VIRTUAL Using Deep Learning to Estimate Greenhouse Gas Emissions Via Satellite Imagery

Overview

Greenhouse gasses (GHG) emitted from fossil-fuel-burning power plants pose a global threat to public health, causing 8.7 million deaths per year through air pollution and raising the frequency of natural disasters. Quantifying GHG emissions is crucial for the success of future climate action. However, current methods to track emissions cost upwards of $520,000 per power plant. These methods are cost prohibitive for developing countries, and are not globally standardized, leading to inaccurate estimations in emissions reports from nations and companies. I developed a novel, low-cost solution via a end-to-end deep learning pipeline that utilizes observations of emitted smoke plumes in satellite imagery to provide an accurate, precise system for quantifying GHG emissions at an global scale by 1) segmentation of power plant smoke plumes 2) classification of the type of fired fuel 3) algorithmic prediction of power plant CO2 emissions. The pipeline was able to achieve a segmentation Intersection Over Union (IoU) score of 0.924, fuel classification accuracy of 96%, and quantify power generation and CO2 emission rates with a R-Squared (R2) value of .91 and a Mass Absolute Error (MAE) within 6.3%, indicating high performance across global regions. The results of this work are significant because they enable the identification of major sources of GHG emissions and their temporal monitoring on a global scale at a low-cost. This enables the development of more-effective climate policy and transparency regarding compliance with the Paris Climate Agreement and COP26 goals, revolutionizing the way we tackle climate change.

Competition history

• JSHS 2022 Category not listed

Resources

• National JSHS abstract book ↗

Related projects

Closest projects by meaning, across every fair and year in the corpus.