Data-Driven Predictive Analysis of Global Temperature Changes with Deep Learning

With growing concerns over rising carbon emissions and global temperatures, there is a pressing need for addressing climate change. As the repercussions of climate change are severe – such as acidification of oceans and extreme weather, it is integral to identify the key contributors leading to the problem to effectively combat it. In this paper, we intend to model the temperatures of specific regions towards predicting the future trends in temperature by leveraging historical data, and thus address global warming. We aim to utilize a data-driven approach to analyze integral parameters such as land surface temperature, ocean surface temperature, percentage change in temperature over the years etc., and use the inference as a basis to interpret the variations in climate policies/climate actions in tackling climate change. We leverage a network of Long Short-Term Memory (LSTM) – a popular deep learning model, for time-series forecasting of global temperature changes. This is coupled with a comprehensive analysis of the model’s predictions based on multiple data mining techniques. Comparison of the LSTM’s performance with the rolling mean baseline model shows that based on temperature changes, 19% of areas across the planet suffer from extreme climatic conditions. This affects the average temperature of the globe drastically. In future, we aim to refine our analysis based on larger datasets and generalize the outcomes of other factors playing a role in climate change beyond global temperature changes.