In recent years, natural catastrophes and extreme weather events have increased in frequency and severity, posing a serious threat to people and governments throughout the globe (Leng et al., 2023). Such occurrences may result in fatalities, infrastructural damage, and interruptions of economic activity (Ruidas et al., 2022a; Chen et al., 2022). Therefore, mitigating the effects of these catastrophes depends on our capacity to properly forecast and prepare for them. In this context, the use of machine learning algorithms has shown promise for improving weather forecasting and natural catastrophe predictions, which may help in disaster preparation and response operations (Linardos et al., 2022, Powers et al., 2023). A subset of artificial intelligence (AI) known as machine learning algorithms (MLAs) enables computers to discover patterns in massive datasets without having to be explicitly programmed. These algorithms are able to scan large data sets and spot patterns that human analysts would miss. Machine learning algorithms may be used to examine a range of data sources, including satellite data, atmospheric data, and historical weather and catastrophe data, to provide precise forecasts in weather prediction and natural disaster forecasting (Aybar-Ruiz et al., 2016; Moosavi et al., 2021).

Weather prediction utilizing MLA’s involves the use of different types of algorithms, such as neural networks (NN), decision trees (DT), and random forests (RF) (Fowdur and Nassir-Ud-Diin Ibn Nazir, 2022; Xu et al., 2023). The number of research on weather prediction using machine learning algorithms from 2008 to 2022 is shown in Figure 1. The data shows that the number of studies has been increasing over time, which has led to the noticeable increase in studies from 2014 to 2022. This pattern’s conclusion demonstrates the increased interest in applying machine learning algorithms for weather forecasting, which might improve forecast accuracy and catastrophe preparation. These algorithms may look at a variety of data sources, including satellite photography and atmospheric data, to find trends and forecast weather. The possibility of a heatwave or a drought, for instance, may be predicted by MLAs by examining data on temperature, humidity, and precipitation (Ruidas et al., 2022c; Kim et al., 2022).

Various kinds of algorithms, including clustering algorithms, regression algorithms, and support vector machines, are used in the prediction of natural disasters (Zhang et al., 2022b). The number of works on utilising machine learning algorithms to predict natural disasters during the last 15 years is summarised in Figure 2. The figures show a consistent rise in the number of investigations every year, with the most research occurring in the most recent year (2022) (57). This shows that there is increased interest in and awareness of the potential advantages of using machine learning algorithms to anticipate natural disasters (Chakrabortty et al., 2023).

These algorithms can analyze various data sources, such as seismic data, historical disaster data, and weather data, to make predictions about natural disasters. For example, they can predict the likelihood of an earthquake or a hurricane by analyzing data such as seismic activity, wind speed, and sea surface temperature (Barrera-Animas et al., 2022). The use of these algorithms in catastrophe preparation and response may have a number of advantages. For instance, precise weather forecasts and natural catastrophe predictions may assist governments and organisations in taking preventive measures to be ready for and react to these occurrences. People may be informed of imminent natural catastrophes through early warning systems, allowing them to flee to safer areas (Karir et al., 2022). By focusing resources on the regions most likely to be impacted by natural disasters, resource allocation may be improved. Finding the most effective escape routes and organising the flow of people may both help with evacuation planning. The use of MLAs in disaster planning and response is not without its difficulties and restrictions, however. The precision with which these algorithms provide predictions is one of the major difficulties. False alarms brought on by inaccurate forecasts may erode public confidence in the system (Jia et al., 2022; Saha et al., 2023). The accuracy of the predictions may also be impacted by the quality of the data utilised to train these algorithms. Additionally, if the data used to train them was biassed, they can be prejudiced as well. Their use in predicting weather and natural disasters has become a potential strategy to enhance disaster preparation and response operations (Esrafilian-Najafabadi and Haghighat, 2022). In this study, we adopt a robust methodology to explore the application of machine learning algorithms in forecasting weather patterns and predicting natural catastrophes, with a focus on enhancing global disaster preparedness and response efforts.

Using precise keywords and phrases linked to meteorological events, natural catastrophes, and machine learning, our method comprises a thorough literature analysis as depicted in Figure 3. From January 2012 to January 2023, we conduct a thorough search and analysis of peer-reviewed publications from renowned databases including SCOPUS, Science Direct, and Google Scholar. Multiple reviewers conduct the screening process, which guarantees consistency and relevance while upholding strict academic standards with the inclusion and exclusion criteria (Ruidas et al., 2022b). To offer a thorough overview of how machine learning algorithms contribute to effective catastrophe prediction, material from chosen papers is extracted and then synthesised (Ruidas et al., 2021; Ruidas et al., 2022a). This information includes technique specifics, input data sources, output results, case studies, and important conclusions (Jaydhar et al., 2022; Ruidas et al., 2023). This rigorous and thorough approach strengthens the reliability and validity of our conclusions, providing a current knowledge of the critical role that machine learning plays in reducing the effects of natural catastrophes on communities and nations throughout the globe.

Agriculture, transportation, and the energy sector are just a few of the businesses that depend heavily on weather forecasting. To make wise judgments and avert weather-related problems, accurate weather forecasts are essential. MLA’s have emerged as a viable strategy to increase the accuracy of weather forecasts in recent years (Fischer et al., 2023; Heddam, 2023). MLAs are capable of analysing enormous volumes of data to find patterns and make predictions. They come in several varieties, and each has advantages and disadvantages (Lan et al., 2020; Quan et al., 2022; Shafiei et al., 2022). Some of the most popular methods for predicting the weather include NN, DT, and RF (Figure 4).

NN are a type of MLA’s inspired by the structure of the human brain (Yang et al., 2022). Applications for these techniques include picture recognition and natural language processing. A comparison of several algorithms for predicting weather and natural disasters is shown in Table 1. The algorithms fall into three primary categories: reinforcement learning, unsupervised learning, and supervised learning. Each algorithm’s advantages and disadvantages are described, along with examples of how it has been used to forecast weather and natural disasters. It attempts to provide a thorough analysis of various machine learning algorithms and assist researchers and practitioners in choosing the best algorithm for their particular application.

In weather prediction, NN are used to analyze multiple data sources, such as satellite imagery, atmospheric data, and historical weather data, to identify patterns and make predictions (Brester et al., 2023). They can handle non-linear relationships between variables and make predictions with high accuracy (Buyrukoğlu et al., 2021; Seon Park et al., 2022). DT is another type used for weather prediction (Li et al., 2023a). These algorithms are built on a set of if-then rules that let the algorithm anticipate outcomes depending on certain circumstances (Pereira et al., 2022). Due to their simplicity and readability, they are great for a variety of applications. They may be used to analyse many data sources, including humidity, temperature, and precipitation, to provide weather predictions (van Blokland et al., 2021). Predictions are produced by RF, a collection of decision trees (He et al., 2022). This requires the development of several decision trees, each of which is trained on a distinct subset of the data. The agreement of each DT’s projections forms the basis for the final projection. They can manage complicated variable interactions and are extremely precise (Ghebleh Goydaragh et al., 2021). They may look at a variety of data sources to make weather forecasts. Satellite data is used to give information on temperature, cloud cover, and other weather-related factors. Weather forecasts for large regions may be made using this data, which includes specifics like temperature, pressure, and humidity. A further use of atmospheric data is the forecasting of local weather. Historical weather data offers details on previous weather conditions and may be used to spot patterns and forecast the weather in the future. One example of a successful MLA-based weather prediction model is the Global Ensemble Forecast System of the National Oceanic and Atmospheric Administration (NOAA) (GEFS). A collection of global weather models are used by the GEFS weather prediction model to provide predictions (Haver et al., 2018). In order to provide predictions, the model examines a variety of data sources, such as atmospheric data and historical weather data. GEFS has been used to forecast a variety of meteorological occurrences, including hurricanes, blizzards, and heatwaves (Jiang et al., 2022). A wide range of businesses, including agriculture, transportation, and energy, rely on GEFS projections. IBM’s Deep Thunder is another reliable weather prediction tool (Afra et al., 2022). It is a very effective model for forecasting the weather, and it has been used to predict a variety of meteorological events, including thunderstorms, hurricanes, and snowstorms. It uses neural networks to evaluate a variety of data sources, including satellite and atmospheric data (Lu et al., 2022). Transportation and energy industries, among others, utilise Deep Thunder’s forecasts.

Accurate prediction and forecasting of natural disasters can aid in minimising their effects and enhancing disaster response. MLAs have demonstrated promise in accurately predicting natural calamities (Jiang et al., 2022; Singh et al., 2023). Random Forest (RF) builds a number of Decision Trees (DT) and then combines their outputs to form a final forecast (Figure 5). This algorithm can handle classification and regression issues. One of the primary benefits of is that it can handle huge datasets with high dimensionality and non-linear variable interactions (Bhoi et al., 2022). It can also manage missing data and outliers and is less susceptible to overfitting than other decision tree-based algorithms, making it more accurate at predicting the outcomes of new data (Liu et al., 2020). In addition, it can provide estimates of variable relevance, which facilitates the identification of the most pertinent factors for producing correct forecasts (Wang et al., 2022b).

RF can analyse multiple data sources, such as satellite imagery, weather data, and historical disaster data, to predict the likelihood of a natural disaster occurring. For instance, an RF algorithm can analyse data on temperature, humidity, wind speed, and precipitation to predict the likelihood of a wildfire (Rahman et al., 2021). Support vector machine (SVM) is another technique often employed for natural catastrophe prediction. SVM is a supervised learning algorithm used to evaluate data and detect patterns. It may examine numerous data sources, such as past disaster data, weather data, and satellite images, to identify trends and predict the probability of a natural catastrophe occurring (Singh et al., 2023). It has various characteristics that make it effective for natural disaster forecasting, it can handle both linear and non-linear data by utilising different kernel functions, such as polynomial, Gaussian, and sigmoid (Li et al., 2023b) and can handle high-dimensional data and escape the curse of dimensionality (Chen and Li, 2020). Also, it is less affected by overfitting, as it tries to maximize the margin between classes (Liu et al., 2021a) and can handle imbalanced datasets, which are common in natural disaster forecasting, by adjusting the weights of the classes. Finally, SVM can be used for both classification and regression tasks, making it versatile for different types of forecasting problems (Chen et al., 2022) (Figure 6).

Deep learning algorithms, such as convolutional neural networks (CNNs), have also shown promise in natural disaster forecasting as it has several advantages when it comes to natural disaster forecasting like it can handle high-dimensional and complex data, such as satellite and radar images, which are often used in natural disaster monitoring (Ullah et al., 2022). They can extract relevant features from the images, such as cloud cover or precipitation, and use these features to predict the occurrence and intensity of natural disasters (Akter et al., 2021). It can learn spatial dependencies between the pixels in an image, allowing them to identify patterns and anomalies that may be indicative of an impending natural disaster and can be trained on large datasets, which can improve their accuracy and ability to generalize to new data (Guha et al., 2022) (Figure 7).

CNN can analyze satellite imagery to predict the likelihood of a hurricane occurring. One of the most critical applications of MLA’s in natural disaster forecasting is predicting the intensity and path of hurricanes. Hurricane forecasting involves analyzing multiple data sources, including satellite imagery, atmospheric data, and historical hurricane data, to generate predictions (Najafi et al., 2022). NN and RF, for instance, can assess various data sources to generate accurate predictions of storm intensity and course, as well as the probability of floods (Najafi et al., 2022; Rani et al., 2022). Flooding is a significant natural disaster that can cause severe damage to life and property. SVM and DT can analyse various data sources, such as weather data and historical flooding data, to predict the likelihood of flooding occurring in a specific region (Danish, 2022). Similarly, SVM and DT can analyse various data sources, such as historical earthquake data, tectonic plate movements, and seismic activity, to predict the likelihood of an earthquake occurring in a specific region (Teodoro and Duarte, 2022).

These algorithms analyse vast amounts of data from numerous sources to predict natural catastrophe frequency and severity. Therefore, emergency responders may be notified in advance and resources can be deployed properly, improving disaster response. Additionally, they may be used to keep track on natural catastrophes in real time, enabling quick and efficient response. Organizations involved in catastrophe planning and response may also achieve improved cooperation and communication by using these algorithms (Islam et al., 2020).

For effective emergency planning and response, accurate weather and catastrophe predictions are crucial. Accurate forecasts are essential for reducing the consequences of natural disasters since inaccurate or delayed predictions may cause significant losses in life and property. Planning your evacuation is an essential part of being prepared for and responding to a catastrophe. The planning of an evacuation may prevent fatalities and speed up the relocation of individuals from dangerous places. For instance, accurate forecasting may help emergency responders and authorities prepare shelter locations, evacuation routes, and other logistical issues in the event of storms or floods. Another way that accurate forecasting might help with disaster planning and response is via resource allocation (Johnson et al., 2022). Authorities may organise the distribution of resources like food, water, medical supplies, and rescue equipment by using accurate forecasting to make sure that they are delivered in a timely and efficient way to the areas that need them the most. As a result, there will be less waste and resources will be accessible when and where they are most needed. Exact predictions may help in disaster response in addition to organising evacuations and allocating resources. For example, accurate forecasting may help emergency responders plan and carry out suitable response strategies, including the deployment of fire-fighting gear or flood barriers to mitigate the consequences of the catastrophe. The ability to identify and locate people who may be trapped or stranded as a consequence of the catastrophe is another benefit of accurate forecasts for search and rescue operations. MLAs are an essential tool for emergency management organisations and first responders because their ability to increase forecast accuracy may significantly strengthen efforts to reduce the impact of catastrophes, react to them, and aid in recovery (Yu et al., 2021).

MLAs may help authorities allocate resources more effectively and efficiently by doing so. Machine learning algorithms can discover and distribute resources to areas with the largest resource needs by analysing data on previous catastrophes, available resources, and potential crisis scenarios. By doing this, waste may be decreased and resources will be distributed where they are most needed. Planning evacuations is another area where machine learning algorithms may be quite useful (Ganguly et al., 2019). These algorithms may help in the planning and execution of efficient evacuation procedures by analysing data on population density, infrastructure, and potential crisis circumstances. With the use of these algorithms, disaster response efforts, resource allocation, evacuation planning, and catastrophe forecast accuracy may all be improved. They should thus be promoted and used more often since they are an essential resource for disaster management organisations and first responders (Chamola et al., 2021).

One instance is the use of machine learning algorithms to predict floods in Vietnam. The Vietnamese government and the World Bank worked together to create a flood warning system that uses machine learning algorithms to analyse data from meteorological stations, river gauges, and satellite images. The technology sends out early warnings of impending floods, enabling authorities to evacuate citizens and more effectively allocate resources. Since its beginnings, the system has successfully lessened the impacts of floods on the local community. Using predictive analytics to find California wildfires is another useful use of machine learning techniques. The state’s fire service uses machine learning algorithms to analyse data on weather patterns, vegetation, and other factors that affect the likelihood of wildfires. The algorithms provide early warnings of potential wildfires, enabling emergency responders to mobilise assets and create response strategies before the fire spreads (Linardos et al., 2022). The approach has been successful in protecting property and reducing the impact of wildfires on the neighborhood’s residents. The effective use of machine learning algorithms in earthquake prediction has also been noted. A system developed by the Japan Meteorological Agency analyses seismic data from hundreds of sensors spread around the country using machine learning techniques. The technology gives early warnings of potential earthquakes, enabling officials to deploy resources and carry out evacuation preparations more successfully. The technique has successfully lessened the impact of earthquakes on nearby populations and cut down on the likelihood of fatalities. Machine learning algorithms have been used in a range of disaster preparation and response projects, including hurricane forecasting, disease outbreak forecasting, and emergency response planning, in addition to these cases. Overall, the effective use of machine learning algorithms in planning and responding to disasters shows the potential advantages of these algorithms in reducing the consequences of natural disasters, saving lives, and protecting property.

Massive amounts of data are needed for MLAs to analyse and provide accurate predictions, and the accuracy of the algorithm’s predictions may be greatly influenced by the quality and completeness of the data. This may be particularly challenging in remote or rural places with poor data access. Another challenge is finding qualified personnel to design, operate, and maintain the systems (Algiriyage et al., 2021). This may be challenging in areas with little resources or skilled labour. Furthermore, machine learning algorithms are not foolproof and may still produce false positives or inaccurate predictions. For example, using algorithms to decide how to distribute resources or how to evacuate might cause issues with justice and equality, particularly if the system has biases against certain communities. In addition to these challenges, there are limitations to the use of machine learning algorithms in disaster planning and response (Johnson et al., 2022). It is difficult to predict the path of a hurricane since doing so requires the examination of a number of different factors, such as wind speed, temperature, and atmospheric pressure. If the criteria are too complex or the data is inadequate, MLAs may have trouble correctly forecasting the course of a cyclone. Despite these challenges and limitations, using machine learning algorithms to disaster prevention and response has significant advantages.

An algorithm created to predict the effects of flooding might be biassed towards low-income areas if it was trained on data that did not accurately represent their specific vulnerabilities and needs (Ganguly et al., 2019). Machine learning algorithms may be a useful tool for boosting disaster preparedness and response efforts, saving lives, and protecting property with careful planning and execution (Yu et al., 2021). Utilizing current, high-quality data that correctly represents current weather patterns and natural disaster occurrences is crucial to lowering the chance of inaccurate predictions. To counteract prejudice, it may be important to establish standards for data collection and analysis and to continuously evaluate the algorithm to find and remove any biases (Algiriyage et al., 2021). Additionally, to make sure that the algorithm truly represents the needs and priorities of the communities being served, it may be beneficial to engage community stakeholders in its design and implementation. In the event of natural disasters, we can better protect communities and save lives by using the potential of machine learning algorithms to enhance disaster planning and response (Islam et al., 2020).

The use of MLAs emerges as a key element in enhancing our capacity to forecast and manage natural catastrophes as we look to the future of disaster planning and response. It is impossible to emphasise the importance of MLAs in this context; nonetheless, there are a number of critical areas that need more study and development in order to fully realise the potential advantages of this technology. For machine learning systems to be useful in predicting disasters, their accuracy must be improved. Future efforts should concentrate on creating algorithms that are more resistant to unforeseen circumstances, combining current data together with past data to improve accuracy. Additionally, in order to guarantee dependability in disaster prevention and response, efforts to improve the quality and completeness of training data are crucial (Jaydhar et al., 2022; Chakrabortty et al., 2023).

The possible bias that might be present in their algorithms is a serious issue when using MLAs. In order to lessen the effects of current biases, future research should adopt detailed criteria for data gathering and analysis. To make sure that algorithms are reflective of the particular needs and vulnerabilities of the communities they serve, engagement with community stakeholders is essential. It is crucial to ensure that the advantages of this technology are available to all communities, regardless of their financial or technical capacity, in order to maximise the social effect of MLAs. To promote the long-term use of these algorithms on a worldwide scale, creative finance methods and collaborations with the commercial sector might be investigated (Ruidas et al., 2022a; Ruidas et al., 2023).

MLAs may be equipped to provide up-to-date information on changing catastrophe situations by integrating real-time data from many sources, such as weather sensors, social media, and satellite imaging. Harmed communities are less affected as a result of the quick detection and reaction to potential dangers made possible by this real-time analysis. Emphasizing data fusion, the synthesis of several data sources, may result in a more thorough and precise knowledge of certain occurrences. MLAs may provide more accurate forecasts of upcoming catastrophic occurrences by merging information from meteorological forecasts, historical records, and real-time sensor data. There are fresh opportunities when MLAs are compared to other upcoming technology like drones and driverless cars. For instance, drones with sensors may gather information on catastrophe circumstances, which MLAs can then examine to provide insightful information that will guide response activities. It is essential to make sure that MLAs are useable and accessible to communities with little access to resources or technical know-how. These technologies may be made more accessible and effective at the community level by creating user-friendly interfaces and partnering with regional groups.

The need of precise weather and natural catastrophe forecasting in efficient disaster preparation and response operations is stressed in this article. Its worldwide repercussions highlight the need of using cutting-edge technology on a global basis. Machine learning algorithms have a revolutionary influence on disaster management because they allow efficient resource allocation and early warnings, thereby saving lives and reducing socio-economic effects. The ubiquitous application of machine learning techniques is what gives our research its worldwide significance. These algorithms, capable of assessing a variety of data sources, provide insights essential for well-informed response activities across various geographies and climatic conditions. They act as a cornerstone in dealing with natural catastrophes all over the globe, highlighting the need for a coordinated, technology driven strategy to global concerns. Our study emphasises the dynamic character of this subject, despite ongoing issues like accuracy and data quality. Continuous innovation and research serve as catalysts for addressing these issues, assuring ongoing advancement and resistance to unanticipated setbacks. By recognising the limits of the present, we open the door to a future in which technology adapts and develops to satisfy the needs of a constantly changing environment. Looking forward, there is a large and fascinating range of possible uses for machine learning algorithms in disaster planning and response. Critical components include prioritising innovation, encouraging cooperation, and interacting with communities. By doing this, we are able to safeguard communities from natural calamities and ultimately save lives by leveraging the revolutionary potential of this technology.