Scatterometry and radiometry are used to obtain measurements of Earth properties with extensive spatial coverage at daily or near-daily temporal resolution. Their measurements are used in many climate studies and weather applications, such as iceberg tracking, ocean wind estimation, and volumetric soil moisture measurements. The spatial resolution of these data products ranges from a few kilometers to tens of kilometers. Techniques to enhance the spatial resolution of these products help reveal finer scale features, but come at the cost of increased noise. This thesis explores the application of neural networks as a possible method to handle the noise and uncertainty in enhanced resolution scatterometer and radiometer data products. The specific sensors discussed are the Advanced Scatterometer (ASCAT) and its Ultrahigh Resolution (UHR) winds, and the Soil Moisture Active Passive (SMAP) radiometer and its soil moisture measurements. ASCAT UHR winds have already been validated in previous studies [1], but inherent ambiguity in the wind retrieval model couples with higher noise levels to decrease overall accuracy. Neural networks are tested as an alternate modeling method to possibly improve the accuracy compared with the current method. It is found that the feed forward neural networks tested are able to accurately estimate winds in most calculations, but struggle with the same ambiguity that occurs in the current model. The neural networks handle this ambiguity inconsistently, which results in worse overall network performance compared to the current wind retrieval method. For the SMAP soil moisture measurements, the radiometer form of the Scatterometer Image Reconstruction algorithm is validated as a method to enhance resolution. While the increased noise at higher resolution does worsen overall accuracy, the performance remains within about 0.04 cm^3 cm^−3 RMSE of a validated soil moisture product, suggesting that fine scale features revealed as resolution is enhanced are accurate. Corrections to the soil moisture extraction model used in these tests could further improve these results. Neural networks are then applied and compared with the theory-based approach to extract soil moisture from the brightness temperature measurements, and are found to give slightly more accurate results than the theoretical model, though with somewhat higher error variance.
| Identifer | oai:union.ndltd.org:BGMYU2/oai:scholarsarchive.byu.edu:etd-10122 |
| Date | 05 December 2019 |
| Creators | Brown, Jordan Paul |
| Publisher | BYU ScholarsArchive |
| Source Sets | Brigham Young University |
| Detected Language | English |
| Type | text |
| Format | application/pdf |
| Source | Theses and Dissertations |
| Rights | https://lib.byu.edu/about/copyright/ |
Page generated in 0.002 seconds