Anomaly Detection in Meteorological Data Using a Hierarchical Temporal Memory Model: A Study on the Case of Kazakhstan

Abstract

In meteorology, which studies atmospheric events, data representing various properties such as temperature, rainfall, and wind speed are collected regularly over a certain period. Unexpected trends in the data may indicate that an abnormal situation is approaching. Therefore, time series (TS) data play an essential role in the early detection of potential meteorological risks. However, applying effective models by considering many complex parameters in performing accurate analysis and anomaly detection (AD) is an important criterion. In this study, machine learning-based AD is performed using a dataset containing meteorological data on different features collected between January 1, 2019, and June 30, 2023, for Kazakhstan, which has the ninth-largest surface area in the world. The Hierarchical Temporal Memory (HTM) model was used for AD, which can provide more accurate forecasts by modeling long-term dependencies and producing effective results in solving TS problems. Detected anomalies are reported at various levels depending on threshold values. In addition, to analyze the ADs more precisely, correlations are calculated using the Spearman model, which allows us to determine the strength and direction of the monotonic relationship between variables. The study's findings show that the HTM is an effective model for AD using TS data on meteorological features.