Advanced methods for earth observation data synergy for geophysical parameter retrieval
The first part of the thesis focuses on the analysis of relevant factors to estimate the response time between satellite-based and in-situ soil moisture (SM) using a Dynamic Time Warping (DTW). DTW was applied to the SMOS L4 SM, and was compared to in-situ root-zone SM in the REMEDHUS network in Wes...
Main Author: | |
---|---|
Other Authors: | , , |
Format: | Doctoral or Postdoctoral Thesis |
Language: | English |
Published: |
Universitat Politècnica de Catalunya
2022
|
Subjects: | |
Online Access: | http://hdl.handle.net/2117/376085 http://hdl.handle.net/10803/675942 https://doi.org/10.5821/dissertation-2117-376085 |
Summary: | The first part of the thesis focuses on the analysis of relevant factors to estimate the response time between satellite-based and in-situ soil moisture (SM) using a Dynamic Time Warping (DTW). DTW was applied to the SMOS L4 SM, and was compared to in-situ root-zone SM in the REMEDHUS network in Western Spain. The method was customized to control the evolution of time lag during wetting and drying conditions. Climate factors in combination with crop growing seasons were studied to reveal SM-related processes. The heterogeneity of land use was analyzed using high-resolution images of NDVI from Sentinel-2 to provide information about the level of spatial representativity of SMOS data to each in-situ station. The comparison of long-term precipitation records and potential evapotranspiration allowed estimation of SM seasons describing different SM conditions depending on climate and soil properties. The second part of the thesis focuses on data-driven methods for sea ice segmentation and parameter retrieval. A Bayesian framework is employed to segment sets of multi-source satellite data. The Bayesian unsupervised learning algorithm allows to investigate the ‘hidden link’ between multiple data. The statistical properties are accounted for by a Gaussian Mixture Model, and the spatial interactions are reflected using Hidden Markov Random Fields. The algorithm segments spatial data into a number of classes, which are represented as a latent field in physical space and as clusters in feature space. In a first application, a two-step probabilistic approach based on Expectation-Maximization and the Bayesian segmentation algorithm was used to segment SAR images to discriminate surface water from sea ice types. Information on surface roughness is contained in the radar backscattering images which can be - in principle - used to detect melt ponds and to estimate high-resolution sea ice concentration (SIC). In a second study, the algorithm was applied to multi-incidence angle TB data from the SMOS L1C product to harness the ... |
---|