Landslide mapping from multi-sensor data through improved change detection-based Markov random field

Lü, Ping; Qin, Yuanyuan; Li, Zhongbin; Mondini, Alessandro Cesare; Casagli, Nicola

doi:10.1016/j.rse.2019.111235

Cited by 136 publications

(90 citation statements)

References 75 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Finding the extent of an existing landslide is difficult using this approach, as a landslide is better represented by a heterogeneous polygon (i.e., a collection of pixels). Detection of landslides activity using image correlation [55,56] and change detection [39,57] are also included in pixel-based methods, but they require a time-series of multi-temporal images.…”

Section: Pixel-based Methodsmentioning

confidence: 99%

Mapping Landslides on EO Data: Performance of Deep Learning Models vs. Traditional Machine Learning Models

2020

View full text Add to dashboard Cite

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 0.495 and a POD (probability of detection) rate of 0.72 .

show abstract

Section: Pixel-based Methodsmentioning

confidence: 99%

Mapping Landslides on EO Data: Performance of Deep Learning Models vs. Traditional Machine Learning Models

2020

View full text Add to dashboard Cite

show abstract

“…Landslides frequently damage buildings, communication systems, agriculture, natural vegetation, and the environment, and they are a major cause of fatalities (Froude and Petley, 2018;Petley et al, 2005). A landslide inventory forms the basis for studies of landslide hazard, risk, and prevention studies (Fan et al, 2019;Guzzetti et al, 2012;Marcelino et al, 2009;Moosavi et al, 2014). Critical elements of analysis include their spatial distribution pattern (Duman et al, 2005;Galli et al, 2008;Xu, 2015), their occurrences with respect to landform evolution (Guzzetti et al, 2012;Rosi et al, 2018), and a range of other environment factors (Duman et al, 2005), susceptibility mapping (van Den Eeckhaut et al, 2009), triggering factors (Li et al, 2016), community risk assessment and mitigation (Marcelino et al, 2009), and land use planning and risk management (Colombo et al, 2005).…”

Section: Introductionmentioning

confidence: 99%

Constructing a complete landslide inventory dataset for the 2018 monsoon disaster in Kerala, India, for land use change analysis

Hao

Westen²,

Martha

et al. 2020

Earth Syst. Sci. Data

View full text Add to dashboard Cite

Abstract. Event-based landslide inventories are important for analyzing the relationship between the intensity of the trigger (e.g., rainfall, earthquake) and the density of the landslides in a particular area as a basis for the estimation of the landslide probability and the conversion of susceptibility maps into hazard maps required for risk assessment. They are also crucial for the establishment of local rainfall thresholds that are the basis of early warning systems and for evaluating which land use and land cover changes are related to landslide occurrence. The completeness and accuracy of event-based landslide inventories are crucial aspects to derive reliable results or the above types of analyses. In this study, we generated a relatively complete landslide inventory for the 2018 monsoon landslide event in the state of Kerala, India, based on two inventories that were generated using different methods: one based on an object-based image analysis (OBIA) and the other on field surveys of damaging landslides. We used a collaborative mapping approach based on the visual interpretation of pre- and post-event high-resolution satellite images (HRSIs) available from Google Earth, adjusted the two inventories, and digitized landslides that were missed in the two inventories. The reconstructed landslide inventory database contains 4728 landslides consisting of 2477 landslides mapped by the OBIA method, 973 landslides mapped by field survey, 422 landslides mapped both by OBIA and field methods, and an additional 856 landslides mapped using the visual image (Google Earth) interpretation. The dataset is available at https://doi.org/10.17026/dans-x6c-y7x2 (van Westen, 2020). Also, the location of the landslides was adjusted, based on the image interpretation, and the initiation points were used to evaluate the land use and land cover changes as a causal factor for the 2018 monsoon landslides. A total of 45 % of the landslides that damaged buildings occurred due to cut-slope failures, while 34 % of those having an impact on roads were due to road cut-slope failures. The resulting landslide inventory is made available for further studies.

show abstract

“…This process is very time-consuming and can be subjective [31,32]. Recent advancements in remote-sensing technologies have significantly increased our ability to rapidly map landslides of various sizes, with less in situ surveys or human interaction [33][34][35]. Remote sensing of landslides can be categorized into two groups: pixel-based and object-based image analysis (OBIA).…”

Section: Introductionmentioning

confidence: 99%

Use of Very High-Resolution Optical Data for Landslide Mapping and Susceptibility Analysis along the Karnali Highway, Nepal

2019

View full text Add to dashboard Cite

The Karnali highway is a vital transport link and the only primary roadway that connects the remote Karnali region to the lowlands in Mid-Western Nepal. Every year there are reports of landslides blocking the road, making this area largely inaccessible. However, little effort has focused on systematically identifying landslides and landslide-prone areas along this highway. In this study, landslides were mapped with an object-based approach from very high-resolution optical satellite imagery obtained by the DigitalGlobe constellation in 2012 and PlanetScope in 2018. Landslides ranging from 10 to 30,496 m2 were detected within a 3 km buffer along the highway. Most of the landslides were located at lower elevations (between 500–1500 m) and on steep south-facing slopes. Landslides tended to cluster closer to the highway, near drainage channels and away from faults. Landslides were also most prevalent within the Kuncha Formation geologic class, and the forested and agricultural land cover classes. A susceptibility map was then created using a logistic regression methodology to highlight patterns in landslide activity. The landslide susceptibility map showed a good prediction rate with an area under the curve (AUC) of 0.90. A total of 33% of the study arealies in high/very high susceptibility zones. The map highlighted the lower elevated areas between Bangesimal and Manma towns with the Kuncha Formation geologic class as being the most hazardous. The banks of the Karnali River, its tributaries and areas near the highway were also highly susceptible to landslides. The results highlight the potential of very high-resolution optical imagery for documenting detailed spatial information on landslide occurrence, which enables susceptibility assessment in remote and data scarce regions such as the Karnali highway.

show abstract

Landslide mapping from multi-sensor data through improved change detection-based Markov random field

Cited by 136 publications

References 75 publications

Mapping Landslides on EO Data: Performance of Deep Learning Models vs. Traditional Machine Learning Models

Mapping Landslides on EO Data: Performance of Deep Learning Models vs. Traditional Machine Learning Models

Constructing a complete landslide inventory dataset for the 2018 monsoon disaster in Kerala, India, for land use change analysis

Use of Very High-Resolution Optical Data for Landslide Mapping and Susceptibility Analysis along the Karnali Highway, Nepal

Contact Info

Product

Resources

About