Automatic Segmentation of Sinkholes Using a Convolutional Neural Network. Issue 2 (15th February 2022)

• Record Type: Journal Article
• Authors: Rafique, Muhammad Usman; Zhu, Junfeng; Jacobs, Nathan
• Is Part Of: Earth and space science. Volume 9:Issue 2(2022)
• Page Start: n/a
• Page End: n/a
• Publication Date: 2022-02-15
• Subjects: image segmentation -- U‐Net -- sinkhole -- LiDAR -- DEM -- aerial image; Space sciences -- Periodicals; Geophysics -- Periodicals; 500.5
• DOI: 10.1029/2021EA002195
• Languages: English
• ISSNs: 2333-5084

Abstract

Abstract: Sinkholes are the most abundant surface features in karst areas worldwide. Understanding sinkhole occurrences and characteristics is critical for studying karst aquifers and mitigating sinkhole‐related hazards. Most sinkholes appear on the land surface as depressions or cover collapses and are commonly mapped from elevation data, such as digital elevation models (DEMs). Existing methods for identifying sinkholes from DEMs often require two steps: locating surface depressions and separating sinkholes from non‐sinkhole depressions. In this study, we explored deep learning to directly identify sinkholes from DEM data and aerial imagery. A key contribution of our study is an evaluation of various ways of integrating these two types of raster data. We used an image segmentation model, U‐Net, to locate sinkholes. We trained separate U‐Net models based on four input images of elevation data: a DEM image, a slope image, a DEM gradient image, and a DEM‐shaded relief image. Three normalization techniques (Global, Gaussian, and Instance) were applied to improve the model performance. Model results suggest that deep learning is a viable method to identify sinkholes directly from the images of elevation data. In particular, DEM gradient data provided the best input for U‐net image segmentation models to locate sinkholes. The model using the DEM gradient image with Gaussian normalization achieved the best performance with a sinkhole intersection‐over‐union (IoU) of 45.38% on the unseen test set. Aerial images, however, were not useful in training deep learning models for sinkholes as the models using an aerial image as input achieved sinkhole IoUs below 3%. Plain Language Summary: Sinkholes are very common in areas with limestone rocks. Sinkholes can damage roads, buildings, and other infrastructure and sometimes even cost human lives. Sinkhole maps are needed for land use planning and hazard mitigation. Because sinkholes often occur in large numbers, often in the thousands, accurately mapping each of them manually is expensive and laborious. In this study, we applied deep learning, a form of artificial intelligence, to build computer models to automatically locate sinkholes from images created from elevation data. These models used the image segmentation technique to label every pixel in an image as either sinkhole or non‐sinkhole. We used images of elevation, slope, elevation gradient, and shaded relief as inputs to models. Model results suggested that deep learning offered a viable way to automatically locate sinkholes from elevation data. In particular, models using elevation gradient information performed the best. We also evaluated aerial imagery to train the models and found that aerial images were not useful in training deep learning models for sinkhole identification. Key Points: Image segmentation models using elevation and aerial images were trained to locate sinkholes Model's out‐of‐distribution generalization was assessed Elevation gradient images provide the best input for training sinkhole segmentation models