The Application of Machine Learning for Predicting Global Seismicity

Introduction

Since ancient times cataclysmic disasters such as droughts, floods, earthquakes, volcanic eruptions, storms, and many other types of natural catastrophes, have had a profound impact on humans, at the cost of countless lives. These disasters are classified as natural disasters (Wirasinghe et al., 2013). The most severe natural disaster in recent history was the flood of the Yangtze–Huai River in China, in summer 1931. Up to 25 million people were affected by the effects of this flood (National Flood Relief Commission, 1933), hence it is considered the deadliest natural disasters since 1900 excluding epidemics and famines.

The number of deaths from natural disasters may change depending on the type of disaster and the affected area. But, from the average point of view, around 40,000 people per year are killed by natural disasters. For example, Figure 1 shows the yearly average of global annual deaths from natural disasters between 1900 and 2010s. The graph was created based on data from (OFDA/CRED International Disaster Data, 2021).

Figure 1.

Yearly average global of annual deaths from natural disasters, by decade.

As seen in Figure 1 the three deadliest natural disasters are droughts, floods, and earthquakes. However, in the last decades, the most dangerous natural disasters for people are considered to be earthquakes, extreme temperature, and floods. Even though the average global death toll from natural disasters in the 21^st century is less than in the previous century, the average death rate is still high.

Most of the Earth's meteorological processes are localised, and they make good weather forecasts only in a limited area. Space weather is always global on the planetary scale (Koskinen et al., 2001).

Further, the assumption that solar activities could have an influence on Earth’s natural disasters is not new. Back in 1853 the astronomer Wolf (1853) suggested sunspots might influence earthquake events. Since then, several studies, using statistical methods, have showed the correlation between solar activity and earthquakes. Odintsov et al. (2006) reported that seismic activity is related to the sunspot maximum during the solar cycle. Marchitelli et al. (2020) showed a correlation between solar activity and earthquakes with a magnitude(M) M>5.6.

Modern solar activity data and natural disaster data, as well as worldwide data, which exponentially increase every year with improved or new technologies – they contain a plethora of different parameters for solar but also natural disaster events. Reinse et al. (2018) stated that the International Data Corporation predicts an increase of the global dataset from 33 ZB in 2018 to 175 ZB by 2025. To work with such a huge amount of data, computer processing power must be faster but also algorithms more intelligent. There is a part of computer science that tries to achieve this goal by employing artificial intelligence. Studies focussing on intelligence of animals (Thorndike, 2000) and plants (Calvo et al., 2020) proved that one of the most crucial requirements for intelligence is learning. High intelligence is based on comprehensive learning and artificial intelligence is not an exception. Therefore machine learning is one of the most important and vital parts of artificial intelligence (Dunjko & Briegel, 2018).

One of the first occasions that machine learning was mentioned, was back in 1959, in the Samuel (1959) study. Samuel (1959) created a checkers programme, where two “machine-learning procedures” were used, and the study provided a start for the development of learning methods that would exceed average human abilities and would solve real life problems. A quote from the original article best describes machine learning: “Programming computers to learn from experience should eventually eliminate the need for much of this detailed programming effort.”