This study examines hydrological trend estimations derived from GRACE (Gravity Recovery and Climate Experiment) mascon solutions at the GSFC (Goddard Space Flight Center) in terms of the Equivalent Water Thicknesses (EWT) for the world's major river basins. The estimation of hydrological trends in mascon time series involves two steps: deterministic modeling and stochastic modeling. A deterministic model is characterized as a harmonic regression that incorporates trend and seasonal signals. For stochastic modeling, the EWT observables are typically considered to be equally weighted and uncorrelated, hence the term "white noise-only model". This interpretation, however, is misleading because typical discrete geodetic time series have temporal correlations, which generate colored noise. To explain the role of temporal correlations, we employ two different methodologies. Applying the least squares variance component estimation (LS-VCE) to time series for a combination of colored noise (i.e., flicker noise) and white noise is one of them. The second methodology uses autoregressive noise models to model the time series. Finally, the white noise-only model is compared to stochastic modeling incorporating temporal correlation. The findings show that a white noise-only model in a GRACE time series underestimates the uncertainty of the hydrological trend.

How to cite: Gunes, O. and Aydin, C.: Role of temporal correlations in the uncertainties of the GRACE hydrological trend, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-9408, https://doi.org/10.5194/egusphere-egu23-9408, 2023.

GIA from forward models suffer from large uncertainties due to approximations and assumptions on the Earth rheology and ice load history. These uncertainties propagate to uncertainties in ice-sheet mass balance and sea level budget studies. Therefore, GIA estimates from contemporary geodetic datasets are gaining interest. The challenges of obtaining data-driven GIA include solving a geophysical inversion that does not have a unique solution and often requires a-prior information and several approximations. In this work, a novel geophysical framework is developed that uses GPS and GRACE data to estimate GIA signal. The method relies on geophysical relations between geopotential and vertical land movement (VLM) caused by GIA and present-day mass changes. For example, the elastic response of solid Earth to the positive surface mass load results in a negative VLM, while a positive GIA mass leads to a positive VLM. We use these relations to express GPS observed VLM and GRACE observed gravity field anomalies in terms of GIA and present day mass change. The method is first shown to work in a closed-loop synthetic experiment and then applied to the NGL provided GPS velocities and GRACE spherical harmonic coefficients provided by ITG Graz. Our GIA estimates differ significantly from commonly used GIA models (such as ICE-6G) over Alaska and central Greenland. Our GIA rates over Alaska are around 5 mm/yr, which matches with several regional studies over Alaska. Similarly, there is a lot of ambiguity over Greenland ice-load history and our results may provide informative input to the ongoing debate. Our estimates are data-driven and are therefore able to pick them up. We also discuss the uncertainties, caveats and limitations of our method and its implications. The published GIA product is made openly available at one degree grid resolution. 

How to cite: Vishwakarma, B. D., Ziegler, Y., Royston, S., and Bamber, J. L.: A data-driven framework for estimating GIA from GPS and GRACE data, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-11697, https://doi.org/10.5194/egusphere-egu23-11697, 2023.

Recent observations from Global Navigation Satellite Systems (GNSS) displacement time-series have demonstrated that the motion of tectonic plates is ubiquitously not steady-state. GNSS displacement time-series can be described as the sum of expected motions such as long term tectonic interseismic motion or annually and semi-annually seasonal oscillations, and unexpected “transients”, such as slow-slip events or volcanic deformations. Although the number and quality of globally available GNSS time-series have increased dramatically in recent years, detecting and modeling these signals remains challenging, because of their elusive nature, masked by the presence of noise. The most popular approach for filtering such noise is the application of common-mode-filters (CMFs) that use weighted averages of residuals after fitting individual time-series with trajectory models. CMFs exploit the spatial coherence of higher frequency noise in both Precise Point Positioning and network double difference solutions to systematically reduce the noise of each time-series. However, the application of CMFs is limited in the case of sparsely distributed GNSS stations. Additionally, CMFs can potentially map local transients into noise when the original trajectory model fits are made suboptimally.

Here we propose an alternative method to CMFs by exploiting Deep Learning (DL) techniques. Our supervised learning regression method aims to remove the noise present in each GNSS time-series regardless of its geographical location on a station-by-station basis. Our dataset consists of nine thousand time-series: all GNSS time-series available on the Nevada GNSS repository with at least 4 years of contiguous data. Our approach is defined by two subroutines. Firstly, the Greedy Automatic Signal Decomposition (GrAtSiD) algorithm is used to fit each time-series and leave behind a residual. GrAtSiD is a sequential greedy linear algorithm that decomposes the time-series into a minimum number of transient basis functions and some permanent functions. The residual time-series after signal decomposition is defined as noise, without qualifying its nature. Once this noise is identified, a DL model is trained to recognize this noise from the raw time-series, without the need for any trajectory modeling. The supervised learning regression model predicts what the residual to a trajectory model would be. Although GrAtSiD is very effective in isolating the high frequency noise of GNSS time-series, its fitting is dependent on the temporal length of input time-series. Additionally, GrAtSiD needs to set arbitrary thresholds which control the convergence of the inversion routine to avoid under- or over-fitting input data. In this context, our DL approach proposes a generalization of GrAtSiD solutions, by exploiting a weakly supervised training based on millions of examples - essentially, the model generalizes an optimal fit having seen many examples of GrAtSiD trajectory fits. This generalization allows the DL model to preserve apparent transient features of the time-series. A multitude of DL architectures are tested in both sequence-to-single and sequence-to-sequence regression framings. This exploration allows us to identify the best framing, architecture, and related hyperparameters for our method to be successful. With the best performing models, we demonstrate the effects of the DL high frequency noise removal and compare it to the CMF approach. 

How to cite: Mastella, G., Bedford, J., Corbi, F., and Funiciello, F.: A Deep-Neural-Network-Based Denoising Method For GNSS Displacement Time-Series, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-5560, https://doi.org/10.5194/egusphere-egu23-5560, 2023.

The EUREF Permanent GNSS Network Central Bureau (EPN CB, www.epncb.eu ^[1]) monitors the quality of the daily GNSS observations of the EPN stations covering the period 1996-today. The associated data quality indicators (the number of observed versus expected observations in dual frequency, the lowest elevation cut-off observed, the number of missing epochs, the number of satellites, the number of maximum observations, and the number of cycle slips) are used to assess EPN stations' performance and as input for outlier detection in the daily position time series of the 400+ GNSS reference stations in Europe. Due to the increasing number of GNSS stations, the development of an automated algorithm to identify coordinate outliers caused by degraded GNSS data quality would allow to reduce the effort of human interpretation of the data quality indicators. We investigate the correlations between daily GNSS data quality metrics and daily position estimates to achieve this.

This study assesses several machine-learning classification algorithms to find a suitable data-driven model based on the correlation between degraded GNSS data quality metrics and quality degradation in position time series. Based on this investigation, the Random Forest algorithm proved to be the most precise algorithm, with an Area Under the Receiver Operating Characteristic Curve (AUC) equal to 0.95, and a correct classification of almost 90% of the test dataset. The GNSS data quality indicators ‘maximum observations‘ and ‘cycle slips’ are also shown as the most important parameters driving this model, which is in accordance with the data quality criteria for GNSS stations that IGS has determined ^[2]. Our model will be implemented in our EPN CB operation routine to develop a new monitoring system that can detect quality degradation on the GNSS station that will enable to improve the reliability of the EPN reference frame product by detecting position outliers due to degraded GNSS data quality. Here, we will present the current development of this automated algorithm, the challenges we faced, and the preliminary results of this work.

References:

^[1] C. Bruyninx, J. Legrand , D. Mesmaker, A. Moyaert, A. Fabian (2022): EUREF Permanent Network Central Bureau (EPN CB) Information System, https://doi.org/10.24414/ROB-EUREF-EPNCB

^[2] IGS (2019) Questions about the data quality graphs. https://kb.igs.org/hc/en-us/articles/204229743-Questions-about-the-data-quality-graphs

How to cite: Bamahry, F., Legrand, J., Bruyninx, C., Pottiaux, E., and Fabian, A.: Correlation Analysis of GNSS Data Quality Indicators and Position Time Series using Machine-Learning Algorithms, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-14585, https://doi.org/10.5194/egusphere-egu23-14585, 2023.

Radio signals sent by Global Navigation Satellite System (GNSS) satellites are received by GNSS stations on Earth. The signals get delayed as they propagate through the troposphere, this delay can be measured and thus, tropospheric properties can be estimated.

The total tropospheric delay in zenith direction is split into a zenith hydrostatic delay (ZHD) and a zenith wet delay (ZWD). While the ZHD can be modelled analytically with high accuracy, the ZWD is more difficult to model and is therefore typically estimated empirically. Estimating ZWD with high accuracy is important because it is one of the major error sources for GNSS positioning. Furthermore, the ZWD is highly correlated to the water vapour content along the signal path and thus, interesting for GNSS meteorology. Therefore, many studies have investigated new methods to improve state-of-the-art ZWD models. Recently, also machine learning (ML) approaches have been used to create tropospheric delay models. In addition to modelling ZWD, forecasting of ZWD is of great importance. Due to the relation of ZWD to water vapour, accurate ZWD forecasts would be essential for weather forecasting.

The aim of this work is to develop a global ML-based model capable of forecasting ZWD for the next 24 hours at any point on Earth. It is trained on ZWDs, provided by the Nevada Geodetic Laboratory,  from over 10'000 GNSS stations and evaluated on ZWDs of 2700 test stations. The model utilizes the geographical location of the GNSS station and meteorological data from the ERA5 data set as its input features. To make hourly ZWD forecasts for the next 24 hours, forecasts of the meteorological data are taken from the Integrated Forecasting System of the European Centre for Medium-Range Weather Forecasts (ECMWF). Preliminary results using Extreme Gradient Boosting (XGBoost) show an average root mean squared error of around 1.5 cm over all testing stations for a forecasting horizon of 24 hours per day.

How to cite: Crocetti, L., Schartner, M., Schindler, K., and Soja, B.: Forecasting of tropospheric parameters using meteorological data and machine learning, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-3453, https://doi.org/10.5194/egusphere-egu23-3453, 2023.

Atmospheric weighted mean temperature, T_m, is an important parameter in the Earth’s atmospheric water vapor sounding with the Global Navigation Satellite System (GNSS) technique. In this study, considering spatial distribution, time-varying characteristics, and the correlation with surface meteorological variables, T_m modeling is realized based on the random forest (RF) machine learning and global atmospheric profiles from radiosonde (RS) data and GPS radio occultations (RO) measurements. Comparisons of modeled results and numerical integrations of atmospheric profiles in 2020 show that the RF-based T_m model with surface meteorological parameters generally obtains a good accuracy with overall RMS errors of 2.8 K in comparison with RS data and 2.6 K in contrast to GPS RO data.

How to cite: Li, Q., Böhm, J., Yuan, L., and Weber, R.: Modeling of the weighted mean temperature based on the random forest machine learning approach, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-17204, https://doi.org/10.5194/egusphere-egu23-17204, 2023.

High-precision global ionospheric modeling is important for radio communication, navigation, or studies on space weather. Traditional spatial ionospheric modeling approaches include spherical harmonics and trigonometric B-splines. The Ionospheric Associated Analysis Centers (IAAC) of the International GNSS Service (IGS) use these methods to model vertical total electron content (VTEC) globally, and generate Global Ionospheric Maps (GIMs). Due to the limitations of spatial modeling approaches, conventional GIMs cannot comprehensively describe the spatial feature of the ionosphere. With the capability of capturing complex and non-linear relationships of diverse data, machine learning (ML) has been increasingly applied to ionospheric modeling. Currently, most of the existing ML-related studies focused on temporal prediction of ionospheric states and rarely considered the aspect of the spatial modeling of VTEC. Although some studies predicted global ionosphere maps using machine learning, they used conventional GIMs as inputs, implying that the precision of the ML-based spatial modeling could be limited by traditional methods and quantity of input GNSS observations utilized to generate GIMs.

The goal of this study is the spatial interpolation of VTEC using ML methods for the generation of ML-based GIMs. We first determine VTEC using carrier-to-code levelling through Kalman filter and based on geometry-free multi-GNSS observations from GNSS stations of the IGS network. The derived satellite-specific VTEC time series are then used to train the ML models. Several algorithms, such as extreme gradient boosting and random forest, are applied and their performance is evaluated. Moreover, VTEC from satellite altimetry is used as an additional means to assess the quality of the generated ML models. Finally, we compare the acquired ML-based GIMs with conventional GIMs to investigate the advantage of using the proposed approach for global VTEC modeling.

How to cite: Mao, S., Kłopotek, G., Awadaljeed, M., and Soja, B.: Machine learning for global modeling of the ionosphere based on multi-GNSS data, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-9260, https://doi.org/10.5194/egusphere-egu23-9260, 2023.

The International GNSS Service (IGS) global ionospheric maps (GIMs) are one of the primary sources of information on the ionospheric state. They are used in many research and GNSS positioning applications. IGS GIMs are created using the weighted average of the products derived from the selected IAAC. This method allows for efficient mapping of the state of the ionosphere, especially on days without major disruptions. However, ionospheric disturbances could be more problematic to map correctly. To improve GIMs quality, we used a machine learning (ML) approach to combine individual IAAC GIMs into one product. We used total electron content (TEC) data from Jason altimetric satellite with a 5-minute interval as reference. To improve the modeling, we used auxiliary parameters such as solar and geomagnetic indices, e.g., F10.7 index. The training process was performed on the 2005-2020 dataset. 

This study presents some preliminary results of VTEC modeling using the ML approach. We show inter-validation and inter-comparison with IGS GIMs, and Jason-derived VTEC. We also used pseudorange code and carrier phase single-frequency GNSS observations to show positioning accuracy improvement achieved using ML-based GIMs. For this purpose, we used 34 evenly distributed IGS stations for the selected calm period and strong geomagnetic storms. The results showed that for both calm and stormy days, the differences between the coordinates obtained from our model and those using the IGS product were up to a few centimeters for most stations for the northern and eastern components of the topocentric coordinates. Additionally, for altitude, we noticed accuracy improvement for most stations during the storm periods relative to results obtained using the final IGS product.  

How to cite: Poniatowski, M., Nykiel, G., and Szmytkowski, J.: Application of machine learning to combine global ionospheric maps from IGS analysis centers, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-11443, https://doi.org/10.5194/egusphere-egu23-11443, 2023.

Airborne laser scanning (ALS) point clouds are commonly acquired to describe a 3D shape of terrain and attributes of objects and landforms located on it. They proved to be useful in a variety of applications, including geometrical characterization of both man-made and natural objects and landforms. Usually, the first step of point cloud processing is classification. As a result, its accuracy highly influences the results of subsequent processing. Therefore, reliable and automatic point cloud classification is of key importance in most of the ALS data applications.

Recently, deep learning techniques attracted the attention of the community in the context of point cloud classification. However, the reproducibility of the trained deep learning networks remains unexplored because there is too little accessible and precisely classified 3D training data. Recently many countries have published their national ALS data sets. This initiative leads to promising options for deep learning classification of point clouds as it allows for comprehensive training of deep networks.

In this study we present the investigations that aimed at creation of a universal, deep learning based classifier that will be able to classify point clouds of varying characteristics. The experiments were carried out using selected parts of data sets that have been made available by three European countries: Poland, Austria and Switzerland. The point clouds were classified into four classes: ground and water, vegetation, buildings and bridges, and others. The results of the experiments showed that it is possible to achieve high overall accuracy of the classification for the ground and water (above 98%), vegetation (92-97%, depending on a test site), and building and bridges (92-96%, depending on a test site). A lower overall accuracy was achieved for class others because of a very high variability of geometry of objects that belong to this class. Furthermore, in some cases, adding training data from a different country to the initial training data resulted in improved classification accuracy in selected classes and reduced dataset-specific errors.

As a result, this study proves that it is possible to create a universal, deep learning based classifier that will be able to maintain high classification accuracy while it processes data sets of different characteristics.

How to cite: Walicka, A. and Pfeifer, N.: Deep learning based classification of multinational airborne laser scanning data, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-8538, https://doi.org/10.5194/egusphere-egu23-8538, 2023.