AI – Based Climate Scenario Generation for Environmental Forecasting

Abstract One of the greatest challenges facing society today is climate change, which is resulting in more frequent and severe weather, an increase in the level of the oceans, and destruction of the environment. While traditional climate models (which rely on physics) can be very effective at providing us with valuable information, their effectiveness is somewhat limited because they are computationally intensive to create (i.e. very long runtime) and have a coarse resolution (i.e. they only provide us with general information over large areas of land). The use of artificial intelligence (AI), machine learning (ML) or deep learning (DL) is beginning to revolutionize how we model and predict future climate conditions. By being trained on large volumes of observation, satellite and reanalysis data, AI models can find nonlinear relationships between climate variables, enhance forecast accuracy, and decrease the cost of computing (to name just a few). AI-enabled tools (e.g. FourCastNet, Atmo AI, and QuickClim) provide improved climate forecasts at a speed and granularity much quicker than traditional models, thereby improving early warning systems (e.g. sudden floods, cyclones and heatwaves). This paper will review the architecture and performance metrics of current AI-enabled climate models, discuss case studies of their use in the real world, and demonstrate how they can improve the capacity of policymakers, urban planners, agricultural managers, and emergency responders to respond to climate disasters. While the use of AI models in the study of climate is promising, there are also several challenges associated with the use of AI (i.e., data quality, explainability, ethics, and the cost of training models). It is critical that these challenges are adequately addressed before we can deploy AI in an ethical and responsible manner. Therefore, this review will summarize the latest research on AI climate models, the latest performance metrics and case studies of their application in real time and provide recommendations to improve the use of predictive analytics for enhancing environmental resilience around the world.[2],[5], [11],[17]. Keywords – Artificial Intelligence, Machine Learning, Deep Learning, Climate Modeling, Climate Forecasting.