Classification of aerosols in CALIOP data using unsupervised semantic segmentation
(2024) FYSK04 20241Department of Physics
Combustion Physics
- Abstract
- This thesis investigates the feasibility and efficacy of employing unsupervised semantic segmentation for classifying features in CALIOP data, aiming to address significant bias inherent in current classification methods. By exploring various preprocessing techniques, dimensionality reduction methods, and classification algorithms, the study evaluates the potential of semantic segmentation in improving the accuracy of aerosol classification. Despite computational limitations imposed by working on a standard laptop, the research produces promising results, demonstrating the capability of certain model configurations to identify important features and maintain continuity in segmented images. The implications of mitigating bias in CALIOP data... (More)
- This thesis investigates the feasibility and efficacy of employing unsupervised semantic segmentation for classifying features in CALIOP data, aiming to address significant bias inherent in current classification methods. By exploring various preprocessing techniques, dimensionality reduction methods, and classification algorithms, the study evaluates the potential of semantic segmentation in improving the accuracy of aerosol classification. Despite computational limitations imposed by working on a standard laptop, the research produces promising results, demonstrating the capability of certain model configurations to identify important features and maintain continuity in segmented images. The implications of mitigating bias in CALIOP data are profound, with potential improvements in understanding aerosol radiative forcing and enhancing climate model predictions. While this study represents a significant step towards replacing current classification methods like SIBYL, further research is warranted to explore and refine the approaches introduced here. (Less)
- Popular Abstract
- Imagine peering into the skies and seeing not just clouds but a complex tapestry of aerosols - tiny particles that play a pivotal role in our climate. In our quest to understand this intricate puzzle, researchers have turned to a new approach: unsupervised semantic segmentation.
Our journey begins with the challenge of distinguishing between smoke layers and ice clouds in satellite data. Traditional methods often fall short due to inherent biases. However, unsupervised semantic segmentation offers a fresh perspective, promising to overcome these limitations.
Through meticulous experimentation, we explore various techniques to process the data, reduce its complexity, and classify aerosols. Despite facing computational constraints, we... (More) - Imagine peering into the skies and seeing not just clouds but a complex tapestry of aerosols - tiny particles that play a pivotal role in our climate. In our quest to understand this intricate puzzle, researchers have turned to a new approach: unsupervised semantic segmentation.
Our journey begins with the challenge of distinguishing between smoke layers and ice clouds in satellite data. Traditional methods often fall short due to inherent biases. However, unsupervised semantic segmentation offers a fresh perspective, promising to overcome these limitations.
Through meticulous experimentation, we explore various techniques to process the data, reduce its complexity, and classify aerosols. Despite facing computational constraints, we discover promising avenues for improvement, showcasing the potential of certain model configurations to enhance accuracy while maintaining image coherence.
But beyond the technicalities lies a deeper significance. By refining aerosol classification, we gain valuable insights into their behavior and its impact on climate dynamics. As we unravel the complexities of smoke and ice distribution, we contribute to more reliable climate projections and informed decision-making in climate policy.
Looking ahead, our study points to the need for further refinement and exploration. Optimizing model performance, investigating advanced segmentation techniques, and securing additional computational resources are crucial steps towards fully realizing the potential of unsupervised semantic segmentation in climate research.
In conclusion, our research represents a step forward in revolutionizing aerosol classification methodologies. By harnessing the power of unsupervised semantic segmentation, we deepen our understanding of atmospheric dynamics and pave the way for more accurate climate predictions - a journey with profound implications for our planet’s future. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/9169374
- author
- Rosager, Sixten LU
- supervisor
- organization
- course
- FYSK04 20241
- year
- 2024
- type
- M2 - Bachelor Degree
- subject
- language
- English
- id
- 9169374
- date added to LUP
- 2024-07-02 08:03:53
- date last changed
- 2024-07-02 08:03:53
@misc{9169374,
abstract = {{This thesis investigates the feasibility and efficacy of employing unsupervised semantic segmentation for classifying features in CALIOP data, aiming to address significant bias inherent in current classification methods. By exploring various preprocessing techniques, dimensionality reduction methods, and classification algorithms, the study evaluates the potential of semantic segmentation in improving the accuracy of aerosol classification. Despite computational limitations imposed by working on a standard laptop, the research produces promising results, demonstrating the capability of certain model configurations to identify important features and maintain continuity in segmented images. The implications of mitigating bias in CALIOP data are profound, with potential improvements in understanding aerosol radiative forcing and enhancing climate model predictions. While this study represents a significant step towards replacing current classification methods like SIBYL, further research is warranted to explore and refine the approaches introduced here.}},
author = {{Rosager, Sixten}},
language = {{eng}},
note = {{Student Paper}},
title = {{Classification of aerosols in CALIOP data using unsupervised semantic segmentation}},
year = {{2024}},
}