Probabilistic fusion of Ku and C band scatterometer data for determining the freeze/thaw state / von Simon Zwieback
VerfasserZwieback, Simon
Begutachter / BegutachterinWagner, Wolfgang ; Bartsch, Annett ; Melzer, Thomas
Umfangx, 87 S. : zahlr. graph. Darst., Kt.
HochschulschriftWien, Techn. Univ., Mag.-Arb., 2011
Zsfassung in dt. Sprache
Schlagwörter (DE)Fernerkundung / Scatterometer / Frier/Tau Zustand / Hidden Markov Model / Graphisches Modell / Expectation Maximization Algorithmus / Rückstreumodellierung
Schlagwörter (EN)Remote Sensing / Scatterometer / Freeze/Thaw State / Hidden Markov model / Graphical Model / Expectation Maximization algorithm / Backscatter modelling
URNurn:nbn:at:at-ubtuw:1-41872 Persistent Identifier (URN)
 Das Werk ist frei verfügbar
Probabilistic fusion of Ku and C band scatterometer data for determining the freeze/thaw state [1.55 mb]
Zusammenfassung (Englisch)

The transition of the landscape from frozen to non-frozen conditions has far-reaching consequences on numerous geo- and biophysical processes such as plant growth and the hydrologic cycle.

Microwave remote sensing has been shown to be an apt tool for monitoring the landscape freeze/thaw state. As the measured signal sigma0 is sensitive to different factors at different radar frequencies, the combination of distinct data sources can potentially lead to improved results.

In light of this observation, a novel sensor fusion algorithm is proposed -- it estimates the F/T state based on scatterometer data: SeaWinds on QuikScat in Ku band and ASCAT on MetOp in C band. In addition, a widely used backscatter model for snow packs is extended, whose purpose is twofold: firstly, it can give insight into the dependence of sigma0 on various factors and secondly, it facilitates the parameterization of the aforementioned sensor fusion model.

The sensor fusion approach is based on a probabilistic model, an adaptation of the well-known Hidden Markov model (HMM). The F/T state is assumed to be a Markov chain, whose value is not directly observable. At each epoch, however, its current state influences the measurements: sigma0 at both frequency bands. The simple structure assures that inference can be done efficiently, e.g. the calculation of the probability of the state on a given day. The algorithm does not use training data; the parameters are estimated for each time series in an unsupervised fashion. This is achieved by maximizing the marginal likelihood in the framework of the Expectation Maximization algorithm.

The algorithm is analyzed and tested in a study area in Russia and northern China, which encompasses the region of 120 - 130 E and 50 - 75 N. The time series of the probability of the state are validated with in-situ snow and temperature data as well as global climate models. In general, the accuracy exceeds 90%, but the algorithm can fail in agriculturally used land (fields, pastures) and bare rock outcrops in mountainous regions. On a more qualitative level, the study affirms the importance of using two distinct frequencies, as particularly dry snow, vegetation and the freezing of the soil water manifest themselves differently at Ku and C band.