Ariza

Foreword by Sentinel Hub

This post is part of a series of guest blog posts written by script authors, talking about their entries to the Sentinel Hub Custom Script Contest. Alexander Ariza and Norma Davila are among the winners in the third round of the Contest. Their winning script with detailed description is available on our GitHub repository.

In this blog post, authors are explaining their winning custom script for detecting landslides using space-based data provided by the Sentinel-2 satellite in Sentinel Playground and EO Browser. Its goal is to facilitate the landslide inventory and rapid mapping in disaster management.

“We are ending a really dramatic year in terms of natural disasters added to the COVID pandemic. An annual record has been set for the number of major storms forming in the Atlantic since 2005 when there were 28 named storms. There have been so many big storms in 2020 that meteorologists exhausted their list of English names and had to turn to the Greek alphabet, which quickly burned from Alpha to Eta and more recently Theta, bringing with it large losses due to floods and landslides.”

Landslides by ETA hurricane in San Cristóbal de Verapaz — Guatemala. Sentinel 2 acquired on 10/11/2020
Landslides triggering by ETA hurricane in San Cristóbal de Verapaz — Guatemala. Sentinel 2 acquired on 10/11/2020. (See at EO Browser)

Rapid Mapping Landslides with Sentinel Hub

Earth Observation technologies are being increasingly utilized in disaster response situations. For example, to direct logistical and emergency support to areas affected by landslides and map the damage they have caused in order to plan for recovery. Mapping of risk areas and monitoring of landslides can be conducted using satellite and airborne imaging platforms, with new methodologies constantly being developed.

Tropical storm “ETA hurricane” in San Cristóbal de Verapaz — Guatemala — Central America, 08/11/2020 (Euronews)

General Description of the Algorithm

The script for landslide detection for rapid mapping is based on the response of the Barren Soil Index BSI for the detection of the traces of the soil movements. The results allow us to extract the shapes of the landslides in the terrain and to calculate their direction of movement. In the same way, The script uses the NDWI, NDVI, and B11 for differentiating between water with sediment, built-up areas, barren areas, and vegetated areas.

The spectral signature of bare soils associated
Barren Soil Index BSI

Most of the pixels related to landslides present high BSI values (> 0.7), with high reflectivity in the B11 but less than 0.8 and with little presence of vegetation (NDVI less than 0.15).

The script returns an RGB combination, where the BSI is a red channel; the NDVI and B11 values to green and the NDWI to blue. Results show landslide zones in brown to orange tones; urbanized or built-up areas showed in orange to yellow; vegetation appears dark green (NDVI> 0.25). Finally, the water is blue with NDWI pixel values higher than 0.15, with some brown tones depending on the sediment load present.

Landslide detection for rapid mapping by the tropical storm “ETA hurricane” (In the left, Queja — Alta Verapaz — Guatemala. CONRED. In the right, KML layer exported from EO Browser over 3D model in Google Earth, in Queja — Alta Verapaz — Guatemala, 10/11/2020.)

High reflectivity of cloud pixels can be confused with bare soil, which is why the script includes a Cloud mask computed using the CLP (cloud probabilities) of s2cloudless, which is available at Sentinel Hub. The script that returns three bands (R/G/B and CLM) will result in a real image, as well as the fraction of cloudy pixels per each observation.

A screenshot from EO Browser with a Cloud mask computed using the CLP (cloud probabilities) of s2cloudless for masking out the clouds

The outcomes are shown according to R:G:B composite image where the yellow color represents bare soils, while orange and brown colors are linked to surficial soils or debris slides, which have suffered a remotion or emplacement motion. Particularly, for an active landslide, both colors will match with vertical erosion towards the proximal area of the landslide (headwerd) or over talud deposits.

Landslide detection in San Cristóbal de Verapaz — Guatemala — Central America (10/11/2020). The yellow color represents bare soils, while orange and brown colors are linked to surficial soils or debris slides

The preliminary results showed that the use of BSI, NDVI and NDWI in combination with the B11 reflectance are sufficient to identify recent traces of soil movements useful for rapid mapping response during a disaster event.

Landslides detection on Sentinel-2 images before and after (01/09/2020–10/11/2020) to the landslides in the town of Queja in Alta Verapaz

However, the use of the BSI index itself cannot characterize the landslide configurations on terrain, due to after landslide event, the spectral response changes shortly especially over zones with fast reactivation in the vegetation cover. Therefore, in order to improve the interpretation of the results, we recommend a temporal analysis (pre and post events) to get better outcomes of BSI values linked to landslide targets. These advances will be available in the next script version.

Landslide Hazard Response

Remote detection of landslides and erosion using optical imagery can be challenging due to the lack of unique spectral signatures associated with the mass movement. The presence of exposed soil or a change in vegetation can be identifying factors; however, these features are heterogeneous at all spatial scales. In old landslides, the ‘sharp’ features rapidly erode and with weathering and re-vegetation the distinctions become blurred. Often the only obvious visible feature remaining will be a slightly round scarp/head region and a hollow, which is difficult to see using satellite imagery.

It is also difficult to use single date spectral information to determine the scale of movement as this does not allow for observation of changes in the landscape. Contextual knowledge of the geology and soils of an area along with multi-temporal data can aid greatly in landslide identification using optical imagery.

The Sentinel Hub team would once again like to thank Alexander and Norma for their participation in our Contest. You can find the authors, Norma Davila on Twitter and LinkedIn, and Alexander Ariza on LinkedIn.

We would also like to invite you to take a look at the other scripts submitted to the Sentinel Hub Custom Script Contests, available here.

For more about remote sensing and how custom scripts may help you tackle different remote sensing problems, visit our Education page.

Sentinel Hub Blog

Stories from the next generation satellite imagery platform