Tomographic reconstruction of the ionospheric electron density as a function of space and time

Electron density distribution is the major determining parameter of the ionosphere. Computerized Ionospheric Tomography (CIT) is a method to reconstruct ionospheric electron density image by computing Total Electron Content (TEC) values from the recorded Global Positioning Satellite System (GPS) signals. Due to the multi-scale variability of the ionosphere and inherent biases and errors in the computation of TEC, CIT constitutes an underdetermined ill-posed inverse problem. In this study, a novel Singular Value Decomposition (SVD) based CIT reconstruction technique is proposed for the imaging of electron density in both space (latitude, longitude, altitude) and time. The underlying model is obtained from International Reference Ionosphere (IRI) and the necessary measurements are obtained from earth based and satellite based GPS recordings. Based on the IRI-2007 model, a basis is formed by SVD for the required location and the time of interest. Selecting the first few basis vectors corresponding to the most significant singular values, the 3-D CIT is formulated as a weighted least squares estimation problem of the basis coefficients. By providing significant regularization to the tomographic inversion problem with limited projections, the proposed technique provides robust and reliable 3-D reconstructions of ionospheric electron density.     

An ensemble deep learning based shoreline segmentation approach (WaterNet) from Landsat 8 OLI images

Shorelines constantly vary due to natural, urbanization and anthropogenic effects such as global warming, population growth, and environmental pollution. Sustainable monitoring of coastal changes is vital in terms of coastal resource management, environmental preservation and planning. Publicly available Landsat 8 OLI (Operational Land Manager) images provide accurate, reliable, temporal and up-to-date information about coastal areas. Recently, the use of machine learning and deep learning algorithms have become widespread. In this study, we used our public Landsat 8 OLI satellite image dataset to create a majority voting method which is an ensemble automatic shoreline segmentation system (WaterNet) to obtain shorelines automatically. For this purpose, different deep learning architectures have been utilized namely as Standard U-Net, Dilated U-Net, Fractal U-Net, FC-DenseNet, and Pix2Pix. Also, we have suggested a novel framework to create labeling data from OpenStreetMap service to create a unique dataset called YTU-WaterNet. According to the results, IoU and F1 scores have been calculated as 99.59% and 99.79% for the WaterNet. The results indicate that the WaterNet method outperforms other methods in terms of shoreline extraction from Landsat 8 OLI satellite images.     

Non-parametric regional VTEC modeling with Multivariate Adaptive Regression B-Splines

In this work Multivariate Adaptive Regression B-Splines (BMARS) is applied to regional spatio-temporal mapping of the Vertical Total Electron Content (VTEC) using ground based Global Positioning System (GPS) observations. BMARS is a non-parametric regression technique that utilizes compactly supported tensor product B-splines as basis functions, which are automatically obtained from the observations. The algorithm uses a scale-by-scale model building strategy that searches for B-splines at each scale fitting adequately to the data and provides smoother approximations than the original Multivariate Adaptive Regression Splines (MARS). It is capable to process high dimensional problems with large amounts of data and can easily be parallelized. The real test data is collected from 32 ground based GPS stations located in North America. The results are compared numerically and visually with both the regional VTEC modeling generated via original MARS using piecewise-linear basis functions and another regional VTEC modeling based on B-splines.     

Regional VTEC modeling with multivariate adaptive regression splines

Different algorithms have been proposed for the modeling of the ionosphere. The most frequently used method is based on the spherical harmonic functions achieving successful results for global modeling but not for the local and regional applications due to the bounded spherical harmonic representation. Irregular data distribution and data gaps cause also difficulties in the global modeling of the ionosphere. In this paper we propose an efficient algorithm with Multivariate Adaptive Regression Splines (MARS) to represent a new non-parametric approach for regional spatio-temporal mapping of the ionospheric electron density using ground-based GPS observations. MARS can handle very large data sets of observations and is an adaptive and flexible method, which can be applied to both linear and non-linear problems. The basis functions are directly obtained from the observations and have space partitioning property resulting in an adaptive model. This property helps to avoid numerical problems and computational inefficiency caused by the number of coefficients, which has to be increased to detect the local variations of the ionosphere. Since the fitting procedure is additive it does not require gridding and is able to process large amounts of data with large gaps. Additionally the model complexity can be controlled by the user via limiting the maximal number of coefficients and the order of products of the basis functions. In this study the MARS algorithm is applied to real data sets over Europe for regional ionosphere modeling. The results are compared with the results of Bernese GPS Software over the same region.