Forecasting steppe fires using remote sensing data of time series

• Published in: IOP conference series. Materials Science and Engineering Vol. 1047; no. 1; pp. 12092 - 12097
• Main Authors: Goryaev, V M, Basangova, E O, Goldvarg, T B, Bembitov, D B, Djachnaeva, E N, Lidzhi-Garyaev, V V
• Format: Journal Article
• Language: English
• Published: Bristol IOP Publishing 01.02.2021
• Subjects: Data mining; Dust storms; Fires; Forecasting; Forecasting techniques; Internet of Things; Machine learning; Meteorological data; Model testing; Regression analysis; Remote sensing; Statistical analysis; Statistical methods; Time series; Weather stations
• ISSN: 1757-8981; 1757-899X
• DOI: 10.1088/1757-899X/1047/1/012092

In fire monitoring systems, Internet of Things sensors allow predicting the area of fire in combination with machine learning. This article provides an insight into the use of machine learning models in predicting steppe fires. The work uses statistical methods for forecasting time series, methods of regression analysis and forecasting techniques. This project explores a data mining approach for predicting steppe fires. Several different data mining methods are considered and at least five different special methods of feature selection (spatial, temporal, Fire Weather Index components, weather attributes and our own developed classes). At the moment, data mining software systems have been tested on real data collected in the southern regions of Kalmykia. The study used open data from meteorological stations, our own database time series collected over the past 3 years, tested various data mining methods and at least five meteorological inputs, which will allow predicting a burnt area of small and medium fires, an area of dust storms, etc. The results show that the k Nearest Neighbor model is the most appropriate choice for the steppe fire forecasting model. The chosen model takes into account both meteorological data and images for early forecasting of FIRMS fires.