Estimating soil freezing from passive L-band microwave measurements

• Main Author: Parkkinen, Tiina
• Other Authors: Matematiikan laitos - Department of Mathematics, Luonnontieteiden tiedekunta - Faculty of Natural Sciences, Tampere University of Technology, Ali-Löytty, Simo
• Format: Master Thesis
• Language: English
• Published: 2014
• Subjects: Teknis-luonnontieteellinen koulutusohjelma; Sodankylä
• Online Access: https://trepo.tuni.fi//handle/123456789/22600

This master of science thesis presents two approaches to estimate soil freezing from passive L-band microwave measurements. The first is the Canny edge detector algorithm which is used to gain estimates on the dates of seasonal changes occurring in the brightness temperature data measured by the ELBARA-II radiometer and the SMOS satellite. Secondly, regression analysis is applied to the ELBARA-II brightness temperature data and frost tube observations to estimate the frost and thaw depths. The process of soil freezing and the effects of it are shortly introduced. More focus is on the developed and tested observation methods and instruments used at the ground level and by remote sensing methods. From the remote sensing methods the passive L-band ELBARA-II radiometer and the SMOS satellite's MIRAS instrument are discussed in more detail. The Canny edge detector algorithm and the criteria used when deriving it are described in detail. The focus in the multiple linear regression analysis is on the testing and validation of the results obtained. The Kalman filter algorithm used to filter diurnal variations from the brightness temperature data is introduced shortly. The Canny edge detector algorithm proved out to have a good performance when applied to the local ELBARA-II brightness temperature data. The accuracy was not as good for the SMOS brightness temperature data measured over Sodankylä, since the SMOS data has both lower temporal sampling and larger spatial resolution. Nevertheless, the test performed for the SMOS data measured over Finland in the fall of 2010 showed promising results and expected behavior. All the three approaches chosen for the regression analysis showed similar results, the most accurate being the model where Kalman filtered data was used, while the simplified linearized model and the model fitted using moving average filtered data followed.