Investigations on Predicting Global Climate Change using Machine Learning Models

Abstract

Historically, climate prediction models have simulated the dynamics of the climate system by using intricate physical
equations; however, these models frequently need large computer resources and lengthy calculation times. Machine
learning approaches have demonstrated significant promise for pattern detection and prediction in recent years. In
particular, machine learning models' benefits in handling massive datasets have made them a popular area of study in
the field of climate science. In this work, we present a convolutional neural network-based (CNN) model that can
process and analyse multi-dimensional climatic data, such as temperature, air pressure, humidity, and CO2
concentration, from large-scale satellite datasets. Historical climate data is used as the input. The convolutional layer
extracts the spatial features, while the fully connected layer performs feature fusion and produces the final forecast
output. Last but not least, we trained the model using historical climate data collected across a variety of time periods.
The findings demonstrate that CNN-based models outperform traditional physical models in terms of accuracy and
prediction errors when it comes to forecasting changes in the world's average temperature, precipitation, and extreme
weather occurrences