Projects

Post-Processing & Automated Reporting of Emissions Data

Extend a Python prototype into a full post-processing tool for vessel emissions and fuel-consumption data — automated reporting, trend analysis and visualization, developed with Alfa Laval and tested at their Aalborg Innovation Center.

Gamification of Shipping Decision-Making: From Proof of Concept to Strategic Value

Copenhagen Merchants Group is developing a proof-of-concept shipping management game, where players take on the role of a commercial decision-maker: hiring vessels, booking cargoes, managing market exposure, and responding to freight market shifts. The project explores how gamification can teach shipping economics through experience and feedback rather than classroom instruction alone.

This thesis project asks how the company should deploy the game to create real value - whether through employee training, employer branding, or not at all - and will deliver an evidence-based recommendation covering target audience, positioning, deployment model, KPIs, risks, and implementation roadmap.

Foundation Models for Forecasting Water Time Series

Recent breakthroughs in AI have led to large, pre-trained foundation models for time series, such as TimeGPT, Chronos, and Lag-Llama. These models promise robust, general-purpose forecasting capabilities across domains. This project will evaluate how well these models perform on water-related time series problems, such as wastewater inflow, harbour water levels, or urban runoff. The student will apply the models and compare the performance of these models against traditional forecasting methods (e.g., ARIMA).

Reservoir detection using deep learning approaches

Refinement of flow accumulation raster using information from water surface frequency and digital elevation models

Estimation of reservoir volume beneath recorded DEM using ML or geostatistical approaches

Downscaling AI weather models to local areas

In the past couple of years many AI meteorological models have emerged, many of these have similar accuracy as physics-based models but are around 100 times faster. Several of these models are open-source, e.g., Microsoft’s Aurora https://github.com/microsoft/aurora and could probably be used by DHI to develop our own downscaled local weather forecasts (or hindcasts) as input to our water models. There are already several different approaches and it does require some deep learning knowledge to develop and run these models. We imagine a student from ML/AI/data science/computer science/maths could carry out a project with us testing out e.g. two different AI downscaling approaches.

Sensor-agnostic coastline delineation using foundational models

Remote sensing provides a unique opportunity to observe coastlines at a large scale and throughout history to analyze changes in shape and structure. These coastlines may be used to assess stability, the effect of coastal protection and/or nature-based solutions, the impact of land cover changes, etc.  

Several tools currently exist to extract coastlines from single cloud-free images from specific sensors and resolutions. However, most of these do not generalize well outside of very specific domains or provide accurate results in the presence of clouds. Additionally, cloud-masking algorithms typically do not capture clouds very well close to the water and often incorrectly remove bright sandy shores. 

Foundational models are generalized across a vast number of different tasks and sensors and have shown improved accuracy over task-specific models when only a limited number of labels are available. This project proposes utilizing these foundational models to create a robust sensor-agnostic model that can accurately separate land, clouds, and water in marine environments globally.

Building a Global Multimodal Earth Observation Dataset by Integrating Mapillary Street-Level Imagery with Sentinel-1/2 Satellite Data

Earth observation (EO) has become a fundamental tool for monitoring environmental processes, land use dynamics, and water systems at regional and global scales. Satellites such as the Sentinel-1 and Sentinel-2 provide high-quality data with global coverage and frequent revisit times. Despite their strengths, observations from satellites often lack fine-scale semantic information about objects and structures visible at ground level.

Street-level imagery platforms such as Mapillary provide billions of geotagged images collected worldwide through crowdsourced contributions. These images contain detailed information about roads, drainage infrastructure, coastal protection structures, vegetation, riverbanks, urban waterways, and flood-prone environments. Such ground observations complement satellite data by providing detailed contextual information about landscapes and infrastructure interacting with water systems.

Integrating satellite and street-level imagery therefore presents a unique opportunity to develop multimodal environmental datasets that combine large-scale Earth observation with detailed ground-level perspectives. Despite the availability of large satellite archives and massive collections of geotagged street imagery, there is currently no global dataset systematically linking Sentinel-1, Sentinel-2, and street-level imagery.

This thesis aims to address this gap by developing a global multimodal dataset that integrates Mapillary street-level images with Sentinel-1 and Sentinel-2 satellite observations, with a particular focus on environmental and water-related applications relevant to hydrological and coastal research.

Uncertainty-Aware Earth Observation Foundation Models

Remote sensing provides unprecedented opportunities for observing Earth from space. Traditionally, Earth Observation (EO) workflows use satellite data, such as Sentinel-1 and Sentinel-2, to extract actionable insights. Recently, EO Foundation Models have emerged as a powerful alternative to conventional workflows. These models are pre-trained on extensive archives of EO data, encoding substantial domain knowledge within their learned representations [1].

However, the critical nature of EO applications demands trustworthy and reliable deep learning models. Uncertainty estimation can significantly enhance model reliability by providing confidence measures for predictions [2]. Current geospatial Foundation Models primarily focus on delivering large-scale deterministic representations while overlooking their uncertainty. Recent research has introduced the concept of representation uncertainty in EO, incorporating an uncertainty layer into model representations [3]. Nevertheless, this work has been limited to single-modality training with supervised datasets.

This thesis will extend this foundational work to multi-modal EO Foundation Models. The student will leverage state-of-the-art pre-trained multimodal EO Foundation Models and investigate integrating an uncertainty layer into their representations. The research will examine how different combinations of input modalities influence estimated uncertainty and explore how zero-shot uncertainty behaves across diverse downstream tasks.

Lightweight Foundation Models for Earth Observation in Low-Data Regimes

Remote sensing provides unique opportunities for observing Earth from space. Traditionally, Earth Observation (EO) workflows use satellite data such as Sentinel-1 and Sentinel-2 to extract insights. Recently, EO Foundation Models have offered an alternative to these traditional workflows. These models are pre-trained on large archives of EO data, encoding vast amounts of knowledge in their representations [1].

However, training these models typically requires enormous computational time and resources. Recent research has shown that lightweight Foundation Models can also perform well on downstream tasks [2]. Following this direction, this project will further explore whether smaller models trained on smaller datasets can achieve comparable performance to their larger counterparts.

This thesis will investigate the use of lightweight Foundation Models in low-data scenarios. The student will explore questions like: How small can we go while maintaining performance? Can methods such as supervised EO Foundation Models work effectively? [3] What techniques make lightweight models competitive?

Hydrology-Aware Deep Learning for Physically Consistent Wetland Classification

Wetlands are critical ecological buffers, yet they remain one of the most difficult land-cover types to map accurately due to high seasonal variability and extreme spectral similarity to upland land-cover. This thesis explores a novel approach using Alpha Earth Foundation Embeddings—highly compressed, 64-dimensional numerical representations of Earth's surface that capture long-term vegetation patterns and moisture cycles. However, when spectral features and topographic features are stacked directly, standard deep learning models are often "lazy," frequently overfitting to simple elevation data. This leads to physically inconsistent results and poor generalization across varying landscapes.

The core objective of this thesis is to investigate advanced spectra-topo fusion techniques that move beyond simple feature stacking. Rather than treating topography as just another input channel, this research proposes novel architectures that use topographical features—specifically Height Above Nearest Drainage (HAND), slope etc.)—as physical constraints to dynamically gate the Alpha Earth embeddings. In this framework, topography generates an attention mask that identifies hydrologically favorable zones. The model is thus trained to prioritize complex spectral-temporal analysis only within these regions. Conversely, in hydrologically unfavorable areas where wetland formation is physically impossible, the topographic constraint automatically filters out false positives, preventing the model from being misled by spectral mimics.

*The expected outcome is to develop a "hydrologically intelligent" segmentation model that is not only more accurate but truly generalizable across diverse terrains, from coastal lowlands to inland uplands. *

This project is tailored for students with a strong background in large data processing, Earth Observation (ideally Google Earth Engine) and Machine Learning (particularly PyTorch), and experience in GPU environments is a plus.

Satellite-derived bathymetry from wave kinematics using radar.

Develop a robust machine learning (ML) model that uses satellite-based radar (Sentinel-1) observations of ocean wave kinematics (e.g., wave height, wavelength, phase velocity) to estimate water depth in the intermediate range (10–50m).

This range is particularly challenging due to the limitations of both traditional optical and deep-water wave inversion methods.

Time Series Forecasting for Water Applications Using N-HiTS

N-HiTS is a recent deep learning model for time series forecasting that has shown state-of-the-art performance across many domains. This project explores how N-HiTS performs on typical water-related time series such as wastewater inflows, harbour water levels, or urban runoff. The student will compare deep learning performance with traditional time series forecasting methods across 3–5 real-world datasets.

Modelling and monitoring of compound flooding in deltas and estuaries using satellite earth observation

Recent satellite earth observation missions such as SWOT and ICESat-2 datasets provide continuous water surface elevation datasets across the river-coast-ocean interface in estuaries and river deltas for the first time. Such datasets are extremely valuable for the calibration and validation of integrated hydrodynamic models used to simulate compound flooding events, when flooding is caused by a combination of high river flows, storm surges in the ocean and extreme local precipitation. Low lying deltas, such as the Mekong River delta, are particularly exposed to compound flooding events, events that are expected to be more frequent and harmful with future climate change and sea level rise. At KU-IGN, we explore and exploit the value of these datasets using the Mekong River delta in Vietnam as a case study. This work is funded by an industrial postdoc project carried out in collaboration with DHI. 

AI Hydrological Forecasting with Reservoir Operations Integration

Neural hydrology models (e.g., LSTMs) can perform as well as or better than traditional physics-based hydrological models in data-rich basins. However, their effectiveness drops in regulated rivers and data-scarce (ungauged) basins, largely because they lack explicit inputs for human-controlled processes (like dam releases) and are typically designed for hindcasting, not forecasting.

Possible Research Questions:

• Reservoir Data Integration: How can dynamic reservoir operations data (e.g., dam release schedules, storage levels) be effectively integrated into LSTM-based hydrological models to better simulate regulated flow patterns?  
• Forecasting Architectures: What model architectures or strategies (e.g., sequence-to-sequence LSTMs, transformers, or other recurrent models) are most effective for accurate short to medium range (7 day ahead) streamflow forecasting?  
• Generalization to Ungauged Basins: How well do the enhanced AI models (with forecasting and reservoir inputs) generalize to ungauged or data-scarce basins, and can transfer learning (pretraining on data-rich regions and fine-tuning) improve predictions in these regions?   

Hybrid surrogate of urban collection system

Traditional physics-based hydrodynamic models for urban collection systems (e.g., sewer and stormwater networks) are highly accurate but computationally expensive, limiting their use in real-time control (RTC) and early warning systems. While data-driven surrogate models can offer massive speedups, they often struggle with the complex spatial topology of pipe networks and the sudden, discrete flow changes caused by active human-controlled infrastructure (like pumps and gates), leading to unstable long-term forecasts.

Possible Research Questions GNN and Control Integration: How can Graph Neural Networks (GNNs) be effectively coupled with rule-based control logic to accurately simulate both the spatial network hydrodynamics and the discrete state changes of active assets (e.g., pumps, controllable weirs)? * Stable Multi-Step Forecasting: What training strategies or architectural modifications (e.g., autoregressive training, temporal attention mechanisms) are most effective for preventing error accumulation and ensuring stable multi-step predictions over extended forecast horizons? * Scenario Robustness:* How robust is the hybrid surrogate model when subjected to unseen operational scenarios and extreme rainfall events, and how does its computational efficiency and accuracy benchmark against a full-scale traditional hydrodynamic model?

Global Hydrological Modelling Development and Advancement

Large-scale hydrological modelling is an essential tool for governments and river basin organizations to ensure the security of the population in our changing climate. DHI has developed a Global Hydrological Model (GHM), which has already been used all over the world: from transboundary river studies in Vietnam, to water scarcity studies in Nigeria, to macroplastic pollution forecasting for UNEP. We are just scratching the surface of possibilities with the model, and are looking for interested students to investigate the research questions that have opened up with this new tool.

There are a number of relevant research questions for a student project. Depending on the student, the project can be tailored to the fit the student's background and interests. A student could choose one of the following topics:

* DHI-GHM produces short and long-range forecasts, which are essential for flood and drought preparedness activities. How do the DHI-GHM forecasts perform, and what can be done to increase the performance? (modeling/statistics/programming)   * DHI-GHM depends on many global input datasets that describe the catchment characteristics. What is the significance of each of these datasets, and how does the model result change as these datasets are updated? This could be relevant through over a timespan of many years (e.g. de-/re-forestation) or over a growing season (root zone depth). (modeling/statistics/hydrology)   * DHI-GHM has been developed to cover river basins in their entirety, but coastlines are sometimes covered less thoroughly. A student project could involve expansion of the model to cover these areas, which are sometimes essential for freshwater inputs to marine environments. (modeling/hydrology)   * The resolution of DHI-GHM is standard at 0.1x0.1 degrees over the entire globe. But sometimes there is finer resolution data available that could inform finer resolution results - which in turn yield critical information for industries such as hydropower. Developing a downscaling routine that can minimize manual inputs is another potential model development project. (modeling/hydrology)   * Water use is a modelling element that most substantially impacts the model performance. A pilot reservoir modelling scheme based on earth observations has been developed. What is the consequence of applying this scheme more broadly in the model, and does the method hold when applied broadly?(modeling/water resources/hydrology)   * Eutrophication of fresh and marine water bodies is a high interest topic in Denmark, the EU, and many other places in the world. What could a global nutrient loading and transport model look like, in the modelling framework of DHI-GHM? (modeling/water quality/contaminant transport)   * Modelling only attains relevance with clear and timely dissemination of the model results. This is essential in disaster risk management, where understanding the extent of floods and droughts in time to act are essential. What could a GUI for the GHM look like, and what do stakeholders (both engineers using the model, and decision-makers viewing results) need to know? (UX design/stakeholder engagement/front-end development)   * Climate change is impacting all planning decisions in government and industry. How can DHI-GHM be developed as a climate change adaptation tool, what are the weaknesses that need to be addressed and strengths that should be maintained? (modeling/climate change/hydrology)

Integrated Hydrological-Hydrodynamic Modelling for High Groundwater, Surface Water and Sewers

As a result of climate change, we are experiencing more annual rainfall in Denmark and greater weather variability with more frequent and intense storms and rainfall events. The increased rainfall affects the interaction between surface and near-surface groundwater, shifting the balance that our systems are designed to handle.

Higher groundwater levels and more water on the surface challenge us both in open landscapes and urban areas. At the same time, the risk of flooding increases when even everyday rain can no longer be absorbed but ends up being stored, often in places that cause significant inconvenience and costs.

There are no integrated tools that correctly mechanistically describe the dynamics between different water domains on a scale relevant to sewers systems and building drainage.

The goal is to investigate the integration between two industry-standard systems, MIKE+ and MIKE SHE.

Data Assimilation in Mike1D models for urban collection systems

In its core, Data assimilation (DA) is "manipulate the calculated states of a numerical model, by using observations to guide the correction". Urban models have many structures as pumps, valves and orifices. Some of these structures are sources of discontinuities in the model - think of an on/off-pump, a valve which closes or a weir which diverts parts of the water in another direction, but only when its crest level is reached. When DA tinkers with the model states it will have adverse effects on a pump which for instance starts and stops at certain levels.

Mike1D has two implementations of a Kalman filter approach - one where the user specifies a stationary gain, and one where the gain is derived from an ensemble of simulations. The former is computationally cheap, because it only runs a single simulation - however, the user-specified gain may not reflect the system dynamics. The latter is computationally expensive, because it needs to run an ensemble of simulations - but has the advantage that the gain "stays true" to the model dynamics (in both space and time).

So far, we have some urban applications that use the stationary gain method, but only to inject/extract water at single points in the system. We would like to investigate if the use of ensemble methods can improve the corrections.

Accurate identification of anomalies in wastewater sensor - Integrating univariate and multivariate methods

Wastewater treatment plants (WWTPs) are increasingly provided with online sensors and analyser to monitor the performance in real-time and timely identify anomalies in process operations and performance. The quality of sensor data is to be ensured by continuous sensor maintenance by WWTP staff, combined with the application of processing techniques to clean time series data. Nevertheless, deviations from normal sensor behaviour (drift, outliers) are uncommon, affecting the reliability of the collected data.

The goals of this thesis project are to

(i) identify commonly used techniques to identify anomalies in sensor data as well as in process operations

(ii) apply and integrate these techniques for anomaly identification in a real WWTP

Nature-based solutions for wastewater - Improving modelling approaches

Nature-based solutions are widely used as low-cost and low-environmental impact approaches for treating wastewater globally. Common nature-based solutions include, e.g., wetlands for decentralized treatment, aerated / non-aerated lagoons, and ponds for handling of stormwater.

Despite the widespread use of these solutions, there are still gaps in the knowledge of the underlying processes and mechanisms - and therefore limited consensus on the modelling approaches to be adopted (Deghani Tafti et al. 2024).

The goals of the current thesis project are to

(i) evaluate the current state of the art in the use of nature-based solutions for wastewater treatment and the modelling of these solutions

(ii) identify and implement a generalized modelling approach and test it for one or more selected use cases, targeting a specific objective (BOD removal, nutrient removal, micropollutant removal)

Tackling micropollutant removal in wastewater treatment plants - Integrating impacts on recipients

Municipal wastewater contains a multitude of micropollutants (MPs), e.g., pharmaceuticals, PFAS, pesticides, that can pose risk to water environments.

The recent approval of the new EU Urban Wastewater Treatment Directive (EU UWWTD) has introduced stringent requirements for the removal of MPs through upgrade and retrofitting of existing treatment facilities. These requirements are defined for a small set of target substances, and do not take into account the impact on recipients.

The goals of this thesis projects are:

(i) to collect existing information and data on MP removal by existing advanced treatment technologies (ozonation, activated carbon)

(ii) simulate and predict removal of a larger set of MPs (beyond those defined in the EU UWWD)

(iii) predict improvements in the environmental quality of recipients as a result of the implementation of advanced treatment technologies

Integrated Air and Water Phase Modelling of H2S Transport in Sewer Systems

Hydrogen sulphide (H2S) is a major problem in many sewer networks where it can cause severe economic cost in maintaining the infrastructure due to corrosion, odor nuisances and dangerous working environment.

During the transport time to the wastewater treatment plant, the wastewater is transformed by ongoing biological processes. Aerobic heterotrophic activity leads to the growth of biomass, which attaches to the sewer pipe walls as biofilm and begins to corrode the dissolved oxygen in the wastewater. Anaerobic conditions can occur particularly in pressure mains, promoting the growth of bacteria that convert sulfate to sulfide. When sulfide is released as the gas H2S, it enters the sewer system atmosphere, causing odor and corrosion issues.

Many sewer systems are primarily designed to collect and transport sewage effectively but without consideration to prevent formation of H2S.

Super-Resolution for Groundwater Models Using Multi-Resolution MIKE SHE Grids

Structured-grid groundwater models like MIKE SHE are often run at high spatial resolution to capture important hydrogeological variations. However, this increases runtime, especially in large or operational setups. This project explores whether techniques inspired by computer vision super-resolution (e.g., CNNs or deep learning upsampling) can be applied to map coarse-resolution groundwater model results to fine-resolution outputs. The student will experiment with deep learning models that learn this mapping on selected MIKE SHE test cases.

Flood mitigation potential by saltmarshes

Recently, Jacobsen (2026) presented a new wave dissipation method in flexible vegetation that naturally accounts for the preferential dissipation on select frequencies (high frequencies for emergent vegetation) and low frequencies for low vegetation in deeper waters.

This master thesis project will investigate the applicability of the Jacobsen (2026) method for wave period transformation across a laboratory saltmarsh in a small-scale facility (Dermentzoglou et al., 2026) and compare the outcome against past wave dissipation formulations, e.g. Suzuki et al. (2012) and Jacobsen (2019).

The research objective is to verify whether the simple modelling approach by Jacobsen (2026) employing a fixed set of force coefficients rather than the variable coefficient set required for e.g. Suzuki et al. (2012) and Mendez and Losada (2004) is applicable to all of the 74 cases by Dermentzoglou et al. (2026). This will greatly improve on accurate predictions of the civil engineering functioning of natural habitats and nature-based solutions so they can become a trusted and quantifiable component in coastal engineering methods for climate adaptation and flood mitigation.

Environmental risk assessment of ocean-based CO₂ removal (Direct Ocean Capture)

Environmental risk assessment of ocean-based CO₂ removal (Direct Ocean Capture). The project examines potential environmental impacts of a pilot-scale technology developed by FlowMinerals that captures CO₂ from seawater through an electrochemical process where carbon dioxide is mineralised into calcium carbonate. The technology operates without chemicals and is powered by electricity.

The impact of operating wind farms in the wave and wind stress field over the North Sea

As a key renewable energy source, wind energy is crucial for mitigating climate change and supporting the transition to a sustainable energy future. Offshore wind capacity has been steadily growing in recent years and is expected to expand significantly in the coming decades.

This thesis will investigate how large-scale wind farms influence the surrounding wave field through two key mechanisms: wake effects, where wind turbines reduce wind speeds and turbulence, affecting wave generation, and scattering effects, where turbine monopiles disrupt incoming waves. Additionally, the study will explore how wave conditions, in turn, impact near-surface wind stress and sea surface wind field.

The research focuses on improving numerical modeling approaches to accurately capture these complex air-sea interactions, providing valuable insights for offshore wind energy planning and policy making.

Spatial chunking strategies for unstructured mesh data

Zarr is a modern specification for large N-dimensional typed arrays.

A key component to enable fast subsetting in any dimension is to use store the data in chunks.

For time or cartesian grids, it is fairly easy to design an optimal chunking strategy based on typical usage patterns.

For data from MIKE 21/3 using unstructured meshes in the horizontal plane and optionally with a structured vertical discretization, finding an optimal chunking strategy is more challenging.

Focus is on enabling fast data access for e.g. model validation comparison with observations or downscaling.

Hierarchical meshing

Meshes can be quite big, and result files on those meshes even bigger, as they contain results for a number of time steps. We are looking for techniques to optimise the transfer and visualisation of results on meshes.

For world maps, viewing the entire world all down to a single house, the technique for map tiling is well defined and straight forward, defining tiles on a number of zoom levels.

We are looking for a similar technique for traingular meshes, so that the amount of data handling and visualization processing is optimized for different zoom levels.

For triangular meshes there is no coarse mesh that aligns perfectly with the fine mesh, so some details wille be lost on coarse levels, due to simplication of the boundary and interpolation of element values.

Multi-Resolution Modeling for Flexible Mesh Coastal-Ocean Models

Flexible mesh models allow for varying spatial resolution, but running high-resolution models in operations is often too costly. This project explores techniques for mapping from coarse- to fine-resolution models, such that only the coarse model needs to be run in operations, with the fine-resolution results being inferred via a mapping. The focus is on testing this in the Southern North Sea and Øresund using MIKE 21 HD.

Koopman Autoencoders as surrogates for MIKE 21 coastal models with integration of observed data

Koopman autoencoders are deep learning models that learn a linear latent space where nonlinear dynamical systems can be evolved using linear dynamics. This project explores implementing a Koopman autoencoder with external forcing, inspired by Morton et al. (2018), and applying it to MIKE 21 HD simulations in the Øresund and Southern North Sea. The student will investigate how observations (e.g., water levels) can be integrated during training to improve predictions. Additional focus areas may include multi-step forecasting, stability analysis, or formulation as a gResNet-type model.

Neural Operators as Surrogates for Coastal Ocean Modeling with MIKE 21 HD

Neural operators are a new class of machine learning models that learn mappings between function spaces (e.g. inputs and solutions to PDEs), making them well-suited as surrogates for physics-based models. This project explores whether neural operators like the Fourier Neural Operator (FNO) or U-NO can emulate MIKE 21 HD simulations on flexible meshes. Two test cases, Southern North Sea and Øresund, will be used for model training and validation.

GNN-Autoencoders as Surrogates for Coastal Ocean Modeling with MIKE 21 HD

Graph Neural Networks (GNNs) have shown great potential in learning from structured spatial data, such as water levels and currents on unstructured meshes. In this project, we propose to investigate whether a GNN-based autoencoder can be trained to emulate MIKE 21 HD simulations. The idea is to learn a latent representation of the dynamics and use it for fast surrogate predictions. Two real-world test cases, the Southern North Sea and the Øresund, will be used for model training and validation.

Self-Supervised Deep Learning in Remote Sensing for Wetland Classification based on Satellite Time Series

LIDAR-corrected Digital Terrain Model for Hydrological Applications using Deep Learning

Digital Elevation Models (DEMs) are critical inputs for hydraulic modeling, particularly in data-scarce regions where ground-based measurements are limited. However, many widely available DEMs, such as those derived from the Copernicus GLO-30 dataset, are Digital Surface Models (DSMs), which include surface features like buildings and vegetation. This discrepancy presents challenges for 2D hydraulic modeling, as accurate flood propagation analysis depends on reliable terrain elevation data that excludes these surface features. 

Spaceborne LiDAR missions like ICESat-2 and GEDI offer valuable complementary data by providing high-precision elevation measurements of ground, canopy, and structures. By leveraging these datasets, there is a promising opportunity to develop methodologies that transform DSMs into Digital Terrain Models (DTMs) for hydrodynamic applications. 

This project aims to improve the accuracy of terrain elevation models for hydraulic modelling by predicting the elevation difference between DSMs and DTMs using spaceborne LiDAR observations, multi-source satellite data, and deep learning techniques. The goal is to provide a scalable and hydrologically sound solution for hydraulic modelers.

Simulation of liner shipping network execution

Heuristics for Terminal friendly Liner Shipping Network Designs

Inspection Prediction 2.0

Our vessels fall under several inspection regimes. The results of such inspections are an important parameter our customers as they indicate the risk our assets poses to their business. Hence successful vessel inspections are key for our commercial team to trade the vessels. In Navigator Gas we want to develop a risk management approach to support in the preparation for such inspections. Based on the vessel’s condition, people onboard, location, etc, we want to determine a risk indication for a successful inspection. This project was initiated by ITU students as part of their Master's thesis. In this second stage, the objective is to further refine the model by improving its accuracy through the incorporation of additional relevant information and free-text data, as well as by adjusting the existing thresholds. Re-evaluating the model's overall approach is also an alternative if deemed necessary.

Technical Fault-finding Assistant

Each of our vessels is packed with mechanical equipment and electrical equipment. This means that sooner or later it breaks down. Faultfinding issues onboard is often time consuming and requires several interactions between crew, shore, service companies and makers. We would be interested in exploring a tool that will provide technical assistance, based on previous records, maker's manuals and emails systems

Effectiveness of energy efficiency investments

How much do individual energy-efficiency technologies actually reduce vessel fuel consumption — and how can we separate their true effects from other operational factors? This project aims to answer that question using real-world fleet data from approximately 57 vessels, each with detailed fuel consumption records and installation timelines for different technologies (e.g., propeller modifications and hull coatings)

Ship particulars and parameters prediction tool - Based on a machine learning approach.

The interest is to develop a functional and user-friendly tool that can predict ranges of parameters and particulars of a ship, such as main dimensions, tonnage, tanks capacities, cargo capacities and more. All of these depend on the type of vessel of interest (ferries, cruise ships, work boats, etc.) With the purpose of facilitating the initial spiral design process of the conceptualization of a vessel. To achieve this goal, it will be necessary to first work out a database from which the tool can be trained continuously, so the predicted values are inside an acceptable range and category.

Cargo flow and operation tool - Based on a machine learning approach.

Ro-Ro and Ro-Pax vessels perform a daily operation of onboarding and outboarding wheel cargo for their transportation. This daily operation requires to be properly carried out, to distribute the inner loads in the correct way. However, this process can become cumbersome and in case it gets delayed, the idle time of the ship at port can be easily translated into fuel consumption for both being at shore and due to the need to compensate for wasted time (acceleration during the voyage).

Given this problem, it is of interest to develop a tool or model that can simulate loading scenarios to carry the most proper distribution. This model should be robust enough to be able to consider different cargo types, dimensions, weight distribution and deck arrangement.

A 3D BIM-Enabled Workflow for Ship Outfitting Design

The outfitting of ships—covering machinery, deck equipment, and interior furnishing—requires complex design processes where multiple disciplines must coordinate across geometry, weight, and functional requirements. Traditionally, ship designers rely on disconnected tools for structural modeling, equipment arrangement, and reporting, leading to inefficiencies and data inconsistencies. This thesis proposes the design and implementation of a prototype system that leverages Building Information Modeling (BIM) principles to create an integrated 3D workflow for the first steps of a Digital Twin.

Biofouling Pressure Model

Hull fouling causes speed loss and increased fuel consumption, posing a significant challenge to vessel efficiency. It is a multifactorial issue, where biofouling pressure represents the environmental potential for fouling based on oceanographic conditions. The proposed thesis is to use the EDITO platform to develop a biofouling pressure model by integrating data from Copernicus Marine and EMODnet sources.

Bayesian event classification from AIS data

The Automatic Identification System (AIS) is a tracking technology used by ships to broadcast information such as position, speed, and course. AIS provides large-scale longitudinal information that can reveal operational and spatial patterns associated with maintenance activity. Since such events are rare and only indirectly observable, both inference and validation require methods that can handle uncertainty explicitly. This thesis therefore proposes a Bayesian framework for probabilistic event identification using AIS-derived operational and spatial features.

Sequence-based detection of fouling-relevant vessel inactivity from AIS data

Automatic Identification System (AIS) data is used to identify periods where vessels are inactive and therefore exposed to increased biofouling risk. Existing idle detection approaches typically rely rules like speed thresholds together with contextual information such as distance to land, anchorage areas, and OpenStreetMap features.These approaches are less suitable for drifting vessels, where an extended inactive period may be interrupted by short periods of steaming to reposition the vessel before drifting again. These interrupted periods may represent one continuous exposure event. This thesis proposes a data-driven approach for identifying drifting behaviour and combining interrupted vessel activity into fouling-relevant inactivity periods.

Data-driven prediction of spatial hull fouling and its resistance impact

Biofouling develops as a result of environmental exposure, vessel operation, coating condition, and time. Its hydrodynamic effect subsequently depends on both the severity and location of the resulting hull roughness.The proposed reduced-order resistance model can represent different roughness conditions across bottom, side, and boot-top regions and calculate their individual 
contribution to frictional resistance. This thesis proposes to connect environmental and operational vessel data with the hydrodynamic model by developing a data-driven model for the temporal and spatial evolution of hull fouling.

Detection of latent engine and fuel regimes from vessel operational data

Vessel performance monitoring systems provide high-frequency measurements of engine and propulsion parameters such as shaft power, torque, engine speed, fuel consumption, and vessel speed. Changes in fuel type or fuel properties can alter the relationship between fuel consumption and produced power, but these changes are often not explicitly recorded in operational datasets. At the same time, apparent changes in fuel efficiency may also be caused by different engine operating conditions, propulsion settings, or measurement effects. The proposed thesis is therefore to develop a data-driven method for identifying latent engine and fuel regimes from operational vessel data and to investigate whether changes in these regimes can be associated with fuel switches or changes in fuel properties.