|
ИСТИНА |
Войти в систему Регистрация |
ИСТИНА ФИЦ ПХФ и МХ РАН |
||
AN EXPERIMENT OF AUTOMATIC CLASSIFICATION AND MAPPING OF THE LANDFORMS OF THE YAMAL PENINSULAстатья
- Автор: Kharchenko S.V.
- Сборник: SGEM International Multidisciplinary Scientific GeoConference EXPO Proceedings
- Том: 24
- Год издания: 2024
- Место издания: STEF92 Technology
- Первая страница: 97
- Последняя страница: 104
- DOI: 10.5593/sgem2024/2.1/s08.13
- Аннотация: This study investigates the potential of automated geomorphological mapping using geomorphometric analysis and machine learning on the Yamal Peninsula, Russia. The research aims to classify landforms based solely on geomorphometric characteristics, bypassing traditional manual interpretation of aerial imagery and digital elevation models (DEMs). The study utilized a DEM of the Yamal Peninsula and a reference geomorphological map, including 10 distinct landform types. A total of 119 geomorphometric variables, including spectral characteristics of the terrain, were calculated and used for training a Random Forest classifier. The results demonstrate that the model achieved a 65.4% overall accuracy, significantly exceeding the baseline accuracy of 10%. While some landforms, like the first river terrace, were accurately classified with 98% precision, others, such as floodplains, showed lower accuracy. The study identified key geomorphometric variables contributing to the classification, highlighting the importance of "focal" characteristics reflecting the texture and pattern of topographic dissection. The findings suggest that automated classification based on geomorphometric analysis holds promise for geomorphological mapping. It can be used to expedite the creation of geomorphological maps and assist in identifying areas of uncertainty for further investigation. However, future research is necessary to improve the accuracy of specific landform classifications, particularly those with high spatial variability.
- Добавил в систему: Харченко Сергей Владимирович