Assessing firn processes withinGreenland and Antarctica is important in recent decades, as melt–refreezing processes can result in accelerated meltwater runoff and land-ice discharge. Meanwhile, surface and depth hoar crystal formation have an impact on the surface warming and surface mass balance (SMB) of the ice sheets. Typically, these processes are monitored using in situ firn core measurements, or estimated using climate models. However, the in situ measurements are sparse due to the harsh conditions of the polar regions, while the climate models are based on simplified assumptions which introduce various uncertainties. The recent advancements of satellite remote sensing techniques provide the opportunity to monitor the firn processes over the ice sheets, due to a vast spatial coverage and a frequent revisit time.

This thesis explores the capability of satellite radiometers, scatterometers and altimeters to assess firn processes, including firn density variation and melt–refreezing processes. Conventionally, satellite radiometers and scatterometers are used in detecting melt events over the ice sheets, based on the principle that melt events change the dielectric constant within the firn layer, while the satellite altimeter is typically used for estimating surface elevation changes over ice sheets. This thesis, on the contrary, explores the feasibility of using radiometers and scatterometers to assess the dry-firn density; meanwhile, it assesses the potential of using the altimeters to observe the melt– refreezing events of firn. The rationale behind this thesis is that the long-term variations in satellite radiometer and scatterometer observations depend on the changing scattering properties due to variations in near-surface firn densities, which provides the opportunity for using satellite radiometer and scatterometer observations to estimate long-term changes in firn densities. Meanwhile, the shape of the waveform obtained by a satellite radar altimeter can be influenced by volume and surface scattering of firn. By assessing the variation of firn scattering properties using the waveform information, the occurrence and impact of melt–refreezing processes within the firn layer can be assessed. Therefore, more potentials lie in the application and interpretation of remote sensing data in the studies of the cryosphere. To study the aforementioned application…

The floating extent of the Antarctic Ice Sheet -- the ice shelves -- play a critical role in stabilizing the ice sheet through a process known as buttressing. This effect slows the flow of grounded ice into the ocean and thereby helps regulating the ice sheet's sea level rise contribution. However, ice shelves are highly sensitive to (climate-driven) changes, which can cause thinning and structural weakening. This, in turn, can diminish their stabilizing influence and accelerate ice loss from the ice sheet. Given Antarctica’s vast potential to contribute to sea level rise, understanding the processes affecting ice shelf stability is essential for predicting future changes and reducing associated uncertainties.

Ocean-driven melting at the base of an ice shelf significantly influences its stability by driving ice thinning, grounding line retreat, and through basal channel formation. These channels, formed by meltwater plumes carving pathways along the ice base, are shaped by ice draft geometry, ocean dynamics and temperature. Basal channels concentrate melting and can weaken ice shelves by acting as structural weak points and promoting fractures that may lead to calving and retreat. On the other hand, basal channels can also stabilize ice shelves by localizing melt, potentially reducing overall thinning. Their evolution -- including changes in size, location, and intensity of melting -- is influenced by changes in ice flow and the availability and temperature of circumpolar deep water, which is expected to increase under climate change. Understanding basal channels and their role in ice shelf (in)stability is thus essential for accurately assessing the future behavior of the Antarctic Ice Sheet and its contributions to sea level rise.

In this thesis a method for detecting basal melting at high spatial resolution, called BURGEE (Basal melt rates Using REMA and Google Earth Engine), was developed and described in Chapter 2. BURGEE combines stereo-imagery from the Reference Elevation Model of Antarctica (REMA) with CryoSat-2 elevation data to obtain high-resolution ice shelf elevation changes, which through a mass conservation approach can be translated into basal melt rates. BURGEE's 50 m posting allows for capturing detailed melt patterns previously unresolved in coarser remote sensing products. Applied to the Dotson Ice Shelf, BURGEE revealed spatial variability within a major melt channel, influenced by a pinning point that affects ocean plume pathways. This method was developed to be scalable allowing for applications to other ice shelves to better understand ice shelf melt dynamics and stability across several ice shelves.

Using BURGEE in Chapter 3, high-resolution basal melt maps revealed that melt rates within ice shelf channels have been underestimated by 42-50% in products relying on altimetry-only. This underestimation has a significant impact on ice shelf stability assumptions, for which channel breakthrough times can be used as a proxy. As breakthrough times are highly controlled by the melt rate within the channels, these altimetry-only studies also significantly underestimate the time it would take for a channel to break through. While so far channels have not been observed to actually break through, they have been observed to cause significant fracturing once they reach a thin and vulnerable state. Channel-induced fracturing has further been observed to lead to ice shelf calving and retreat. The faster-than-previously-assumed channel breakthrough times -- and thus weakening -- exacerbates the vulnerability of ice shelves to channelized melting and consequent fracturing and retreat. Incorporating basal melting at high resolution into ice-sheet and ocean models is thus crucial for improving projections of ice shelf stability and global sea level rise.

In Chapter 4, BURGEE has further revealed sudden changes within the basal channel system on George VI Ice Shelf, marked by a 23 m surface lowering over just nine years. This rapid development coincided with increased ocean temperatures and salinity during the 2015 El Niño event, highlighting the influence of large-scale climate patterns on basal melting. The high resolution further revealed subtle shifts in ice flow indicative of fracturing, suggesting a combined weakening effect from basal melting and structural integrity causing changes and possible re-routing of the channel system. Such findings underscore the importance of monitoring dynamic ice shelf channels at a high resolution to better understand and predict their role in ice shelf weakening.

Together, these findings represent a significant advancement in our understanding of basal melting and its impact on ice shelf stability. This thesis has provided the tools and insights needed to detect, quantify, and analyze the spatial variability of basal melting at high spatial resolution. By uncovering the underestimation of channelized melting, identifying key drivers of channel evolution, and linking these processes to ice shelf weakening and retreat, this work has filled critical knowledge gaps. It emphasizes the importance of high-resolution observations and models in capturing the complex interactions between ocean dynamics, basal melting (especially within channels), and ice shelf integrity.

Antarctica, the coldest, windiest, and most remote continent on our planet, plays a crucial role in the global climate system. Its ice mass loss is a major driver of rising sea levels, with projections indicating a potential rise of several meters in the coming centuries. However, there remains considerable uncertainty about the future trajectory of Antarctic mass loss. A major area of uncertainty is the fate of ice shelves—floating extensions of land ice that surround much of Antarctica and act as barriers, slowing the flow of glaciers into the ocean. Ice shelves are affected by warm water from below, which thins them and increases their vulnerability to cracking, as well as by warm air from above, which melts the surface and forms ponds of meltwater.

This research focuses on surface melt, a phenomenon where meltwater forms and either refreezes or accumulates on the ice shelf surface. When the water accumulates, it can seep into cracks, causing them to deepen and widen, which can weaken the ice shelves. In today’s era of abundant satellite imagery and advanced deep learning techniques, we can efficiently process large volumes of data, enabling more comprehensive research on surface melt dynamics. The aim of this dissertation is to enhance the mapping and understanding of surface meltwater on Antarctic ice shelves using remote sensing and deep learning methods.

The introductory chapter provides an overview of the Antarctic Ice Sheet, emphasizing the continent's immense scale and importance. Written in an accessible style, it presents key concepts about Antarctica and explores how ice shelves and surface melt influence the continent. The chapter also describes the use of satellite data to map surface melt and discusses advancements in computational resources and deep learning, which have significantly improved our ability to analyze the expanding catalog of satellite data. It concludes with an overview of the research questions addressed in the thesis.

In the second chapter, various remote sensing datasets are compared to illustrate how and why satellite observations of surface melt differ. Using state-of-the-art melt detection algorithms, we analyze surface melt patterns and observe large differences, especially in icy areas, regions with subsurface melt, and during winter. These differences arise from factors such as satellite overpass times, spatial resolution, signal penetration, cloud cover, and detection methods. Despite these challenges, the variations create opportunities to combine data from multiple satellites, enhancing the overall accuracy of surface melt detection across Antarctica.

The third chapter builds on the previous findings and addresses the challenge of balancing spatial and temporal resolution in satellite observations. Surface melt in Antarctica is highly dynamic and varies regionally, making high-resolution mapping essential. To tackle this, we develop UMelt, a surface melt dataset for all Antarctic ice shelves with high spatial (500 m) and temporal (12 h) resolution, covering the period from 2016 to 2021. Our deep learning model integrates data from multiple satellites, allowing for detailed detection of surface melt while maintaining high temporal resolution. UMelt offers the potential for new insights into how ice shelves respond to changing atmospheric conditions.

In the fourth chapter, we shift from mapping the presence of surface melt to estimating its volume. Since surface melt is mainly driven by local processes, high-resolution regional climate models (RCMs) are necessary. However, current RCMs have a coarse resolution (25--30 km) that is insufficient for capturing small-scale melt processes. To address this, we introduce SUPREME, a deep learning method that downscales surface melt to 5.5 km resolution using a physically-informed super-resolution model. This model combines remote sensing data on albedo and elevation with a 27 km resolution Regional Atmospheric Climate Model (RACMO), accounting for the diverse drivers of surface melt across Antarctica. SUPREME demonstrates the potential of super-resolution techniques with physical constraints for high-resolution surface melt mapping, providing valuable insights into localized melting patterns.

The fifth chapter examines the hydrology of surface meltwater lakes on Antarctica, investigating whether they refreeze or drain into fractures at the end of the melt season, potentially destabilizing ice shelves. Monitoring these lakes with optical satellite imagery is often limited by cloud cover, complicating the tracking of their changes over time. To overcome this, we develop a spatiotemporal deep learning model using radar imagery from Sentinel-1, which allows us to classify the evolution of meltwater lakes regardless of cloud conditions. Our findings reveal no clear connections between lake evolution and ice shelf parameters, highlighting the need for further research and model refinement. The study is an initial step in using deep learning and Sentinel-1 data to monitor the evolution of supraglacial lakes on Antarctic ice shelves.

The sixth and final chapter reflects on the research and outlines future directions. It begins by summarizing the state of Antarctic surface melt research at the start of my PhD. The chapter then highlights the key contributions of this thesis and concludes with three proposed research ideas aimed at advancing our understanding of surface melt processes in Antarctica.

Climate change, with global temperatures rising over the past decades, is a primary driver of sea level rise through the thermal expansion of seawater and the melting of the Antarctic and Greenland Ice Sheets (AIS and GIS). These ice sheets are crucial for predicting future sea level changes, as increased melting forms supraglacial lakes. These lakes can induce hydrofracture, leading to ice shelf instability and accelerated ice flow into the ocean, further elevating sea levels and affecting global climate systems. This study focuses on the AIS and GIS, emphasizing the development and application of a deep learning model to detect and classify the behavior of summer supraglacial lakes using Sentinel-1 SAR satellite data.

The methodology involved normalizing SAR imagery data to enhance data consistency, training a deep learning model using a U-Net architecture on a labeled dataset of Greenland lakes for semantic segmentation, and evaluating its performance on both Greenland and Antarctic datasets using metrics such as accuracy, precision, recall, F1-score, and SSIM. The model is trained to distinguish between draining and refreezing lakes based on backscatter intensity patterns captured in the satellite images. Furthermore, a sensitivity analysis is conducted by creating ten different perturbations of the testing dataset, which included variations in intensity, rotation, and zoom levels. The trained model is then applied to Antarctic data to create an Antarctic-wide map of lake behavior.

The deep learning model exhibited high performance, achieving an accuracy of 90.6%, precision of 90.9%, recall of 90.6%, and an F1-score of 90.0%. It showed high classification accuracy for non-lake pixels (97.5%) and draining lake pixels (84.4%), but lower accuracy for refreezing lake pixels (55.4%) due to class imbalance. Sensitivity analysis revealed optimal performance on the 'zoomed out' dataset with an overall accuracy of 92.9%. Applying the model to Antarctic data successfully identified regions of draining and refreezing lakes, providing a starting point for monitoring ice sheet dynamics and their implications for climate change.

This study underscores the potential of deep learning models to enhance supraglacial lake monitoring, contributing to a better understanding of ice sheet stability and the impacts of climate change. Future work should address class imbalance and explore further model optimizations to improve classification accuracy across both lake types.

The timing and magnitude of global sea level rise remains difficult to predict, driven for a large part by the potential instability of ice shelves in Antarctica. Ice shelves, the floating extension of the Antarctic ice sheet, govern the mass loss of the ice sheet by providing resistance (buttressing) to the grounded ice — thereby modulating the ice flow to the ocean. The short-term collapse or long-term weakening of ice shelves can result in drastic increases of ice discharge and Antarctic mass loss. Understanding the processes that affect the weakening, retreat, and instability of ice shelves is therefore essential in order to improve sea level rise predictions.

Damaged areas on ice shelves, consisting of fractures, crevasses and/or rifts, are first indicators of its weakening. As ice shelves weaken, they can provide less buttressing to the ice sheet, causing accelerated ice flow, heightened internal stress, and increased strain rates. This creates a feedback loop, further promoting damage development and ice mass loss through increased discharge. Moreover, the propagation of crevasses or rifts through the ice shelf eventually leads to calving of (often large) ice bergs. Observable damage is therefore an important precursor to this mode of mass loss. Damage has been considered key for the collapse of the Larsen B ice shelf and the retreat of Pine Island Glacier and Thwaites Glacier. Despite its significance for future ice shelf stability, damage processes remain one of the least understood in marine ice sheet dynamics. This dissertation therefore aims to improve our understanding of damage impacts on ice shelf weakening and retreat from an observational perspective.ined...

The Alps are experiencing a gradual reduction in snow cover due to rising temperatures, impacting the landscape and dependent ecosystems. While several models have been developed to study snow cover in the region, there is a lack of visual representations. This research employs a Conditional Generative Adversarial Network (cGAN) to generate a multispectral Landsat-8 image of the Alps using environmental data inputs. The study utilizes elevation, monthly precipitation, and monthly temperature data from November to March to produce an end-of-winter Landsat-8 image. The resulting multispectral image is then used to calculate the Normalised Difference Snow Index (NDSI) and determine the snow extent by counting pixels with NDSI values above 0.4. Two climate scenarios are considered, and the generated images are compared to actual Landsat-8 imagery. Findings indicate that while the generated imagery closely resembles the real imagery, the snow extent is generally underestimated in the current model configuration, and the snow reflectance is consistently overestimated across all training steps of the cGAN. Additionally, it is observed that increasing the spatial distance between the training and testing locations leads to increased error in the results. The thesis demonstrates the feasibility of using a cGAN to generate snow extent, but suggests that enhancements to the training dataset and cGAN architecture are necessary for improved accuracy. By leveraging the cGAN, future climate scenarios can be visually represented through multispectral imagery, enabling a more detailed understanding of potential future landscapes under different climate conditions.

Atmospheric rivers transport 90% of all atmospheric moisture in the mid-to-high latitudes, while covering only 10% of the Earth’s surface at any given time. Atmospheric rivers occur infrequently, and atmospheric river frequency in the polar areas is especially low, but they can have a large impact on the cryosphere when they make landfall. They have been linked to various processes affecting the surface mass balance, such as extreme precipitation events as well as surface melt.
This thesis addresses the importance of atmospheric rivers in the polar regions. The study aims to compare atmospheric river precipitation estimates obtained from reanalysis data with ICESat-2 satellite altimetry observations in Antarctica between 2019 and 2021, to determine whether the two different types of data sets show similar amounts of atmospheric river precipitation.
To achieve this goal, an atmospheric river detection algorithm designed specifically for polar regions was used to identify atmospheric rivers in Antarctica. All precipitation falling within an atmospheric river footprint for the first 24 hours after detection is attributed to the atmospheric river. The detection algorithm and precipitation attribution are performed to MERRA-2 and ERA5 reanalysis data. The resulting atmospheric river precipitation anomalies were then compared to ICESat-2 height change observations using correlation analysis and a metric based on variance reduction. A detailed analysis
is presented of specific drainage basins that show promising results based on the comparison of the reanalysis data and ICESat-2 observations, using time series.
The results show a high degree of correlation between the atmospheric river precipitation anomalies from reanalysis data and ICESat-2 height changes in multiple drainage basins. Variance reduction shows that atmospheric river precipitation can explain a significant part of the variance of the ICESat2 height change observations in these drainage basins. This suggests that in select locations, the atmospheric river precipitation expected based on reanalysis data is indeed observable in ICESat-2 data. A challenge is the coarse temporal resolution of ICESat-2 data. ICESat-2 data has a temporal resolution of 91 days, whereas atmospheric rivers typically last between a few hours up to a few days
at most, and occur very infrequently (up to ∼2% of the total time over the time period 2019-2021).
Nonetheless, this thesis provides a comprehensive analysis of atmospheric river precipitation in Antarctica using both reanalysis data and satellite observations, contributing to a better understanding of the impact of atmospheric rivers on the surface mass balance of Antarctica. Additionally, it suggests that as long as its limitations are taken into account, ICESat-2 data can be a valuable tool to use in addition to reanalysis data in the study of atmospheric rivers in the polar regions.

Fire both shapes and destroys forests. Forest fires are therefore essential to ecological processes and vegetation as we know them. With climate change, forest fires are expected to increase in severity and frequency. To maintain the functioning of the forests, it is important to understand the vegetation response to and recovery from forest fires. Within this field of study, remote sensing techniques are common, and optical indices such as the NDVI are most prevalent.

With recent developments, the ASCAT variables slope and curvature have become of increasing interest in structural vegetation monitoring. These variables are a second-order Taylor polynomial’s first and second derivatives used to normalise the ASCAT backscatter-incidence angle relationship. This study explored the possibility to use these novel variables in forest fire research.
The focus was to discover to what extent the variables responded to a major forest fire.

To do so, grid points affected during the 2009 Australian Black Saturday Fires are compared to unaffected control grid points utilizing Z-scores. These control grid points have been selected based on time series similarity in the two years before the fire. Time series from 2007 to 2021 are used to investigate fire impact and recovery. The ASCAT variables are compared to a similar
NDVI time series to aid in interpreting the results.

The findings show that both slope and curvature are sensitive to the major forest fire. Both variables show an impact shortly after the fire, which can be explained by the loss of scattering elements in the vegetation due to the fire. In the following years, there is a notable recovery which can be explained by the vegetation regrowth forming new scatterers. The NDVI showed similar
behaviour but the recovery was differently timed, suggesting that the signal recovery is driven by something else or that the regrowth of leaves is different from the regrowth of the structural elements that the ASCAT variables represent.

The results help in understanding the ASCAT variables and their interpretation in terms of vegetation scatterers. Especially for the curvature, the clear change in signal deflects the discussion of whether the variable has information potential. Nevertheless, for the ASCAT variables to be applicable in forest fire research, they need to be better understood and additional research is necessary. Suggestions are for example a ground validation study or a global forest fire study, which would suit the coarse resolution of the ASCAT variables better and improve the understanding of the interaction between forest fires and the variables.

Although using the ASCAT variable for forest fire research is in its infancy, the results about its suitability are promising, both for ASCAT product development, as well as for the forest fire research field. Exploring the possibilities further is worthwhile, especially considering the fact that the slope and curvature time series cover a long continuous period starting as early as 1991. This long time series makes the ASCAT variables suitable for long-term forest fire monitoring, and the daily nature of the data might also make them interesting for short-term monitoring.

Icebergs drifting through the Southern Ocean release fresh water and nutrients. This has local impacts on surrounding ecosystems and sea ice formation. On a global scale, salinity patterns and ocean circulation are affected. In addition,
tudying icebergs as a proxy for ice shelves in a warming climate can help predict future climate impacts and sea level rise. Furthermore, drifting icebergs can pose a threat to ship navigation and offshore projects. In the past, icebergs have been tracked mostly manually, a time-consuming and labour-intensive task. The most widely used data source for this is Synthetic Aperture Radar (SAR), as icebergs often have a much higher backscatter than their surroundings. A few attempts have been made to automatically track icebergs, but these methods do not allow tracking of icebergs that are only partially visible in a satellite image. In this study, a new method is proposed based on partial contour recognition using the contours’ curvature, a technique derived from the matching of ancient pottery fragments. Since the automatic tracking of multiple icebergs requires a large amount of data and computational resources, the web-based environment of Google Earth Engine is used. The new method, called the Contour Curvature (CC) method, is based on three main steps. (1) Detection of icebergs using Simple Non-Iterative Clustering (SNIC) in combination with a threshold function. (2) The icebergs targets are filtered using an area and solidity filter. (3) Among the remaining targets, the best match is selected by comparing the curvature function of the contour with the reference iceberg. The performance of the algorithm is tested by automatically tracking 15 icebergs and comparing
the results to the existing Centroid Distance Histogram (CDH) method. The overall performance of the CC method can be attributed in large part to the inclusion of the area and the solidity filter, with the latter serving as an overall shape filter. For small icebergs (< 10 km2), both the CC and CDH method perform poorly, due to the abundance of icebergs in this range. For medium to large icebergs (10 to 1000 km2), the methods show similar performance with one method occasionally outperforming the other method. For large icebergs (> 1000 km2), the CC method performs better. Since these icebergs are often only partially visible, this leads to strong deviations in the histogram used in the CDH method, making this method less suitable for these situations. Since the CC method allows for partial contour recognition, these icebergs can still be identified. Furthermore, due to the wide variety of backscatter conditions, the detection method occasionally fails to distinguish icebergs from their surroundings.

Transforming the global energy sector from fossil-fuel based to renewable energy sources is key to limiting global warming and efficiently achieving climate neutrality. The decentralized nature of the renewable energy system allows private households to install photovoltaic (PV) systems on their rooftops. In this context, planning an efficient grid expansion is becoming increasingly difficult. Therefore, deep learning (DL) techniques, such as convolutional neural networks (CNNs), can support collecting meta data about PV systems from aerial or satellite images, as research in the field of remote sensing has shown. However, previous research lacks the consideration of ground truth data-specific characteristics of PV panels.

This thesis aims to implement a semantic segmentation model that detects PV systems in aerial imagery to emphasize the relevance of area-specific characteristics for the training data and convolutional neural network (CNN) hyperparameters. A CNN with U-Net architecture is employed to analyze the impacts of land use types, rooftop colors, near-infrared (NIR) data, and lower-resolution images on the detection rate of PV panels in aerial imagery. The results indicate that a U-Net is suitable for classifying PV panels in high-resolution aerial images (10 cm) by reaching F1-scores of up to 91.75% while demonstrating the importance of adapting the training data to area-specific ground truth data in terms of urban and architectural properties.

Disintegration of Antarctic ice shelves can induce devastated consequences for the environment and human infrastructure in the form of an increase of the global mean sea level. One of the causes of an ice shelf break down is hydrofracturing due to the mass load of supraglacial lakes. The top of the snowpack melts and the meltwater flows to a local depression where it accumulates and forms melting ponds. Detecting and quantifying the depth of these supraglacial lakes will increase the knowledge on the evolution of supraglacial lakes. Satellite remote sensing techniques are able to determine the volume of individual ponds. However, these methods have their limitations in calculating the depth and area per lake. Regional climate models are capable of estimating the total volume of meltwater within a certain area of interest, but have until now not been able to measure the depth and area of separate supraglacial lakes with certainty. This research study focused on highlighting the limitations of and developing possible improvements for three climate-based and satellite-based methods for comparing them to one another. The first method made use of Sentinel-2 scenes and a threshold-based classification to calculate the water extent and the water depth was calculated by use of band values and by knowing both the volume can be estimated. The second method made use of Sentinel-1 and Sentinel-2 images to classify areas containing water over a biweekly interval. A new method, denoted as the kernel method, was developed for measuring the depth of each detected lake with a lake mask and a digital elevation model. The volume is subsequently derived from the water extent and depth. Snowmelt, refreezing, precipitation and snowfall from a regional climate model, RACMO, was applied to estimate the total volume of meltwater within a catchment. With a digital elevation model a routing is determined to visualize where the calculated meltwater accumulates and subsequently the depth and area were computed. Based on comparing the water extents over the period of 2016-2021 on the Nivlisen Ice Shelf the following can be concluded: the climate model-based method cannot produce realistic water extents (the results were ten times larger than the satellite-based methods); the different classification methods have similar outcomes, the thresholds of the method using solely Sentinel-2 are preferred; The satellite methods are limited by clouds and frozen ice lids. The results of the water extents indicated unnatural large depths (average +30 $m$) and that the satellite remote sensing methods produce water extent in the same order of magnitude. In addition, the kernel method showed potential, since it can be executed without non-optical satellite data. However, in order for it to improve, the size of the kernel needs to be optimised. The total volumes are in the same magnitude range, but the climate data method overestimates ( getal) due to the fact that the maximum value is chosen within the biweeks. In addition, this study's resulting volumes are close to the values computed by Van der Zalm (2020) and Dell et al (2020), which increases the confidence in the results. However, a problem that arose is the absence of a ground truth to accurately compare the results with and therefore it is recommended to possibly compare the data to that of an altimeter. Additional improvements can be made in kernel size optimization based on the middle line of the lakes and developing a method to correctly locate and calculate the depth of water using a total runoff and a digital elevation model.

Antarctic ocean temperatures are rising due to climate change, causing land ice to melt at increasingly higher rates. Ice shelf bottom melt is a key factor responsible for Antarctic ice mass loss and as such understanding melt processes in the Antarctic is therefore key to more accurately predict how the global sea level will respond to climate change in the foreseeable future. Basal melt results in the formation of both basal melt channels underneath an ice shelf and persistent sea ice wakes (named plume-driven polynyas) at the ice shelf shoreline. The goal of this research is to develop a method that can help to automatically infer basal melt locations along the Antarctic shoreline with significantly increased spatio-temporal resolution compared to previously researched basal melt detection methods.
We infer basal melt locations by detecting plume-driven polynyas. We used dual-pol (HH/HV) Sentinel-1 EW SAR data (40x40m resolution) in combination with GLCM textural features as input for a random forest classification that differentiates images as water or ice in four sub-classes: undisturbed ’open’ water, disturbed ’rough’ water, sea ice and (floating) land ice. We assessed what the advantages and limitations of this approach were for plume-driven polynya detection by performing water-ice (sub-class) classifications and examining which GLCM features proved most useful, what GLCM window size is preferred, and how classification can be aided by post-processing classified images.
We computed GLCM textures for window sizes w = [5,11,21] and created a classifier for each choice (GLCM5, GLCM11 and GLC21) and compared results to a classifier based on original dual-pol SAR data (BASE). Via cross validated recursive feature elimination we determined that ’sum average’ (HH and HV polarization) and ’difference variance’ (HV polarization) were most useful for separation of water and ice classes (HH_savg, HV_savg and HV_dvar). Our results have shown that using GLCM texture based dual-pol classifiers improves water-ice classification significantly compared to dual-pol only classifiers, although using HH_savg and HV_savg instead of orignal dual-pol data comes at a cost of reduced spatial resolution. Water-ice classification accuracy of BASE was 92.2% (kappa = 84.4%) was increased to 95.9% (kappa = 91.5%) for GLCM5, 96.3% (kappa = 92.7%) for GLCM11 and to 96.5% (kappa = 93.0%) for GLCM21. From a spatial context, GLCM21 showed an insufficient ability to detect small-scaled bodies of water at a sub-kilometer scale. GLCM5 showed unsatisfactory results in terms of sea ice classification. GLCM11 showed highest robustness in both these performance aspects and proved to be most successful classifier for the application of polynya detection. Using an area filter as a post-processing step proved successful when a classifier is based on GLCM data with a window size no larger than w=11. Noise output (small regions of falsely classified open water pixels) was heavily reduced via this form of post-processing and significantly increased polynya detection performance.
The final classified product however still contained too many incorrectly classified water regions of similar spatial scales as plume-driven polynyas to be able to apply this algorithm as a reliable automated polynya detection method. We urge to build upon this SAR-based detection method, by using additional non-GLCM input features or using extra post-processing steps, such as temporally filtering water body presence, until results are satisfactory for a fully automated plume-driven polynya detection algorithm. The method presented here has the potential to make detection significantly faster, easier and more accessible than the current methods available. Lastly, in its current state, this method can already be used to validate predicted locations of basal melt by ocean-ice sheet models and DEM-based methods.

This research aims to analyse the sensitivity of the YOLOv5 object detection algorithm to current issues related to the tracking of icebergs in SAR imagery. To this end a sensitivity study was done on (1) the sensitivity of the algorithm to variations in input image resolution, (2) the sensitivity of the algorithm to variations in contrast between an iceberg and its surroundings and (3) the sensitivity of the algorithm to variations in icebergs size. The results show that the algorithm is very robust against variations in contrast between iceberg and surroundings, but is significantly sensitive to iceberg size. Furthermore, it seems that only by using high resolution images, the spatial features of icebergs can be well distinguished from features of other objects in the ocean. The YOLOv5 algorithm thus shows great potential for iceberg detection applications, but it should be explored if the
current sensitivity to size can be overcome if a more evenly distributed training dataset is used. On top of this, it should be noted that this research only serves as an exploratory analysis on the application of the algorithm and it should thus still be explored if our results based on augmented data, also apply on real data.

Satellite data, such as optical and Synthetic Aperture Radar imagery, can provide information about the location and level of destruction caused by natural hazards. This information is essential to optimise the rescue mission logistics by humanitarian aid organisations and save people in need. Currently, many Automatic Damage Assessment (ADA) methods exist, designed explicitly for one data type with corresponding spatial resolution. However, the weather and satellite coverage conditions can hinder rapid and complete data acquisitions after large events. Therefore, it is important to identify the limits and capabilities of novel methodologies testing various data availability scenarios and adjusting them to become robust and widely deployable.
In this research, the Convolutional Neural Network Caladrius of 510 ­ an organisation of the Red Cross Netherlands is selected to perform experiments. Initially, the model was designed to input high-­resolution imagery and based on the Siamese Architecture, including two Inception­-V3 modules fol­lowed by three connected layers. The multiple experiments are based on single­, dual­, and cross­mode scenarios, representing data characteristics with varying resolutions, satellite sources and observation sensor types. The xBD dataset provides pre- and post-­event high­-resolution optical imagery of numer­ous disasters with corresponding validated damage labels of the included buildings. Subsequently, this dataset is replicated in three down­sampled versions and using Sentinel-­2 1C and Sentinel­-1 GRD data. With the use of the Macro F1-­score and the Cohen’s Kappa coefficient the performances are compared and the predictions’ reliability is determined in operational situations.
The results indicate that a lower resolution of the input data has a negative effect on the correct classi­fied buildings. A linear relation does not express the loss in performance, as most damage propertiesare captured between 0.5­ and 2.5-meter. Consequently, this implies that the Sentinel 10­-meter res­olution datasets provide little recognisable features. The Sentinel­-2 1C experiment outperforms the Sentinel­-1 GRD, which equals the output of a random classifier. However, no final conclusion is drawn between the true prediction rate of the model compared to the input data type; optical and SAR im­agery due to the non-­optimal experiment circumstances and limited included datasets. Furthermore,the results from the dual-­mode mapping showcase the importance of identical data characteristics be­tween train and test datasets. Conversely, with the use of the cross-­mode experiments, it is found not essential to match the pre-­ and post­-event resolution imagery. This latter is very promising for the Red Cross and creates flexibility to construct datasets quickly after the disaster has struck.

It is well known that warming of deep Atlantic Water in recent decades resulted in extensive retreat of marine terminating glaciers in Northwest Greenland and increased their discharge, which contributed significantly to sea level rise. Here we use data and model resources over a wide range of space and timescales to determine how the pathways of deep Atlantic Water, through the complex bathymetry of Melville Bay, increased the vulnerability of glaciers over the observed ocean warming period. New observations of salinity and temperature of the ocean water and bathymetry from NASA’s Ocean Melting Greenland mission as well as Mankoff’s discharge estimates are combined with FESOM and HYCOM ocean model results. We have shown that these pathways of Atlantic Water are crucial for understanding the increase in discharge of certain glaciers over the ocean warming period. More specifically, the vulnerability of a marine terminating glacier in Northwest Greenland to Atlantic Water depends on its latitudinal position, the location of the fjordal channel entrance along the Southern or Northern canyon head and whether the fjordal channel is deep enough to be a pathway for Baffin Bay Intermediate Water. The Upernavik N and C glaciers, which are in the most vulnerable position, contributed 10% to the total discharge change of Northwest Greenland. In addition, the glaciers that exhibited the largest normalised discharge change showed correspondence between their discharge estimates and the observed changes in fjord geometry during the retreat of the glacier calving front. Warming of deep Atlantic Water impacted the normalised discharge estimates, but the sensitivity of the fjord geometry also controlled large parts of the observed trends. With this study, new insights in the vulnerability of marine terminating glaciers were obtained, which showed that the pathways of Atlantic Water should not be overlooked when developing climate models.

Subsurface firn processes play a crucial role in ice sheet mass loss mechanisms. On Greenland surface meltwater percolates to deeper layers where porous firn retains it, directly inhibiting runoff. However, secondary effects such as the formation of impermeable ice slabs may indirectly and irreversibly accelerate runoff and with it global sea level rise. Microwave remote sensing offers opportunities to monitor these processes, but due to the simplicity of their underlying snow models retrieval methods fail over areas subject to melt and refreezing - areas where the firn's (in)ability to buffer meltwater is critical. This study presents a new forward model which given initial conditions and atmospheric forcing first solves for the firn state through a full-complexity snow model (SNOWPACK) and then simulates multifrequency brightness temperature (Tb) time series (using radiative transfer model SMRT). As part of a comprehensive sensitivity analysis three ensembles of multi-decade Tb time series (19 and 37 GHz) were modelled for the DYE-2 site in the percolation area of the Greenland Ice Sheet. Model performance based on RMSE w.r.t. independent Tb satellite observations was found to be sensitive to biases introduced in the atmospheric forcing record (with air temperature, precipitation and relative humidity controlling variance) and snow model settings (new snow grain size and albedo settings) and not to initial profile conditions. However, computed RMSEs were high (min. 17.8 K at 37V and 19.4 K at 19V) due to trends in modelled Tb consistently underestimating observed trends when taken over an accumulation season. It is shown that this can only be explained by the constant-with-time stickiness assumption used to link the snow model’s microstructure representation to the sticky hard sphere model employed for the radiative transfer scheme. A seasonal stickiness signal is made evident for the conditions at DYE-2 and linked to its yearly melt-refreeze-accumulation cycle. These results demonstrate that earlier approaches to forward model microwave satellite observations based on a constant-with-time stickiness assumption (or that lack a third snow microstructure parameter altogether) are not valid for ice sheet areas prone to melt. This study is expected to be the starting point for a more sophisticated implementation that estimates a snow layer's stickiness from microstructure information already available in the snow model. If successful it would be the first of its kind and open the door to satellite-based retrieval of subsurface firn properties and processes from areas where observations are currently lacking, greatly reducing uncertainty in ice sheet mass loss and global sea level rise projections.

The Mara Wetland in Tanzania has an important role in regulating the quality, timing and magnitude of the flow of water into Lake Victoria. In addition, the wetland provides natural resources for local communities and habitat for variety of species. The planned dam construction upstream of the wetland and projected changes in the local climate could affect the physical and ecological equilibrium of the system. Baseline information on seasonal inundation dynamics is necessary to sustainably manage these potential threats. The wetland is sparsely instrumented, which has hampered a thorough temporal and spatial understanding of the local water balance. In addition, the highly vegetated nature of the wetland, and relatively frequent cloud-coverage, motivates multi-source integration of remotely sensed data to capture flood patterns at a high resolution.
In this study, the spatiotemporal inundation pattern of the Mara Wetland in Tanzania is reconstructed using optical remote sensing data. The annual fluctuations in aerial wetland extent are analysed in parallel to the fluctuations of local water balance components: downstream water level of Lake Victoria, upstream discharge, direct precipitation and evaporation. The analyses aims to shed light on the underlying mechanisms and hydrological processes that control the hydric status of the wetland. Comparing the temporal changes in extent with surrounding physical processes provides insight on the responsiveness of the wetland to specific water balance components.
The intra- and inter-annual trends in inundation of the Mara Wetland are reproduced for the years 2017, 2018, 2019. The Random Forests (RF) algorithm is trained bi-seasonally (using bands and derived water and vegetation indices from Sentinel-2 data and a Digital Elevation Model (DEM) as input variables), and used to classify the land-covers of the wetland region in a semi-automated way for a total of 73 Sentinel-2 scenes. The scenes are classified into 7 individual land-cover classes; 3 wetland classes (open water, flooded vegetation, wet floodplain) and 4 dryland classes (dry floodplain, wet agriculture, dry agriculture, bare land). The overall classification accuracy achieved (based on an independent validation set, not used to train the classification algorithm) is 98.6 %. The spatiotemporal variability of the inundated area is used in combination with available hydrological field-data to reproduce the local water balance.
The seasonal expansion and contraction of the wetland follows a consistent bi-modal regime, and the results from the waterbalance affirm the importance of local precipitation in the seasonal expansion and contraction of the wetland. The base-flow supplied by the Mara River, together with the backwater from Lake Victoria appear to be at equilibrium at the extent of the permanent swamp during the dry season, insinuating the importance of the riverflow during these low-rainfall months. The occasional yet extreme flood events induced by high discharge rates are expected to play a specific ecological role in the wetland, and should be accounted for during future dam operations upstream.

The coastal area provides important services such as valuable habitats for wildlife, resources for regional development, and buffer zones for the land against natural disasters such as storm surges. But these narrow coastal areas experience pressure from both land and ocean side. In order to regulate sustainable coastal development, protect coastal areas from natural forces, implement ecosystem protection strategies and mitigate the impacts of climate change on coastlines, it is crucial to observe and quantify the changes along coastlines that are vulnerable to these pressures. Among the various coastal environments, the focus of this master thesis is on muddy coasts. A muddy coast is defined as a coastal depositional environment that is usually formed along an unsheltered coastline exposed to low energy conditions and that consists mainly of fine sediments, which are smaller than 63 micrometres in diameter. So far, few methods have been developed to detect sandy coasts and the rate of change of sandy coastlines at global scale, but little is known about the detection and behaviour of muddy coasts. The main objective of this thesis is to develop an automated classification method to identify muddy coasts along all the coastlines of the world. The main objective can be translated into the following research question: How can a reliable global mud classification be obtained by analysing the characteristics of muddy coasts using publicly available satellite remote sensing techniques and globally available coastal datasets? To obtain a reliable global mud classification, a method based on the spectral properties of individual mud patches and on the physical geographical characteristics of muddy coasts is proposed.
The spectral properties of individual mud patches are studied at pixel level using multispectral satellite images and the physical geographical characteristics of a muddy coastal system are studied using seven globally available coastal datasets. The information from the multispectral images and the information from the coastal datasets are used as the input of the development of a hybrid coastal transect classification model, in order to obtain a more reliable and robust classification model that can detect muddy coasts at global scale. With supervised machine learning the hybrid coastal transect classification model is developed, which can classify global coastlines into five coastal types: plain beaches, muddy coasts, coastal cliffs, vegetated coasts and other. 85 percent of the muddy coasts are correctly recognised by the hybrid classification model. The accuracy of the hybrid model is still increasing significantly, meaning that the model could perform even better if more training data is added to train the model. The large amount of training and validation data at both pixel and transect level resulted in a reliable, robust global mud classification model. With the addition of the coastal datasets, the model performs better than when only the multispectral satellite images are used for classifying muddy coasts. The hybrid classification model is used to classify 100000 global coastal transects from which 12 percent are classified as muddy coasts; 60 percent of the classified muddy coasts are in the tropics.

Recent research suggested that the digital elevation models can be considered, as an alternative to the altimetry data. However, the water prohibits the ability of monitoring the construction the waterbed, due to the loss of the returning signal. The elevation models have therefore a flat surface, which prohibits the extraction of the water level and volume variability.

The goal of this thesis was to develop a tool, that uses the elevations where no flatting had occurred, and create a linear representation of the shape of the lake or reservoir. The following research question was formulated:

What is the potential of applying elevation models to monitor the volume levels of lakes and reservoirs, when replacing the flat elevations with extrapolated depths?

The case study comprised of two parts. First, the validation of the water level extraction was conducted for pre-selected lakes and reservoirs. The results showed an average RMSE of 4.10 [m]. When analysing the distribution of lakes and reservoirs data, the RMSE was centred at 1.27 [m] (NED and ALOS) and 3.70 [m] (SRTM). The comparison between the three considered elevation models, showed a minor improvement regarding the relative water level. At the same time, it demonstrated that the number of waterbodies that can be monitored increases when applying the developed tool.

The second part of the case study aimed at providing an insight in the potential of applying the DEM for estimating the volume variability. An analysis of the USGS in-situ data compared to the results, showed that the overall RMSE decreased for each considered model. From the same analysis, the relative time-series could be determined for each model i.e., in case the minimum amount of surface area was available and in case the water level variation was significantly noticeable.

The results from this work indicated the potential of the global elevation models to monitor the volume levels of lakes and reservoirs. The developed tool demonstrated that elevations can be extrapolated, which would fit a more realistic shape of the reservoir at areas where the bathymetry was flattened or not found.

Accurate information for extracting the water level and assessing the volume variability of lakes and reservoirs currently depends on extensive hydrographic surveys. This work provides guidelines for an alternative method: the Linear Bathymetry for Digital Elevation Models tool.

Dikes hold back water and protect the land behind it from flooding. Due to rising sea levels, land subsidence and more extreme weather patterns, the function of dikes become increasingly important. To ensure dike safety, dikes are regularly inspected. With about 22,500 kilometers of dikes in the Netherlands, this is a very slow, costly and time consuming process. Remote sensing could contribute to dike inspections as it can screen large areas in a short time period and more continuously monitor inspection parameters. Several studies have already assessed the use of remote sensing for different inspections parameters such as deformation, grass cover quality and seepage detection. An important parameter that affects dike stability is soil moisture, as effective stress and shear strength are directly related to soil moisture content. Intense periods of drought can lead to low soil moisture values which consequently decreases dike stability. On the other hand, excessive soil moisture can lead to excess pore water pressure and to a decrease in shear strength. Remote sensing would be an ideal way to monitor soil moisture within grass-covered dikes on large scale. In this study, it was assessed if remote sensing data can give a proxy for soil moisture for grass-covered dikes. This was investigated by using open- access optical and SAR remote sensing data, as this would be an ideal data source since it is freely available. Remote sensing data was obtained from satellite missions Landsat 7/8 and Sentinel 1/2. The majority of the research was conducted for two grass-covered regional dikes. First of all, it was assessed if a lagged relationship could be found between the average soil moisture value of a pixel, extracted from in-situ soil moisture sensors at 20 cm depth, and retrieved vegetation indices (GRR, MSR, NDVI, RVI and NDII) of a pixel. Pearson correlation coefficients were calculated for the harmonized Landsat 7 and 8 data set as the number of data from the single satellite missions was limited. Results show that (1) at lo- cation Bermweg a weak correlation was found (R=0.32-0.40) for the MSR, NDVI, RVI and NDII when the optimal lag of around 30 days was applied. A negligible correlation was found for the GRR (R=0.19); (2) at location Geer- weg, for one pixel, a negligible (R=0.12-0.16) correlation was found for all vegetation indices, except the NDII, when the optimal lag of 23 days was taken into account. A negative correlation was found for the other pixel. For the NDII a negligible correlation (R=0.13-0.28) was found for the two pixels when the optimal lag of 31 day was applied. The grass-cover at location Bermweg was maintained by grazing whereas at location Geerweg the grass was maintained by both grazing and mowing. Secondly, it was investigated if a (lagged) relationship could be found between SAR backscatter and in-situ soil moisture measurements at 20 cm depth. An increase in soil moisture results in an increase in backscatter. Since SAR measures only the top few centimeters of the soil, a lag was taken into account. In addition, it is known that there is a lagged correlation between root-zone soil moisture and LAI, which is also sensitive to backscatter (Jamalinia et al., 2019). A Pearson correlation analysis was performed to assess if there was a (lagged) relationship between soil moisture and retrieved backscatter. Only negative and negligible positive correlations were found, showing that SAR backscatter cannot give a proxy for soil moisture, as a positive correlation was expected. Lastly, a relationship was examined between cumulative precipitation deficit, which can give a proxy for soil moisture, and vegetation indices. The sample size of Landsat 8 and the harmonized Landsat 7 and 8 data set were large enough to demonstrate statistical significance (N > 31). Results show that (1) at location Bermweg the optimal correlation was found for both data sets when a cumulative precipitation deficit period of around 15 days was taken into account. The correlation was negligible (R=0.24-0.36) for the harmonized Landsat 7 and 8 data set (statistically significant, with the exception of the NDVI and NDII) and moderate (statistically significant) for Landsat 8 (R=0.38-0.56); (2) at location Geerweg the cumulative period resulting in the optimal correlation was different for each satellite mission. A statistically significant weak correlation (R=0.47-0.57) was found for the harmonized Landsat 7 and 8 data set when a cumulative period of 20 days was taken into account. For Landsat 8 a statistically significant moderate correlation (R=0.49-0.62) was found for a cumulative period of 90 days. The overall pattern of the calculated correlation coefficients, when different cumulative periods were taken into account, vary largely for each satellite mission. All in all, no universal relationship could be found. The study has shown that vegetation indices and SAR backscatter cannot give an indication of soil moisture within dikes. No strong relationship was found between soil moisture and vegetation indices which can be assigned to noise introduced by various factors like management practices (i.e. mowing, grazing), other key fac- tors influencing vegetation state (i.e. nutrient availability, radiation), low spatial resolution, and scene-to-scene variability. These factors also influence the vegetation index and overrule the true soil moisture conditions. Moreover, the results show that cloud contamination hinders the use of optical remote sensing data for dike inspections as satellite imagery might not be available for extended periods of time, disabling to gain insight into the dynamics of vegetation indices over time. The insensitivity of SAR backscatter to soil moisture can be assigned to the fact that several parameters (i.e. surface roughness, vegetation, dike slope) can affect backscatter as much, or more than, soil moisture. Furthermore, the backscatter signal was extracted from a relatively small area and thus contains a large amount of noise. This also explains why backscatter was unable to give an indirect proxy of soil moisture by estimating the LAI.