Possibilities to make measurements of ground subsidence more effective, using dInSAR, GNSS and levelling

(1)

INOM EXAMENSARBETE TEKNIK, GRUNDNIVÅ, 15 HP , STOCKHOLM SVERIGE 2017

Possibilities to make

measurements of ground

subsidence more effective, using

dInSAR, GNSS and levelling

EVELINA ÖSTBLOM

KTH

(2)

I

ABSTRACT

Ground subsidence is today very common. It can occur due to man-made or natural causes. Today, it is most common that subsidence occurs in urban areas, and there the causes are almost exclusively man-made, including groundwater extraction, construction under or above ground. This can lead to damage of buildings or other constructions and lead to large costs for reparation. To avoid this, subsidence must be detected in an early stage.

Therefore, this thesis work will study the most effective way to detect and measure subsidence using dInSAR, GNSS and levelling. The thesis work will contain a literature study, a compilation of cases comparing the methods and a quantitative comparison of data, called case Stockholm. The main focus of case Stockholm is to compare the RMS error for the datasets and to determine how well the linear regression of the datasets cohere. The literature study and the case compilation presents the strengths and weaknesses of the different methods, where dInSAR’s strength is the ability to measure large areas at once while the weakness is the inability to detect small movements within a large movement. The possibility to measure individual points of interest is the strength of both GNSS and levelling, where the most time-consuming method, levelling, also has the highest accuracy. In case Stockholm, the linear regression for dInSAR mostly follows the linear regression for GNSS and levelling. However, irregular levelling measurements that do not follow the general ground subsidence is missed by dInSAR and the amplitude of the dInSAR measurements differ from both GNSS and levelling measurements. This confirms the strengths and weaknesses mentioned in the literature study. The conclusion that can be drawn from this is that the most effective way of using dInSAR, GNSS and levelling is to first screen large areas for any movement using dInSAR. Later only the areas that display movement of any sort is measured with either GNSS or levelling depending on demands on accuracy. Keywords: dInSAR, GNSS, levelling, subsidence

(3)

II

SAMMANFATTNING

Sättningar i mark är något som idag är väldigt vanligt. De sker antingen av naturliga orsaker eller skapas av människans påverkan på marken. Idag är den vanligaste förekomsten av sättningar i eller i närheten av stora städer där upptagning av grundvatten samt konstruktion ovan och under mark påverkar marken i så stor utsträckning att skador på byggnader och konstruktioner kan uppstå. För att undvika stora reparationskostnader så är det nödvändigt att upptäcka sättningarna i ett så tidigt skede som möjligt.

Denna studie kommer därför behandla möjligheten att på ett så effektivt sätt som möjligt upptäcka och mäta sättningar med hjälp av dInSAR, GNSS och avvägning. Detta kommer göras genom en litteraturstudie, en sammanställning av andra praktiska fall där de tidigare nämnda metoderna jämförts med varandra samt en kvantitativ jämförelse av data över Stockholm. I den kvantitativa jämförelsen kommer vikten ligga på att jämföra metoderna baserat på RMS fel samt hur väl de linjära approximationerna följer varandra för de olika metoderna.

Litteraturstudien tillsammans med sammanställningen av de praktiska fallen ger en bred bild av metodernas styrkor och svagheter, där dInSARs styrka ligger i förmågan att läsa av stora områden men dess svaghet är att små individuella rörelser inom en stor rörelse inte kan fångas upp. Styrkan för GNSS och avvägning är punktinmätning, där avvägning har den högsta noggrannheten, men också är mest tidskrävande. Den kvantitativa jämförelsen av data bekräftar styrkorna och svagheterna för metoderna då det var tydligt att dInSAR till största del fångar upp samma markrörelse som både GNSS och avvägning. Det som skiljer dInSAR från GNSS och avvägning är amplituden av mätningarna samt det faktum att vissa avvägda mätningar som visar en annan rörelse än den generella inte fångas upp av dInSAR. Slutsatsen som kan dras från detta är att det mest effektiva sätt att upptäcka och mäta sättningar är att till en början grovt granska stora områden för eventuell rörelse med användning av dInSAR och att sedan, där rörelse uppmäts, göra noggrannare punktmätningar med antingen GNSS eller avvägning, beroende på önskad noggrannhet. Nyckelord: dInSAR, GNSS, avvägning, sättningar, markrörelser

(4)

III

ACKNOWLEDGEMENTS

First, I would like to thank everybody at the WSP office for welcoming and supporting me, making the writing of this thesis a great experience. A special thanks to my supervisor Johan Vium Andersson for his time, patience, remarks and guidance throughout the project. Also, Jacob Ekblad for introducing me to the topic of subsidence measurement using dInSAR, helping me formulate the problem and for his constant support.

Finally, I would like to express my sincerest gratitude to my supervisor and teacher at KTH, Milan Horemuz for his time and help during the writing of this thesis. His inputs and comments has been invaluable for the final results.

(5)

IV

1 INTRODUCTION

It is widely known that urbanization has taken place for hundreds of years, in most of the world’s countries. This, among other things, is caused by the industrialism that dragged people towards cities with austerity about work opportunities and more tolerable lives, unlike the hard-working life of the countryside, lined with starvation and disease. Even today, there is a lot of evidence that urbanization continues, Stockholm is no exception. Swedish agglomerations have increased by 550 000 inhabitants in ten years (Svanström, 2015-06-09) and Stockholm specifically has a population increase of 35 000 – 40 000 inhabitants per year (Trafikverket, 2017-01-05).

This urbanization means that housing is required to a greater extent than before and cities are densified, expanded and changed. New buildings and larger infrastructure are required and this can cause major strains on the ground, including subsidence, building deformation and, in worst case, security risks for the inhabitants of the city.

1.1 SUBSIDENCE AND THE NECESSITY OF MONITORING

Subsidence of the ground, in urban areas, is often seen as cracks or destruction of buildings. The cause of the subsidence can be either man made or natural. The man-made sources are common in urban areas and includes ground water extraction, mining, underground construction or, as mentioned above, increased burden on the ground due to construction. The natural sources however, are mostly geological and includes sinkholes, caves and similar phenomena (Zeitoun & Wakshal, 2013, pp. 9-23). Due to the possible effect on built constructions, it is important to monitor the subsidence rate, this to avoid and prevent any safety hazards caused by a possible collapse in the constructions.

Several methods of measuring subsidence rate are available today. All methods have different strengths, weaknesses and preferred areas of usage. The methods being handled in this report is traditional levelling, GNSS (Global Navigation Satellite System) and the relatively new method of SAR Interferometry (Synthetic Aperture Radar), also called dInSAR.

1.2 OBJECTIVES

The purpose of this study is to compare dInSAR, levelling and GNSS to determine the most effective way of detecting and determining subsidence, today and in a future perspective. This is done by a theoretic presentation and comparison with precision levelling and GNSS, a compilation of geographically spread cases that compare the methods and lastly by application and comparison of the methods in data retrieved from different places in Stockholm.

1.3 DISPOSITION

The report starts with Section 2 which is a presentation of previously performed studies within the area of levelling, GNSS and SAR Interferometry, i.e. dInSAR, as well as comparisons between the methods. This is followed by a theoretical description of the methods in Section 3. Section 4 describes the methods and limitations of the two main parts of the work, a literature study and a quantitative comparison of data over Stockholm. Later the results of both the study and the comparison are presented and discussed in Section 5 and 6, respectively. Lastly, in Section 7, a conclusion of the suitability of the different methods for subsidence measurement is drawn and further work on the subject is mentioned.

(8)

7

2 RELATED STUDIES

2.1 PREVIOUS WORK ON THE MONITORING METHODS

As early as 1853, the first Swedish educational book

about levelling, “Handledning uti nivellerings-konsten” by Adolf Helander was released. Since then, levelling has been used, developed and

improved, and publications such as “Modeling interseismic deformation field of North Theran Fault

extracted from precise leveling observation” by Amighpey, Voosoghi and Arabi (2016), “Ground uplift detected by precise leveling in the Ontake earthquake swarm area, central Japan in 2002-2004” by

Kimata et al. (2004) and “Contemporary vertical surface displacement in Yellowstone National Park” by Pelton and Smith (1982) show the usage of levelling for monitoring ground movement.

In 1957, however, a new era begun. The first satellite, Sputnik-1, was launched (Williams, 2017) and this was later going to lead to the usage of GNSS, where GPS was the first system that could be used in positioning, this in the 1970s (Lantmäteriet, 2017). Today, with GNSS being used worldwide, countless research, articles and reports have been done on the subject, and “Monitoring large-area

mining subsidence by GNSS based on IGS stations” by Bian, Zhang, Zhang and Zheng (2014),

“Subsidence and current strain patterns on Tenerife Island (Canary Archipelago, Spain) derived from

continuous GNSS time series (2008-2015)” by Sánchez-Alzola, Martí, García-Yeguas and Gil (2016),

“Preliminary results of land subsidence monitoring project in Konya Closed Basin between 2006-2009

by means of GNSS observations” by Ustun, Tusat and Yalvac (2010) and “Monitoring the tectonic plate movement in Turkey based on the national continuous GNSS network” by Uzel et al. (2013) is only a

few examples of publications studying ground monitoring with GNSS measurements.

In the 1990s, radar images enabling SAR Interferometry, and thereby dInSAR, began to be collected and in connection with that, research on the area increased (Alessandro Ferretti, Monti-Guarnieri, Prati, & Rocca, 2007). Much research has been done on the subject since then where research teams in Italy, China and the Netherlands has been recurring during the research of this work.

2.2 PREVIOUSLY MADE COMPARISONS BETWEEN THE METHODS

Several studies comparing the methods have been performed. The comparisons are focused on different types of ground movement, such as landslides, earth quakes or subsidence caused by other natural or man-made phenomena. A selection of studies focused on subsidence with different causes is presented in the Result section. The compilation includes studies from Finland, Spain, Germany, Iran, Indonesia, China, Turkey and France and for each of the studies the conclusion from is presented and discussed.

3 THEORETICAL DESCRIPTION OF THE METHODS

3.1 DINSAR

In this Section, Section 3.1, the method of dInSAR will be handled. To begin with, in Section 3.1.1, the basic of SAR Interferometry and data acquisition will be handled and after that, in 3.1.2, the error sources that commonly occur during data acquisition will be handled. Later, in Section 3.1.3, the basic method for data processing using the dInSAR method is presented and here the elimination or mitigation of the error sources is mentioned and lastly, in Section 3.1.4, the error sources for interpreting the processed data is handled.

(9)

8

3.1.1 SAR

INTERFEROMETRY

SAR is based on the RADAR principle of using an active sensor that sends out radio signals in pulses that reflects of the ground and travels back to the sensor. The advantage of using RADAR instead of optical images is that it is not limited by clouds or lack of sunlight. SAR is as mentioned, a type of RADAR sensor called Synthetic Aperture Radar, which uses a synthetic antenna, which is created by a combination of stable data signals(Hanssen, 2001, pp. 10-11).

The RADAR sensors are mounted on satellites which, depending on the type of satellite, orbits at a height of 600-800 m above ground. (European Space Agency, Accessed: 2017-06-08a, Accessed: 2017-06-08b) When traveling along the orbit, the satellite scans the area beneath it and saves the information as images. These images consist of a grid of pixels associated with an area on the ground, called the resolution cell. Each pixel contains a complex value that represents the information about the phase and amplitude, i.e. the energy strength of the reflected signal from its associated resolution cell. The amplitude changes depending on the character of the resolution cell, and is high in urban areas with a high number of hard, reflective surfaces, as where the amplitude of vegetated areas is close to zero, due to the reflection being scattered in the volume of the vegetation instead of reflected to the instrument on the satellite. The phase can be considered to be a sinusoidal signal with a known wavelength and is one of the key factors to interferometry, allowing the measurement of small height differences (Alessandro Ferretti et al., 2007, pp. A6-A8).

The fundamental idea of interferometry is to measure the phase difference between SAR images acquired at different times. Due to the phase being a measurement of distance, this would, in theory, represent the change in height of the point reflecting the signal (A. Ferretti, Prati, & Rocca, 2001, p. 8). Unfortunately, reality is a bit more complex. The complexity lies in the phase being affected by different error sources (Tomás et al., 2014, p. 6). This to the extent that a comparison of the phases without any corrections of the errors would result in such an inaccurate result that it would be unusable. Therefore, different methods have been developed to exclude the error sources from the phase measurement and by that get significantly more accurate results. The two main categories in interferometry are based on the reflectance in the resolution cell. If there is one scatterer, i.e. target of reflection, that is significantly stronger than all other scatterers, one could say that the cell has a persistent or permanent scatterer, this due to the fact that these targets often remain and does not change over time. The other case is when the resolution cell consists of scatterers that have approximately the same reflectance, and the scatterers are distributed (Maljaars, 2017-03-31; Sansosti, Casu, Manzo, & Lanari, 2010, pp. 2-3).

Over time the methods of excluding errors from the phase have morphed into different methods, formed to work with the different types of reflection, but in this report only the general method of the differential interferometry of SAR acquisitions will be presented. The general method is a description of the dInSAR method that includes all different types of methods.

3.1.2 T

HE MAIN ERROR SOURCES OF RETRIEVING DATA

The application of SAR interferometry that is of interest in this study is deformation monitoring and all other effects on the phase is considered an error source. 3.1.2.1 IMPACT OF THE ATMOSPHERIC CHANGE When being sent out from the instrument, the signal passes through both ionosphere, which with its dry gasses affects the signal, and the troposphere. In the troposphere the humidity, temperature and pressure is inconsistent both in time and space. This along with local weather changes, the atmospheric

(10)

9 effect on two signals acquired at different times will differ significantly (Alessandro Ferretti et al., 2007, p. A24). The atmosphere´s change in pressure and level of water vapor will directly affect the phase by slowing it down. Due to water vapor being such a big influence on the signal, acquisitions during night is often preferred prior to acquisitions during the day, due to the air being cooler and having less ability to hold water vapor (Hanssen, 2001, p. 45). 3.1.2.2 IMPACT OF ORBITAL INCONSISTENCY

When monitoring deformations using differential interferometry, a great amount of SAR images, separated in time, is key; this is enabled by repeat-pass acquisition of the images. Unfortunately, the orbits of the satellites differ for every time it passes over the area. This means that the angle from which the image is acquired, i.e. the look angle, along with the distance from the satellite to the target will change. This leads to a non-real elongation or foreshortening of the phase signal that is different between acquisitions. The effect of the inconsistency of the satellite orbits can be of the magnitude between tens and hundreds of meters (A. Ferretti et al., 2001, p. 9). 3.1.2.3 IMPACT OF MULTIPATH

Multipath can be explained as the signal being a combination of all reflected signals within the resolution cell, and thereby the inability to decide from where the signal reflects (Alessandro Ferretti et al., 2007, p. B64). This problem affects the two main categories, persistent and distributed scatterers, differently. If using the persistent scatterer technique, multipath error results in the inability to determine where the dominant scatterer is located in the resolution cell. If the distributed scatterer technique is used, multipath impact will lead to the inability to determine what the combination of signals is composed of (Kampes, 2006, p. 8).

3.1.2.4 IMPACT OF THE TOPOGRAPHY

Since the measurement is performed on the signal that reflects of the surface of the earth, the topography’s form will affect the distance between the satellite and the ground target. When monitoring deformations, the height is not of interest and is therefore removed during the generation of the differential interferograms, see “The general method for interferometric processing” below (Alessandro Ferretti et al., 2007, p. A23; Tomás et al., 2014, p. 6). This causes an error due to the elevation model not being accurate enough to completely eliminate the topography. The reason for the inaccuracies in the elevation model is due to the reflected signal being a combination of all returned signals, called multipath, and placing the exact signal on the exact location in the resolution cell is therefore impossible (A. Ferretti, Prati, & Rocca, 2000, p. 2202; Kampes, 2006, p. 8).

(11)

10

3.1.3

D

I

N

SAR

DATA PROCESSING

FIGURE 1: THE GENERAL FORMATION OF DIFFERENTIAL INTERFEROGRAMS.

During the dInSAR process many of the earlier mentioned error sources are removed or mitigated. The atmospheric effect on the phase signal is detected and later filtered out, this due the fact that it changes characteristics over time and looks different in every image used in the analysis (A. Ferretti et al., 2000, p. 1). Any topography artifacts is mitigated due to redundancy in the space-time data and a compensation of the orbital errors is made when creating the differential interferograms, where the data for the specific orbital path is subtracted (Sansosti et al., 2010, p. 3). One error source that, at this stage in the development of the technique, cannot be resolved is the multipath error. There is no way knowing how many times the signal has bounced during the traveling from the ground to the sensor, e.g. the signal can reflect of a road, bounce via a wall, a car, a rooftop or similar (Kampes, 2006, p. 8). This can result in different responses and signal strength for the pixel by only a small change in the angle from which the signal is sent out from the satellite, i.e. the orbital changes.

After eliminating and mitigating all possible error sources, the precision of a deformation rate determined by dInSAR measurements would be approximately 1 mm/year (Chen, Xu, Li, Li, & Li, 2015, p. 5; Lanari, Casu, Manzo, & Lundgren, 2007, p. 20).

3.1.4 T

HE MAIN ERROR SOURCES OF INTERPRETING DATA

The final data set that is obtained from the calculations represents locations where the height changes over time is determined. The density of the captured locations depends on the character of the ground, whereas a character resulting in an amplitude close to zero will not be captured. If looking at a large dataset over an urban area, the center of the dataset will have a lower amount of vegetation than the areas in the outer regions. This results in a higher amount of captured locations in the center of the dataset.

Since the captured locations are used as reference points of the calculation of subsidence and subsidence rate, the frequency of the locations is important, where the larger amount of points yields an over estimation of the subsidence rate.

(12)

11

3.2 LEVELLING

Levelling is a well-known, optical method of measuring height differences between established benchmarks (Ketelaar, 2009, p. 15). The area different accuracy levels of the measurements, but from here on forward, precise levelling is referred to when the term levelling is used. The measurements are performed by measuring the height difference between a known, fixed, point and an unknown point of interest, as seen in Figure 2. If the unknown point is far from the known point, the measurements can be performed in a train. This will result in an absolute height of the points of interest. It is not until at least two measurements have been performed that a comparison of height differences can be made (Avdelningen för geodesi och satellitpositionering, 2015, p. 40). FIGURE 2: FIGURE SHOWING HOW LEVELLING IS PERFORMED.

3.2.1 E

RROR SOURCES AND CORRECTIONS

Although the technique seems straight forward, and the instruments are assumed to be controlled and corrected, there are some limitations of the method and several error sources affecting the measurements. The biggest limitation of the method is the inability to measure long distances in one measurement. This since the method is optical and therefore having a maximum distance between the rods of 100 m (Harrie et al., 2013, p. 101), but the normal distance used between instrument and rod is 60-80 m-. The error sources can be divided into systematic and random errors. The systematically occurring errors are refraction, rod scale and collimation error. Both affects the sight axis of the instrument, refraction by the variations in the density in the atmosphere near the earth surface which bends the sight axis (Harrie et al., 2013, p. 106) and collimation error by the sight axis not being completely horizontal (Engeltoft, 1996, p. 7). The collimation error is only noticeable when the instrument is not centered between the rods. Both these errors can be mitigated, the refraction error by not measuring any closer to the ground than 50 cm and the collimation error by calibration of the instrument, called peg test.

(13)

12

FIGURE 3: COLLIMATION ERROR WITH CENTERED INSTRUMENTS (LEFT) AND NON-CENTERED INSTRUMENTS (RIGHT).

FIGURE 4: REFRACTION ERROR IN LEVELLING.

The error sources that can be considered random is errors sources that is unpredictable and does not necessarily occur for every measurement. These errors are often hard or impossible to correct and will therefore affect the measurements, unlike the systematical errors. Some of the random errors is generated by an external force, such as vibrations, gradual subsidence of the arrangement in soft ground (Engeltoft, 1996, p. 13) or other movement caused by a disturbed environment. Other sources of the errors are caused inside the instrument, such as temperature changes that causes expansion or contraction in the metal and optic in the instrument. When using levelling to determine heights, a demand on the precision per measured distance is set and if the result of the measurement does not meet the requirements, it is redone. The precision is dependent on both the length of the train that has been measured and the precision of the instrument. For precise levelling the requirement is 1 mm/km1/2_{or less (Lantmäteriet, 2014, p. 64).} The precision of the height difference in a point is based on a combination of the precision of the different measurement that is the basis of the comparison of height difference. This does not necessarily have to be the same as the precision per measured distance but can still be on a sub-mm level (Avdelningen för geodesi och satellitpositionering, 2015, p. 42).

(14)

13

3.3 GNSS

GNSS, like dInSAR, uses phase measurements from satellites. That, however, are the only similarities regarding the techniques. There are several types of GNSS measurements and many different systems that falls under the term GNSS, such as GPS, GLONASS, Galileo among others (Hofmann-Wellenhof, 2008, pp. 309-365). The different types of measurements use instruments which are more or less accurate and are adapted to fit different types of requirements such as simple navigation to high accuracy measurements of landslides or deformation in constructions such as bridges and houses. The most accurate type of measurement is the static relative positioning. Here several receivers are used by being placed over points where they are static during the session. To determine a position using GNSS, the receiver must be connected to at least four different satellites. This to measure four pseudoranges simultaneously and thereby determine the position in 3D and a clock error (Hofmann-Wellenhof, 2008, p. 8). Leick, Rapoport and Tatarnikov (2015) describe the details of how GNSS measurements are performed well in “GPS Satellite Surveying”, and due to that being one of many publications handling this subject, it will not be described any further in this work.

3.3.1 E

RROR SOURCES AND CORRECTIONS

During good conditions, static relative positioning can generally provide coordinates with a precision of approximately 3 mm, (Leica Geosystems, 2016) but if there are too many error sources affecting the measurement it will be significantly lower. Errors that are affecting the GNSS measurement are those related to the satellite, such as orbital errors and clock error, those related to the signal propagation, such as atmospheric errors and multipath error and lastly those errors that are related to the receiver. Using the formation with several receivers and performing a dual frequency measurement, error sources related to the satellite and atmosphere are eliminated (Hofmann-Wellenhof, 2008, p. 9).

The only error that is hard to remove is, as in the method of dInSAR, the multipath error that is unpredictable and differs in every measurement. It is reduced by using special antennas on the receiver that blocks the bouncing signals and by trying to choose a location with the least amount of reflective objects (Harrie et al., 2013, p. 169).

4 METHODOLOGY

4.1 LITERATURE STUDY

A literature study of published sources was made to describe the methods and to theoretically compare them. Articles, books and reports have been gone through and the important information for the subject has been handpicked to present the methods and the comparisons. The chosen literature was primarily scientific and from sources that originated from different authors and geographical places. To begin with, the three methods used to determine height differences could have been described more in depth. However, this has not been done since it is not required to understand the comparison between the data and the best applications of each method. Other limitations are that the results and conclusions of the concluded cases does not account for a result that is comparable. Not all reports

(15)

14

give a numeric result and not all numeric results that are comparable. This will be mentioned in the result section as well.

4.2 QUANTITATIVE COMPARISON OF DATA IN STOCKHOLM

To be able to perform a quantitative comparison a common denominator of all datasets must be determined and the general trend of all datasets over the same time span needs to be examined. RMS (Root Mean Square) error is suitable as a common denominator due to it describing a standard deviation between values. It is most commonly used to statistically determine how much a set of measurements deviates from the mean value of the set. In this case, and as seen in formula 1, it is used to compare how well the result of measurements of the same object with two different methods cohere. Assuming that one of the methods generate results that are correct and free from errors, RMS error can be used to determine how trustworthy the results from the other method is. This due to the deviation of the second method showing the error with which it has measured the common object. 𝑹𝑴𝑺_{𝒓𝒆𝒍𝒂𝒕𝒊𝒗𝒆} = 𝑵𝒊0𝟏 𝒙𝒊-𝒙𝒊 𝟐 𝑵 1 where:

𝑥₅ = the height change of point 𝑖 between the current time epoch and the reference time epoch, measured with method 1, equal to ∆ℎ = ℎ9− ℎ;<=, 0 < 𝑗 ≤ 𝑛

𝑥5 = the height change of point 𝑖 between the current time epoch and the reference time

epoch, measured with method 2, equal to ∆ℎ = ℎ9− ℎ;<=, 0 < 𝑗 ≤ 𝑛

𝑁 = the number of measured points

However, the methods used in this thesis work is not errorless and therefore, the RMS that is relative two methods, equation 1, contains error sources caused by both methods, seen in equation 2, and to determine how trustworthy one of the methods is, the errors from the other method must be removed. 𝑹𝑴𝑺_{𝒓𝒆𝒍𝒂𝒕𝒊𝒗𝒆}𝟐 = 𝑹𝑴𝑺_{𝒎𝒆𝒕𝒉𝒐𝒅 𝟏}𝟐 + 𝑹𝑴𝑺_{𝒎𝒆𝒕𝒉𝒐𝒅 𝟐}𝟐 2 Due to the data obtained from dInSAR measurements containing very large amounts of points, a mean value of the subsidence for all points in the vicinity of a point of another dataset is determined. The mean value of the subsidence is later used to calculate the RMS error between dInSAR and the other dataset. The character of the different data sets differs greatly, where the GNSS data is in singular points that are widely spread, the levelling data is spread in clusters and the dInSAR data is, as mentioned, a large amount of closely placed points. To enable the determination of the RMS error and the general trend, several selections and calculations had to be made. A general workflow is presented as a flowchart in Figure 5, below. The key step is the calculation of the RMS error, which was done according to equation 1 and then the individual RMS error of GNSS was removed from that result, using the relationship between the relative and individual RMS error presented in equation 2.

The subsidence trends from the methods were estimated and later, the trends from the different methods used over the same area were presented in the same graph with a common time scale.

(16)

15 FIGURE 5: THE WORKFLOW OF THE QUANTITATIVE COMPARISON OF DATA.

4.3 DELIMITATIONS AND LIMITATIONS OF THE THESIS

This thesis work is delimited in several ways, mostly due to time limitations, resulting in room for a lot of further work on the area. This will be mentioned in the section “Conclusions and further work”. Some of the general limitations are the source of the height changes. There are several sources such as subsidence, landslides, any movement caused by the tectonic plates, climate change, man-made destruction such as war destruction or similar. Here the main focus is on subsidence, and the other sources are not handled.

(17)

16

4.3.1 L

ITERATURE SELECTION

–

DELIMITATIONS IN THE LITERATURE STUDY

The publication ”InSAR Principles: Guidelines for SAR Interferometry Processing and Interpretation” by Ferretti et al. (2007) lacks some of the features that is desired to identify a published work as scientific, this is most likely due to it being a guideline. Despite that, it is used as a source in the study, explaining the InSAR methodology. The source is used due to it sharing the authors as many of the scientific works published and it being published by ESA, the European Space Agency, which is an international organization with 22 member states (European Space Agency, 2017-01-24). Regarding the sources of the compilation of earlier performed comparisons of the techniques InSAR, Levelling and GNSS, they can seem to be secondary sources. But this is due to the interest laying in the comparison between the two methods and not the methods themselves. Authors such as A. Ferretti, P. Berardino and R. Lanari is recurring in several writings used as sources in this study to describe the different methods of using InSAR. They are all based in Italy, but at different institutions (Berardino, Fornaro, Lanari, & Sansosti, 2002; A. Ferretti et al., 2001) and due to them describing the methods in a good way, and the theoretical conclusions being denied or confirmed by real cases from widely spread locations.

4.3.2 L

IMITATIONS OF THE QUANTITATIVE COMPARISON

There are several limitations of the quantitative comparison, case Stockholm. One of the limitations with the greatest impact on the result is the differing character of the datasets, leading to the comparison of points that are not in the exact same location but only arbitrary and that points measured by levelling is measured on ground level while dInSAR and GNSS is measured on rooftops in urban areas. Another limitation of great impact is the amount of data, which due to time constraints, had to be limited. A more comprehensive study with larger amount of data could give a more accurate and reliable result. When determining the RMS, measurements performed on the same dates are compared. This works well with GNSS and dInSAR, but due to the lower frequency of the levelling measurements, RMS will not present a good result to show how much dInSAR and levelling differs from each other. Therefore, it was not included in this comparison.

5 RESULTS

5.1 THEORETICAL RESULTS

The theoretical results consist of both the theoretical comparison of the methods characteristics and the compilation of publications that has compared the methods in real cases. This to give an idea of how the methods work, what their greatest strengths and weaknesses are.

5.1.1 T

HEORETICAL COMPARISON OF D

I

N

SAR,

L

EVELLING AND

GNSS

TABLE 1: A SHORT THEORETICAL COMPARISON OF THE CHARACTERISTICS OF THE METHODS DINSAR, LEVELLING AND

GNSS. dInSAR Levelling GNSS Resources At least 20 SAR images and the software that enables the processing of the images The instruments to perform the measurement, including at least two persons Instruments to perform the measurements, access to a SWEPOS reference point and

(18)

17 working with it and software to process the data software to process the data Ability of usage Cannot be used in vegetated or agricultural areas Demands optical visibility between instrument and rod Needs connection with at least four satellites, which means that usage in forested areas and among high constructions in urban areas could be limited Temporal 3-35 days between acquisitions, for an ESR satellite, but archived data can be used as well, processing takes up to a few days (European Space Agency, 2017) The measurements are performed manually and is limited by height changes and the number of and distance between points. A rough estimation is that 20-50 points could be measured in one day Depends on the number of instruments available, which can be automated by permanently placing instrument on places of interest. Another factor is the length of the session, which also affects the accuracy and therefore differs. Spatial Areas in the size of

hundreds of km2 Lines in the size of tens _of km

Several points, depending on the number of instruments Location of measurement On rooftops On height fixes on the ground On specific points on rooftops Resulting data Height differences between epochs Height differences relative known points Points with known location Detected

displacement Line of sight Height

3D points in a global reference system

Achievable

accuracy Millimeter level Sub-millimeter level

Centimeter to millimeter level Future Due to its ability to measure on a distance, it will be increasingly important for inaccessible areas Will continue to be the most accurate method Development will lead to higher accuracy and ability to automate the process of measurement. Therefore, it will be available at a lower cost and the everyday use will be increased

5.1.2 C

OMPILATION OF QUANTITATIVE COMPARISONS OF REAL CASES

Based on the theoretical comparison, it is clear that all methods have their specific weaknesses and strengths. Many studies have been performed where two or more methods have been compared, some of them are presented in Table 2. Several of the presented cases have drawn the conclusion that a combination of the methods would result in a more accurate result than what could be acquired by one method alone (Abidin et al., 2016, p. 441; Hastaoglu et al., 2017, p. 14). However, the combining of the methods is nontrivial. One of the most prominent problems is the combing of datatypes, this

(19)

18

due to GNSS measuring 3D points in a global reference system and dInSAR measuring height displacement along the line of sight. When transforming data into another type, for example 3D to 1D, some data may be lost or affected (Hastaoglu et al., 2017, p. 10).

TABLE 2: A COMPILATION OF RESULTS OF COMPARISONS BETWEEN INSAR, LEVELLING AND GNSS PERFORMED IN

DIFFERENT CASES. Information about the specific case Results Building subsidence in Turku, Finland Included methods: Levelling, InSAR Timespan for measurements InSAR May 1992 – August 2005 Levelling 1990 – 2003, test scene 2005

Levelling and InSAR follows the same trends but on average, the subsidence measured with the InSAR method is 1.0 mm slower per year than the subsidence measured with the levelling method. This could be caused by non-linearity of the subsidence of the buildings due to renovations or due to high levels of noise affecting the InSAR measurements (Karila et al., 2013, pp. 810-811). Subsidence due to underground construction in Barcelona, Spain Included methods: Levelling, InSAR Timespan for measurements InSAR March 2013 – October 2014 Levelling March 2014 – May 2016 Levelling and InSAR both captured the instant subsidence and differed only by the magnitude, which for InSAR was 5 mm higher for the average subsidence and about 3-6 mm higher for the maximum subsidence.

The difference in measurements most likely are caused by the fact that the reference points of the levelling was affected by subsidence as well. This leading to measurements showing subsidence rates lower than what took place. Considering that together with the building characteristics, the difference between the magnitude of the measurements can be corrected, and therefore lowered to approximately 2 mm (Serrano-Juan, Pujades, Vazquez-Sune, Crosetto, & Cuevas-Gonzalez, 2017, pp. 6-8). Uplift due to geothermal drilling in Staufen in Breisagu, Germany Included methods: Levelling, InSAR Timespan for measurements InSAR July 2008 – July 2011 Levelling January 2018 – November 2011

For most measurements, the results follow the same trend but the magnitude differs by the InSAR results being both smaller and greater than the levelling results. The difference between the results are 2.4 cm at most and is most likely caused by horizontal movement. This movement affects only the InSAR measurement and this by increasing or decreasing the magnitude of the measurement (Lubitz, Motagh, Wetzel, & Kaufmann, 2013, p. 3093). Subsidence due to over extraction of groundwater in Mashhad Valley, Iran Included methods: Timespan for measurements InSAR June 2003 – November 2005 Levelling 2002 – 2005 GNSS 2005 – 2006

(20)

19

Levelling, InSAR, GNSS The location is one of the places with fastest subsiding landmasses in

the world. With regards to the average subsidence rate, GPS measurements differs the most from the other results, but looking at the maximum rate, InSAR and GPS is very similar. The reason for this result is the fact that the time span for the measurements do not overlap (Motagh et al., 2007, pp. 520-523). Subsidence due to construction in Jakarta, Indonesia Included methods: Levelling, InSAR, GNSS Timespan for measurements InSAR 1993 – 1998 Levelling 1991 – 1997 GNSS 1997 – 2002

The average subsidence rate of levelling and InSAR differs by approximately 4-9 cm during the time of measurement. For the next time span of measurement InSAR is compared to GPS and here InSAR is within the given interval of the average subsidence. These results show a relatively good correspondence of all three methods (Abidin et al., 2005, pp. 561-566). Subsidence of the Bohai Building and the China Theater, China Included methods: Levelling, InSAR Timespan for measurements InSAR May 2014 – December 2014 Levelling May 2014 – December 2014

For the Bohai Building the biggest difference between levelling and InSAR is 1.95 mm while the maximum difference of the China Theater is 2.36 mm. Both indicating an accuracy of approximately 1 mm for InSAR measurements. The high accuracy is due to the usage of corner reflectors on the buildings (Yang, Yan, Huang, Chen, & Wu, 2016, pp. 11-12). Landslides in Koyulhisar, Turkey Included methods: InSAR, GNSS Timespan for measurements InSAR 2006 – 2008 GNSS 2006 – 2008 When measuring landslides, GPS shows every detail of the movement, for the points monitored, while InSAR, due to being bound to one direction, shows the general movement of the main landslide while local movement is missed. Here the correlation between InSAR and GPS was as low as 10%, caused by earlier mentioned problems (Hastaoglu et al., 2017, pp. 10-11). Creeping and subsidence due to mining in Vauvert, France Included methods: Levelling, InSAR Timespan for measurements InSAR January 1993 – March 1999 Levelling 1995 – 1998 This area is somewhat vegetated with agricultural fields and therefore a lower accuracy of the InSAR method is expected. On the other hand, a lot of effort was put into filtering it out. The InSAR measurements stretched over a longer timespan than the levelling measurements but both methods agreed that the subsidence rate of the area was higher than 2 mm per year. This result was attested by a RMS error of 0.198 cm per year, which means that the standard deviation between the measurements are relatively low, intending a good result. (Raucoules, Carnec, Mouelic, King, & Maisons, 2003, pp. 2939-2941).

(21)

20 Subsidence due to groundwater extraction, construction and tectonic activities in Jakarta, Indonesia Included methods: Levelling, InSAR, GNSS Timespan for measurements InSAR 2005 – 2011 Levelling 1982 – 2011 GNSS 1997 – 2011 The time span of the different methods does not overlap completely, as only InSAR and GPS overlap and levelling being performed previously. Despite that, the results are similar for all methods, except for the maximum subsidence rate of InSAR, which is half the rate of the maximum rate measured with GPS and levelling (Abidin et al., 2016, p. 438). Subsidence due to groundwater extraction, construction and tectonic activities in Bandung, Indonesia Included methods: InSAR, GNSS Timespan for measurements InSAR 2007 – 2011 GNSS 2000 – 2011 The measurements were performed during the same time period and gives a result of very similar magnitudes for average, minimum and maximum rate (Abidin et al., 2016, p. 438). Subsidence due to groundwater extraction, construction and tectonic activities in Semarang, Indonesia Included methods: Levelling, InSAR, GNSS Timespan for measurements InSAR 2007 – 2011 Levelling 1999 – 2011 GNSS 2008 – 2011 The time periods of the different methods are only overlapping with about a year, but a greater overlap of InSAR and GPS is available. Here the InSAR result for maximum rate is approximately half the magnitude as of GPS and levelling and the typical rate is slightly lower (Abidin et al., 2016, p. 438). Looking at the different cases, almost all cases shows that dInSAR measurements coincided with both levelling and GNSS. The common denominator for these cases is that the cause for ground movement was man-made, and therefore somewhat uniform. However, the case in Turkey, comparing dInSAR and GNSS on measurements of a landslide showed zero to none coherence. This explained by the landslide being a complicated movement and dInSAR only capturing the large movement of the landslide while the point measurements performed by GNSS captured all small individual movements of the points.

5.2 QUANTITATIVE COMPARISON – CASE STOCKHOLM

Due to confidentiality of some of the data used in the comparison, no data will be presented with exact location, instead, the datasets will be numbered.

5.2.1 GNSS

COMPARED TO D

I

N

SAR

The comparison between GNSS and dInSAR included both calculation of RMS error and the analysis of the linear regression of the data. In the comparison of GNSS and dInSAR data, only dInSAR data calculated with the Persistent scatterer method was used due to the lack of arbitrary points between the GNSS data and the dInSAR data calculated with the Distributed Scatterer method.

As mentioned in the Method section, the RMS error between GNSS and dInSAR contains noise and error sources from both GNSS and dInSAR and to obtain the individual RMS of dInSAR, the RMS of GNSS must be eliminated. The RMS error of the GNSS measurements used in this case is 3 mm (Lantmäteriet, 2015, p. 82), which gives the following result of the RMS of the dInSAR measurement.

(22)

21

TABLE 3: THE RESULTS FOR DETERMINING THE RELATIVE AND INDIVIDUAL SQUARED RMS FOR GNSS AND DINSAR.

Dataset Relative RMS2 _{Individual RMS}2_GNSS _{Individual RMS}2_dInSAR

1 41.9 mm 9.0 mm 32.9 mm

2 30.9 mm 9.0 mm 21.9 mm

3 34.0 mm 9.0 mm 25.0 mm

TABLE 4: THE RESULT FOR DETERMINING THE RELATIVE AND INDIVIDUAL RMS FOR GNSS AND DINSAR. Dataset Relative RMS Individual RMS, GNSS Individual RMS, dInSAR

1 6.5 mm 3.0 mm 5.7 mm 2 5.6 mm 3.0 mm 4.7 mm 3 5.8 mm 3.0 mm 5.0 mm The height changes were individually represented in graphs for all datasets (see Appendix A) and the linear regression of the data was graphically presented in a graph over the same time span.

DIAGRAM 1: THE COMPARISON OF THE LINEAR REGRESSION FOR DATASET 1 OF GNSS AND DINSAR.

-0,007 -0,006 -0,005 -0,004 -0,003 -0,002 -0,001 0 2011 September 2016 August Su bs id en ce [m ]

Linear regression - Dataset 1

L GNSS L dInSAR

(23)

22

As seen in Diagram 1, 2 and 3, the height changes measured by GNSS and dInSAR does not resemble each other. Instead the GNSS data shows a rise of the ground while the dInSAR data shows subsidence for the same areas.

After this the dInSAR data was examined more in depth. By looking at the measurements of locations that is nearby the area of interest and that is known to be fixed without any movement, an overestimation of the subsidence rate of the dInSAR dataset could be determined to approximately 0.5 mm per year, for all three locations. Due to the points being without any movement and dInSAR still measuring as moving, then this can be considered faulty, and can be removed from the datasets. The resulting, and final, comparison of the datasets is presented in Diagram 4, 5 and 6. -0,004 -0,003 -0,002 -0,001 0 0,001 0,002 0,003 0,004 0,005 2013 July 2016 August Su bs id en ce [m ]

Linear regression - Dataset 2

L GNSS L dInSAR -0,006 -0,005 -0,004 -0,003 -0,002 -0,001 0 0,001 0,002 0,003 0,004 2014 February 2016 August Su bs id en ce [m ]

Linear regression - Dataset 3

L GNSS L dInSAR

(24)

23

DIAGRAM 4: THE COMPARISON OF THE LINEAR REGRESSION FOR DATASET 1 OF GNSS AND DINSAR, WHERE A REDUCTION OF LOCAL EFFECTS HAS BEEN MADE IN THE DINSAR DATA.

-0,007 -0,006 -0,005 -0,004 -0,003 -0,002 -0,001 0 0,001 0,002 2011 September 2016 August Su bs id en ce [m ]

Corrected linear regression - Dataset 1

L GNSS L dInSAR -0,003 -0,002 -0,001 0 0,001 0,002 0,003 0,004 0,005 2013 July 2016 August Su bs id en ce [m ]

Corrected linear regression - Dataset 2

L GNSS L dInSAR

(25)

24

5.2.2 L

EVELLING COMPARED TO D

I

N

SAR

The comparison between levelling and dInSAR included only the analysis of the graphic representation of the data and the graphic representation of the linear regression of the data. In the comparison of levelling and dInSAR data, only dInSAR data calculated with the Persistent scatterer method was used due to the lack of arbitrary points between the levelling data and the dInSAR data calculated with the Distributed Scatterer method. The height changes were individually represented in graphs for all datasets (see Appendix B) and the linear regression of the data was graphically presented in a graph over the same time epoch.

DIAGRAM 7: THE COMPARISON OF THE LINEAR REGRESSION FOR DATASET 1 OF LEVELLING AND DINSAR.

-0,005 -0,004 -0,003 -0,002 -0,001 0 0,001 0,002 0,003 0,004 2014 February 2016 August Su bs id en ce [m ]

Corrected linear regression - Dataset 3

L GNSS L dInSAR -0,005 -0,004 -0,003 -0,002 -0,001 0 0,001 0,002 0,003 0,004 2012 April/May 2016 August/September Su bs id en ce [m ]

Linear Regression - Dataset 1

(26)

25

-0,004 -0,0035 -0,003 -0,0025 -0,002 -0,0015 -0,001 -0,0005 0 2012 December 2015 May Su bs id en ce [m ]

Linear Regression - Dataset 2

L1 Levelling L4 Levelling L dInSAR

-0,008 -0,007 -0,006 -0,005 -0,004 -0,003 -0,002 -0,001 0 0,001 2011 November 2016 October Su bs id en ce [m ]

Linear Regression - Dataset 3

L1 Levelling L dInSAR -0,012 -0,01 -0,008 -0,006 -0,004 -0,002 0 0,002 2011 September 2016 October Su bs id en ce [m ]

Linear Regression - Dataset 4

(27)

26

As seen in Diagram 4-8, the height changes measured by levelling and dInSAR resembles each other. Also clear is that the subsidence rate of dInSAR is higher, resulting in a larger final subsidence, and that small movements, occurring in only some points is not captured by dInSAR.

After this the dInSAR data was examined more in depth. By looking at the measurements of locations that is nearby the area of interest and that is known to be fixed without any movement, an overestimation of the subsidence rate of the dInSAR dataset could be determined. Due to the points being without any movement and dInSAR still measuring as moving, then this can be considered faulty. The faulty subsidence rate was 0.5 mm per year, 0.2 mm per year, 0.8 mm per year, 0.4 mm per year and 0.7 mm per year, respectively for dataset 1-5. After the removal of the faulty subsidence rate, the resulting, and final, comparison of the datasets is presented in Diagram 12-17.

DIAGRAM 12: THE COMPARISON OF THE LINEAR REGRESSION FOR DATASET 1 OF LEVELLING AND DINSAR, WHERE A REDUCTION OF LOCAL EFFECTS HAS BEEN MADE IN THE DINSAR DATA.

-0,014 -0,012 -0,01 -0,008 -0,006 -0,004 -0,002 0 2011 November 2015 November/December Su bs id en ce [m ]

Linear Regression - Dataset 5

-0,003 -0,002 -0,001 0 0,001 0,002 0,003 0,004 2012 April/May 2016 August/September Su bs id en ce [m ]

Corrected linear regression - Dataset 1

(28)

27

-0,004 -0,0035 -0,003 -0,0025 -0,002 -0,0015 -0,001 -0,0005 0 2012 December 2015 May Su bs id en ce [m ]

Corrected linear regression - Dataset 2

-0,0035 -0,003 -0,0025 -0,002 -0,0015 -0,001 -0,0005 0 0,0005 0,001 2011 November 2016 October Su bs id en ce [m ]

Corrected linear regression - Dataset 3

L1 Levelling L dInSAR

(29)

28

6 DISCUSSION

6.1 LITERATURE STUDY

As mentioned, the theoretical description and comparison of the methods dInSAR, GNSS and levelling, it is clear that they differ greatly and that they have specific weaknesses and strengths.

In the case of subsidence monitoring, the obvious strength of levelling is the accuracy that cannot be met by any other method. The method has been used for a long time and corrections for several error sources affecting the measurements has been developed, resulting in measurements reaching an

-0,009 -0,008 -0,007 -0,006 -0,005 -0,004 -0,003 -0,002 -0,001 0 0,001 0,002 2011 September 2016 October Su bs id en ce [m ]

Corrected linear regression - Dataset 4

L1 Levelling L6 Levelling L8 Levelling L dInSAR

-0,012 -0,01 -0,008 -0,006 -0,004 -0,002 0 2011 November 2015 November/December Su bs id en ce [m ]

Corrected linear regression - Dataset 5

(30)

29

accuracy of sub millimeter. The combination of the short geographic distance between measurements and the possibility to correct the measurements on such a level is the reason of the high accuracy in the measurements. However, the short distance between the measurements is also the biggest limitation of the method since it prevents quick measurements of larger areas. A method that is not limited by distance in the same way as levelling is GNSS, which only condition to determine a three-dimensional location is connection with at least 4 satellites. The cost of being able to perform measurements over larger areas is the level of accuracy that is possible to be achieved. GNSS measurements is affected by several other error sources than levelling, which of some is not possible to correct and eliminate completely with today’s technology. Even if GNSS can perform measurements without distance limitations, the measurements only generate one point for every measurement. To be able to measure several points over a large area dInSAR is the favorable method. This due to its ability to measure many points over a large spatial area. The accuracy of dInSAR measurements is similar to the measurements performed by GNSS, depending on the character of the measured area. dInSAR’s obvious limitation is the inability to perform measurements over heavily vegetated areas. Since the one methods strength being another methods weakness, the methods complement each other well and no method would be able to replace any other method. Instead they could be used together to make detection and measurement of subsidence more effective. Looking at the results from the compilation of cases where the methods have been compared, the theory of a combination of the methods is strengthened. To be able to use the methods together, it is necessary that the measurements gives approximately the same result, i.e. that the measurements follows the same trends and that they are of approximately the same magnitude. Both theoretically and seen in the cases in the compilation, those kinds of measurements are possible to achieve.

6.1.1 S

HORTCOMINGS IN THE COMPILATION OF QUANTITATIVE COMPARISONS OF REAL CASES

In the compilation of other work that have compared either GNSS, levelling or both with dInSAR, a summary of the conclusions is presented. This presents a wide spectrum of results drawn by others but lacks a common denominator and therefore a comparison between the cases cannot be performed. It would have been desirable to be able to compare them on equal conditions and to compare them to the quantitative comparison made in this study. If each comparison in the compilation would have presented an RMS error for the methods being compared, it could have been compiled in a more accurate way.

6.2 QUANTITATIVE COMPARISON – CASE STOCKHOLM

The quantitative comparison strengthens the theory of dInSAR being able to perform measurements that follows the same trend as measurements performed with both GNSS and levelling, but also shows that neither of the methods are flawless. Looking at the comparison between dInSAR and levelling, it shows that levelling reaches a higher level of accuracy than dInSAR, which only captures the rough movement of the area while levelling captured the small movements as well. This phenomenon was presented in the case of monitoring of landslides in Koyulhisar in Turkey (Hastaoglu et al., 2017, pp. 10-11) in the compilation and also in case Stockholm,

(31)

30

as seen in Diagram 12 and 13. However, it is important to keep in mind that levelling is almost exclusively performed before and after a reinforcement of an area affected by subsidence and therefore, the trends of the levelling results can be somewhat affected. As mentioned, the measurements are non-trivial, and as seen in the comparison between GNSS and dInSAR, faulty measurements can occur. Looking at the measurements that have been corrected for the overestimation of dInSAR, Dataset 3 is still showing a result that was unexpected, see Diagram 6. Knowing that the building that measurements have been performed on stands on solid rock and has not been exposed to any modifications during the time of measurements, the measurements would be expected to show no movement and therefore a subsidence rate of 0 mm per year. The measurements on the other hand shows an uplift for GNSS and a small subsidence for dInSAR. After ruling out the post glacier uplift as a reason of the differing results, the reason of the difference can only be caused by the previously mentioned, characteristic error sources of the measurement methods. Determining the cause for the different results can due to time limitation not be done in this thesis work.

6.2.1 S

HORTCOMINGS IN THE QUANTITATIVE COMPARISON

–

CASE

S

TOCKHOLM

One of the shortcomings in case Stockholm was the impossibility to compare all three methods simultaneously with the datasets used. This since the measurements performed with levelling and GNSS were not in vicinity of each other. Another shortcoming is that more datasets could have been used to present more results.

7 CONCLUSIONS AND FURTHER WORK

7.1 CONCLUSIONS

All three methods studied in this thesis work can individually detect and measure subsidence, however, today the methods area not utilized in the best way. To be able to densify and expand the urban areas of today, it is crucial to determine if the ground on which new buildings and infrastructure will be built on will remain stable. If this is not done correctly and in time, damage will occur on the buildings and the reparation cost will be large. To make the process of measuring subsidence in an area as effective as possible, the strengths of all methods should be considered and put together to develop a way of working that will reduce unnecessary measurements performed by the most time-consuming methods. dInSAR´s ability to measure large spatial areas often makes the method perfect for screening areas for any possible movement. Since the method follows the trends of subsidence presented by levelling, the method can rule out areas that are stable and does not show any signs of subsidence. Thereafter, the areas that are affected can be measured more closely on points of interest by either GNSS or levelling, depending on the accuracy level needed for the area.

7.2 FURTHER WORK

Since there were several limitations in this thesis work, it leaves room for much further work. Proposals of what could elaborate or complement this thesis work will be presented in the following sections.

7.2.1 I

NCLUDE SUB

-

METHODS

The literature study in this thesis work only describes the methods briefly and does not include any sub-methods. Any further work that could be made on this area is presenting a literature study that

(32)

31 compares not only the general method of dInSAR but all different types of methods of handling dInSAR data, such as Distributed Scatterers, Persistent Scatterers and all combinations of those two to other methods for determining height changes in the ground.

7.2.2 O

THER CAUSES OF GROUND MOVEMENT

Since the focus of this thesis work has been on small scaled subsidence, similar comparisons of dInSAR, GNSS and levelling with focus on other causes of ground movement would be of interest. These causes could be landslides, movement of tectonic plates, earthquakes, other natural disasters or man-made destruction of surfaces. This could in turn lead to further development of dInSAR and the discovery of new applications of the method.

7.2.3 Q

UANTITATIVE COMPARISONS

As mentioned earlier in the report, the amount of data used for the quantitative comparison was limited, due to limitations in time. Future work that would be of interest would be a much more extensive comparison of data where locations with different types of landcover could be compared against each other.

7.2.4 E

XAMINING THE UNEXPECTED RESULT IN CASE

S

TOCKHOLM

As mentioned, the reason for the differing result of dataset 3 in case Stockholm cannot be examined in this thesis work due to time limitations. In further work however, an examination of the location using more precise methods such as levelling could be used to detect any errors affecting the GNSS and dInSAR measurements.

(33)

32

REFERENCES

Abidin, H. Z., Andreas, H., Gamal, M., Djaja, R., Subarya, C., Hirose, K., . . . Rajiyowiryono, H. (2005). Monitoring Land Subsidence of Jakarta (Indonesia) Using Leveling, GPS Survey and InSAR Techniques. In F. Sansò (Ed.), A Window on the Future of Geodesy: Proceedings of the International Association of Geodesy IAG General Assembly Sapporo, Japan June 30 – July 11, 2003 (pp. 561-566). Berlin, Heidelberg: Springer Berlin Heidelberg. Abidin, H. Z., Andreas, H., Gumilar, I., Yuwono, B. D., Murdohardono, D., & Supriyadi, S. (2016). On Integration of Geodetic Observation Results for Assessment of Land Subsidence Hazard Risk in Urban Areas of Indonesia. In C. Rizos & P. Willis (Eds.), IAG 150 Years: Proceedings of the IAG Scientific Assembly in Postdam, Germany, 2013 (pp. 435-442). Cham: Springer International Publishing. Amighpey, M., Voosoghi, B., & Arabi, S. (2016). Modeling interseismic deformation field of North Tehran Fault extracted from precise leveling observation. Tectonophysics, 679, 169-179. doi:https://doi.org/10.1016/j.tecto.2016.04.051 Avdelningen för geodesi och satellitpositionering. (2015). Analys av mätosäkerhet. Paper presented at the Analys av mätosäkerhet, Introduktionskurs, Stockholm. Berardino, P., Fornaro, G., Lanari, R., & Sansosti, E. (2002). A new algorithm for surface deformation monitoring based on small baseline differential SAR interferograms. Ieee Transactions on Geoscience and Remote Sensing, 40(11), 2375-2383. doi:10.1109/tgrs.2002.803792 Bian, H.-f., Zhang, S.-b., Zhang, Q.-z., & Zheng, N.-s. (2014). Monitoring large-area mining subsidence by GNSS based on IGS stations. Transactions of Nonferrous Metals Society of China, 24(2), 514-519. doi:http://dx.doi.org/10.1016/S1003-6326(14)63090-9 Chen, S., Xu, H., Li, C., Li, J., & Li, W. (2015, 16-18 Sept. 2015). A novel analysis method of deformation accuracy for spaceborne PS-DInSAR. Paper presented at the 2015 IEEE International Conference on Imaging Systems and Techniques (IST). Engeltoft, T. (1996). Data Analysis and Adjustment in Precise Levelling with Applications to the Third Swedish Precise Levelling. (Doctor), Royal Institute of Technology, Stockholm. (1040) European Space Agency. (2017). Satellite ERS. Retrieved from https://earth.esa.int/web/guest/missions/esa-operational-eo-missions/ers European Space Agency. (2017-01-24). What is ESA? European Space Agency. (Accessed: 2017-06-08a). Sentinel-1. Retrieved from https://earth.esa.int/web/guest/missions/esa-operational-eo-missions/sentinel-1 European Space Agency. (Accessed: 2017-06-08b). Sentinel-3. Retrieved from https://earth.esa.int/web/guest/missions/esa-eo-missions/sentinel-3 Ferretti, A., Monti-Guarnieri, A., Prati, C., & Rocca, F. (2007). InSAR Principles: Guidelines for SAR Interferometry Processing and Interpretation. The Netherlands: ESA Publications, ESTEC. Ferretti, A., Prati, C., & Rocca, F. (2000). Nonlinear subsidence rate estimation using permanent scatterers in differential SAR interferometry. IEEE Transactions on Geoscience and Remote Sensing, 38(5), 2202-2212. doi:10.1109/36.868878 Ferretti, A., Prati, C., & Rocca, F. (2001). Permanent scatterers in SAR interferometry. IEEE Transactions on Geoscience and Remote Sensing, 39(1), 8-20. doi:10.1109/36.898661 Hanssen, R. F. (2001). Radar Interferometry, Data Interpretation and Error Analysis (1 ed.): Springer Netherlands. Harrie, L., Andersson, B., Persson, C.-G., Horemuz, M., Boberg, A., Olsson, P.-O., . . . Reshetyuk, Y. (2013). Geodetisk och fotogrammetrisk mätnings- och beräkningsteknik. In H. i. G. Lantmäteriet, Kungliga Tekniska Högskolan, Lunds Universitet, Kartografiska (Ed.): Lantmäteriet. Hastaoglu, K. O., Poyraz, F., Turk, T., Yılmaz, I., Kocbulut, F., Demirel, M., . . . Balik Sanli, F. (2017). Investigation of the success of monitoring slow motion landslides using Persistent Scatterer Interferometry and GNSS methods. Survey Review, 1-12. doi:10.1080/00396265.2017.1295631

Possibilities to make measurements of ground subsidence more effective, using dInSAR, GNSS and levelling