Over the years, InSAR has become an indispensable tool in the study of ground deformation, including volcanic deformation, and this continues to be the case in times of improved technology. Since the volcanoes on the Caribbean islands of Saba and St. Eustatius are active, the implementation of an InSAR-based monitoring system is crucial to enhance the spatial resolution of volcano monitoring beyond the capabilities of the ground-based monitoring network, for instance in the case of localized deformations such as dike intrusions. However, technical challenges arise in these tropical settings, caused by dense rainforest, atmospheric artifacts and terrain variability, posing serious challenges to the use of InSAR. Time series InSAR, including SBAS and PSI, can be used to overcome these limitations. A previous study has explored the use of PSI for monitoring, using the already available DePSI software. In this research, an SBAS approach within the Delft InSAR software framework is developed using state-of-the-art Python packages, including (sar)xarray, dask and zarr, and is used to assess whether there is capability to develop SBAS into a volcanic monitoring tool for Saba and St. Eustatius. In addition, a preliminary comparison between the SBAS and PSI methodologies is performed based on a theoretical and (semi-)quantifiable approach. This study combines data from two satellites operating at different wavelengths: Sentinel-1 (C-band) and ALOS-2 (L-band).

Assuming no ongoing deformation, based on GNSS results, the variability of the results around zero can be used as an indicator of precision. The results obtained through SBAS are promising, in particular for L-band, on account of e.g., extensive spatial coverage, efficiency and relatively low variability even with the presence of atmospheric and DEM components. Overall, the results reveal mm order deviations. In the event of volcanic activity, the expected deformation signals are in the range of cm-dm's and can therefore be detected, i.e., with an estimated minimal detectable deformation of 1.5 cm/year in the worst-case scenario. The implementation of three different coherence-based masking approaches-water, single and individual-give an indication of the level of robustness and reliability of the results. Generally, a relatively high level of consistency can be observed among the different masking results of ALOS-2 for both islands, for St. Eustatius following the correction of the unwrapping errors using two testing approaches: an interferogram removal approach and an adaptive approach based on the DIA procedure. The latter procedure allows for retaining all observations and their residuals and is therefore preferred. In contrast, the Sentinel-1 results reveal a lower level of consistency. It is suspected that this inconsistency mainly arises on account of the numerous unwrapping errors within the single masking approach. The individual masking approach appears to be less susceptible to unwrapping errors, however is more prone to outliers than the single masking approach. Further research, following the correction of the atmospheric component and DEM errors, may offer insights into the preferred masking approach. Overall, the use of L-band imagery shows potential, offering spatial coverage where C-band does not, even with limited ALOS-2 data availability and large temporal baselines. The preliminary comparative analysis with the PSI approach, based on the respective strengths and limitations from literature, spatial coverage, processing steps, precision and computational requirements, suggests that a hybrid method could prove to be advantageous to minimize (potential) signal loss, e.g. either from limited spatial coverage or spatial resolution, and enhance volcanic risk assessment. SBAS excels in the extensive spatial coverage, especially using L-band, providing nearly homogeneous coverage of St. Eustatius, even on the flanks of the Quill, and on the outer flanks Mt. Scenery on Saba. However, regardless of the mask, wavelength or method, acquiring coverage around the summit of Mt. Scenery on Saba remains challenging.

The study contributes to advancing InSAR time series analysis for the volcanic monitoring on Saba and St. Eustatius through the successful implementation of an SBAS approach within the Delft InSAR software framework based on state-of-the-art packages, the implementation and evaluation of new approaches to enhance the method in terms of the efficiency and robustness and a comparison with existing software. In addition, the software can be applied in a generic sense for various applications and can be extended for further improvements.

How to deal with the presence of weather affected data is an unavoidable topic in the processing of optical imagery. Clouds and cloud shadows significantly alter the spectral signatures obtained from satellite data, which often leads to problems for any kind of scientific analysis. In this research there has been elaborated on two different kind of problems: The detection of clouds and cloud shadows and the mitigation of the effect caused by cloud shadows. Most of existing operational cloud detection algorithms are so-called rule-based. Their performance is highly variable and they have their limitations. A new promising research was done by Mohajerani and Parvaneh (2019), where a convolutional neural network (CNN) named ’Cloud-Net’ was developed. In this study we have elaborated on this CNN, by converting the analysis to Sentinel-2 data and making significant modifications on the model setup. The results have been compared to the ESA Scene Classification Map (SEN2COR algorithm). It was found that for the detection of clouds the overall CNN accuracy outperforms the ESA Scene Map (95.6% vs. 92.0% respectively). For the detection of cloud shadows the modified Cloud-Net model also gave better results (90.4% vs. 84.4%). Previous work on cloud shadow correction algorithms show rather complex and inconvenient methods, where the only goal was to remove the effect of the shadow. If one is interested to also correct for illumination effects, to make it more aligned to a predetermined ground truth, new possibilities arise which allows for simpler and more direct methods. Two proposed methods have been investigated in this study. The first method, called ’decomposition of components’, investigated the use of a single formula. The affected cloud shadow pixel is corrected based on the RGB difference with a ground truth image, and a single correction factor that was determined based under the assumption that cloud shadows cause a homogeneous alteration effect in a small area. The second method, called the ’CNN based method’, presents a totally new idea by changing the Cloud-Net model to a regression model, in order to correctly alter cloud shadow affected pixels. The performance of both methods was quantified by the structural similarity index measure (SSIM). It was found that the decomposition of components method has the most potential, showing significant improvements on the correction of cloud shadow affected areas.

In this study an analysis of the efficacy of using satellite data, in the form of InSAR measurements, as an extension of the volcanic monitoring network on Saba and St. Eustatius is performed. For this research, data from three different satellites that operate at three different wavelengths are available: ALOS-2 (L-band SAR), Sentinel-1 (C-band SAR) and PAZ (X-band SAR). The data are analysed through the formation of interferograms that are obtained using the Delft Object-oriented Radar Interferometric Software (DORIS) and Persistent Scatterer Interferometry (PSI) performed following the Delft Persistent Scatterer Interferometry (DePSI) algorithm.
The interferograms and PSI results differ strongly per satellite and are affected by the combined impact of several factors. In this study the impact of the misalignment of the master image used in the generation of the interferograms with respect to the Digital Elevation Model (DEM) is discussed, as well as the impact of the incidence angle, the spatial resolution, the temporal resolution, the perpendicular baseline, the number of available images and the wavelength.
The interferograms of ALOS-2 are of a good quality, however the low temporal resolution makes studying fast surface deformation difficult. However, they could be used to study surface changes in retrospect or to study slower processes, such as the pressurisation of a magma chamber, causing gradual surface deformation. The
low spatial resolution makes the interferograms of Sentinel-1 difficult to interpret and the interferograms for the PAZ data currently show too large amounts of decorrelation to study surface deformations.
The PSI analysis produces reliable results for Sentinel-1. The estimated linear deformation for the Persistent Scatterers (PS) shows constant values over both islands, which are centred around 0 mm/y and have low standard deviations. Therefore it is assumed, based on the data and prior knowledge about the area, that there is currently no deformation on either of the islands. The PSI analyses for the other two satellites do not provide reliable results, because the number of available images in the stacks is too low (only 10-12 images compared to the 116-123 available images for Sentinel-1). The PAZ data might be used in the future, when more images are available, however the low temporal resolution of the ALOS-2 data means that an appropriate stack cannot be acquired within the design lifetime of the satellite.
The ALOS-2 interferograms and the PSI analysis for Sentinel-1 could thus at present be a useful addition to the ground-based monitoring network. When a larger stack of data for PAZ is available, the PSI analysis could potentially be conducted again in order to determine its use as a volcanic monitoring tool.

In the Netherlands, the Delta Programme aspires to adjust spatial planning climate-proof and water-resilient, in order to be prepared for extreme weather in 2050. To achieve this ambition, municipalities, provinces, regional water authorities and central governments conduct stress tests to map out the vulnerabilities in their areas of authority by no later than 2019. The stress tests comprise four themes: pluvial flooding, drought, heat and floods. In addition, the Delta Programme 2019 acknowledges the mitigation of and adaptation to land subsidence as an important tasking. The municipality of Rotterdam faces the challenge of adding land subsidence as stress test theme and assessing its influence on pluvial flooding. Contrary to the stress test pluvial flooding, no consended methodology exist on how to map out vulnerabilities concerning land subsidence. In addition, only few studies have numerically investigated the spatial-temporal effect of land subsidence on pluvial flooding in urban areas. However, the advent of techniques to measure ground level (LiDAR) and land subsidence (InSAR) and advances in high resolution flood modelling (3Di) enable the numerical modelling of urban pluvial flooding influenced by land subsidence. This research explores the investigation of the influence of land subsidence on pluvial flooding in Rotterdam by supplementing the conducted stress test pluvial flooding with a land subsidence assessment. The conducted stress test pluvial flooding in Rotterdam is based on a 3Di-simulation of standardised rain events, based on a DEM 2016.

To asses the current influence of land subsidence on pluvial flooding, a Digital Elevation Model (DEM) that approximates the sub-neighbourhood Tuinenhoven at design level is created and used as 3Di-input. When comparing 3Di-results based on this design DEM to the stress test pluvial flooding, it becomes clear that the total volume of water during extreme rainfall stored on the streets is not affected by land subsidence. The bathymetry of the DEM does affect the water's distribution however. The Tuinenhoven case-study demonstrates that currently land subsidence increases the severity of the impact of pluvial flooding but that the main cause of pluvial flooding during extreme rainfall is the limited capacity of the drainage system.

Land subsidence in Rotterdam complex. The conducted land subsidence analysis based on an InSAR data-set supports this complexity. It illustrates that the subsidence behaviour of Rotterdam is influenced by foundation type, land use classification, the presence of dredge in the anthropogenic layer and top soil type. This respectively indicates the occurrence of pole rot and shallow foundations, anthropogenic compression and compaction of shallow soft layers caused by loading, landfill subsidence as a result of land fillings that contain dredging spoil and consolidation of the Holocene clay layer as a result of drainage. However, the land subsidence analysis failed to identify location-specific dominant land subsidence processes. This failure was primarily caused by the limitation to only one linear subsidence rate between 2009 and 2014 per point. To demonstrate how land subsidence can be translated to pluvial flooding based on a land subsidence analysis, land use classification was selected as the most dominant influencing factor and used in a linear land subsidence prognosis until 2030. The linear assumptions largely obstructs results to be interpreted location-specific.

The IJsselmonde case-study shows that land subsidence is expected to decrease the passability of roads and decrease the risk per building in the future. These decreases are caused by the fact that roads relatively subside fast and buildings relatively slow. The biggest influence on the risk per building classification is the assumed threshold value per building. Simulated road maintenance results in an increase of the passability of roads and an increase of buildings at risk of water nuisance. The loss of the water-storing function of the road after reconstruction increases the water levels in gardens and puts buildings at an increasing risk.

In conclusion, the most challenging part of investigating the influence of land subsidence on pluvial flooding is the crucial identification of the different occurring land subsidence processes. It is demonstrated that the possibilities with InSAR-data are promising, when used with sufficient competency, although the available InSAR data should be divided in shorter intervals to detect the subsidence rate trends. Land subsidence rate trends are crucial in the identification of land subsidence processes and assessing influences like groundwater variations and increased loading due to maintenance or construction works. When the land subsidence analysis is improved, so will the land subsidence and threshold height per building prognosis. When the relative prognosed decrease of the threshold value per building is improved, it can be quickly assessed whether buildings classified at risk in the stress test pluvial flooding are at future increasing risk during extreme rainfall, without conducting a full 3Di-simulation.

Accurate information on Arctic sea ice thickness has been historically limited both spatially and temporally to spare submarine sonar measurements until the advent of satellite altimeters such as ICEsat (operating from 2002 to 2008) and CryoSat-2 (operating from 2011 to present day). This study aims to use the historical record of normalized radar backscatter measurements from satellite Ku-band and C-band scatterometers (ERS, QuikSCAT and ASCAT), which have been continuously operating since 1992, to homogenize the satellite altimeter record and extend the record of Arctic sea ice thickness measurements backwards in time. This study is structured so as to first derive a set of empirical relationships between normalized backscatter measurements and wintertime sea ice thickness estimates in the Arctic using existing satellite altimeter records as a reference. Two separate scatterometer sea ice thickness models are produced using coincident scatterometer and altimeter observations, one for C-band sea ice thickness estimation using ASCAT and CryoSat-2 collocations, and another for Ku-band sea ice thickness estimation using QuikSCAT and ICESat collocations. Based on the agreement to the altimeter records, the estimation of wintertime sea ice thickness using the C-band and Ku-band scatterometer models is uncertain to within 0.5 m (1-sigma), that is, a precision similar to that of the original altimeter references. The homogenization of the satellite altimeter records cannot be done directly, because the ICESat and CryoSat-2 instruments operate in different periods, but it can be done indirectly by comparing the sea ice thickness estimates obtained from Ku-band (based on ICESat) and C-band (based on CryoSat-2) estimates during the years that the Ku-band and C-band scatterometers operate simultaneously. These overlap years have been used to verify the consistency between the C-band and Ku-band relationships, and to correct for a 0.55 m bias in the CryoSat-2 reference, having considered the earlier ICESat record as absolute standard. After removing this bias from the CryoSat-2 reference, the sea ice thickness estimates from C-band and Ku-band records agree to within 0.15 m (1-sigma). Moreover, the resulting scatterometer sea ice thickness models allow the introduction of new thickness thresholds from a previously existing backscatter-based classification of Arctic sea ice types, providing a thickness threshold of 1.54 m to define first year ice (FYI), and a thickness threshold of 2.25 m to separate second year ice (SYI) from older multiyear ice (old MYI). Ancillary datasets were used to investigate the correlations between backscatter-based sea ice thickness and physical variables, such as snow depth and surface deformation, in order to investigate possible sources of systematic error, which otherwise appear to be bound within 0.30 m (1-sigma). The maps of differences between scatterometer and altimeter sea ice thickness estimates were analysed in terms of collocated sea ice convergence, sea ice shear and snow depth parameters using a multiple regression model. The results show that both the Ku-band and C-band models underestimate ice thickness in areas of high convergence such as the Fram Strait, and overestimate ice thickness in areas with high shear such as the Beaufort Gyre. These correlations may be interpreted as led by increases in backscatter due to surface deformation with (case of convergence) or without (case of shear) associated increases in ice thickness. In addition, the Ku-band model is found sensitive to snow load, with overestimation interpreted as led by an increase in backscatter without associated ice growth, and the C-band model is sensitive to marginal rough ice. An unphysically large dependence on snow depth was found for the C-band estimate, which we conjecture is due to problems with the CryoSat-2 reference. Finally, a reconstruction of Arctic sea ice thickness in the wintertime has been made by combining the Ku-band and C-band sea ice thickness models with the normalized radar backscatter record of ERS, QuikSCAT and ASCAT from 1992 through 2017. These showed a decline average Arctic sea ice thickness of -0.28 m / decade, with a steady decline in second-year ice and high variability in the mean multi-year ice thickness.

Almost four million people in the Netherlands who live below sea level are protected by 17000 km of dikes against the water. Since deformation can be an indicator of a developing failure mechanism, information about this parameter can help authorities to detect weak spots in the dutch water defense structures.
In this research, Radar Interferometry (InSAR) is used to estimate deformation time series on the vegetated part of the dike. Using Point Scatterers (PS), it is already possible to estimate deformation time series with mm precision on objects with a coherent reflection over time. Apart from objects, also surfaces provide a reflection of the radar signal. These so-called Distributed Scatterers (DS), which can for instance be found on dikes, are often strongly affected by temporal decorrelation. Although conventional approaches to detect and estimate these DS do not work, there may still be coherent information for interferograms with small temporal baselines. In this research project, a methodology is presented to detect coherent DS on the vegetated part of a dike and to estimate their deformation time series.
To do so, the phases of PS are subtracted from the DS on the vegetated part of the dike to reduce the phase noise due to different states of the atmosphere. The estimated coherence matrix is then used to only select coherent interferograms which are used to unwrap the phases with respect to PS. Using data obtained from the TerraSAR-X (X-band) and Radarsat-2 (C-band) satellites, it was shown that DS could be detected on some parts of the grass-covered inner slope of the Marken dike.

We use ground and space geodetic data to study surface deformation and gravity change at Kīlauea volcano from January to September 2015. This period includes an episode of heightened activity in May 2015, which we refer to as ’the May 2015 event’. The data set consists of Global Navigation Satellite System (GNSS), tilt, visual and seismic time series along with 25 descending and 15 ascending acquisitions of the Sentinel-1a satellite in Interferometric Wide swath mode and microgravity surveys taken a few years before and just after the May 2015 event. We identify four different stages of surface deformation and volcanic activity during the May 2015 event which we attribute to the movement of magma and pressure changes in response to a magma supply and withdrawal imbalance in the shallow plumbing system. In particular, we model the deformation sources attributed to the Halema’uma’u reservoir (HMMR) and South caldera reservoir (SCR). The SCR was best described by inflation of a spheroidal at 2.8 (2.65-3.07) km depth below the Southern caldera region. The HMMR source was modelled by a point source deflation located East of the Halema’uma’u crater at 1.5 (0.95- 2.62) km depth. The surface microgravity changes which would result from changes in these reservoirs are significantly lower than the actually observed microgravity changes. We attribute this to the lack of complexity of the single point source model used. Mechanisms that add/remove mass from the subsurface without accompanying surface deformation, which are not part of the point source model, played a significant role. More frequent microgravity campaign surveys, if needed with a smaller network, are the only way to improve our understanding of these processes and help to quantify them.

Flooding is a very costly natural disaster especially when it hits urban areas. Yet synthetic aperture radar (SAR) based flood mapping barely works in urban areas. Buildings and man made objects have similar backscatter signatures as still standing water. This makes them hard to distinguish from one another. Structures can block the visibility of the ground surface for side looking SAR satellites making large parts of potentially flooded ground going unseen by SAR satellites. Smooth surfaces and limited ground visibility make it hard to produce accurate flood maps using SAR in urban environments. Here we have shown how the use of a temporal stack can improve the result of urban flood detection with
SAR. Traditionally SAR based flood mapping uses a single image or an image pair to classify flooded and non-flooded pixels. This study found these methods unable to detect flooded pixels in an urban setting. By using a temporal stack of SAR images more pixels are correctly classified as flooded while keeping false positive classifications low. However the number off correctly classified pixels remains too low to be useful on its own, by adding ancillary data in the form of a high resolution DEM an accurate flood map for a very specific area is produced. This means that SAR images are not suitable for flood mapping in urban areas as a single source of information. When they are combined with other data they have the potential to produce accurate flood maps useful for First responders when the next flooding disaster hits.

Repeat-pass acquisitions with coherent Synthetic Aperture Radar (SAR) systems, preserving both phase and amplitude, are more readily available than ever (Bruzzone, 2016). Phase measurements from SAR systems have seen widespread use in the Interferometric Synthetic Aperture (InSAR) technique to measure deformations and elevations since the late 1980’s (Hanssen, 2002). Since the late 1990’s an increase in radar-based change detection is observed, mainly relying on amplitude measurements (Ajadi et al., 2016; Dekker, 1998). The unpredictive multiplicative noise-like speckle, inherent to coherent SAR,makes change detection in SAR imagery difficult (Bamler, 2015). However, the advantages in the all-weather mapping capabilities and object penetrating properties of SAR make it a suitable remote sensing technique for certain applications, such as natural disaster damage assessment (Bruzzone and Prieto, 2000). Broadly speaking, change detection in SAR-based images usually consists of applying an operator on two spatially filtered SAR images to create a difference image (DI), which is then analysed for change points by thresholding and/or clustering (e.g. Alphonse and Biju, 2015). However, such an approach completely neglects the long-term stability of a pixel. When taking the temporal evolution of a pixel into account, the steep increase in data volume puts an emphasis on finding an optimal (’best practice’) approach to the multitemporal change point detection problem. Here it is shown that change point detection methods that properly take the temporal evolution of a pixel into account can provide good segmentation results in multi-temporal SAR data stacks, even in unfiltered stacks that preserve the complete spatial resolution and without considering the spatial context in a pixel’s direct neighbourhood. Moreover, it is found thatmore sophisticated change point detection algorithms don’t
necessarily yield superior segmentation results for various discontinuity functions. This means algorithm selection has to be application driven. The results demonstrate that the suitability of algebraic methods in heterogeneous areas is limited, whereas proper time series analysis yields fairly consistent results over different land covers within the same image. Often, little effort is spend on finding an optimal approach; neglecting data selection and storage, a sensitivity analysis and/or the post-processing analysis procedure, all of which are shown or known to increase the success rate, efficiency and understanding of the segmentation results. It is anticipated that change point detection in SAR imagery will shift away from the classical bi-temporal methodology and multi temporal approaches will become the norm, be it by decomposing multi-temporal stacks or time series analysis.

Recent studies have shown that processes like the thinning of the ice shelves, the acceleration of the outlet glaciers and the collapses of the ice shelves are connected to surface melting. Several ice shelves at the Antarctic Peninsula, a region that has experienced an atmospheric warming much larger than the global average, have retreated and disappeared. In order to assess the stability of the ice shelves, snowmelt amount can be used. Microwave remote sensing instruments are used in studying surface melting. The presence of liquid water causes a decrease in volumetric scattering that is dominant over absorption in dry snow, and increases absorption due to the increase of the imaginary part of snow permittivity. Thus, active microwave scatterometers can detect melt because of the decrease in backscatter. In this study, data from QuikSCAT (2000-2009) and ASCAT (2010-2016) are used for the investigation of the Antarctic surface melting dynamics for the period 2000-2016. The methodologies of Trusel et al. (2012) and Bothale et al. (2015) and two other threshold-based melt detection methodologies are applied and evaluated using in-situ meteorological records and are compared to passive microwave threshold-based melt duration results from AMSR-E and AMSR2. In addition, an attempt for a retrieval of a consistent melt time series is performed. The results support the use of a dynamic threshold-based melt detection approach for scatterometer data, but it is not possible to achieve consistency in melt duration results from the two scatterometers mainly due to their differences overpass timing. Overall, a high interannual variation in melt extent and melt index is found and results show that the extent of the areas that experience morning and afternoon melt consistently remains constant through the years. However, morning measurements underestimate melt and thus midday observations are important for melt detection studies, as melting varies throughout the day. In addition, although most of the large scale melting phenomena are captured by AMSR-E, the higher sensitivity of QuikSCAT to melt and its finer resolution result in differences in the derived melt metrics.

This thesis explores the use of InSAR data for building monitoring. The area of main interest is the city centre of Amsterdam, where quay wall replacements are necessary on a large scale. In Amsterdam, the foundation quality needs to be checked before construction activities nearby are performed, such as a quay wall replacement. Levelling benchmarks are usually installed to determine building deformation. In case large deformation rates are observed, additional measures are taken to ensure no building damage occurs during construction. Interferometric Synthetic Aperture Radar (InSAR) measures deformations from space and provides multiple years of deformation measurements. This can therefore reduce required monitoring time before construction.
In this thesis, the use of InSAR for building deformation monitoring is investigated by analysis of occurring deformations of three areas in the city of Amsterdam. Furthermore, deformation in the Spoorsingel area in Delft is analyzed since large deformations are observed here. In all cases, levelling benchmarks are available for comparison of deformations.
The results from the Amsterdam cases indicate similar building deformation behaviour by InSAR and levelling. Differences between InSAR and levelling are suggested to be the result of the measurement location and non-uniform building movement. In the city centre of Amsterdam, buildings with more than 2 mm/y settlement require additional attention. In this regard, a probabilistic approach is proposed. The probability of exceeding the 2 mm/y threshold is calculated by combining the calculated deformation rate with the variance of this estimate. This method is not only useful for InSAR displacement time series but also for other monitoring methods such as levelling, where the amount of measurements is often limited.
Furthermore, the InSAR data enables the identification of building deformation mechanisms such as rotation, as is observed for the Spoorsingel in Delft. To identify this movement, rigid building behaviour is assumed. For buildings in plane-strain situation a method is devised to calculate the three dimensional deformation, using the ascending and descending track of a Synthetic Aperture Radar.
All in all, the InSAR data is of added value to determine the building deformation behaviour. For this, the location precision, the quality of the time series, the line-of-sight measurement and the variance of the deformation rate estimate should be taken into consideration. A flow chart is proposed which describes the use of InSAR for building monitoring.

Knowing 'where I am' is always essential and a prior to answer for a moving vehicle. Among numerous onboard sensors, a GNSS receiver for single-frequency Precise Point Positioning and camera are competitive due to the fairly lower cost and the potential to provide a lot of useful information.

However, due to the degraded GNSS solution performance in city valleys, a tight integration is considered combining the two sensors at the observation level, ie. processing the GNSS ranges and the vision measurements in the image of the camera. The availability of High Definition Maps (HD Maps) aids vehicle positioning by providing extra information on the environment. In this project, landmark positions are retrieved through vision and the HD map, and can complement GNSS in city valleys. Additionally, the project focuses on building the mathematical model for the integration of observed landmark position (using a single camera, considering the ease of implementation and cost) and GNSS measurements, analyzing the performance as well as the feasibility for vehicle positioning. The project emphasizes the feasibility study of the proposed mathematical model, which is flexible and capable of using all available input automatedly, and providing a position solution with the best precision.

The uncertainty in the available landmark positions (for instance errors in the HD maps) is handled in two different ways: one is to include the landmark position coordinates as measurements into the model, the other one projects the uncertainty onto the measurements in the camera image. The latter method turns out to be much more efficient. To integrate vision and GNSS measurements, a conversion between an ECEF (earth-centered, earth-fixed coordinate frame), typically used for GNSS, and a world coordinate frame for the camera measurements, is required. A position offset between the GNSS antenna and camera is considered, since the camera lens center does not coincide with the GNSS antenna center. In the simulation and experiment, an extended integration is also presented and discussed which leaves out the position offset, for instance when the GNSS antenna is very close to the camera, which can further improve the redundancy and lower the computational load.

From the simulation and experiment, we conclude that the integration model is able to produce a position solution when one of the sensors is unable to produce a position solution and the other one still can; the extended integration model is able to produce a position solution even when both sensors individually fail to produce a position solution. Among these scenarios, the one when GNSS fails and vision operates, the integration model can produce a position solution within a quarter of a meter in local horizontal coordinates, and the GNSS measurements do not contribute much to the position solution. Compared to the integration model, the extended integration improves the model by reducing or eliminating the (typically heavy) correlation between the estimates, in particular those for the camera-antenna position offset, the GNSS receiver clock error and the vertical coordinate. Under the same scenarios, the extended integration improves the standard deviation in vertical coordinate and receiver clock error, within a quarter of a meter and one-third of a meter respectively.

Further study is recommended in the direction of applying full image processing procedures to obtain more realistic vision measurements, to include GNSS carrier phase observations to replace the current GNSS positioning based on Precise Point Positioning, in order to have a position solution of similar quality as the vision part. The dimension of system gets larger when carrier phase measurements (phase ambiguities) are added as well as two additional rotations for a camera; the extra rotations introduce a significant amount of nonlinearity in the model. A larger model with increased nonlinearity may call for an alternative model formulation.

Nowadays, many people and organizations depend on Global Navigation Satellite Systems (GNSS. Monitoring of GNSS is important to ensure the quality of GNSS measurements and products. The availability of the signals-in-space (SiS) is an essential part of the monitoring of GNSS, but it is not clear how availability is defined and standards for monitoring are lacking. The main research question for this thesis is therefore: Which are the key performance indicators related to availability that unambiguously describe sensor station and system performance in time, how can these be computed in an operational manner, and how can they be presented in a condensed form to the stakeholders? This includes an objective of defining unambiguous performance parameters for sensor station and system, and address the considerations related to the definition. A prototype software tool is created to study the algorithms and compute the key performance indicators.

Availability is in the basis a binary operation: a signal is available or unavailable. When this is applied on daily measurements, daily statistics can be computed. A signal is considered available if the code, carrier phase and C/N0 measurements are present, and meet certain standards. A signal is said to be expected if the satellite is expected to transmit that signal, the receiver is configured to receive that signal, and the signal is not blocked by objects in the signal’s path to the sensor station. For this it’s needed to define and compute an elevation mask for each station.

The sensor station and system performance parameters are computed from a network of sensor stations, using files in the Receiver Independent Exchange format (RINEX). Four key performance indicators are defined for the sensor station performance. The Daily Station Availability describes the part of the day that the station is operational, the Daily Station Total Availability gives how well the station receives, and the Effective Mean Elevation quantifies the elevation mask and thus the location of the sensor station. These three parameters are summarized into the Overall Station Quality parameter, which gives and overall performance class to the sensor station.

For the system performance a satellite is considered available if all signals are received by a sensor station of the monitoring network and the health status is healthy. The satellite is considered unavailable if the signals are received by none of the monitoring stations, while expected by at least two stations, or the health status is unhealthy. The Daily GNSS Availability parameter gives the percentage of the day that the satellite was available and the Daily Available number of Satellites tells how many satellites were available during the day. 

The parameters are computed for a period of 100 days. Results are presented using color codes and by showing only detailed information in case of anomalies or specific investigations. The proposed key performance indicators showed to be very useful at pointing out good performances or anomalies. While SiS availability gives much insight in the performance, a monitoring tool can be improved when combined with other performance aspects.

 

Since GPS tends to fail for indoor positioning purposes, alternative methods like indoor positioning systems (IPS) based on Bluetooth low energy (BLE) are developing rapidly. Generally, IPS are deployed in environments covered with obstacles such as furniture, (partition-) walls, people and electronics influencing the signal propagation. The major factor influencing the system performance and to acquire optimal positioning results is the geometry of the beacons. The geometry of the beacons is limited to the available infrastructure that can be deployed (number of beacons, basestations and tags), which is dependent on the budget and deployment effort of the customer.
This leads to the following challenge: Given a limited number of beacons, where should they be placed in a specified indoor environment, such that the geometry contributes to optimal positioning results?

This challenge is approached by using theoretical design computations. The design computations require the definition of a chosen 3D space, the number of beacons, possible user tag locations and a performance threshold (e.g. required precision). For any given set of beacon and receiver locations, the precision, internal- and external reliability can be determined on forehand. The results of a given geometry have been validated by deploying an IPS of BlooLoc and comparing the observed precision with the modeled precision for a chosen set of user tag locations. The theoretical model showed a precision pattern with several equivalent precision patterns of the measured data, however, some significant differences could not be explained physically and the model has to be adapted further. Besides determining the precision based on a set of beacon and receiver locations, the model is able to select the optimal geometric configuration based on a performance threshold (e.g. required precision). Depending on the performance threshold, the optimal configurations can either be a single solution or consist of multiple solutions that satisfy the performance threshold of the customer. The performance threshold varies depending on the use case and the user requirements. Therefore, the amount of possible combinations in terms of 3D space, amount of available beacons, possible beacon locations, user tag locations and performance thresholds are limitless and the model can thus be used for all kind of applications.

All in all, the initial model (design computations) can be used by IPS customers for all kind of applications. The model is able to select the optimal geometric configuration in terms of precision based on a performance threshold specified by the user. Furthermore, the model requires user specified input parameters and the amount of possible combinations are therefore unlimited. Although the initial model has to be adapted further to account for the differences in modeled and measured data as a consequence of environmental factors, the initial model is a good initiative for the rising indoor positioning market and its customers. Therefore, the model can be adapted further such that it can explain significant differences between the modeled and the measured data by including factors that influence the system performance in real life, such as materialistic properties, signal attenuation, interference, multipath, NLOS etc.