Search

Remote sensing data acquired from various sensors have been used for decades to monitor sea ice conditions over polar regions. Sea ice plays an essential role in the polar environment and climate. Furthermore, sea ice affects anthropogenic activities, including shipping and navigation, the oil and gas industry, fisheries, tourism, and the lifestyle of the indigenous population of the Arctic. With the continuous decline of sea ice in the Arctic the presence of human-based activities will grow. Therefore, reliable information about sea ice conditions is of primary interest to protect the Arctic and to ensure safe and effective commercial activities and polar navigation. Currently, sea ice services produce operational ice charts manually using the knowledge of sea ice experts. However, with an increasing number of various data sources that provide different information regarding sea ice, it is important to develop automatic methods for sea ice characterization. Robust and automatic ice charting can not be achieved using only one satellite mission. It is fundamental to combine information from various remote sensors with different characteristics for more reliable sea ice monitoring and characterization. However, how do we know that all the information is actually relevant? It may be redundant, corrupted, or unnecessary for the given task, hence decreasing the performance of the algorithms from the required processing time and accuracy point of view. Therefore, it is crucial to select an optimal set of features that provides the relevant information content to enhance the efficiency and accuracy of the image interpretation and retrieval of geophysical parameters. The work in this dissertation specifically focuses on the development of such a method. In this thesis, we employ a fully automatic, flexible, accurate, efficient, and interpretable information selection method that is based on the graph Laplacians. The proposed approach assesses relevant information on a global and local level using two metrics simultaneously and selects relevant features for different regions of an image according to their physical characteristics and observation conditions. Moreover, it is linked in a common scheme with a classification algorithm that helps to properly evaluate the performance of the information selection and provides sea ice classification maps as an output. Accordingly, in recent studies, we investigate and evaluate the robustness and effectiveness of the proposed method for sea ice classification by testing several data combinations with various sea ice conditions. Experiments illustrate the flexibility and efficiency of the proposed scheme and clearly indicate an advantage of combining various sensors. Moreover, the results demonstrate the potential for operational sea ice monitoring that should be further thoroughly examined in future studies.

Remote sensing is the discipline that studies acquisition, preparation and analysis of spectral, spatial and temporal properties of objects without direct touch or contact. It is a field of great importance to understanding the climate system and its changes, as well as for conducting operations in the Arctic. A current challenge however is that most sensory equipment can only capture one or fewer of the characteristics needed to accurately describe ground objects through their temporal, spatial, spectral and radiometric resolution characteristics. This in turn motivates the fusing of complimentary modalities for potentially improved accuracy and stability in analysis but it also leads to problems when trying to merge heterogeneous data with different statistical, geometric and physical qualities. Another concern in the remote sensing of arctic regions is the scarcity of high quality labeled data but simultaneous abundance of unlabeled data as the gathering of labeled data can be both costly and time consuming. It could therefore be of great value to explore routes that can automate this process in ways that target both the situation regarding available data and the difficulties from fusing of heterogeneous multimodal data. To this end Semi-Supervised methods were considered for their ability to leverage smaller amounts of carefully labeled data in combination with more widely available unlabeled data in achieving greater classification performance. Strengths and limitations of three algorithms for real life applications are assessed through experiments on datasets from arctic and urban areas. The first two algorithms, Deep Semi-Supervised Label Propagation (LP) and MixMatch Holistic SSL (MixMatch), consider simultaneous processing of multimodal remote sensing data with additional extracted Gray Level Co-occurrence Matrix texture features for image classification. LP trains in alternating steps of supervised learning on potentially pseudolabeled data and steps of deciding new labels through node propagation while MixMatch mixes loss terms from several leading algorithms to gain their respective benefits. Another method, Graph Fusion Merriman Bence Osher (GMBO), explores processing of modalities in parallel by constructing a fused graph from complimentary input modalities and Ginzburg-Landau minimization on an approximated Graph Laplacian. Results imply that inclusion of extracted GLCM features could be beneficial for classification of multimodal remote sensing data, and that GMBO has merits for operational use in the Arctic given that certain data prerequisites are met.

Classic target decomposition methods use scattering space in their approaches. However, the goal for this project is to investigate whether a different approach to retrieve accurate and reliable estimates on the earth composition is possible when using the feature space with covariance matrix-based features. The approach consists of four steps. Generating multidimensional feature space data from sea ice scenes, extracting endmembers, finding the optimal number of endmembers in the scene and finding the contribution for the endmembers to each of the polarimetric feature pixels in the scene. In order to validate the performance of the approach several validation steps where conducted. Classification of the endmembers, calculating the average reconstruction error, classification of the scene and studding the abundance coefficients were some of these steps. Also, generation of synthetic data was conducted as an additional review of the approach. The system in this approach does not take in to account the variability of the polarimetric feature values in the different classes. It also assumes that the pixels are linearly mixed, something they probably not are. As a consequence, the approach is not able to retrieve accurate and reliable estimates on the earth composition for scenes consisting of sea ice. However, the approach gave good results on the synthetic datasets. Further work and investigation on the approach would include adapting the approach to consider the variability all sea ice data suffers from. Further, the methods considering linear mixing should then be replaced with methods considering nonlinear mixing.

This thesis investigates speckle properties from Sentinel-1 GRDM data to explore its dependency of incidence angle. The data is processed using the SNAP-tool such that a uniform region can be selected and estimation of speckle variance, skewness and kurtosis can be carried out. These speckle properties are calculated over ocean and sea ice for comparison and modeled to fit a regression line. Hypothesis testing is carried out to determine the statistical significance. Due to the GRDM product having different number of multi-looks depending on the swath, a normalization is done to fit the regression line. It is shown that there is a slight decrease in variance over incidence angle with statistical significance for both ocean and sea ice. The response is however low. It is further shown that the trend in texture measurements such as skewness and kurtosis have a lower statistical significance and may be deemed to have little to no variation over incidence angle.