GRSG Conference 2022: Orbit to Outcrop

Title: SOILRISK

Author: Georgia Karadimou

Abstract:

In the last 100 years, 43 people have died in San Miguel Island (Azores, Portugal) due to rainfall-triggered landslide occurrences. The most tragic one was in the municipality of Ribera Quente on the 31st of October 1997, killing 29 people. It is scientifically proved that global warming is increasing the apparition of extreme meteorological events, such as an intensification of rainfall episodes which has a direct impact in the severity of landslides.

Within this context, the systematic monitoring of landslides is vital in order to prevent and mitigate the consequences they might have. For this reason, we created and present SOILRISK, a high-resolution Predictive Model Application for Landslides, that gathers atmospheric and environmental data and converts it into valuable intelligence in order to forecast landslide occurrence and provide a warning system for prevention.

SOILRISK is based on an innovative Machine Learning (ML) approach and a proprietary early-warning algorithm using open-source data. SOILRISK’s feasibility study was co-funded by ESA through a kick-start Activity in Portugal. A Landslide Susceptibility Map (LSM) was produced, including more than 20 conditioning factors. The model was validated through ROC curve and Kappa index.

The areas under curves (AUC) were 95% and the Kappa index showed an accuracy of 96%. Moreover, the sensitivity of the model to identify landslides was very high, around 98%, for a total of 1700 landslide observations. SOILRISK uses the LSM as a base layer for the algorithm. Then, the algorithm is daily fed with meteorological forecast data coming from meteorological stations and forecasting models. The dynamic datasets introduced within the model are: daily rainfall, accumulated rainfall over the last 3 months, wind speed and wind direction.

This way, SOILRISK produces a 3-day forecast, highlighting the areas with the highest probability for a landslide to occur in the coming days. Additionally, the algorithm is also indicating the areas with the highest displacements for a 6-day period, thanks to the InSAR analysis using Sentinel-1 data.

SOILRISK is built with the most advanced predictive landslide algorithm, comparing to the existing ones from the available literature. It provides a very detailed picture of the conditions for a potential landslide event to happen, making it an asset for on-time predection in order to save lives. The algorithm is currently tested successfully for the Azores region, but thanks to the use of open-source data as its basic element, it is available to be applied globally. More information can be found here https://www.soilrisk.com/#/