APPLICATION OF SAR TIME-SERIES AND DEEP LEARNING FOR ESTIMATING LANDSLIDE OCCURRENCE TIME

We found that the altitude (< 240 m) and distance to rivers (< 300 m) emerged as important factors for the cause of landslides. LR-MLP-Boosting achieves the highest prediction accuracy. The coupling models outperform the corresponding single models and Boosting algorithm performs better than the Bagging algorithm. High-quality non-landslide samples enhance the accuracy of LSM. They can be effectively obtained by using the LR model to constrain its selection range.

Nepal’s second-largest city has been rapidly growing since the 1970s, although its valley has been affected by rare, catastrophic floods in recent and historic times. We analyse potential impacts of such floods on urban areas and infrastructure by modelling 10 physically plausible flood scenarios along Pokhara’s main river. We find that hydraulic effects would largely affect a number of squatter settlements, which have expanded rapidly towards the river by a factor of up to 20 since 2008.

Tsunamis caused by volcanoes are a challenge for warning systems as they are difficult to predict and detect. In Southeast Asia there are many active volcanoes close to the coast, so it is important to identify the most likely volcanoes to cause tsunamis in the future. For this purpose, we developed a point-based score system, allowing us to rank volcanoes by the hazard they pose. The results may be used to improve local monitoring and preparedness in the affected areas.

Prairie wetlands are important for biodiversity and water availability. Knowledge about their variability and spatial distribution is of great use in conservation and water resources management. In this study, we propose a novel approach for the classification of small water bodies from satellite radar images and apply it to our study area over 6 years. The retrieved dynamics show the different responses of small and large wetlands to dry and wet periods.