UAV photogrammetry was used to obtain the spatial distributions of ground height and vegetation height in a sandy, oligotrophic wetland in Naruto-Togane, where carnivorous plants grow. The correspondences between these ground and vegetation height distributions, groundwater quality, and vegetation distributions of Iris ensata, Patrinia scabiosifolia, and Phragmites australis were analyzed. The entire wetland is burned totally at the end of January every year and irrigated after the burning in order to preserve the carnivorous plant community in this wetland. Immediately after the burning, the 3D ground surface was measured using stereo photographs from a UAV. The 3D plant canopy surfaces of the entire wetland were also measured in June and September, when Iris ensata and Patrinia scabiosifolia were in bloom, respectively. The ground resolution was 15 mm. As a result, the ground surface elevation values were obtained with the average error of 33 mm. The spatial distribution of vegetation height was obtained from the ground and vegetation surface measurements. The wetland characteristics shown in this study provide the basis for the conservation of Naruto-Togane wetland.

This paper shows uncertainty evaluation and proposals for performance improvement methodologies in image generation of landslide mass by deep learning models. Model ensembles and Monte Carlo Dropout (MCD) were used to evaluate uncertainty. Three methodologies were examined as performance improvement methodologies. The methodologies were slide processing, recall/precision emphasized models, and transfer learning with an inherent factor of landslide. The recall/precision emphasized models were developed by the improved loss function. The result showed that MCD could not be an alternative to model ensembles. In performance improvement methodologies, the transfer learning with geology distribution scored at 80% of precision. The recall/precision emphasized models inferred the distribution of landslide mass adequately. The effectiveness of the slide processing was found to be dependent on the performance of the trained model.