Year of Award


Document Type


Degree Type

Master of Science (MS)

Degree Name

Computer Science

Department or School/College

Computer Science

Committee Chair

Jesse Johnson

Commitee Members

Doug Brinkerhoff, John S. Kimball


Machine Learning, Brightness Temperature, Super Resolution


University of Montana

Subject Categories

Data Science


Steady improvements to the instruments used in remote sensing has led to much higher resolution data, often contemporaneous with lower resolution instruments that continue to collect data. There is a clear opportunity to reconcile recent high resolution satellite data with the lower resolution data of the past. Super-resolution (SR) imaging is a technique that increases the spatial resolution of image data by training statistical methods on simultaneously occurring lower and higher resolution data sets. The special sensor microwave/imager (SSMI) and advanced microwave scanning radiometer (AMSR2) brightness temperature data products are well suited to super-resolution imaging, and SR can be used to standardize the higher resolution across the entire record of observations. Of the methods used in super-resolution imaging, neural networks have led to major improvements in the realm of computer vision and have seen great success in the super-resolution of photographic images. We trained two neural networks, based on the design of the Resnet, to super-resolution the 25 kilometer resolution SSMI and AMSR2 brightness temperature data products up to a 10 kilometer resolution. The mean error over all frequencies and polarizations for the AMSR and SSMI models’ predictions is 0.84% and 2.4% respectively for the years 2013 and 2019.

Included in

Data Science Commons



© Copyright 2021 Kellen A. Donahue