Deep Learning and AI-Powered Natural Catastrophes Warning Systems

Introduction

Natural disasters (e.g. landslides, floods, earthquakes, tropical cyclones, etc.) are complex phenomena that affect not only the environment of the area but also assets of that area like agriculture, infrastructure, and economic assets. Worldwide, governments and citizens alike have been increasingly threatened by the growing number and intensity of natural disasters and extreme weather occurrences in the past several years (Kim et al., 2018). As the number of people living in cities continues to rise, it is more important than ever to take measures to keep people safe and improve their living conditions. The globe is experiencing global urbanisation as more and more people choose to live in cities due to the comfortable living conditions there. Urban regions are home to about half of the world's population. This sort of expansion happens more quickly in developing nations (Ream et al., 2020). Figure 2 shows that between 1980 and 2021, the percentage of Chinese citizens residing in urban areas rose from 19.37% to 64.72% of the total population. As the number of people living in urban areas rises, more and more cities are being built or existing ones are being enlarged. An example of a developed/expanded mega city with more than 25 million residents is Shanghai, China. Within this framework, the security and regular functioning of buildings and infrastructures are significantly jeopardized by both natural and man-made catastrophes (Pekar et al., 2020). The frequency of floods and landslides caused by heavy rain in metropolitan areas has been on the rise in recent years, perhaps due to the increased occurrence of extreme weather events brought about by climate change. Early prediction and warning systems can lessen the impact, while technical solutions can stop them from happening altogether. Nevertheless, earthquakes can still not be predicted with enough precision to prevent damage. Therefore, it is essential for sustainable city development to develop and implement earthquake-resistant, vibration-damping, and seismically-isolated structures. One “natural” calamity that humans have brought about is land subsidence, which is mostly caused by the over-pumping of groundwater. Consequently, in order to properly respond to such crucial situations, updated solutions are required for disaster management and built assets operations. Government agencies, reaction groups, and healthcare institutions must communicate effectively and make quick decisions in order to respond to public health emergencies. Crucial components of disaster risk reduction include understanding situational risk, increasing governance, improving readiness for effective response, and investing in steps to build resilience (Lopez et al., 2018).

More accurate disease outbreak predictions, better evacuation plans, and more efficient resource distribution are all possible thanks to the proliferation of AI and ML, which allow for real-time data monitoring and decision-making in high-pressure situations. As shown in Figure 1, ML models are normally trained on massive amounts of task-specific data and then applied to new test data without the need for explicit programming or manually-crafted decision limits. In order to train, these algorithms often change the model's parameters iteratively; subsequent predictions and performance on the target task are based on these modifications.3 When it comes to the ability to draw conclusions or make predictions, ML is comparable to statistics. On the other hand, ML algorithms focus on prediction, whereas statistical models excel at inferring correlations between variables (Najafi et al., 2022).

Figure 1.

Number of research studies on disaster prediction in the last thirteen years