Identification of unique lakes of different origin by machine learning methods
Автор: Rasulova Anna, Izmailova Anna
Журнал: Бюллетень науки и практики @bulletennauki
Рубрика: Биологические науки
Статья в выпуске: 12 т.8, 2022 года.
Бесплатный доступ
At present, more than ever, the issue of developing criteria for selecting lakes for inclusion in the lists of protected areas, as well as assessing natural ecosystems that have undergone significant anthropogenic impact and require special attention from environmentalists, has become relevant. However, peer review of each individual ecosystem requires significant research and economic resources. Taking into account the area of Russia and the inaccessibility of some regions, it becomes almost impossible. For preliminary assessments and narrowing the search for candidates for protected areas, cameral methods can be used. One of them includes various methods for identifying anomalies in databases of morphometric, hydrochemical, hydrological, and hydrobiological characteristics of lakes. This paper discusses some machine learning methods aimed at identifying anomalous values for lakes of karst, volcanic and glacial origin. The main goal of this study is to find optimal mathematical methods for establishing the uniqueness of a particular lake ecosystem. The paper considers test samples of lakes obtained on the basis of the WORLDLAKE database. The following methods were used in the analysis: 1) local outlier factor, 2) isolated forest, and 3) one-class support vector machine. The features of the application of various methods depending on the morphometric origin of lake basins are revealed. The resulting anomalous objects were compared and then subjected to expert evaluation for their unique properties in various parameters. The expert assessment confirmed that most of the identified lakes can be considered unique, taking into account other features that characterize lake ecosystems.
Protected areas, ecosystem conservation, lakes, identification of anomalies, local outlier factor, isolated forest, one-class support vector machine
Короткий адрес: https://sciup.org/14126157
IDR: 14126157 | DOI: 10.33619/2414-2948/85/23