Repository logo
 

Super-resolution generative adversarial network for weather radar applications

Simple item page
dc.contributor.authorLesher-Garcia, Jacob T., author
dc.contributor.authorChandrasekar, V., advisor
dc.contributor.authorCheney, Margaret, committee member
dc.contributor.authorPopat, Ketul, committee member
dc.date.accessioned2024-01-01T11:24:13Z
dc.date.available2025-12-29
dc.date.issued2023
dc.description.abstractWeather radars are vital to ensuring the safety of society by providing timely, accurate products used to forecast the development of weather phenomena. To this end, high spatiotemporal resolution data is paramount. Collecting high-resolution polarimetric observation data necessitates scan strategies with a slow scan rate. This thesis proposes the use of an established deep learning model in order to augment the current weather radar operational paradigm. Specifically, this thesis focuses on evaluating the efficacy of the super-resolution generative adversarial network (SRGAN) in generating physically realistic, pseudo-high-resolution radar scans – referred to as super-resolution (SR) scans – from low-resolution (LR) weather radar scans. With this, weather radar systems would be able to collect LR scans at faster scan rates while maintaining the quality of high-resolution (HR) scans by using the SRGAN to generate SR scans. This thesis aims to assess the generating capabilities of the SRGAN within the scope of generating SR scans from a pseudo-LR scan, processed from an actual HR scan. In order to accomplish this task, multiple experiments are setup, designed to test the SRGAN's capabilities in conducting SR for different architectural configurations, scan types, resolution scaling factors and downsampling methods, one of which simulates the characteristics of actual LR weather radar scans. The experimental SRGANs' performances are assessed both quantitatively and qualitatively, comparing between the SR scans and the baseline interpolation methods. The results of this thesis have found that the SRGAN model can outperform the baseline methods, specifically for the higher resolution scale factors and especially for the RHI radar scan type. Furthermore, the SRGAN is able to generate a physically representative SR image that reflects the natural features of a HR image. This is significant as it suggests that the SRGAN model is more effective when applied to practical applications.
dc.format.mediumborn digital
dc.format.mediummasters theses
dc.identifierLesherGarcia_colostate_0053N_18084.pdf
dc.identifier.urihttps://hdl.handle.net/10217/237354
dc.languageEnglish
dc.language.isoeng
dc.publisherColorado State University. Libraries
dc.relation.ispartof2020-
dc.rightsCopyright and other restrictions may apply. User is responsible for compliance with all applicable laws. For information about copyright law, please see https://libguides.colostate.edu/copyright.
dc.rights.accessEmbargo expires: 12/29/2025.
dc.subjectdeep learning
dc.subjectremote sensing
dc.subjectweather radar
dc.subjectgenerative adversarial network
dc.subjectdata processing
dc.subjectsuper-resolution
dc.titleSuper-resolution generative adversarial network for weather radar applications
dc.typeText
dcterms.embargo.expires2025-12-29
dcterms.embargo.terms2025-12-29
dcterms.rights.dplaThis Item is protected by copyright and/or related rights (https://rightsstatements.org/vocab/InC/1.0/). You are free to use this Item in any way that is permitted by the copyright and related rights legislation that applies to your use. For other uses you need to obtain permission from the rights-holder(s).
thesis.degree.disciplineElectrical and Computer Engineering
thesis.degree.grantorColorado State University
thesis.degree.levelMasters
thesis.degree.nameMaster of Science (M.S.)

Files

Original bundle
Now showing 1 - 1 of 1
No Thumbnail Available
Name:
LesherGarcia_colostate_0053N_18084.pdf
Size:
11.83 MB
Format:
Adobe Portable Document Format
Download

Collections

2020-
Theses and Dissertations