1
Monitoring the trajectories of tidal flat
morphology on a global scale
An assessment of semi-automated remote sensing techniques performed in Google Earth Engine, and construction of a series of models to determine the long term behaviour of tidal flats over a decadal timescale.
Author: BSc. M. C. de Blok
Student ID: 10646477
Supervisor (Uva): Dr. K.F. Rijsdijk
Examinator (Uva): Dr. A. Tietema
Internship institution: Nederlands Instituut voor Onderzoek der Zee (NIOZ)
Supervisor at institution: Msc. T. Grandjean
Period of internship: February – July 2020
Date: 13/11/2020
2
Internship report – assignment contents
At the end of the Internship the student should reserve enough time to write a final report. This final report has to meet the following requirements.
I. Thematic summary of the internship (½ page)
II. II. Description of the company/institute and the type of the activities that were performed during the internship. (± 1 page, see also point 4).
III. A personal reflection on what was learned during the internship (± 1 page, see also point 4).
IV. IV. A detailed report of the content of the internship (10-15 pages (18 ECTS) or 15-20 pages (24 ECTS).
3
Thematic summary of the internship
The coast, the interface between the earth’s landmasses and its oceans, is of unparalleled beauty and one of the most essential ecosystems globally. Several important functions such as tourism and foraging areas for many bird- and fish species are provided by coastal ecosystems. Due to the wide range of environmental circumstances that occur near coastlines, a large range of landforms exists with highly variating characteristics in terms of morphological development (Christopherson & Birkeland, 2015). In this research, the primary focus lies towards the landforms that become regularly inundated; tidal flats, marshes and mangroves. These coastal ecosystems pose several other important ecosystem functions, including a critical primary production, high species richness and biodiversity and coastal protection (Tong et al., 2020). Being constantly subject to the diurnal cycle of the tides, tidal flats are the most dynamic of the coastal ecosystems, and a major part of the coastal defence (Leuven et al., 2019). Increasing anthropogenic pressures have impacted river basins and their estuaries worldwide through changes in the suspended sediment load carried by rivers (Murray et al., 2019). Both the development of sediment transport- and delivery to the estuary, as well as urban expansion at the estuaries themselves have resulted in diversification of tidal flat behaviour.
Considering future expectations in the perspective of ongoing climate change, rising sea levels, enhanced subsidence and ongoing urban expansion at the cost of tidal areas and -dynamics, tidal flats will continue to be increasingly subject to environmental pressures (Ladd et al., 2019). These arguments of essentiality, lacking scientific knowledge and vulnerability reconfirm the need for a proper assessment tool. Identifying tidal flat development for a large number of estuaries worldwide may provide valuable insights into the response of tidal flats to these environmental and anthropogenic changes over the past decades. The essential question; whether tidal flats are sufficiently dynamic and resilient to cope with the rapidly changing conditions, will be one step closer to being answered.
Recently, research groups such as Murray et al. (2019), have made progress regarding the calculation of global tidal flat extension, as derived from enormous collections of satellite imagery. However, Murray et al. (2019), only determined the changes in size, while the changes in
morphology will provide more detailed information on how these tidal flats respond to their environments.
The focus on morphological development has led to the formation of the following research question and sub-research questions:
“How can the morphological development of tidal flats be assessed over a decadal- to
annual timescale, through the application of cloud-based platform Google Earth Engine and
other remote sensing techniques?”
1) What kinds of data and imagery should be collected, and what timeframe should be applied? 2) Which techniques should be used, and what steps taken to determine the morphological
developments of tidal flats?
3) What kind of requirements are there in terms of software, skills and programming needs? 4) How does the GEE-platform function and what are its options for studies on different
timescales?
4
Description of the institute and activities
The internship was organised at NIOZ, the Dutch Institute for Maritime Research. NIOZ is part of the NWOI, the collection of scientific institutes of the Netherlands.
The mission of NIOZ can be summarised in threefold; (1) to perform academically excellent
multidisciplinary fundamental and frontier applied marine research addressing important scientific and societal questions pertinent to the functioning of oceans and seas. (2) To serve as a facilitator for the (marine) scientific community of the Netherlands which is supported by the non-scientific National Marine Facilities Department. (3) To stimulate and support multidisciplinary fundamental and frontier applied marine research, education and marine policy development in both the national and international context.
NIOZ consists of four scientific departments. The internship research is on tidal flat dynamics, and is allocated to the Estuarine & Delta Systems department. Collaboration with the Coastal Systems department is certainly not ruled out.
Estuarine & Delta Systems Coastal Systems
Ocean Systems
Marine Microbiology & Biogeochemistry
The four departments are divided over two locations in the Netherlands; On the Wadden island of Texel and in the town of Yerseke on the boarder of the Eastern Scheldt. During the internship, visits were mostly scheduled to the establishment in Yerseke, since the Estuarine & Delta Systems department is based there. Visits to the Texel establishment were ruled out due to the developments surrounding the COVID-19 measures.
At NIOZ, I joined the research team of MSc. T. Grandjean, an enthusiastic young professional with brilliant ideas and a sixth sense for programming. The PhD project of T. Grandjean aims to explain the transition of tidal flats that cope with anthropogenic pressures, and is constructed to explain the behaviour over several different timescales. My activities have focussed on explaining tidal flat dynamics on the longest timescale considered in the research project, the decadal timescale. This is a remote-sensing project that uses state of the art cloud-based modelling through Google Earth Engine. This interactive platform is quite user-friendly, and allows the researchers to work with very large amounts of data on the Google Earth Engine servers, where over 10,000 CPU’s are linked together into a powerful cluster (Gorelick et al., 2017). Furthermore, all of the datasets collected over decades are readily available on the Engine servers.
For my internship assignment, I was tasked with the selection and analysis of case studies. An amount of 60 estuaries was determined to provide a sufficient level of variation and robustness. The selection of cases was dependent on several requirements; a meso- to macro tidal range (> 2 meters) was required for significant tidal flat areas to exist. Furthermore, an attempt was made to maximise the variation within the sixty case studies. Estuaries of different formation types, different sizes and different levels of management and anthropogenic pressures were chosen to produce a set of case studies as representative as possible for the entirety of the world. A standard workflow was developed to ensure the quality of the analysis and the reproducibility. A brief conceptual summary of the workflow is displayed in formula 1. The workflow will be elaborated on further in this report.
5 𝐸𝑠𝑡𝑢𝑎𝑟𝑦 𝑖𝑑𝑒𝑛𝑡𝑖𝑓𝑖𝑐𝑎𝑡𝑖𝑜𝑛 → 𝑃𝑟𝑒𝑝𝑎𝑟𝑎𝑡𝑖𝑜𝑛 𝑓𝑜𝑟 𝐺𝐸𝐸 → 𝐺𝐸𝐸 𝑑𝑜𝑤𝑛𝑙𝑜𝑎𝑑 →
𝑃𝑟𝑜𝑐𝑒𝑠𝑠𝑖𝑛𝑔 𝑜𝑓 𝑖𝑚𝑎𝑔𝑒𝑠 → 𝐶𝑎𝑙𝑐𝑢𝑙𝑎𝑡𝑖𝑛𝑔 𝑁𝐷𝑉𝐼 𝑓𝑜𝑟 𝑣𝑒𝑔𝑒𝑡𝑎𝑡𝑖𝑜𝑛 𝑚𝑎𝑝𝑠 →
𝐶𝑎𝑙𝑐𝑢𝑙𝑎𝑡𝑖𝑛𝑔 𝑁𝐷𝑊𝐼 𝑓𝑜𝑟 𝐷𝐸𝑀′𝑠 → 𝑃𝑟𝑜𝑑𝑢𝑐𝑖𝑛𝑔 & 𝑖𝑛𝑡𝑒𝑟𝑝𝑟𝑒𝑡𝑖𝑛𝑔 𝑡ℎ𝑒 ℎ𝑦𝑝𝑠𝑜𝑚𝑒𝑡𝑟𝑖𝑐 𝑐𝑢𝑟𝑣𝑒𝑠 F.1
Personal reflection
The internship is a non-mandatory segment of the Earth Sciences master’s programme. Throughout the Future Planet Studies and the Earth Sciences Master study programmes, the preliminary focus has been on theoretic knowledge development, short practical assignments and fieldwork experience. In my opinion, there is too little of a connection between the educational programme and the corporate society, and that is why I immediately decided to include the internship into my study programme. Experience in working at a company is in my opinion an essential part of preparing oneself for their professional career.
Naturally, every individual has their strengths and their weaknesses, and for me, the internship has functioned as a perfect possibility to focus and improve on my weaknesses. Programming and remote-sensing were one of my lesser developed skills, and thus one of the primary learning goals was to gain experience in python, ArcMap, and google earth engine. Especially getting familiar with Engine has opened a new mental window for me. Many applications that I could only dream of before my internship are now within reach, and have been performed in the past five-month period.
My personal interests have always been focussed on hydrology, climate change and the interaction between nature and society. Tidal flats are a perfect example of complex hydrological features that can be highly impacted by human behaviour, and may be at risk of degradation in the near future. Identifying issues early on, and developing a standard method to assess the morphological development of these tidal flats is critical for the preservation of these areas and all of the functions that our society relies on.
At the beginning of the internship, I discovered that the level of programming in the research was quite complex. Several tutorials and theoretical studies on python and google earth engine were required to reach a sufficient level of knowledge to be able to work with the scripts. Now, four months later, I still do not have the illusion that I would be able to write such complex scripts from scratch. However, reading, understanding, adapting and working with the scripts have improved considerably, and python feels quite familiar by now. The main objective of my internship within the PhD project of T. Grandjean was to contribute to the development of a (semi-automated) assessment method to determine the morphological developments of tidal flats. Building further on recent work by for example Murray et al., (2019), Leuven et al., (2019), Ladd et al., (2019) and Tong et al., (2020) has unveiled the desire for a consistent method that is able to cope with enormous amounts of data while minimising the manual labour required. I believe that this project has made huge progress in that sense, and that a solid first step towards a watertight method has been made. However, several steps in the progress did still require extensive manual feedback, which is a point of interest for further automatisation.
Finally, one of the great features of this internship project was the broad focus, both in terms of timescale and in sense of methodology. Linking detailed local field observations, collected through specially designed sensors, with long term observations acquired through the newest remote-sensing techniques would have attributed to a very holistic research methodology. However, unfortunately, the developments regarding the COVID-19 virus drastically limited the options for personal contact. As a consequence, all visits to the research facility in Yerseke, all field work, supervision meetings and laboratory assignments were cancelled. I must really thank and compliment my supervisor T. Grandjean for his never ending efforts to ensure that the lockdown did not delay my internship.
6
Report
Embedding of the research
To clarify the broad outline of this report, the larger embedding of the research will first be
summarized. A strategic scientific alliance was constructed between China (Tsinghua University and SKLEC, State Key Lab for Estuarine and Coastal research) and the Netherlands (Technical University Delft and NIOZ, the Dutch Institute for Maritime Research) to collaborate on the theme of coping with deltas in transition. The leading Chinese governmental partners are the Yangtze Estuary Waterway Administration Bureau, the Shanghai Water Authority and the Shanghai Municipal Bureau of Forestry. In regard to Dutch organizations and institutes, the universities of Amsterdam, Utrecht and Wageningen, Rijkswaterstaat, Deltares, the Wadden-academy and the WWF are part of the strategic alliance.
The mission of the alliance, coping with deltas in transition is a broad subject and is split into four research themes, with each partner focussing on a specific theme.
- Theme I : Estuaries in a changed land-ocean flux continuum (Tsinghua University) - Theme II : Regime shifts in estuarine morphology (Delft University)
- Theme III : Transitions in intertidal bio-geomorphology (NIOZ) - Theme IV : Solutions for deltas in transition (SKLEC)
This internship research assignment took place in the third domain at the Dutch Institute for Maritime Research. At NIOZ, the research team is supervised by Professor Dr. T. Ysebaert, and further consists of three team-members. Among these team members is my personal supervisor, MSc. T. Grandjean. Collaboration is focused on a PhD project with the aim of intending to create a better understanding of the morphological developments of tidal flats. A perspective exceeding approach was adapted; the desire is to explain patterns on a global scale and a decadal timescale, through the support of high-resolution spatial and temporal data of specific case studies and in situ data collection on a short timeframe.
The first segment of the project focuses on the development of a methodological framework for the remote-sensing analysis of tidal flat morphology on a global scale, and on a decadal timescale (1985 – 2019). The application of the Google Earth Engine software combined with a series of operations through multiple open-source platforms provides a unique opportunity for future environmental research on various applications. This is also the subject where the internship has focussed on, as will this report.
The second segment of the project focusses on a revolutionary improvement of in-field data collection in tidal flat ecosystems. Concerning the development of the acoustic SED (Sediment bed Elevation Dynamic) sensors, to improve the timescale of field observation measurements to minutes timescale, while further improving on vertical resolution and enormously reducing the manual labour required comparing to previous methods. Field experiments with these sensors are performed in China (Yangtze) and in the Netherlands (Westerschelde, Oosterschelde and Eems) together with an extensive fieldwork campaign. During the internship, there have been little chances to participate in these activities due to the lockdown in both countries.
7
Introduction and relevance
The importance of coastal ecosystems, and in particular estuaries has long been recognized (Leuven et al., 2019, Murray et al., 2019 & Christopherson & Birkeland, 2015). Owing to the continually varying influences of oceanographic, tidal, sediment transport and –depositional processes, the coastal ecosystems are
exceptionally dynamic (Ladd et al., 2019). Especially tidal flats, the meso- to macro tidal coastal areas that suffer tidal regimes surpassing two meters are unique, but challenging to study due to their constantly changing nature. Estuaries and their most widespread habitat types; mangroves, salt marshes and tidal flats provide several critical functions such as food production (Tong et al., 2020), coastal protection (Murray et al., 2019), valuable ecological habitat provision (Leuven et al., 2019) and access to harbours and waterways (Christopherson & Birkeland, 2015). Coastal areas simultaneously function as a hotspot for human habitation. 21 of the world’s largest cities are located close to estuaries. A negative trajectory of many coastal ecosystems has been observed globally in the past decades because of extensive degradation through multiple causes; coastal development and -urban expansion, reduced sediment supply from major rivers through construction of sediment barriers, subsidence and sinking of riverine deltas, rising sea levels and increased erosion have all magnified the pressures, and are expected to continue to impact coastal ecosystems (Murray et al., 2019). Near future scenarios do not involve a decrease in these coastal stressors, and with sea levels expecting to rise at an accelerating rate, the resilience of coastal ecosystems cannot be sustainable in the long term considering current trends.
Although tidal flats are among the most widespread coastal ecosystems, the global distribution and status remain largely unknown, which hinders efforts to manage, protect and restore coastal ecosystems around the world (Murray et al., 2019). Because, until recently, global-scale data on the distribution and extent of tidal flats were lacking, no studies were able to overcome the difficult problem of remote-sensing a widely distributed ecosystem that is frequently obscured through tidal inundation. Recently, new technologies and modelling techniques have created opportunities for analysis on a global scale. Sophisticated platforms such as Google Earth Engine provide a fresh perspective on global scale remote-sensing research. Murray et al., (2019) took the first step and used over 700.000 satellite images to construct a global multi-decadal distribution analysis of tidal flats over the period 1984-2016. However, we challenged ourselves to develop this method to the next level. We are aiming to identify not only the extent and distribution of tidal flats, but also the
morphological development on an estuary-level scale. A unique feature of this research is that it aims to apply these novel technologies to develop a method that is both semi-automated, open-source, and globally applicable on every type of estuary that suffers a meso- to macro tidal regime. To achieve the goals just described, a main research question and several sub research questions were defined.
Research questions
“How can the morphological development of tidal flats be assessed over a decadal- to
annual timescale, through the application of cloud-based platform Google Earth Engine and
other remote sensing techniques?”
1) What kinds of data and imagery should be collected, and what timeframe should be applied? 2) Which techniques should be used, and what steps taken to determine the morphological
developments of tidal flats?
3) What kind of requirements are there in terms of software, skills and programming needs? 4) How does the GEE-platform function and what are its options for studies on different
timescales?
8
Theoretical background
Tides
Currently, our capability to accurately model and predict tides are well-developed. However, the interaction of the tides with the coastal zone is much more complex and less understood. Tides can be described as the gravitational pull generated by the sun and the moon on every water body on earth (Cristopherson & Birkeland, (2015). This diurnal cycle varies highly around global coastlines, ranging from mere centimetres in the Mediterranean Sea to over 16 meters in the Bay of Fundy in Canada. Several geographical and local conditions play a role in the severity to which a coast suffers to the tidal regime. Firstly, the tides follow a monthly pattern. The variations in solar- and
lunar alignment have a significant impact on the tides. During neap tide, with the minimum tidal amplitude, the sun and moon are not aligned, which counteracts a fair part of the gravitational pull. During spring tide, the sun and moon are aligned, and gravitational pull is at its maximum, thus also creating the largest tidal amplitude (Christopherson & Birkeland, 2015).
Secondly, the existence of amphidromic points or tidal nodes should be explained. Tidal nodes are geographical locations which have a zero tidal amplitude. These nodes are created through the Coriolis Effect and represent the central points around which the tides circulate. The height difference in water
level between high- and low tide increases with distance from the amphidromic point. In figure 1, the tidal nodes are displayed. It can be observed that the coastlines characterised by a meso- to macro tidal regime are located at a large distance from the amphidromic points. The distance to the nearest amphidromic point thus partly determines the main tidal regime at a certain coastline (Kvale et al., 2006).
Secondly, the shape of the coastline has a significant impact on tide magnification or reduction. The baseline of tidal range is determined through the distance from the amphidromic point. However, the shape of the basin can enlarge the tidal regime by several factors. Especially funnels that narrow land-inward can significantly magnify the tides (Kvale et al., 2006). The explanation can be compared to a tsunami wave; when the mass of water is forced into a shallow basin, the only available direction for the water to go is in the vertical direction, and a massive wave is created. Regarding tides, this mechanism works slightly different. Instead of the seafloor becoming shallower, the basin becomes narrower. The result of water being elevated is however identical. For example, the Northern coast of France and the North Sea are quite far away from the central amphidromic point in the Northern Atlantic Ocean, and the geography of Great Britain functions as a bottleneck for the incoming tides which has resulted in high tidal ranges along the North-Western European coast.
The final factor that is a significant contributor to tidal variation are meteorological circumstances, and especially the Aeolian stimulation. In case a sufficient fetch length is available before the wind-generated waves reach the coastline, significant contributions to high sea level can be made. Most of the ocean-related flooding disasters that have occurred in the past centuries were caused by an extreme coincidence of the aforementioned circumstances (Lopes et al. 2017 & Dehenauw, 2008).
Modelling such dynamic environments on a global scale has proved a challenge which was, until recently, impossible to overcome. The creation of global tidal models such as the WorldTides model allows researchers to automatically determine the water level that corresponds to the exact moment the satellite image was taken. Through this method, an interactive coupling between image collections and actual water levels is created, which solves the previously mentioned problem.
9
Tidal flat morphology
During the research project, six types of tidal flat morphological behaviour were distinguished. These types are three complementary sets of characterizations;
1. Steepening / Flattening – This characteristic focusses on the profile of the tidal flat. The steepness of the slope is considered, which can variate from a steep S-curve to approaching a linear distribution. 2. Expansion / Decline – This characteristic focusses primarily on the extension or withdrawal of the
tidal flats. Extension means the sea-ward expansion and thus indicates sedimentation. Decline implies a withdrawal of tidal flats land inward, thus indicating erosion.
3. Stability / Fluctuation – This characteristic focusses on the dynamics of the tidal flats. Some hypsometric profiles of the case studies might display little variation over decades, while others are subject to constant change.
Sediments and suspended sediment concentration
The dynamic nature of tidal flat ecosystems and their remarkable capacity to resilient adaptation is mainly originating from the continuous delivery of sediments to the estuaries, both from marine sources and transported by streams and rivers throughout the river basin (Jaffe et al., 2007). Every tidal cycle, water levels rise and carry sediments onto the tidal flat. Vertical expansion is thus possible provided that sufficient sediments are delivered continuously. A measure to assess the sediment concentration is the Suspended Sediment Concentration (SSC). Over the past decades, declining SSC trends have been observed globally through several mechanisms (Leuven et al., 2019). Over the past centuries, anthropogenic influences in river basins have increased, with dams having the most significant impact. A dam functions as a sediment trap, and thus prevents sediments from reaching the estuary (Wu et al, 2003). Furthermore, the dam creates a new base-level of erosion, reducing the erosional power of the water that flows downstream. For example, in the lower section of the Yangtze River in China, a series of dams were built post 1990, including the three gorges dam, the largest dam structure in the world. This dam has created a reservoir over 600 km long, with a total water holding capacity of over 39 km3 (Wu et al., 2003)The dam was constructed at an elevation of 185
meters above sea level. Creating a height difference with the other side of the dam of 110 meters.
Anthropogenic influences
Growing populations have continuously enlarged the pressures on coastal ecosystems on many levels (Leuven et al., 2019). Historically, civilizations have often settled near estuaries since there are many usefull functions such as trade and navigation, water supply and food production in the form of agriculture and aquaculture. These cities belong to the largest and fastest-growing cities in the world, and many constructions and land reclamation projects occur at the cost of tidal flat ecosystems. Furthermore, the aforementioned measured upstream limit the delivery of fluvial sediments to the deltas. On the contrary, there are also examples of anthropogenic behaviour with positive reinforcement on the distribution of sediments. For example, gold mining activities in Canada and Alaska in the period 1800-1900 have led to elevated SSC concentrations of factor four compared to the natural concentration. Corresponding rapid growth of tidal flats at estuaries associated with large scale gold mining activities were observed (Jaffe et al., 2007).
10
Remote sensing and Google Earth Engine
Remote sensing has become a widely accepted method for planetary-scale environmental analyses over the past decades. However, many of the programmes available required a high level of expertise while also demanding expensive licenses, and were limited in their use because users had to insert their datasets. Furthermore, the increasing level of detail and complexity, together with the desire for upscaling to global coverage has led to a situation in which computation times have reached incredible lengths, even longer than the time of a human lifespan in some instances. Fortunately, the development of open-source cloud-based platforms has revolutionized the industry of remote sensing and created endless opportunities for
environmental research. Pioneer Google launched the Earth Engine platform in 2014, combining a multi-petabyte catalogue of satellite imagery and geospatial datasets with planetary scale analysis capabilities to make it available for scientists, researchers and developers to detect changes, map trends and quantify differences on the earth’s surface.Google Earth Engine is powered by Google’s cloud infrastructure, and runs the analyses on a cluster of powerful CPU’s (Gorelick et al., 2017). With a minimum level of expertise, highly complex, planetary-scale analyses can be performed, such as the project discussed in this report.
The Landsat 5- and 7 satellites used throughout the research, are equipped with state of the art software and photography equipment. The imagery is acquired through a combination of 7 or 8 bands, specific spectrums of wavelengths to optimally observe specific features. The Landsat 5 thematic mapper consists of seven spectral bands with a spatial resolution of 30 meters. Table 1 displays the bands, the corresponding wavelengths, the observed characteristics and its possibilities for application.
Table 1 – Landsat 5- and 7 satellite bands and their applications.
Band Spectrum Wavelength (μm) Usefull in application for:
Band 1 Blue 0.45-0.52 Bathymetry, distinguishing soil from vegetation
Band 2 Green 0.52-0.60 Peak vegetation determination
Band 3 Red 0.63-0.69 Vegetation slopes
Band 4 Near infrared 0.77-0.90 Biomass content and determination of shorelines
Band 5 Short-wave infrared 1.55-1.75 Moisture content of soil and vegetation
Band 6 Thermal infrared 10.40-12.50 Thermal mapping and soil moisture
Band 7 Short-wave infrared 2.09-2.35 Hydrothermally altered rocks and mineral deposits
Band 8 (LT7) Panchromatic 0.52-0.90 Sharpening image definition
The main operations performed during the analyses are the Normalized Difference Vegetation Index (NDVI) and Normalized Difference Water Index (NDWI) operations. Both the NDVI and NDWI are composed from two bands. The formulas for each calculation are displayed below:
F1. NDVI = (NIR-Red/NIR+Red)
The NDVI is thus based on the near-infrared- and red bands. This might seem strange since the studied entity is green vegetation. However, green vegetation is only green because it reflects the light from the green
spectrum and thus seems green. In fact, vegetation required the energy from the red-light spectrum for its growth. Through application of the NDVI formula, an indexed NDVI-score is constructed, and values with an NDVI > 0.1 are considered to be vegetated.
F2. NDWI = (Green-NIR/Green+NIR)
The NDWI is slightly more complex. Two different types of NDVI were constructed in 1996. The first method by Gao et al., (1996) is based on the near-infrared and the short-wave infrared bands. This method is primarily applicable on a small scale, for example to determine the moisture content in leaves. The second method is posed by McFeeters et al., (1996). This method is more applicable on a large scale on the estuary-level. The original formula is displayed in formula 2. However, the requirements of this study have demanded for a personalised NDWI, and thus a new formula was created, which is based on three instead of two bands:
11
As can be observed from formula 3, the reworked-NDWI is based on the green, near infrared and short-wave infrared bands.
Methodology
This section will explain the methodology applied throughout the research. The covered subjects are: selection of case studies, pre-preparing for download, downloading image collections, processing and analysis and finally the visualisation of the workflow.
Case study selection
Several criteria were set during the selection of case studies. The Yangtze estuary in China, the Eastern- and Western Scheldt and the Ems estuary in the Netherlands were automatically included in the selection. During the process, the following criteria (table 2) were set. These criteria, their technical implementation and the motivation are displayed in table 2.
Table 2 – Criteria and technical approach in case study selection
Criterion Technical approach Motivation for criterion
Meso to Macro tidal regime Tidal range > 2 meters Sufficient dynamics Sufficient tidal flat area Tidal flat area > 50.000 m2 Observability on imagery
Sufficient data available Literature and imagery from ’85 onwards
Construct a complete time-series complimented with back-ground knowledge
Distance from previous case >300 km preferably Avoid identical image collections Acquire global coverage Cases on all eligible coasts Produce a global dataset Acquire full profile range Collect various estuary types Broaden the research
An analysis of global eligible coastlines was performed through high-resolution images on Google Earth Pro; these regions of interest were manually screened for eligible cases and consequently tested for the criteria displayed in table 2. An overview of the complete collection of cases and the global distribution of micro-, meso- and macro tidal regimes are displayed in figure 2.
12
A more detailed overview of the cases and their corresponding locations are provided in table 3. The case study locations per region are displayed in figures 3-15.
Southern Europe
Northern Europe
West-African coast
China
Figure 3 - Case studies in Southern Europe (Google Earth Pro) Figure 4 - Case studies in Northern Europe (Google Earth Pro)
13
Alaska
United States East coast
East Africa/Madagascar
Figure 9 - Case studies at the East African coast and Madagascar (Google Earth Pro)
14
South America
Brazil / Guiana
Central America
Bangladesh Canada
Australia
Figure 10 - Case studies in South America (Google Earth Pro) Figure 11 - Case studies in Brazil and Guiana (Google Earth Pro)
Figure 12 - Case studies in Central America (Google Earth Pro)
Figure 13 - Case study in Bangladesh (Google earth Pro)
Figure 14 - Case studies in Canada (Google Earth Pro)
15 Table 3 – complete inventory of the case studies per continent/region.
Estuary Country Coordinates (lat, lon) Research area (km2)
Southern Europe
1. Arcachon France 44.667, -1.167 220.0 2. Santander Spain 43.466, -3.790 29.6 3. Olhao Portugal 36.983, -7.876 100.0
Northern Europe
4. Solway Great Britain 54.833, -3.800 436.6 5. Humber Great Britain 53.631, -0.186 376.4 6. Blackwater Great Britain 51.730, 0.849 18.6 7. Colne Great Britain 51.799, 1.000 34.0 8. Exe Great Britain 50.617, -3.417 23.4
9. Elbe Germany 53.867, 8.717 818.0
10. Ems Netherlands 53.333, 6.933 844.0 11. Western Scheldt Netherlands 51.333, 3.833 271.0 12. Eastern Scheldt Netherlands 51.600, 4.017 310.0
13. Canche France 50.547, 1.589 14.2
14. Authie France 50.376, 1.564 19.8
15. Sommes France 50.235, 1.570 47.4
16. Ay France 49.203, -1.621 8.7
17. Portbail France 49.329, -1.700 6.1
West- African coast
18. Konkoure Guinea 9.809, -13.788 115.0 19. Tagrin Bay Sierra Leone 8.529, -13.099 478.0 20. Muni-Pomadze Ghana 5.318, -0.646 4.3 21. Forcados Nigeria 5.369, 5.453 177.0 China 22. Wulong China 36.618, 120.869 103.0 23. Yangtze China 31.417, 122.233 4286.0 24. Zhujiang China 22.769, 113.622 746.0 Alaska 25. Borough Alaska 56.707, -157.569 5.9 26. Knik arm Alaska 61.401, -149.723 541.0
United States (Eastcoast)
27. Westport United States 41.508, -71.093 12.0 28. Slocums United States 41.521, -70.968 6.9
South America
29. Laguna de Cahuil Chile -34.481, -72.035 0.5 30. Itata Chile -39.387, -72.843 21.9 31. Bustamante Argentina -45.174, -66.524 48.0 32. Rio Negro Argentina -41.616, -65.009 1.5 33. Caleta de los Loros Argentina -41.033, -64.100 34.2
Brasil/ Guiana
34. New Amsterdam French Guiana 6.315, -57.528 21.5 35. Arraial Brazil -2.744, -44.201 422.0 36. Parnaiba Brazil -2.731, -41.780 673.0 37. Tatajuba Brazil -2.845, -40.697 3.0 38. Ceara Brazil -3.692, -38.590 2.7 39. Jaguaribe Brazil -4.423, -37.760 17.6 Central-Americas
40. Colorado Costa Rica 10.142, -85.084 85.7 41. Coto Costa Rica 8.541, -83.162 14.1 42. Chiriqui Nuevo Panama 8.268, -82.451 82.1 43. Chiriqui Viejo Panama 8.554, -83.154 3.8
16
44. Canas Panama 7.411, -80.282 11.3
45. Chame bay Panama 8.642, -79.784 77.2 46. Guayas Ecuador -2.646, -80.333 3287.0
Indian Coast
47. Meghna Bangladesh 22.419, 91.108 5688.0
Canada
48. St. Lawrence Canada 48.133, -69.183 4347.0 49. Ungava bay Canada 58.138, -68.041 318.0
Australia
50. Fitzroy Australia -17.251, 123.513 571.0 51. Herbert Creek Australia -22.339, 149.835 313.0 52. Moreton bay Australia -27.367, 153.167 229.0
East- Africa/ Madagascar
53. Pwani Tanzania -6.469, 38.986 42.4 54. Mnazi bay Mozambique -10.331, 40.356 232.0 55. Ruvuma Tanzania -10.464, 40.442 25.6 56. Atolo Mozambique -13.052, 40.565 27.8 57. Lurio Mozambique -13.521, 40.502 10.1 58. Bombetoka Madagascar -15.889, 46.334 601.0 59. Morombe Madagascar -22.209, 43.281 59.2 60. Androka basin Madagascar -25.000, 44.026 20.9
Workflow visualisation
A complete visual representation of the methodology is provided as workflow diagram in figure 16. The workflow will be discussed in detail in the next chapters of the methodology section.
17
Pre-preparation for download
In case an estuary was approved for all conditions (figure 16, fist column), the estuary was prepared for the download in Google Earth Engine (figure 16, second column). Firstly, a polygon of the research area was cut-out on Google Earth Pro. The polygons included the complete water bodies of the estuaries, and all possible sites for tidal flat formation. The permanent vegetation (salt marshes and mangroves) was not included in the research area polygons. Consequently, the .kmz file as derived from Google Earth Pro was adapted in ArcMap to a shapefile (.shp) format for compatibility with Engine. Due to the automatization of the procedure, a highly specific set-up in the directories in Google Drive was required. A folder of the particular estuary was created, and the shapefiles added to the directory. Next, the shapefile was coupled to Engine as an ‘asset’. These assets are compatible with Engine and contain the area-of-interest on which the platform will select imagery.
Downloading image collections
The process of downloading imagery is displayed in the third column of figure 16. In advance of the internship, the methodological framework was already partly constructed. Because of the desire to explain morphological developments on a decadal timescale, a broad distribution of research periods was chosen. A complete overview of the research periods and corresponding satellites is displayed in table 4. The year 1985 was chosen as a starting point because of the reliability of Landsat 5 data availability from this point in time onwards.
Table 4 – Research periods and corresponding satellites.
Research period Satellite
1985 - 1987 Landsat 5
1994 - 1996 Landsat 5
2003 - 2005 Landsat 5, Landsat 7
2012 - 2014 Landsat 7
2017 – 2019 Landsat 7
In Google Earth Engine applications, a base map, or base data frame should be selected. This base map provides the actual data (Gorelick et al., 2017). The base map set is the Copernicus CORINE land cover map. When initiating the analysis, several parameters had to be provided to the download script; the name of the estuary, the coordinates, the timeframe in which imagery should be selected and the folder in which the images should be stored. Furthermore, the script contains filters and masks, which ensure that each selected image is highly usable during further analysis. Two filters were applied; one for the quality of the image, and one for the coverage of clouds. The image quality threshold was set to 7 (out of 10) and the cloud cover filter was set to 40%, meaning that only images with a cloud cover below 40% were added to the image collections. Moreover, a series of masks were performed throughout the script. For example, masks for oversaturation, atmospheric opacity, snow, cloud shadows and noise on the image edges. Finally, an interpolation method is applied to fill any empty pixels that might be present on the image. Six separate runs were performed for each studied estuary, resulting in varying sized image collections, ranging from 46-514 images, with an average of 165 images per collection. A total of 9924 images were stored and analysed during the project.
18
Processing and analysis
Post download, the images were further processed through a series of scripts ran in Google Collaboratory. The functions of- and operations in each individual script will be briefly summarized. The fourth column of figure 16 displays these operations. One especially unique operation that deserves to be separately mentioned is the methodology for the Digital Elevation Model (DEM) construction. Throughout the scripts, the individual images are combined into a stacked bathymetric model to build the elevation maps. The exact functionality of this method is somewhat difficult to
perceive. However, the piece of art displayed in figure 17 is a proper visual aid. In this comparison, each coloured layer is represented by an individual image. With all of the images and their corresponding data combined, a 3-D elevation model of the area-of-interest is composed.
Script I: Processing
The first script consists of two sections. The first sections function is mainly to create the folder structure and set the path of directories. Consequently, the primary operations are performed on the images; Metadata- and water level documents are created to store all the relevant data and results. Consequently, the link to the WorldTides API is made. This link ensures that the correct tidal-data is attached to each individual location and moment in time. Finally, the NDWI- and NDVI are calculated from the image and water level images are created and stored.
Script II: Compile NDVI images
In the second script, the vegetation maps are created and viewed. In every image, the NDVI is analysed to determine whether a pixel contains vegetation. In order to be designated vegetation in de final vegetation map, a NDVI-count of >20% was set, meaning that a pixel was determined to be vegetated when vegetation was observed more than one-fifth of the time. Towards the end of the script, a section is reserved to plot and manually assess the vegetation maps for possible errors. A summary of vegetated area during each time period is also provided.
Script III: Compile NDWI images
The third script is the most labour-intensive operation throughout the analysis. To identify tidal flat areas, the NDWI is applied to distinguish land from water. For each image in a collection, the NDWI was analysed to determine the water line at that certain moment, and the areas that had fallen dry or had become inundated at that moment. The application of the WorldTides model corrects for the tidal characteristics at the identical moment, creating a model as accurate as currently possible of the actual sea level at that moment in time. In most occasions, multiple misclassified maps were present in each collection, and an assessment of plausibility was necessary. This operation had to be performed manually through a combination of images analysis and logic reasoning. Wrongly classified images were removed from further analysis. Consequently, all of the accurate NDWI-images were merged into a digital elevation model for each period (1985-1987, 1994-1996, 2003-2005, 2012-2014 and 2017-2019). This stacked-bathymetric method is visualised in figure 17.
Script IV: Hypsometric analysis
The fourth script combines all of the previous data to create a sigmoidal function that represents the
morphological development for the tidal flat in the study area as a whole. Several steps are performed during the script; first, the sigmoidal function is defined. A best-fit model approach with 10.000 repetitions was
19
applied. Consequently, the minimum- and maximum area of tidal flat and the Low- and High Water Boundaries (LWB/HWB) are calculated. The concept of the Low- and High Water Boundaries is displayed in figure 18. These boundaries display the minimum range in which data is available for every period. The application of this window of analysis is an absolute requirement. Outside of this window, not all periods have data, which would lead to a misinterpretation of the results. During the final step, the hypsometric curves are plotted and can be consequently interpreted.
20
Results
Revisit of the research questions
In the introduction, a set of research questions was posed to focus the direction of the research, and to ensure a holistic approach on the matter. In table 5, the research questions will shortly be
discussed and provided by a descriptive answer. Table 5 – Research questions and –answers.
Research question Result
1A. What timeframe should be studied? Decadal timescale, three periods of three years, with 9 years in between: 1985-1987, 1994-1996, 2003-2005 and to include the most recent data 2017-2019.
1B. What kinds of data should be collected? The most important data to be collected are the image collections on the 60 case studies. Furthermore, background information on the cases and corresponding tides. The full range of data collected is displayed under the header ‘dataset’.
2A. What techniques should be used? Please refer to the methodology section
2B. Which steps should be taken? Please refer to the workflow visualisation section
3A. What are the requirements regarding software? Access to Google Earth Engine, licenses for ArcMap, excel, google earth Pro, an extensive google Drive and a
license/account for the WorldTides API. 3B. What are the requirements regarding skills and
programming?
Basic python knowledge, know how to work with Engine/ArcMap/Excel/Earth/WorldTides.
4A. How does the GEE-platform work? Please refer to the theoretical background Remote sensing and Google Earth Engine.
4B. What are the options for studies on different timescales in GEE?
The set-up of the research methodology is completely personalised. Research windows of days- decades are possible, as long as there is data available.
5A. How does the morphology develop? The six types of morphological behaviour: Steepening/Flattening,
Expansion/Decrease, Stability/Fluctuation
21
Dataset
An extensive dataset was created in SPSS to function as an inventory overview. Many parameters and other information is displayed and/or calculated. Table 6 displays these characteristics and their corresponding explanations.
Table 6 – Parameters included in the SPSS dataset and their explanations.
Characteristic Explanation
1. ID-number ID-number in the order of case study selection 2. Name of the estuary Estuary name
3. Estuary type Type of estuary
4. Country Country in which the case is located 5. Continent Continent on which the case is located 6. Latitude Latitude coordinate of the case 7. Longitude Longitude coordinate of the case 8. Tidal station Name of most proximate tidal station 9. Tidal station - location Latitude, longitude of this tidal station
10. Average tidal amplitude Amplitude based on Mean High/ Mean Low water 11. Maximum tidal amplitude Amplitude based on Astronomical High/ Low tides 12. Tidal type Micro/ Meso/ Macro tidal
13. Landsat 5 (1985-1987) Size of image collection for corresponding period 14. Landsat 5 (1994-1996) Size of image collection for corresponding period 15. Landsat 5 (2003-2005) Size of image collection for corresponding period 16. Landsat 7 (2003-2005) Size of image collection for corresponding period 17. Landsat 7 (2012-2014) Size of image collection for corresponding period 18. Landsat 7 (2017-2019) Size of image collection for corresponding period 19. Area Total surface area of the studied estuary (km2)
20. Perimeter Perimeter of the studied estuary (km2)
21. Ratio AP Ratio between area and perimeter 22. Ratio PA Ratio between perimeter and area
23. Maximum surface area Maximum area of tidal flats in hypsometric script 24. LWB Low water boundary of research window (m) 25. HWB High water boundary of research window (m) 26. Comments Comments on the LWB/HWB by punctuation mark 27. Explanation Explanation on previous comment
22
Hypsometric curves
Earlier in this report, the six types of tidal flat morphological behaviour have been discussed; Stability,
fluctuation, decline, expansion, steepening and flattening. Each of these type of behaviour will be illustrated by a hypsometric curve plot from of the studied cases.
1. Stability
The hypsometric curve plot of the Colne estuary (figure 19), located in Great Britain, displays a quite stable morphological regime. The slope of the curves does not change considerably over the four research periods, and especially the relative surface at the LWB is remarkably consistent. These results assign the Colne estuary to the top five most stable tidal flats from the project. The tidal flats of the Arraial estuary in Brazil, the Eastern Scheldt in the Netherlands and the Solway estuary in Great-Britain are more examples of estuaries characterised by stable tidal flats.
2. Fluctuation
In contrast to the previously discussed stability, some estuaries were characterised by fluctuation. Figure 20 displays the hypsometric curve plots for the French estuary Arcachon. Large differences between every timeslot can be observed, both in terms of relative surface area and in terms of steepening and flattening. Arcachon proved to be the most dynamic case out of the sixty cases studied. Other estuaries characterised by fluctuating behaviour in tidal flats are the Exe estuary in Great Britain, the Ems estuary in the Netherlands and the Fitzroy estuary in Australia.
3. Decline
The Humber estuary in Great Britain is a proper example of an estuary with declining tidal flats (figure 21). The Humber estuary is one of few cases with such a consistently declining trend. Between every research period, a decline in relative surface is observable. Major jumps occur between 1986 and 1995, and between 2004 and 2013. A cumulative loss at both the HWB and LWB windows of around 30% is observed. Another estuary with clearly declining morphological behaviour is the Wulong estuary in China. The anthropogenic expansion in this estuary is enormous, and a sub-category of
declining through direct anthropogenic expansion might be considered.
Figure 19 - Hypsometric curves of the Colne estuary
Figure 20 - Hypsometric curves of the Arcachon estuary
23
4. Expansion
The Zhujiang estuary in China is also subject to rapid urban expansion and other anthropogenic pressures. Nevertheless, the hypsometric curve plot displayed in figure 22 clearly indicates that the tidal flats of the estuary are expanding quite consistently. In the past decade, a small decline can be observed, but overall the increase in relative surface area is highly significant. Another estuary characterised by expanding tidal flats is the Sommes estuary in France. However less pronounced, the increase is observable in figure 23. Similar to the Zhujiang estuary, a slight decline occurs in the period after 2005.
Some other examples of estuaries that display expanding tidal flats in the period 1985-2014 are the Pwani estuary in Tanzania, the Exe estuary in Great Britain and the Canche estuary in France. However, although these cases display a net growth over the entire period considered, more variation is present throughout the individual four research periods.
5. Steepening
An example of an estuary with steepening tidal flats is the Authie estuary in France. Figure 24 displays the
hypsometric curve plots which display the steepening behaviour. During the first three periods (1985-2005) the curves indicate some expansion, but are mostly
characterised by stability. Consequently, in the period after 2005, a rapid steepening of the profile can be observed, together with a further expansion. Some of the relative surface is lost at the HWB, however, the relative surface gained between the LWB and +2 meters elevation outweighs the loss.
Some other cases that displayed steepening behaviour were the Canche estuary, which was also located in France, and Mnazi Bay in Mozambique.
6. Flattening
The final type of morphological development considered in the research is the flattening of the tidal flats. The
hypsometric curve plots displayed in figure 25 represents the morphological development of the Portbail estuary in France. It can be observed that the steepness of the profile was much more pronounced in the period 1985-1987 than it was in the following three periods. Initially, a flattening of the profile can be observed, without great loss of relative surface area. Consequently, post 2005, the further decline of the tidal flat area is unmistakably present. Another example of an estuary with flattening tidal flats is the Ay estuary, which is also located at the Northern coast of France.
Figure 22 - Hypsometric curves of the Zhujiang estuary
Figure 23 - Hypsometric curves of the Sommes estuary
Figure 24 - Hypsometric curves of the Authie estuary
24
Predicted versus measured tidal range
To assess the fitness of the method, a prerequisite is a sufficiently high ratio between the modelled and the measured tidal range. The modelled tidal range originates from the WorldTides model, which is a very reliable baseline for comparison. The measured tidal range originates from the window of analysis that is the HWB-LWB. These parameters are obtained from the collections of satellite imagery and the operations that were performed on them. Figure 26 displays this comparison.
The results can be considered to be quite successful. Considering all of the 60 cases in the dataset, the observation can be made that the modelled tidal range obtained during this research follows the modelled baseline to a more than usable, rather satisfactory level. As expected, a linear relationship appeared to be the best fit. This best-fit line is displayed in figure 27. With an R2 of 0.793, the
accuracy is certainly within satisfactory levels. Indicating that this method can be justifiably used. The estuaries that differ from the fitted line are predominantly estuaries with a larger tidal regime, allowing for more room for error in the calculations. Nonetheless, even these estuaries together fit the plotted line perfectly.
Figure 26 – Predicted average tides versus model window (HWB-LWB)
25
Discussion
During the internship, the methodology has been thoroughly tested. The goals set in advance were definitely reached, however, for each victory achieved, another question arose. Since the research is still in the early stages, a constant feedback between methodology and desire was installed. Several niches can be distinguished in regard to the discussion; case study selection, data availability and internal methodology.
Case study selection
In regard to case study selection, the preliminary issue can be attributed to the availability of high-resolution imagery on Google Earth Pro. Initially, a literature analysis was performed to identify the largest- and most well-known estuaries and tidal flats globally. However, in many instances, the image collections did not range until 1985 or were of such low quality that they proved unusable since tidal flat features were not observable. Consequently, a switch was made in approach. Instead of identifying cases through literature, every meso-or macro tidal coastline globally was surveyed.
Data availability
Some difficulties were encountered regarding the period 1985 – 1987. In the start-up period of the Landsat programme, the global coverage of ground-stations was not yet fully complete, and thus no imagery is available for some of the case studies in the less densely populated areas of the world. Nonetheless, most cases considered displayed sufficiently large image collections for the analysis. Furthermore, in several instances cases were observed for which there was practically no imagery available. Two reasons were identified; firstly, in rainforests, the evapotranspiration results in rapid cloud formation above them. Secondly, the elevation of air over a mountain range creates an orographic lift and thus cloud condensation, limiting the visibility on satellite imagery. Consequently, some doubts should be expressed regarding the application of the NDWI in such divergent cases. Case studies with highly varying tidal ranges, turbidity characteristics and sediment colours were studied. Nevertheless, only a single script was applied to analyse all of the cases. I suspect that the results will further improve when the parameters of NDWI and NDVI are personalised for every case study. Unfortunately, this specific research will require a lot of extra time and energy, which goes against the fundamental desire to construct a semi-automated methodology.
The research window chosen over this research is quite broad, since the focus lies on the explanation of tidal flat morphological behaviour over multiple decades. For this reason, study periods of three consecutive years were selected, with gaps of 9 years in between. For some of the cases considered, one of the periods studied delivered empty image collections, meaning that no imagery was available for this 3 year period. When this occurs, the time series contain a large gap of around 20 years in width. These gaps undermine the predictive power of the model, and automatically lead to a situation in which no solid conclusion can be drawn.
Internal methodology
Another point of doubt regarding the method springs from the consideration of the HWB-LWB windows. Most of the cases display a fair ratio between predicted tidal range derived from the WorldTides model, and the measured tidal range in the study. However, some examples exist (Olhao estuary, LWB = -0.32 m, HWB = 0.39 m). This range of 71 cm is around one third of the predicted tidal range of 2.10 meters. The hypsometric curves are thus composed only on this 71 cm window, leaving the higher- and lower sections of the tidal flats out of the results. It may be argued that the window of analysis still contains the section that is most dynamic and of primary interest, nevertheless, it seems like part of the data is ignored. In addition to this, the methodology for the construction of the hypsometric curves should be mentioned as a point of discussion. In many cases, tidal flat areas in a case study are a mosaic of plates and sandbars. Our method is designed to assess the tidal flat morphological change on an estuary level. The hypsometric curves display the development of the estuary as a whole, while individual plates may in fact show very different behaviour. Applying the transect-methodology might provide further valuable data. Thus, considering the individual plates in an estuary instead of the estuary as a whole may prove valuable.
Furthermore, this paper was written during the initialisation phase of the research. During my internship period, the first results were achieved. Since there are several novel concepts applied throughout this research (LWB & HWB, the reworked NDWI and the stacked-bathymetric DEM method) there were several teething
26
problems, and more importantly, the missing elements of the research became clearer; For instance during the analysis of the hypsometric curves, the six morphological types were still conceptual, and are not attributed a solid scale to quantitatively assess the degree of stability or fluctuation.
Finally, a recurring issue was observed regarding the plots of hypsometric curves. The period 2003-2005 seemed to be repeatedly overestimated. The exact reason for this is unknown. However, a plausible explanation may be found in the fact that the period 2003-2005 is the only period considered, which is composed of imagery from both Landsat 5 and Landsat 7.
Recommendation and future perspective
Significant progress was made during the internship period. A solid methodology for the assessment of tidal flat morphological change has been created and thoroughly tested. This method will prove valuable on many different applications, with its primary opportunity on the assessment of tidal flat morphological dynamics. This research thus focusses on tidal flats. However, with some minor alterations, the series of scripts could be adapted to assess environmental change in many different landforms. Application of other operators than the NDWI pose a wide range of opportunities. For example, an interesting development was the invitation from Saudi Arabia to apply the method on anthropogenically created islands. This internship involved the test-phase of the methodology, and more time for improvement is available. In the discussion, parameter optimization for individual case studies, and a more detailed view applying the micro-transect methodology were discussed. These and several other improvements such as the manual removal of wrongly classified images in script III require more attention in order to reach the best possible outcome. Nevertheless, the initial results are very promising, and multiple scientific papers are underway.
27
References
- Christopherson, R. W., & Birkeland, G. (2015). Elemental geosystems. Pearson.
- Dehenauw, I. D. (2008). De stormvloed van 1 februari 1953: een historische terugblik met moderne technieken.
- Gorelick, N., Hancher, M., Dixon, M., Ilyushchenko, S., Thau, D., & Moore, R. (2017). Google Earth Engine: Planetary-scale geospatial analysis for everyone. Remote sensing of Environment, 202, 18-27. - Jaffe, B. E., Smith, R. E., & Foxgrover, A. C. (2007). Anthropogenic influence on sedimentation and
intertidal mudflat change in San Pablo Bay, California: 1856–1983. Estuarine, Coastal and Shelf
Science, 73(1-2), 175-187.
- Kvale, E. P. (2006). The origin of neap–spring tidal cycles. Marine geology, 235(1-4), 5-18.
- Ladd, C. J., Duggan‐Edwards, M. F., Bouma, T. J., Pagès, J. F., & Skov, M. W. (2019). Sediment supply explains long‐term and large‐scale patterns in salt marsh lateral expansion and erosion. Geophysical
Research Letters, 46(20), 11178-11187.
- Leuven, J. R., Pierik, H. J., van der Vegt, M., Bouma, T. J., & Kleinhans, M. G. (2019). Sea-level-rise-induced threats depend on the size of tide-influenced estuaries worldwide. Nature Climate
Change, 9(12), 986-992.
- Lopes, C. L., Alves, F. L., & Dias, J. M. (2017). Flood risk assessment in a coastal lagoon under present and future scenarios: Ria de Aveiro case study. Natural Hazards, 89(3), 1307-1325.
- McFeeters, S. K. (1996). The use of the Normalized Difference Water Index (NDWI) in the delineation of open water features. International journal of remote sensing, 17(7), 1425-1432.
- Murray, N. J., Phinn, S. R., DeWitt, M., Ferrari, R., Johnston, R., Lyons, M. B., ... & Fuller, R. A. (2019). The global distribution and trajectory of tidal flats. Nature, 565(7738), 222-225.
- Tong, S. S., Deroin, J. P., & Pham, T. L. (2020). An optimal waterline approach for studying tidal flat morphological changes using remote sensing data: A case of the northern coast of Vietnam. Estuarine,
Coastal and Shelf Science, 236, 106613.
- Wang, Z.B., Ding, P. (2016). Coping with deltas in transition. Application form Strategic Alliance phase of the Programme Strategic Scientific Alliance. (KNAW, the Netherlands)
- Wu, J., Huang, J., Han, X., Xie, Z., & Gao, X. (2003). Three-Gorges dam--experiment in habitat fragmentation?.
