Mapping landslides using automated methods is a challenging task, which is still largely done using human efforts. Today, the availability of high-resolution EO data products is increasing exponentially, and one of the targets is to exploit this data source for the rapid generation of landslide inventory. Conventional methods like pixel-based and object-based machine learning strategies have been studied extensively in the last decade. In addition, recent advances in CNN (convolutional neural network), a type of deep-learning method, has been widely successful in extracting information from images and have outperformed other conventional learning methods. In the last few years, there have been only a few attempts to adapt CNN for landslide mapping. In this study, we introduce a modified U-Net model for semantic segmentation of landslides at a regional scale from EO data using ResNet34 blocks for feature extraction. We also compare this with conventional pixel-based and object-based methods. The experiment was done in Douglas County, a study area selected in the south of Portland in Oregon, USA, and landslide inventory extracted from SLIDO (Statewide Landslide Information Database of Oregon) was considered as the ground truth. Landslide mapping is an imbalanced learning problem with very limited availability of training data. Our network was trained on a combination of focal Tversky loss and cross-entropy loss functions using augmented image tiles sampled from a selected training area. The deep-learning method was observed to have a better performance than the conventional methods with an MCC (Matthews correlation coefficient) score of and a POD (probability of detection) rate of .
Rapid mapping of event landslides is crucial to identify the areas affected by damages as well as for effective disaster response. Traditionally, such maps are generated with visual interpretation of remote sensing imagery (manned/unmanned airborne systems or spaceborne sensors) and/or using pixel-based and object-based methods exploiting data-intensive machine learning algorithms. Recent works have explored the use of convolutional neural networks (CNN), a deep learning algorithm, for mapping landslides from remote sensing data. These methods follow a standard supervised learning workflow that involves training a model using a landslide inventory covering a relatively small area. The trained model is then used to predict landslides in the surrounding regions. Here, we propose a new strategy, i.e., a progressive CNN training relying on combined inventories to build a generalized model that can be applied directly to a new, unexplored area. We first prove the effectiveness of CNNs by training and validating on event landslides inventories in four regions after earthquakes and/or extreme meteorological events. Next, we use the trained CNNs to map landslides triggered by new events spread across different geographic regions. We found that CNNs trained on a combination of inventories have a better generalization performance, with a bias towards high precision and low recall scores. In our tests, the combined training model achieved the highest (Matthews correlation coefficient) MCC score of 0.69 when mapping landslides in new unseen regions. The mapping was done on images from different optical sensors, resampled to a spatial resolution of 6 m, 10 m, and 30 m. Despite a slightly reduced performance, the main advantage of combined training is to overcome the requirement of a local inventory for training a new deep learning model. This implementation can facilitate automated pipelines providing fast response for the generation of landslide maps in the post-disaster phase. In this study, the study areas were selected from seismically active zones with a high hydrological hazard distribution and vegetation coverage. Hence, future works should also include regions from less vegetated geographic locations.
Landslide hazard has always been a significant source of economic losses and fatalities in the mountainous regions. Knowledge of the spatial extent of the past and present landslide activity, compiled in the form of a landslide inventory map, is essential for effective risk management. High-resolution data acquired by Earth observation (EO) satellites are often used to map landslides by identifying morphological expressions that can be associated with past and/or recent deformation. This is a slow and difficult process as it requires extensive manual efforts. As a result, such maps are not readily available for all the landslide hazard affected regions. Fully automated methods are required to exploit the exponentially increasing amount of EO data available for landslide hazard assessments. In this context, conventional methods like pixel-based and object-based machine learning strategies have been studied extensively in the last decade. Recent advances in convolutional neural network (CNN), a type of deep-learning method, has outperformed other conventional learning methods in similar image interpretation tasks. In this work, we present a deep-learning based method for semantic segmentation of landslides from EO images. We present the results from a study area in the south of Portland in Oregon, USA. The landslide inventory for training and ground truth was extracted from the Statewide Landslide Information Database of Oregon (SLIDO). We were able to achieve a probability of detection (POD) greater than 0.70. This method can also be extended to be used for rapid mapping of landslides after a major triggering event (like earthquake or extreme metrological event) has occurred.This work is done in the framework of European Commission's Horizon 2020 project "BETTER”. More information is available on the website https://www.ec-better.eu/.
Landslides are a common secondary hazard triggered by major earthquakes or extreme rainfall events in steep terrains. Rapid mapping of event landslides is crucial to identify the areas affected by damages as well as for effective disaster response. Traditionally, such maps are generated with visual interpretation of remote sensing imagery (manned/unmanned airborne systems or spaceborne sensors) and/or using pixel-based and object-based methods exploiting data-intensive machine learning algorithms. Recent works have explored the use of convolutional neural networks (CNN), a deep learning algorithm, for mapping landslides from remote sensing data. These methods follow a standard supervised workflow involving a model trained using a landslide inventory over a relatively small region and then applied for prediction in the surrounding. Here, we propose a new strategy, i.e., a progressive CNN training relying on multiple event landslide inventories to build a generalized model that can be applied directly to a new, unexplored area. These inventories are spread across different geographic regions. We first proved the effectiveness of CNNs for mapping event landslides in four regions after earthquakes and/or extreme meteorological events. The best MCC score for each region ranged from 0.574 to 0.806. However, when mapping new unseen areas, we found that CNNs trained on a combination of multiple inventories have a better generalization performance with a bias towards higher precision. This combined training model could achieve the highest MCC score of 0.69 when mapping landslides in new unseen regions. Despite the expense of a slightly reduced accuracy, the main advantage of combined training is to overcome the requirement of a local training inventory for mapping event landslides triggered by future events. This can facilitate an automated pipeline providing fast response for the generation of landslide maps in the post-disaster phase.
Rockfalls affect steep slopes in several geographic regions. Different systems from remote to in-situ instruments are used for their detection and study. In this scenario, seismic signals produced by the detachment, bouncing, and rolling of rockfalls are being increasingly used for the detection and classification of such events. This is typically done by using different manual, semi-automatic and/or automatic signal processing strategies. In this work, we applied a new Deep Learning (DL) algorithm in order to test the performance on the automatic classification of seismic signals. We applied the method to seismic data acquired by a low-cost Raspberry Shake 1D seismometer (sampling rate 50Hz) in order to discriminate rockfall from not-rockfall events occurred at the Moosfluh active slope region in Wallis (CH). Here we present the methodology and show the results obtained on a continuous record of more than 2-years of seismic data. The performance accuracy of the DL approach reached values larger than 90%. Our results show that the application of DL strategies in this context can be very useful and save time on seismic data classification.