Automated machine learning for rainfall-induced landslide hazard mapping in Luhe County of Guangdong Province,China

Landslide hazard mapping is essential for regional landslide hazard management.The main objective of this study is to construct a rainfall-induced landslide hazard map of Luhe County,China based on an automated machine learning framework(AutoGluon).A total of 2241 landslides were identified from satellite images before and after the rainfall event,and 10 impact factors including elevation,slope,aspect,normalized difference vegetation index(NDVI),topographic wetness index(TWI),lithology,land cover,distance to roads,distance to rivers,and rainfall were selected as indicators.The WeightedEnsemble model,which is an ensemble of 13 basic machine learning models weighted together,was used to output the landslide hazard assessment results.The results indicate that landslides mainly occurred in the central part of the study area,especially in Hetian and Shanghu.Totally 102.44 s were spent to train all the models,and the ensemble model WeightedEnsemble has an Area Under the Curve(AUC)value of92.36%in the test set.In addition,14.95%of the study area was determined to be at very high hazard,with a landslide density of 12.02 per square kilometer.This study serves as a significant reference for the prevention and mitigation of geological hazards and land use planning in Luhe County.

An open-accessed inventory of landslides triggered by the M_(S)6.8 Luding earthquake,China on September 5,2022

This study constructs a preliminary inventory of landslides triggered by the M_(S) 6.8 Luding earthquake based on field investigation and human-computer interaction visual interpretation on optical satellite images.The results show that this earthquake triggered at least 5007 landslides,with a total landslide area of 17.36 km^(2),of which the smallest landslide area is 65 m^(2)and the largest landslide area reaches 120747 m^(2),with an average landslide area of about 3500 m^(2).The obtained landslides are concentrated in the IX intensity zone and the northeast side of the seismogenic fault,and the area density and point density of landslides are 13.8%,and 35.73 km^(-2) peaks with 2 km as the search radius.It should be noted that the number of landslides obtained in this paper will be lower than the actual situation because some areas are covered by clouds and there are no available post-earthquake remote sensing images.Based on the available post-earthquake remote sensing images,the number of landslides triggered by this earthquake is roughly estimated to be up to 10000.This study can be used to support further research on the distribution pattern and risk evaluation of the coseismic landslides in the region,and the prevention and control of landslide hazards in the seismic area.

Producing consistent visually interpreted land cover reference data: learning from feedback

Reference data for large-scale land cover map are commonly acquired by visual interpretation of remotely sensed data.To assure consistency,multiple images are used,interpreters are trained,sites are interpreted by several individuals,or the procedure includes a review.But little is known about important factors influencing the quality of visually interpreted data.We assessed the effect of multiple variables on land cover class agreement between interpreters and reviewers.Our analyses concerned data collected for validation of a global land cover map within the Copernicus Global Land Service project.Four cycles of visual interpretation were conducted,each was followed by review and feedback.Each interpreted site element was labelled according to dominant land cover type.We assessed relationships between the number of interpretation updates following feedback and the variables grouped in personal,training,and environmental categories.Variable importance was assessed using random forest regression.Personal variable interpreter identifier and training variable timestamp were found the strongest predictors of update counts,while the environmental variables complexity and image availability had least impact.Feedback loops reduced updating and hence improved consistency of the interpretations.Implementing feedback loops into the visually interpreted data collection increases the consistency of acquired land cover reference data.

An Updated Database and Spatial Distribution of Landslides Triggered by the Milin, Tibet M_(w)6.4 Earthquake of 18 November 2017

The M_(w)6.4 earthquake on November 18, 2017 in Milin County, Nyingchi City, Tibet triggered thousands of landslides. By comparing visual interpretation of satellite images acquired shortly before and after the earthquake and field survey, we have created a new landslide database which includes 3 130 coseismic landslides, each with an area of 0.01 to 4.35 km^(2). Six factors(elevation, slope angle, slope aspect, lithology, distance from the epicenter and distance from the seismogenic fault) were selected to correlate with the coseismic landslides. In addition, the area and density of landslides were counted as indicators. Results show that most landslides occurred where the elevation is between 2 000–3 000 m, with a 40°–50° slope angle and S, E or SE slope aspect, schist or gneiss lithologies, 10–15 km from the epicenter, and 5 km within the seismogenic fault. Most of the landslides, triggered by the M_(w)6.4 earthquake, are concentrated near the seismogenic fault rather than at the epicenter, indicating that the seismogenic structure is more influential than the location of the epicenter. Our findings may differ from other landslide database due to temporal image acquisition, interference from weather, and image resolution.

A correction technique for false topographic perception of remote-sensing images based on an inverse topographic correction techniqu

The false topographic perception phenomenon(FTPP)refers to the visual misperception in remote-sensing images that certain types of terrains are visually interpreted as other types in rugged lands,for example,valleys as ridges and troughs as peaks.For this reason,the FTPP can influence the visualization and interpretation of images to a great extent.To scrutinize this problem,the paper firstly reviews and tests the existing FTPP-correction techniques and identifies the inverse slope-matching technique as an effective approach to visually enhance remote-sensing images and retain the colour information.The paper then proposes an improved FTPP-correction procedure that incorporates other image-processing techniques(e.g.linear stretch,histogram matching,and flat-area replacement)to enhance the performance of this technique.A further evaluation of the proposed technique is conducted by applying the technique to various study areas and using different types of remote-sensing images.The result indicates the method is relatively robust and will be a significant extension to geovisual analytics in digital earth research.