On the Impact of GPS Multipath Correction Maps and Post-Fit Residuals on Slant Wet Delays for Tracking Severe Weather Events

in Atmosphere (2023), 14(219), 1-26

Climate change has increased the frequency and intensity of weather events with heavy precipitation, making communities worldwide more vulnerable to flash flooding. As a result, accurate fore- and ... [more ▼]

Climate change has increased the frequency and intensity of weather events with heavy precipitation, making communities worldwide more vulnerable to flash flooding. As a result, accurate fore- and nowcasting of impending excessive rainfall is crucial for warning and mitigating these hydro-meteorological hazards. The measurement of integrated water vapour along slant paths is made possible by ground-based global positioning system (GPS) receiver networks, delivering three-dimensional (3D) water vapour distributions at low cost and in real-time. As a result, these data are an invaluable supplementary source of knowledge for monitoring storm events and determining their paths. However, it is generally known that multipath effects at GPS stations have an influence on incoming signals, particularly at low elevations. Although estimates of zenith total delay and horizontal linear gradients make up the majority of the GPS products for meteorology to date, these products are not sufficient for understanding the full 3D distribution of water vapour above a station. Direct utilization of slant delays can address this lack of azimuthal information, although, at low elevations it is more prone to multipath (MP) errors. This study uses the convective storm event that happened on 27 July 2017 over Bulgaria, Greece, and Turkey, which caused flash floods and severe damage, to examine the effects of multipath-corrected slant wet delay (SWD) estimations on monitoring severe weather events. First, we reconstructed the one-way SWD by adding GPS post-fit phase residuals, describing the anisotropic component of the SWD. Because MP errors in the GPS phase observables can considerably impact SWD from individual satellites, we used an averaging technique to build station-specific MP correction maps by stacking the post-fit phase residuals acquired from a precise point positioning (PPP) processing strategy. The stacking was created by spatially organizing the residuals into congruent cells with an optimal resolution in terms of the elevation and azimuth at the local horizon. This enables approximately equal numbers of post-fit residuals to be distributed across each congruent cell. Finally, using these MP correction maps, the one-way SWD was improved for use in the weather event analysis. We found that the anisotropic component of the one-way SWD accounts for up to 20% of the overall SWD estimates. For a station that is strongly influenced by site-specific multipath error, the anisotropic component of SWD can reach up to 4.3 mm in equivalent precipitable water vapour. The result also showed that the spatio-temporal changes in the SWD as measured by GPS closely reflected the moisture field estimated from a numerical weather prediction model (ERA5 reanalysis) associated with this weather event. [less ▲]

On Improving Slant Wet delays for Tracking Severe Weather Events: An evaluation During Two Storms in Europe

Scientific Conference (2022, December 14)

Climate change has led to an increase in the frequency and severity of weather events with intense precipitation and, subsequently, a greater susceptibility of communities around the world to flash ... [more ▼]

Climate change has led to an increase in the frequency and severity of weather events with intense precipitation and, subsequently, a greater susceptibility of communities around the world to flash flooding. Networks of ground-based Global Navigation Satellite System (GNSS) stations enable the measurement of integrated water vapor along slant pathways, providing three-dimensional (3D) water vapor distributions at low-cost and in real-time. This makes these data a valuable complementary source of information for tracking storm events and predicting their paths. However, it is well established that residual modelling errors and multipath (MP) effects at GNSS stations do impact incoming signals, especially at low elevations and during storms when the atmospheric conditions change rapidly. Until now, the bulk of GNSS products for meteorology are estimates of the more conventional zenith total delays and horizontal gradients, but these products may not be most appropriate for determining 3D distributions of water vapor during convective storm events. In this study we investigate the impact of residual-phase-corrected and multipath-corrected slant wet delay (SWD) estimates on tracking extreme weather events using two events in Europe that led to flooding, damage to property and loss of life. We employed Precise Point Positioning (PPP) with integer ambiguity resolution to generate station-specific MP correction maps. The spatial stacking was carried out in congruent cells with an optimal resolution in elevation and azimuth at the local horizon but with decreasing azimuth resolution as the elevation angle increases. This permits an approximately equal number of observations allocated to each cell. In our analysis we recovered the one-way SWD by adding GNSS post-fit phase residuals, representing the non-isotropic component of the SWD, i.e., the higher-order inhomogeneity. Using the derived MP maps in a final step, the one-way SWD were improved to employ them for the analysis of the weather event. Moreover, we validated the SWD between ground-based water-vapor radiometry and GNSS-derived SWD for different elevation angles. Furthermore, the spatio-temporal fluctuations in the SWD as measured by GNSS closely mirrored the moisture field from the ERA5 re-analysis associated with this severe weather event [less ▲]

Forschung in Geodäsie und Geoinformation an der Uni.lu

Presentation (2022, November 10)

This is a summary of selected research carried out by the team GGE of the DoE in 2017-2022.

kCV-B: Bootstrap with Cross-Validation for Deep Learning Model Development, Assessment and Selection

in kCV-B: Bootstrap with Cross-Validation for Deep Learning Model Development, Assessment and Selection (2022, October)

This study investigates the inability of two popular data splitting techniques: train/test split and k-fold cross-validation that are to create training and validation data sets, and to achieve sufficient ... [more ▼]

This study investigates the inability of two popular data splitting techniques: train/test split and k-fold cross-validation that are to create training and validation data sets, and to achieve sufficient generality for supervised deep learning (DL) methods. This failure is mainly caused by their limited ability of new data creation. In response, the bootstrap is a computer based statistical resampling method that has been used efficiently for estimating the distribution of a sample estimator and to assess a model without having knowledge about the population. This paper couples cross-validation and bootstrap to have their respective advantages in view of data generation strategy and to achieve better generalization of a DL model. This paper contributes by: (i) developing an algorithm for better selection of training and validation data sets, (ii) exploring the potential of bootstrap for drawing statistical inference on the necessary performance metrics (e.g., mean square error), and (iii) introducing a method that can assess and improve the efficiency of a DL model. The proposed method is applied for semantic segmentation and is demonstrated via a DL based classification algorithm, PointNet, through aerial laser scanning point cloud data. [less ▲]

Deep Learning for Ground and Non-ground Surface Separation: A Feature-based Semantic Segmentation Algorithm for Point Cloud Classification

E-print/Working paper (2022)

Precise ground surface topography is crucial for 3D city analysis, digital terrain modeling, natural disaster monitoring, high-density map generation, and autonomous navigation to name a few. Deep ... [more ▼]

Precise ground surface topography is crucial for 3D city analysis, digital terrain modeling, natural disaster monitoring, high-density map generation, and autonomous navigation to name a few. Deep learning (DL; LeCun, et al., 2015), a division of machine learning (ML), has been achieving unparalleled success in image processing, and recently demonstrated a huge potential for point cloud analysis. This article presents a feature-based DL algorithm that classifies ground and non-ground points in aerial laser scanning point clouds. Recent advancements of remote sensing technologies make it possible digitizing the real world in a near automated fashion. LiDAR (Light Detection and Ranging) based point clouds that are a type of remotely sensed georeferenced data, providing detailed 3D information on objects and environment have been recognized as one of the most powerful means of digitization. Unlike imagery, point clouds are unstructured, sparse and of irregular data format which creates many challenges, but also provides huge opportunities for capturing geometric details of scanned surfaces with millimeter accuracy. Classifying and separating non-ground points from ground points largely reduce data volumes for consecutive analyses of either ground or non-ground surfaces, which consequently saves cost and labor, and simplifies further analysis. [less ▲]

Resampling methods for a reliable validation set in deep learning based point cloud classification

in Resampling methods for a reliable validation set in deep learning based point cloud classification (2022, June)

A validation data set plays a pivotal role in tweaking a machine learning model trained in a supervised manner. Many existing algorithms select a part of available data by using random sampling to produce ... [more ▼]

A validation data set plays a pivotal role in tweaking a machine learning model trained in a supervised manner. Many existing algorithms select a part of available data by using random sampling to produce a validation set. However, this approach can be prone to overfitting. One should follow careful data splitting to have reliable training and validation sets that can produce a generalized model with a good performance for the unseen (test) data. Data splitting based on resampling techniques involves repeatedly drawing samples from the available data. Hence, resampling methods can give better generalization power to a model, because they can produce and use many training and/or validation sets. These techniques are computationally expensive, but with increasingly available high-performance computing facilities, one can exploit them. Though a multitude of resampling methods exist, investigation of their influence on the generality of deep learning (DL) algorithms is limited due to its non-linear black-box nature. This paper contributes by: (1) investigating the generalization capability of the four most popular resampling methods: k-fold cross-validation (k-CV), repeated k-CV (Rk-CV), Monte Carlo CV (MC-CV) and bootstrap for creating training and validation data sets used for developing, training and validating DL based point cloud classifiers (e.g., PointNet; Qi et al., 2017a), (2) justifying Mean Square Error (MSE) as a statistically consistent estimator, and (3) exploring the use of MSE as a reliable performance metric for supervised DL. Experiments in this paper are performed on both synthetic and real-world aerial laser scanning (ALS) point clouds. [less ▲]

Scanning the Past: A 3D Model of Trausch's Library

Scientific Conference (2022, May 05)

Today, modern geospatial technologies and methods are widely used in combination with the documentation and preservation of objects of importance to cultural heritage. In this setting, archeologists and ... [more ▼]

Today, modern geospatial technologies and methods are widely used in combination with the documentation and preservation of objects of importance to cultural heritage. In this setting, archeologists and historians alike benefit from the rapid technological developments over the past decades, which have resulted in instrumentation that allows the capture of real objects and the generation of accurate and precise three-dimensional (3D) digital representations, i.e. models, from these sensed data. Here, the object of interest is the villa of the late historian Professor Gilbert Trausch with its library, for which a virtual library should be created for the general public. The building is an 19th century townhouse located on Limpertsberg in the City of Luxembourg and contains a cellar, three floors as well as a loft. Of particular interest were the cellar, stair cases, the first (the location of Professor Trausch’s office) and second floors as most of the bookshelves are situated there. In line with state-of-the-art approaches for 3D building modelling, a broad spectrum of modern geospatial technologies including classical surveying, Global Navigation Satellite System (GNSS), digital close-range photogrammetry and terrestrial laser scanning were employed to capture the Villa Trausch and its 33 bookshelves, i.e. Trausch’s Library, in all its details, while providing all data in one homogeneous coordinate system. Models and more photo-realistic visualizations of the exterior and interior have been obtained using, e.g., indoor images captured during the scanning. These allowed us to explore different virtual reality (VR) pathways employed by the gaming industry, for the generation of a first VR experience of the building in the sense of a digital museum. Currently solutions for the development of a public virtual library using commercial providers are investigated. [less ▲]

A TWO-STEP FEATURE EXTRACTION ALGORITHM: APPLICATION TO DEEP LEARNING FOR POINT CLOUD CLASSIFICATION

in A TWO-STEP FEATURE EXTRACTION ALGORITHM: APPLICATION TO DEEP LEARNING FOR POINT CLOUD CLASSIFICATION (2022, March)

Most deep learning (DL) methods that are not end-to-end use several multi-scale and multi-type hand-crafted features that make the network challenging, more computationally intensive and vulnerable to ... [more ▼]

Most deep learning (DL) methods that are not end-to-end use several multi-scale and multi-type hand-crafted features that make the network challenging, more computationally intensive and vulnerable to overfitting. Furthermore, reliance on empirically-based feature dimensionality reduction may lead to misclassification. In contrast, efficient feature management can reduce storage and computational complexities, builds better classifiers, and improves overall performance. Principal Component Analysis (PCA) is a well-known dimension reduction technique that has been used for feature extraction. This paper presents a two-step PCA based feature extraction algorithm that employs a variant of feature-based PointNet (Qi et al., 2017a) for point cloud classification. This paper extends the PointNet framework for use on large-scale aerial LiDAR data, and contributes by (i) developing a new feature extraction algorithm, (ii) exploring the impact of dimensionality reduction in feature extraction, and (iii) introducing a non-end-to-end PointNet variant for per point classification in point clouds. This is demonstrated on aerial laser scanning (ALS) point clouds. The algorithm successfully reduces the dimension of the feature space without sacrificing performance, as benchmarked against the original PointNet algorithm. When tested on the well-known Vaihingen data set, the proposed algorithm achieves an Overall Accuracy (OA) of 74.64% by using 9 input vectors and 14 shape features, whereas with the same 9 input vectors and only 5PCs (principal components built by the 14 shape features) it actually achieves a higher OA of 75.36% which demonstrates the effect of efficient dimensionality reduction. [less ▲]

INVESTIGATION OF POINTNET FOR SEMANTIC SEGMENTATION OF LARGE-SCALE OUTDOOR POINT CLOUDS

Scientific journal (2021)

Semantic segmentation of point clouds is indispensable for 3D scene understanding. Point clouds have credibility for capturing geometry of objects including shape, size, and orientation. Deep learning (DL ... [more ▼]

Semantic segmentation of point clouds is indispensable for 3D scene understanding. Point clouds have credibility for capturing geometry of objects including shape, size, and orientation. Deep learning (DL) has been recognized as the most successful approach for image semantic segmentation. Applied to point clouds, performance of the many DL algorithms degrades, because point clouds are often sparse and have irregular data format. As a result, point clouds are regularly first transformed into voxel grids or image collections. PointNet was the first promising algorithm that feeds point clouds directly into the DL architecture. Although PointNet achieved remarkable performance on indoor point clouds, its performance has not been extensively studied in large-scale outdoor point clouds. So far, we know, no study on large-scale aerial point clouds investigates the sensitivity of the hyper-parameters used in the PointNet. This paper evaluates PointNet’s performance for semantic segmentation through three large-scale Airborne Laser Scanning (ALS) point clouds of urban environments. Reported results show that PointNet has potential in large-scale outdoor scene semantic segmentation. A remarkable limitation of PointNet is that it does not consider local structure induced by the metric space made by its local neighbors. Experiments exhibit PointNet is expressively sensitive to the hyper-parameters like batch-size, block partition and the number of points in a block. For an ALS dataset, we get significant difference between overall accuracies of 67.5% and 72.8%, for the block sizes of 5m×5m and 10m×10m, respectively. Results also discover that the performance of PointNet depends on the selection of input vectors. [less ▲]

Feasibility of ERA5 integrated water vapor trends for climate change analysis in continental Europe: An evaluation with GPS (1994–2019) by considering statistical significance

in Remote Sensing of Environment (2021), 260(112416),

Although the statistical significances for the trends of integrated water vapor (IWV) are essential for a correct interpretation of climate change signals, obtaining accurate IWV trend estimates with ... [more ▼]

Although the statistical significances for the trends of integrated water vapor (IWV) are essential for a correct interpretation of climate change signals, obtaining accurate IWV trend estimates with realistic uncertainties remains a challenge. This study evaluates the feasibility of the IWV trends derived from the newly released fifth generation European Centre for Medium-Range Weather Forecasts (ECMWF) atmospheric reanalysis (ERA5) for climate change analysis in continental Europe. This is achieved by comparing the trends derived from in-situ ground-based Global Positioning System (GPS)’s daily IWV series from 1994 to 2019 at 109 stations. The realistic uncertainties and statistical significances of the IWV trends are evaluated with the time series analysis on their noise characteristics and proper noise models. Results show that autoregressive moving average ARMA(1,1) noise model is preferred rather than the commonly assumed white noise (WN) or first-order autoregressive AR(1) noise for about 68% of the ERA5 and GPS IWV series. An improper noise model would misevaluate the trend uncertainty of an IWV time series, compared with its specific preferred noise model. For example, ARMA(1,1) may misevaluate the standard deviations of their trend estimates (0.1–0.3 kg m−2 decade−1) by 10%. Nevertheless, ARMA(1,1) is recommended as the default noise model for the ERA5 and GPS IWV series. However, the preferred noise model for each ERA5 minus GPS (E-G) IWV series should be specifically determined, because the AR(1)-related models can result in an underestimation on its trend uncertainty by 90%. In contrast, power-law (PL) model can lead to an overestimation by up to nine times. The E-G IWV trends are within −0.2–0.4 kg m−2 decade−1, indicating that the ERA5 is a potential data source of IWV trends for climate change analysis in continental Europe. The ERA5 and GPS IWV trends are consistent in their magnitudes and geographical patterns, lower in Northwest Europe (0–0.4 kg m−2 decade−1) but higher around the Mediterranean Sea (0.6–1.4 kg m−2 decade−1). [less ▲]