Global warming is accelerating the melting of glaciers and ice sheets, leading to the formation of supraglacial meltwater, which accumulates in supraglacial lakes and significantly impacts glacier stability. Supraglacial channels can redistribute meltwater and improve glacier stability and are currently monitored using optical data from Sentinel-2. Using Sentinel-1 synthetic aperture radar data omits issues related to cloud cover and darkness, which is the primary focus of this study. The study focuses on three glacier site subsections: Nioghalvfjerdsbrae Glacier, Humboldt Glacier, and Russell Glacier. Sentinel-1 data is compared to the well-established optical data from Sentinel-2, which serves as the reference for channel delineation. The methodology encompasses both thresholding techniques and a GLCM-assisted random forest regression for co- and cross-polarized Sentinel-1 data, while Sentinel-2 data utilizes a thresholding technique based on Glen et al. (2024) and a path-opening algorithm developed by Yang et al. (2015) to delineate narrow channels from enhanced NDWI images. Google Earth Engine is employed for data access and preprocessing, whereas Python is leveraged to conduct the subsequent analysis. Preliminary studies have shown that individual Sentinel-1 scenes may only partially capture supraglacial channels. Therefore, individual scenes are aggregated by taking the backscatter minimum (assumed to be indicative of water due to specular reflection) and standard deviation (representative of change) over a melt season. Composite images offer a more comprehensive view of the supraglacial drainage network during the melt season and form the basis for GLCM texture feature calculations. A 3-by-3 kernel is used around pixels to generate an additional 64 input features by calculating the contrast, correlation, heterogeneity, and dissimilarity over 4 discrete angles for the HH and HV minima and variance. The backscatter features are fed through a random forest regressor with the per-pixel water fraction (i.e. the fraction of which each pixel is classified as water over all images) as the target variable derived from the Sentinel-2 reference classification, providing a more flexible regressor than a binary distinction. Results indicate that channels with sufficient cross-sectional area are detectable in Sentinel-1 data for both polarizations, albeit with varying success. Challenges such as speckle noise, channel size, the heterogeneity of pixel elements, and the movement of glaciers complicate the analysis. The study finds that while thresholding outperforms GLCM-assisted regression, both methods struggle with speckle noise inherent in Sentinel-1 data and alignment issues between the model input and reference classification. Despite these challenges, GLCM-assisted regression shows promising results by leveraging textural information and outweighing backscatter minima and variances in terms of feature importance. To conclude, Sentinel-1 data can be incorporated to delineate channels of sufficient size. Channels approaching the pixel resolution are captured partially at best and require image compositing. The variability in terrain and channel features, alongside persistent noise, limits the accuracy and reliability of this approach. GLCM texture feature analysis and random forest regression show promising results despite the multitude of problems. Other machine-learning techniques should be explored prior to continuing with this approach.

The production of cocoa beans contributes to 7.5% of European Union (EU) driven deforestation. For this reason, the recent European Union Deforestation-free Regulation (EUDR) requires producers to perform comprehensive tracking of cocoa farm extents. However, cocoa crops present unique detection challenges due to their complex canopy structure, spectral similarity to forest, variable farming methods, and location in frequently cloudy regions. Previous work employs Multispectral Imagery (MSI) and/or Synthetic Aperture Radar (SAR) for pixel-based classification of satellite images. Convolutional Neural Network (CNN)s offer
a promising approach to semantic segmentation of cocoa parcels that considers = both spectral and spatial characteristics. This thesis aims to evaluate the impact of combining SAR and MSI data in the training of a CNN for cocoa detection, in order to demonstrate the importance of texture, moisture and canopy characteristics in identifying cocoa canopies. A U-NET is employed to evaluate
how prediction results are impacted by the stacking of MSI datasets with different SAR polarizations, seasons and temporality. The results show that the addition of single-day and temporal SAR to a single-day MSI image can improve the predictions, reaching an F1 score of 86.62%. This research demonstrates the influence of SAR measurement season and polarization, and ground truth classes, on the semantic segmentation of cocoa.

The Greenland Ice Sheet (GrIS) is an ice sheet situated on the island of Greenland. It has a surface area of about 1.74 million km² and contains a volume of ice equivalent to 7.4 m of global mean sea level rise. The GrIS is vulnerable to climate disruptions such as anthropogenic climate change. As a result of increased greenhouse gas emissions, the mass of the GrIS is observed to be decreasing starting in the 1990.
Our research uses output from a climate model called Community Earth System Model version 2.1 (CESM2.1). CESM is an earth system model, meaning that it tries to model the whole earth for its major physical, chemical and biological functions. Importantly for our research, it models the GrIS such that its shape can change, so that it can model changes in atmospheric flow. This is known as an “interactive ice sheet”.
This research three scenarios run on an interactive ice sheet to find its results. It uses a historical run to evaluate the model. Then it also uses two hypothetical scenarios called 3xCO2 and 4xCO2. Each starts at the pre-industrial concentration of 285 ppm CO2, and then increases the concentration by 1\% per year until reaching 3 and 4 times pre-industrial concentrations respectively, after which the concentration is maintained constant.
This thesis focuses on the mass balance (MB) of the GrIS, as these differ strongly between 3xCO2 and 4xCO2. In the scenarios we researched, this mass balance is dominated by the surface mass balance (SMB). SMB is in turn primarily driven by the ice melt. Negative SMB is called Ablation Area SMB
The thesis makes two important observations. Firstly, area distribution can be split up into a total surface area component, which depends on time, and a relative elevation distribution component, which depends on elevation. Secondly, the area normalised AASMB is linear with elevation, but the gradient varies with time.
Typically, increased melt is explained as a result of ablation area expansion. However, this does not capture the elevation distribution of the AASMB. Using the two results drawn from above, we create a new mental model which results in a more detailed explanation of the GrIS mass loss.

In the last decades, climate change is causing our environment to change rapidly, unprecedented in recent history. Civil engineering structures are dependent on the deteriorating environment they are situated in. Changes can cause an increase in loading due to, for example, extreme weather events or alter the structure’s resistance by, for instance, accelerated corrosion or an increase in the number of frost days. However, planning for such events depends significantly on the state of climate change, which is not considered in the sequential decision-making optimization for inspecting and maintaining our infrastructures.

Sequential decision-making optimization refers to the process of finding an optimal sequence of actions to be performed to keep the structure safe. Optimality often refers to the plan with the lowest costs. Traditional inspection and maintenance plans seek a solution by applying heuristic-decision rules on, for instance, time- or condition constraints and fail to find an optimal solution. During the last decade, a new method called Deep Reinforcement Learning (DRL) has been applied and proven to find an optimal strategy to beat traditional approaches. Especially the ability of DRL to find an optimal solution for partially observable environments makes it a perfect candidate for the problem at hand.

This thesis has developed a framework that incorporates partial observability over possible climate scenarios in the decision-making of engineering structures’ inspection and maintenance planning. The framework explains the steps required to translate a physical system towards a Partially Observable Markov Decision Process (POMDP), as the mathematical framework needed to seek an optimal se- quence of inspection and maintenance actions for. Then, it is explained how the POMDP can be used to find an optimal policy with a DRL algorithm, which includes benchmarking and testing the policy. A belief state has been incorporated over the climate scenarios to simulate the partial observability of the climate. Updating over the scenarios is provided by Bayesian Inference, using a climate parameter, such as temperature.

The framework has been applied to a case study in the second part of the thesis. The case study con- figures a stochastic deterioration process for different climate scenarios. An optimal policy has been found by applying Proximal Policy Optimization (PPO) with a decentralized policy for the various com- ponents. Two well-established heuristic-based maintenance policies, time-based maintenance (TBM) and condition-based maintenance (CBM) have been configured as benchmarks for the framework. The framework is compared against the benchmarks using lifecycle costs and safety as metrics. The policy provided by the framework has outperformed both benchmarks in terms of costs while maintaining the same safety. The policy beat TBM with 5% and CBM with 1% in terms of lifecycle costs. Another important distinction is that the benchmarks are optimized for the climate scenarios, while the frame- work finds this distinction without prior knowledge. It is therefore concluded that the framework can find an optimal policy under the uncertainties related to climate change. The case study, however, does not fully capture the complexity of engineering structures that the framework can catch. It is therefore recommended to use the framework for a more complex structure in the future.

This study is an analysis of sand mining in the Vietnamese Mekong Delta (VMD) with the use of the optical satellite data set PlanetScope. This is done with a detection and classification model of sand mining vessels in the VMD. The classification model is based on machine-learning and it is trained with three classes: sand mining vessels, other vessels, and background. This study shows that it is possible to distinguish vessel types with a 3 metre spatial resolution and it shows great potential for a vessel classification model based on machine learning.

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

In this thesis, an easily reproducible modeling approach was developed for assessing the climate change impact on streamflow. This approach was tested by using it to assess the impact of climate change on streamflow in 5 different contrasting catchments across the United States. Many studies show that climate change is expected to influence streamflow regimes all over the world. However, these studies are often difficult to reproduce because the modeling approaches used are usually only locally applicable. In the approach used in this study, hydrological model calibration and validation were done using open-accessible ERA5 forcing together with observed streamflow data provided by the GRDC. The model performed best in a mountainous catchment, while the worst performance was found in a dry catchment and a catchment containing several lakes. The low performances here are mainly caused by imperfect forcing data used for calibration and the neglection of lake processes. The climate change impact analysis used forcing from two CMIP6 models with the SSP245 and SSP585 scenarios. The projections showed significant changes in streamflow in colder regions, which are most likely related to changing snow melt processes. The main finding in warmer regions is that streamflow is generally expected to decrease in the drier periods. Changes of streamflow in these regions are most likely related to changes in precipitation and evaporation processes. However, results remain very uncertain due to disagreements between climate models and sometimes doubtful performance of the hydrological, caused by oversimplification of the model and imperfect ERA5 calibration data. The designed modeling approach facilitates reproducibility of climate change impact analyses in a wide range of catchments using different climate models and scenarios. Its use makes it easier to expand similar analyses to a large ensemble of these aspects.

The Greenland ice sheet (GrIS) is an important component of the climate system and is a key contributor to future sea level rise, as it is storing frozen water that would raise sea levels by 7.4 m should it all melt (Bamber et al., 2018). Of particular concern is the amount of global warming we are facing now and in the future, as it is becoming more likely that even if our emissions are significantly reduced, global warming will reach at least 2∘𝐶 (Arias et al., 2021). Much research concerns the future contribution of the GrIS to sea level rise for high emissions scenarios and low emission scenarios, but there are few studies giving the main focus to what is becoming a more likely future, the moderate emissions scenarios. This research aims to quantify the mass loss of the Greenland ice sheet and subsequent contribution to future sea level rise under a moderate CO2 concentration scenario over a multimillennia timescale. An idealised simulation of 3000 years, where CO2 concentrations are increased by 1% annually until reaching two times pre-industrial values and then kept constant, is run with the high-resolution Community Earth System Model version 2.1 (CESM2.1) and Community Ice Sheet Model version 2.1 (CISM2.1). The climate, run with CESM2.1, is simulated for 1000 years. After 500 years, it is assumed that the climate is close to equilibrium, and thus one climate year is used for five years of forcing the ice sheet in CISM2.1, resulting in 3000 years of ice sheet simulation. At the end of the simulation, the global mean annual temperature has increased by 5∘𝐶 and the temperature over Greenland is 9∘𝐶 warmer than pre-industrial. The rate of sea level contribution in the first centuries is lower than the observed contemporary mass loss of 0.7 mm/yr (Shepherd et al., 2020) but increases after year 710 to a rate of 1 mm/yr. Another increase in mass loss is happening from the year 1380 until the end of the simulation where the rate is 2 mm/yr and the total contribution to sea level rise is 4.1 m. The limited mass loss in the period between years 71-400 and its increase thereafter is found to relate to temporal strong weakening and posterior recovery of the NAMOC. This study projects that the GrIS is a major contributor to future sea level rise, even in a moderate warming scenario, and that the changing NAMOC has a noteworthy effect on the GrIS mass budget.

This article uses Sentinel-1 satellite images to identify rice fields in Rwanda from 2017 to 2021 in an attempt to derive the current status of rice in Muvumba catchment, in the northern part of Rwanda. The timing of rice cultivation in each season in that area was identified as not homogeneous, but generally aligned with the local rainy season. The results identified after late 2019 show a large change, with flooding from extreme rainfall as a possible cause. The paddy fields may have been completely flooded and the infrastructure destroyed. Policy changes in Rwanda’s agriculture may also be a contributing factor. Among them, the implementation of land consolidation policies can influence some of the farmers’ options, for example, by withdrawing from the rice planting program. The absence of official data and field data makes this project not being able to provide definite reasons for the results. The open discussion of this project makes it exploratory and offers the possibility for potential follow-up studies. The uncertainty of the results also suggests that more attention needs to be focused on these topics. However, what is certain is that the use of remote sensing images to monitor rice has the potential to be cost-effective.

Seasonal snow is the major water resource of more than a billion people around the world. In a plethora of regions in the Northern Hemisphere, agricultural, industrial, and drinking water supply are highly dependent on seasonal snow. In addition, the melting of seasonal snow regulates the magnitude and timing of high and low flows, controls the length of the growing season, and determines land surface warming via the albedo feedback. The aim of the study is to quantify the local and regional dynamics of snow cover in the Taurus mountain range and to identify the main climatic drivers responsible for the detected snow cover variability. A data-driven approach based on satellite observations is followed for the systematic analysis of large-scale snow cover in the region of interest. Compared to various remote-sensing snow cover studies that focus either on small/catchment scales, with limited spatial context, or on continental scales that cannot provide detailed insights into a specific region, this study investigates local and regional spatial patterns and temporal dynamics of snow cover across a specific mountain range of interest. The objectives of this study are: (a) to quantify the snow cover temporal variability in the different sub-regions of the Taurus mountain range, (b) to analyze the trends and to identify the regional differences, and (c) to examine the sensitivity of annual snow cover duration to inter-annual climatic variability. The Taurus Mountain Range is divided into sub-regions, using the WWF HydroSHEDS Basins Level 3 dataset and the Köppen–Geiger climate classification map, and in100-m elevation bands. The temporal variability of snow cover is quantified by the Regional Snowline Elevation (RSLE). The Regional Snowline Elevation (RSLE) is estimated in Google Earth Engine (GEE) using the methodology developed by Krajci et al. (2014). The temporal trends of the annual number of snow cover days (Dsc), for the different elevation zones in each sub-region, are derived from the RSLE time series and are analyzed using a modified Mann-Kendall (MK) non-parametric test. The sensitivities of Dsc to the inter-annual variability of winter temperature and precipitation and the snow cover duration trends are estimated in each tile and for all elevation bands in all sub-regions. In general, the analysis carried out, considering its limitations and uncertainties, found that there is no reason to be concerned about future water shortages in the area of the Taurus Mountain Range due to the decrease in the amount of water stored as snow. This is a positive outcome that indicates that there are no significant future changes in snow cover patterns and, as a result, the availability of water in the region will not be at risk.

From three coherent SAR images it is possible to estimate three interferograms. Combined in a circular way, the sum of the three interferometric phases is called the closure phase which necessarily adds up to zero on a single pixel level. However, if the interferograms are spatially averaged, phase consistency is not guaranteed. In most of the interferometric studies, those mismatches were assumed to be caused by decorrelation noise alone, and were either not considered or deemed negligible, eluding further investigations of its origin. However, recent publications have confirmed that inconsistent phase closures are systematic and not the exception, pointing to an underlying geophysical cause. Comparisons of the spatial signatures of phase closures with land cover maps suggest a spatial and temporal correlation that is related to the characteristics of different land cover types. Since interferometric measurements are sensitive to variations of the dielectric constant, those similarities have been attributed to dynamics in vegetation and soil moisture. A closure phase significance test developed at the Geoscience and Remote Sensing department at TU Delft aimed to increase the signal-to-noise ratio of this geophysical signal component by providing a significance ratio for phase closures. However, the sensitivity of (significant) phase closures to dynamics in vegetation and soil over different land cover types has not been assessed yet. Here we show that with enough averaging of the interferometric phase, the spatial and temporal characteristics of closure phase can be used to distinguish between different land cover types. We found that the degree of spatial averaging has a significant impact on both the phase closure values and its spatial and temporal consistency. The magnitudes of significant phase closures generally increased over low-vegetated land covers, suggesting that closure phases are most sensitive to soil moisture dynamics, whereas vegetation cover was associated with decreasing phase closure magnitudes and spatial inconsistency. Besides spatial averaging, significant differences were observed between closure phases from different polarizations. Furthermore, we found that amplitude backscatter and closure phase are spatially and temporally correlated, pointing to similar influencing mechanisms. Our results demonstrate the importance of applying a closure phase significance test and describe the effect of spatial averaging on the characteristics of phase closures with respect to different land cover types. We anticipate this study to provide useful steps towards using the closure phase for soil and vegetation monitoring in the future. For example, the findings could be used to further exploit potential synergies with amplitude backscatter for soil moisture retrieval from closure phase or develop more sophisticated methods for land cover mapping using InSAR. If not used for applications linked to land cover, vegetation or soil, being able to better predict the effect of those parameters on the interferometric phase and coherence, eventually enables to separate their contribution from other signals, such as deformation estimates. Additional research is needed to relate significant phase closures to moisture changes in vegetation.

Plastic pollution is one of the most challenging global environmental problems. Currently, more than 1000 rivers transport approximately 80% of the plastic influx into the oceans. Naturally, more and more companies are interested in tackling this problem. One of them is Noria Sustainable Innovators, a company based in Delft (Netherlands). It is focussed on the detection, removal, and reuse of plastic from Dutch waterways. The company has the ambition to automate the detection of plastic for a wide range of applications. The quantification of plastic and understanding its spatiotemporal variability are crucial for the mitigation of plastic pollution. Current monitoring methods (e.g., visual counting) are tedious, time-consuming, and labour-intensive. Furthermore, the detection of different plastic debris objects could provide more insight about the source of plastic pollution. This thesis explores the feasibility of automating plastic detection in waterways using modern deep learning (DL) algorithms named convolutional neural networks (CNNs) with image classification and object detection techniques. To train these models, a large dataset is required. Due to the unavailability of data, images were gathered in a controlled environment with two GoPros and a Huawei P30. The data was aggregated during sunny and cloudy conditions, different camera heights (2.7m and 4.0m) and angles (0 and 45 degrees). For the simplest case (2.7m/0 degrees), a maximum accuracy of 87.6% was obtained for the multiclass classification of plastic debris in images, using the DenseNet121 model. By applying a majority vote for the three best performing models (DenseNet121, ResNet50 and InceptionV3), the accuracy could be increased to 91%. A qualitative and quantitative analysis found that the following factors influence the model performance negatively: presence of organic material, wind, transparent objects, submerged objects, small objects, overlapping and occluding plastic debris, sun glint and reflection of other objects on the water surface. Sunny conditions yielded a lower accuracy (79%) than cloudy conditions (90%), explained by the presence of sun glint. By applying object detection, the error sources influencing the model performance could be reduced. For training and testing data from 2.7m/0 degrees on one class (‘plastic debris’), the YOLOv4 model yielded an accuracy of 95.61% (GoPro). For four classes (e.g., plastic bottles, other plastic, paper, metal tins) an average accuracy of 66.04% was found, indicating that the model experienced difficulties distinguishing different floating debris in water. Furthermore, it was also shown, that the use of a different image source (Huawei P30), does not have a negative effect on the accuracy (96.63%) compared to the original image source (GoPro). Furthermore, due to height differences, discrepancy in object sizes and different camera settings, the trained model had large difficulties generalizing to a dataset from Indonesia (12.23%). On the other hand, training on the dataset from Indonesia and testing on the dataset from 2.7m/0 degrees achieved an accuracy of 63.51%. Although the error sources could be reduced, the model was still negatively impacted by small, transparent objects, submerged objects and the presence of sun glint. This study clearly showed that Deep Learning-based computer vision can detect floating plastic debris with a high accuracy and have the potential to automate the process of plastic detection in the future. Future work would comprise the following aspects: sensor improvements (polarising filter for sun glint and multispectral sensor for continuous monitoring), data collection from the natural environment and different image sources, implementation of guidelines for Citizen Science platforms, addition of an object tracking module for monitoring (YOLOv4) and focussing on the detection of specific plastic debris objects after the removal from waterways.

Protecting forests from agricultural expansion and wildfires while the world population is growing and the climate is warming remains one of the biggest challenges humanity currently faces. While global modelling and regional observation based studies have found significant effects from deforestation on precipitation, leading mostly to drying precipitation trends and shorting rainy seasons, this study represents the first global estimate of first order deforestation effects on precipitation. Using a recently developed precipitationshed database and actual deforestation data, a new measure is developed to quantify potential deforestation impact per grid cell which in turn is related to annual precipitation trends as well as seasonal differences in tropical regions. In seven regions analysed, a majority of subregions suggested a relationship between deforestation impact and a relative drying precipitation trend in the 2001-2018 study period compared to the long term average. While these results provide further evidence of deforestation contributing to a downwind drying precipitation trend across different continents and climate regions, five other regions studied showed no significant relation or suggest relative wetting related to deforestation impact. One of this regions is the South America Tropical (SAT) region, the region most well-known for its widespread and intense Amazonian deforestation. The two regions downwind of the SAT region however are highly impacted by SAT deforestation and experience most relative drying in the areas impacted most impacted by deforestation, suggesting strong teleconnecting effects. In the seasonal analysis, only two out of four tropical regions studied show more subregions linking deforestation impact to relative drying in the first wet month compared to the wettest month. While these results provide new insights into the global influence deforestation can have on moisture availability, more research needs to be done into the indirect and feedback effects related to deforestation. Additionally, a more robust way of including other factors influencing precipitation trends like large scale oscillations could further enhance the understanding of this important issue.

Glacial lake outburst floods (GLOFs) are outbursts caused by the failure of glacial lake moraine dams. Longer ongoing processes, such as moraine dam degradation, or instantaneous events, such as landslides, can trigger dam failure. GLOFs have a catastrophic downstream impact leading to significant economic damages and more than 12000 casualties worldwide until 2015, with Bhutan and Nepal being impacted the most. Climate change causes increasing temperature and precipitation, leading to the expansion of glacial lakes and the destabilisation of glaciers, slopes and moraine dams. Consequently, GLOFs are likely to become more frequent, and glacial lakes require continuous monitoring and analysis to understand and predict GLOF-related hazards.

Since glacial lakes often lie in inaccessible mountainous regions, on-site monitoring is challenging and remote sensing proposes a safe and cost-effective solution. Satellite radar is unaffected by nighttime and clouds, enabling continuous displacement measurements. Interferometric synthetic aperture radar (InSAR) using Sentinel-1 data from 2014 to 2021 was applied at six Himalayan glacial lake areas (Imja, Lunana, Barun, Rolpa, Thulagi and Lumding) to identify potential GLOF hazards and to investigate InSAR's capability as a monitoring tool. Optical, meteorological and topographical data were used to aid in interpreting the InSAR observations; linking displacements to potential hazards and evaluating the limitations of an InSAR-based analysis.

Significant deformation was detected at the terminal moraines of Imja, Thulagi, Rolpa, Lunana and Barun Lakes; on lateral moraines at Rolpa and Lunana Lakes; and on rock glaciers at Imja, Rolpa, Barun and Lunana Lakes. In addition, significant seasonal variation could be distinguished, showing the impact of temperature and precipitation on geomorphological processes and potential hazard developments at glacial lakes. InSAR-related limitations arose in regions with significant topographic variations, extant snow or vegetation covers, and rapid displacements.

This study demonstrates the capability of satellite InSAR as a glacial lake monitoring tool. An InSAR-based analysis is instrumental in highlighting areas from where GLOFs could originate, requiring mitigation measures or further investigation to map the impact of failure. By extending the research frame over multiple years, continuous and long-term monitoring could demonstrate the climatic influence on displacements and GLOF trigger developments.

Droughts are considered to be one of the most damaging, yet least understood, natural hazards of all. Despite their prevalence, a thorough understanding of them lacks because they are such complex phenomena, and their manifestation can differ depending on the region they occur in. Monitoring hydrological variables and processes is imperative for a good understanding of how droughts develop and persist. Backscatter from ASCAT and previous scatterometers has long been used for soil moisture retrieval. The first and second order derivative, slope and curvature respectively, of the backscatter - incidence angle relation in the TU Wien Soil Moisture Retrieval algorithm are used to correct for vegetation effects. Recently, new developments to this algorithm have allowed to account for interannual variations in the slope and curvature. This has given rise to the potential of monitoring vegetation directly with slope and curvature, rather than only using it to correct for vegetation effects in soil moisture retrieval. The long data record of ASCAT and previous scatterometers combined has the potential to provide valuable information for drought monitoring. This study investigates if ASCAT could be used as a self-contained dataset in drought monitoring. The spatial variability, the seasonal cycle, and the drought response of backscatter, slope and curvature across different vegetation types in Australia is assessed. Simulated surface- and root zone soil moisture, LAI and GPP from the land surface model ISBA are used to aid in the interpretation of the ASCAT signal. The results from this study show that backscatter, slope and curvature can adequately capture vegetation dynamics in times of drought across dry semi-arid grasslands and croplands. Over these regions the soil moisture and vegetation anomalies observed with ASCAT and simulated in ISBA correspond well. Considerable information into the vegetatin dynamics can be gained from analyzing the backscatter - incidence angle relationship. Especially the ability to monitor drought in crops with a coarse spatial resolution is promising for future applications. It proved more difficult to accurately capture the propagation from a soil moisture anomaly into vegetation anomaly across forests and mixed vegetation with grasses and trees. The first reason for this is the increased attenuation of the signal by vegetation, which hampers accurate measurements of soil moisture content. The second reason is that it is more difficult to separate the soil moisture and vegetation effects due to the fact that less is known about the scattering mechanisms induced by vegetation structure and moisture distribution. Overall the results support earlier findings the slope can be used as a measure of vegetation wet biomass and confirm that curvature is also a valuable source of information that gives insight into the relative contribution from surface or volumetric scattering to total backscatter. These relations have been shown to also adequately describe vegetation dynamics in times of drought.

Meltwater features play an important role in the stability of the Antarctic ice shelves. They can destabilize ice shelves by exerting additional loading forces due to the weight of the water concentrated on one location. However, it is possible that developed meltwater networks transport water from an ice shelf into the ocean. In this way, they prevent this destabilization. A better understanding of the development of meltwater features on an ice shelf is needed to make predictions on future ice shelf stability. This research investigates the development of meltwater features during one melting season on the Nivlisen ice shelf. The investigation consists of three elements. First, satellite imagery is used to analyse the change of meltwater area and volume. Next, a potential routing network is created to study the influence of topography on the shape of meltwater features. Finally, climate data is compared to the calculated volumes of the meltwater features. This study shows that it is possible to create a routing network that predicts the shape of meltwater features very well. It is essential that surface depressions are correctly incorporated in the flow routing algorithm. The currently available climate data underestimates the meltwater production significantly. The growth of observed meltwater features during the melting season is not reflected in the modelled melt data. To further study the development of meltwater features, the representation of hydrological processes in climate models has to be improved.

This research focuses on using Sentinel-2 optical imagery to provide a means of high-resolution monitoring and evaluation of changes in Total Suspended Matter (TSM) concentration in the Brantas river basin. In situ spectral measurements as well as laboratory results show an extremely turbid nature of the Brantas River surface water. Current monitoring of the river water quality, is done by point measurements representing point estimations of the water quality in time and pace. Interactions within the system are mostly unknown. Having accurate knowledge of near real time water quality information will greatly enhance the effectiveness of the monitoring organizations, especially if this comes in a high
spatial and temporal resolution. The Sentinel- 2 remote sensing platform delivers information which can be used to derive such data with a 10m resolution and revisit time of 5 days. To estimate TSM concentrations a multi-conditional algorithm is developed. It uses linear regression for low to medium
TSM concentrations based on the green and red band reflectance values and polynomial regression for high to extremely high TSM concentrations based on the red edge NIR band. Testing the multi-conditional algorithm on the WISP-3 in situ spectral data shows the model’s performance is good with r2 = 0.79, RMSE
= 66.5 mg/L and NRMSE = 9.7%. Performance of the multi-conditional algorithm is found to be poor when based on Sentinel-2 (S2) bottom of atmosphere data from bands green, red and red edge NIR. However, when recalibrating the polynomial model on Sentinel-2 atmospherically uncorrected top of atmosphere
data, results are more promising: r2 = 0.75, RMSE = 64.2 mg/L and NRMSE = 11.3% . Also, TSM estimates from remote sensing reflectances atmospherically corrected by different processors are compared, from which ACOLITE (RMSE = 5.0 mg/L, NRMSE = 25.3%) performs significantly better than C2RCC (RMSE =
11.3 mg/L, NRMSE = 57.5%) and Sen2Cor (RMSE = 42.8 mg/L, NRMSE = 217%). This study shows that 1) high-resolution spatial and temporal variation of TSM concentration estimation can be made visible within the Brantas river basin, 2) an overview of TSM concentration estimation of the entire basin at one
moment in time can be achieved and visualised, 3) an extensive historical record of TSM concentration estimations can be accessed, and 4) information is provided to prioritize sampling locations and field surveying times.

Remote retrieval of Normalized Difference Vegetation Index (NDVI) over the Earth’s surface is a critical component of monitoring the surface processes of our planet. NDVI is a widely used and useful indicator of vegetation health and quantity ­ however its retrieval using satellite data is hindered by the frequent presence of clouds in the Earth’s atmosphere. Zeng et al. (2016) developed a novel technique that estimates a surface's Bidirectional Reflectance Distribution Function (BRDF) with a Ross­Li­Maignan (RLM) BRDF model from a set of observations. This method, the Changing­Weight Iterative (CWI) method, uses iterative a posteriori estimation of observation errors to reduce the impact of cloud-contaminated measurements in the sample. Its performance was compared to two conventional methods, ordinary­least squares (OLS) and Li­Gao BRDF-fitting. The three different BRDF­fitting methods were compared in a numerical experiment. 6,000 surface types covering a broad range of surface types were modeled using the canopy radiative transfer model PROSAIL. For each surface, sets of pseudo-observations of the surface’s red and NIR band reflectance were generated using realistic sun­target view geometries from the MODIS and MERSI satellite sensors. The effects of cloud­contamination were simulated by adding different numbers of cloud­contaminated observation to the sample, with varying degrees of contamination. The RLM BRDF model was fitted to these samples using the three different methods to estimate the BRDF model parameters. These were subsequently used to calculate a NDVI composite value. Each method’s estimate was compared to a reference ­value generated by PROSAIL. Results for the 6,000 surfaces confirmed that the CWI method is more noise­resistant than OLS and Li­Gao in situations with many observations (i.e. a large sample), and resulted in estimates that more closely matched the reference value from PROSAIL, compared to the conventional Li­Gao and OLS methods. In scenarios of low­cloud contamination, all three methods failed to detect and significantly suppress the impact of noisy observations, which was expected from existing literature. For a large­sized sample of 13 pseudo­observations studied for the validation site Mongu, Zambia, the CWI method was observed to have a very accurate performance, for up to 5 contaminated observations in the sample. With smaller sized samples of 8 and 10 for two other validation sites, it was found that the RMSE of the CWI method would suddenly increase approximately tenfold when the number of contaminated observations increased beyond 2 and 3, respectively. After these ’tipping points’, the Li­Gao method was more accurate and outperformed CWI. The CWI method therefore performed promisingly when given a large enough sample size, and in these cases it was more accurate than the conventional Li-Gao and OLS methods. However, when it fails to correctly identify noisy observations, its accuracy could decrease suddenly, which should be taken into consideration for operational use. Since the results of the experiment were averaged over 6,000 different sampling points of the PROSAIL model's parameter space, it is suggested that the conclusions apply to a wide range of surface types found all over the Earth.

In this study, an unsupervised classification approach is used to investigate and characterize the spatial and temporal variability of MetOp-A ASCAT backscatter (σ◦) and the TUW SMR vegetation parameters across mainland France between 2007 and 2017. Currently, soil moisture data is retrieved from ASCAT backscatter measurements using the TU Wien Soil Moisture Retrieval (TUW SMR) approach. To correct for the influence of vegetation on soil moisture, two so-called ’vegetation parameters’ are also estimated from the backscatter measurements. These vegetation parameters are the slope (σ′) and curvature (σ′′) of a second-order Taylor polynomial which describes the incidence angle dependence of backscatter. Recently, Steele-Dunne et al. (2019) showed that the slope and curvature contain significant information about vegetation phenology and vegetation water dynamics across the North-American grasslands, suggesting that the vegetation parameters are a potentially valuable source of information on vegetation dynamics. This study further investigates the value of the vegetation parameters as a source of information about vegetation dynamics for land cover types present in France. 3492 ASCAT grid points were clustered using agglomerative hierarchical clustering based on σ′ and subsequently analysed. The results show that clusters based on σ′ are contiguous and can resemble distinct land cover features; areas like Paris and the Alps are clearly visible in cluster maps. While the clusters differ in terms of σ′ – which follows from a clustering based on σ′ – the results show that the clusters generally also have distinct σ◦ and σ′′ characteristics. This suggests that the clusters represent ’scattering surfaces’ that differ in terms of their seasonal scattering characteristics. It was found that grid points with a heterogeneous land cover footprint tend to have noisy seasonal backscatter signatures, while homogeneous land cover footprints have more recognizable seasonal behavior. Additionally, certain backscatter signatures tend to correspond to certain land cover footprints; in particular the agricultural area around Paris produced clear σ◦ , σ′, and σ′′ signatures corresponding to specific growth stages of wheat and the rapid land cover change during the agricultural season. In general, the results are consistent with the existing assumptions that σ′ is a measure for vegetation density and σ′′ is a measure for the relative dominance of ground-bounce and direct scattering from vertical vegetation constituents. Finally, clustering was performed on ten years of dynamically estimated σ′ and a measure for robustness was introduced to quantify the clustering certainty for each grid point. Robust grid points are found in areas that have relatively stable land cover like the Alps or Paris, suggesting that these areas exhibit predictable seasonal backscatter behavior with low interannual variability. Poor robustness scores are mainly found in north-west France, where land cover is heterogeneous and seasonal backscatter behavior is highly variable. This study confirms that the TUW SMR vegetation parameters contain valuable information about vegetation phenology across different land cover footprints. Furthermore, it was shown that unsupervised classification methods based on the vegetation parameters are able to identify areas with similar scattering characteristics, and are able to show how these areas change over time.

Methodology
Agriculture is an important sector to provide in our needs. Since 2000, the arable land per farmer increased with 66%. To help farmers managing their increasing amount of arable land, it is important to develop new techniques in example based on remote sensing data. One of the key parameters for agricultural management is the water availability in the area. The water availability can be clarified with help of two different soil water properties, soil water holding capacity and soil moisture content. The aim of this study is to estimate soil water properties at a scale of 30 meters based on bare soil surface reflectance and thermal infrared image analysis. In this study, two newly developed models are proposed to determine the spatial patterns of soil wetness. The first model is based on bare soil surface reflectance, named as SoilGrids30m, a spatial estimation model for clay content and organic matter content in the study area. These estimates are then used as input of pedotransfer functions to obtain spatial estimates of soil water holding capacity in the study area. The second model is based on the relation between the normalized difference vegetation index and the crop temperature, named as the soil wetness indicator. The soil wetness indicator is a representative of soil moisture content in the root zone of crops. Both models are evaluated against the soil moisture content estimates obtained from SEBAL. Soil water holding capacity is not directly related to soil moisture content. The hypothesis is therefore, soils with a high soil water holding capacity will tend to have higher soil moisture content estimates and vice versa during a long period of drought. The year 2018 has been used as reference year because of the extreme drought conditions in the months May until July.

Conclusions
The SoilGrids30m model improved the estimation of clay content and organic matter content in the study area compared to the existing coarser SoilGrids250m and SoilGrids1000m models. Especially, the organic matter content estimates significantly improves with help of the SoilGrids30m model. However, the influence of adding bare soil surface reflectance to the model was relatively low for both clay content (37%) and organic matter content (13%). The influence of bare soil surface reflectance only improved the clay content estimates compared to the SoilGrids1000m model. In all other cases, the target variable estimates did not improve by adding bare soil surface reflectance data. It can therefore be concluded that the improvement of the SoilGrids30m model is due to the use of a larger set of observation data from the study area and not because of the input of bare soil surface reflectance data.
The obtained soil water holding capacity estimates with help of pedotransfer functions did not show reliable results compared to the soil moisture estimates of SEBAL. The highly empirical pedotransfer functions are the main cause of the unreliable results but also seepage and irrigation are important factors that could have played a role. In future work it would therefore be recommended to avoid using pedotransfer functions calibrated in other regions than the study area.
The results of the soil wetness indicator showed a clear correlation with the soil moisture content estimates from SEBAL. The soil wetness indicator is therefore a simple and reliable tool to recognize spatial patterns of wetness. In example for precision agriculture, the soil wetness indicator could be used for irrigation management. The relative representation of the soil wetness indicator could determine the distribution of irrigation within a field.