Применение алгоритма Isolation Forest для обоснования уникальности водоемов в группе карстовых озер

Abstract

Improving the scientific foundations for the development and expansion of the network of specially protected natural areas requires the search for algorithms that could be used to identify unique ecosystems. Algorithmization of the anomaly identification process provides an opportunity not only to process large amounts of data but also leads to obtaining objective and comparable estimates. The purpose of this research is to identify the most optimal mechanisms for identifying anomalous values for the morphometric characteristics of karst lakes, which may indicate the uniqueness of the entire lake ecosystem. Within the framework of this article, the study was carried out based on a mathematical analysis of samples built for various characteristics based on the WORLDLAKE database. Statistical methods and the Isolation Forest (iForest) machine learning algorithm were used as methods of analysis. As a result of applying the iForest algorithm to a sample of morphometric parameters of karst lakes, consisting of 738 objects, 43 anomalous water bodies were identified. An expert assessment of the final set of lakes for the uniqueness of their ecosystems showed that the chosen method for identifying anomalous values is well suited for the task at hand. Many lakes with an anomaly index above 60% can be recognized as unique due to the unusualness of their abiotic characteristics; a number of them also have a peculiar biota. The anomalous objects included such well-known lakes as Tserik-Kol’, Crveno, Salda Lake, Trihonida, Vegoritida, Petron, etc. Moreover, for most of them, anomalies were detected for several parameters at once. Thus, the applied algorithm for identifying anomalous morphometric characteristics of lakes made it possible to obtain interesting samples for further expert analysis of the entire lake ecosystem for its uniqueness.