An ensemble modeling approach for bird nest counting and 3-D cave modelling at Gomantong, Sabah, Malaysia using terrestrial laser scanning

As far back as early 15th century during the reign of the Ming Dynasty (1368 to 1634), Gomantong cave in Sabah, Malaysia, has been known as one of the largest roosting sites for swiftlet birds (Aerodramus maximus and Aerodramus fuciphagus) and wrinkle-lipped bats (Chaerephon plicatus) in very large colonies. Until recently, no study has been done to estimate the colony sizes of these inhabitants in spite of the grave danger posed to them by human activities and potential loss of the cave to post-speleogenetic deformation. In this study, high resolution terrestrial laser scanning data was used to (i) detect and count bird-nests in cave, (ii) model the cave in three-dimensional (3D) geometry at different resolution scales and (iii) to analyze the cave passage geomorphology. The proposed ensemble modelling approach integrates Taguchi and Objective Function optimization techniques to optimize image segmentation parameters - scale, shape, and compactness. The resulting image objects were classified as birds-nest and cave-wall by exploiting the laser return intensity and the size of image objects. Thereafter, the detection accuracy, reliability and transferability of the method were evaluated. For geometric reconstruction, 3D model of the cave was generated at two resolution scales by using meshing and textured 3D rendering. The models allow investigating the passage network structure, morphometry, geologic rock structure and macro- and micro-geomorhpological features. Result of the detection accuracy yielded area under the ROC curve (AUC) of 0.93 and P<0.0001 at 95% confidence level. Similarly, the tests of heterogeneity using Cochran’s Q and Inconsistency I2 produced p-value of 0.384 and I2 value of 5.10% at 95% confidence interval respectively. Also, the transferability assessment produced overall accuracies of 0.93 and 0.94 AUC and p<0.0001 at 95% confidence level using the two image scenes taken from the lower and upper caves respectively. From the nine image scenes used for testing and validation, 36,088 bird-nests were counted. The 3D mesh model allows visualizing and characterising the general geomorphological complexity of the entire cave network, internally and externally in 3D space. Furthermore, mesh model permits extracting detail profiles and cross sections in their geometrically correct dimensions, analyzing the cave’s structure and morphometry. On the other hand, textured 3D rendering offers true-to-life visualization of all the fine geometric details and complete representation of the cave’s interior “infrastructure” at sub-millimetres accuracy. It allows discerning different speleogenetic phases, karstification processes and micro-morphologies such as wall and ceiling seepage, hanging rocks, fractures, scallops, ceiling flush dome, pockets, bell-hole and avens. Besides, the texture model permits identification of cave arts and engravings along the passages. Point cloud-based analysis of passage rock discontinuity shows that the formations of the cave channels have been influenced by strike, dip/direction and bedding planes. Potentially, passage failure is predicted along east-west, west-east and southeast-northwest. This investigation reveals that Gomantong cave passage developments were largely controlled by the geologic rock structure; some of the conduits still retain their original phreatic tube formation while majority have been largely modified.

Preview Text FK 2018 11 IR.pdf Download (987kB) | Preview

Actions (login required)

View Item