Created a fast-running AI model of the Earth’s climate that runs on a laptop or PC using historical climate data and public simulation outputs.
TEAM INFORMATION:
TEAM MEMBER 1:
Full name: Hashmitha Sugumar
Womanium Program Enrollment ID: WQ24-yA5BlnnHSFSfCLy
TEAM MEMBER 2:
Full name: Anthony Vijay M
Womanium Program Enrollment ID: WQ24-2dlGlzr4asYPlpN
PROBLEM STATEMENT:
The goal of this project is to develop a fast-running AI model of Earth's climate that can be executed on a standard laptop or PC. This model will utilize historical climate data and public simulation outputs to predict future climate conditions and understand past climate behavior.
BACKGROUND :
Climate models help us understand the complex interactions within Earth's climate system and predicting future climatic changes. These models are typically large-scale, computationally intensive, and require substantial resources to run, often necessitating supercomputers or dedicated clusters. Advances in ML and AI provide an opportunity to create more efficient models that can approximate the behavior of traditional climatic models with significantly reduced computational requirements. By leveraging the historical climate data and the existing simulation outputs, an AI-based climate model can be trained to emulate the results of more complex models, thereby offering a practical alternative for those with limited access to high-performance computing resources.
As we know, developing a fast-running AI climate model that can run on a laptop or PC democratizes access to climate modeling tools as it enables a broader range of users to engage themselves with the climate science and conduct their own analyses.This model can be used as a preliminary tool for researchers to quickly test hypotheses and explore the effects of different parameters before committing to more time-consuming and resource-intensive simulations on traditional climate models. Our team found it interesting because we thought that by making climate modeling more accessible and efficient, we would get a deeper understanding of climate dynamics and support efforts to address the global challenge of climate change.
In the very ancient times, people developed weather prediction methods based on careful observation of natural phenomena. The Babylonians, who lived in Mesopotamia from the 18th to the 6th century BCE, could track the movements of celestial bodies. They believed that the positions of stars and planets could help predict weather patterns. Several methods and technologies have been explored to create fast-running AI models of Earth's climate like the traditional statistical models, simplified physical models, and more recently, machine learning and AI-based approaches. Statistical models are generally simpler to implement and require less computational power compared to complex physical models. These models can run quickly on standard computers because they do not involve solving complex equations.However these models heavily depend on the quality and quantity of historical data. They might struggle to predict future conditions if the climate system behaves in a non-linear or unprecedented way. Simplified physical models, such as energy balance models (EBMs) and reduced-complexity climate models (RCMs), use simplified representations of the climate system's physical processes.These models retain a physical basis, making their outputs more interpretable and scientifically grounded. However the necessary simplifications can lead to inaccuracies, especially for regional climate predictions or extreme events.
This project involves the development of a regression AI model designed to predict climate conditions across various regions of the Earth based on the concentration of atmospheric gases. The model employs a Linear Regression algorithm trained on historical climate data, ensuring robust and accurate predictions. By integrating this algorithm with a Flask-based web application, the project achieves a high level of efficiency and cost-effectiveness, delivering rapid predictions. The accuracy of the model is notably high, providing reliable climate forecasts that can be utilized for various applications. This web-based solution demonstrates the potential of advanced machine learning techniques in addressing complex environmental challenges.
SOURCES