Master
Themenvorschläge für Masterarbeiten
Mapping gradients of species composition using airborne hyperspectral data
Background: Ecosystems are characterized by the diversity of species they harbor, and monitoring this composition is essential for understanding ecological dynamics and assessing the impacts of environmental changes. In grasslands, there is a particular need to study species composition across different gradients of land use management practices to understand their reciprocal interactions. Hyperspectral remote sensing can be used to track changes in species composition providing a detailed spectral fingerprint of vegetation, allowing researchers to discriminate between different plant species and assess their abundance and distribution.
Here, plant abundance and hyperspectral data recorded in 2015 in the German biodiversity exploratories covering 150 plots are used to explore the potential of hyperspectral remote sensing for grassland monitoring. Additional links of plant and spectral communities to land use intensity across all sites are drawn.
Objectives of this thesis are:
- Extracting and averaging the spectral signal from hyperspectral airborne datasets
- Apply (existing) functions to normalize and smooth spectral reflectance values (post-processing) in R
- Use of ordination methods to handle multidimensional data (e.g., Principal component regression, redundancy analysis)
Requirements:
- Experience in scripting using R
- Experience of using big data in R is helpful (parallel computing)
- Experience with hyperspectral data is helpful
- Interest in ecological data and ordination methods
Project:
- SensGrün
Contact:
Using NDVI-time series from very high spatial resolution PlanetDove satellites in combination with Phenocam time-series to learn about grassland dynamics
Background: Utilizing NDVI-time series data from multi-spectral satellites in combination with Phenocam time-series offers an innovative approach to gain insights into the dynamics of grassland ecosystems. NDVI data on a bi-daily or weekly basis, provide detailed and frequent observations of vegetation health, while Phenocams capture high-resolution images of phenological events in grasslands. By integrating these datasets, researchers can explore the intricate relationships between environmental factors, grassland growth, and seasonal vegetation changes, enabling a deeper understanding of ecosystem dynamics at fine spatial scales. This integrated approach has practical implications for agriculture, conservation, and land management, providing tools to monitor grassland, optimize land use practices, and adapt to environmental changes. Here, we aim to test the relationship of remotely sensed grassland dynamics using very high spatial resolution PlanetDove time-series and PhenoCam data from open data networks.
Objectives of this thesis are:
- Selection of 4-8 possible sites across diverse grassland ecosystems in Europe
- Using Phenocam data from open source projects such as PhenoCam (https://phenocam.nau.edu/webcam/) or EUROPhen (https://european-webcam-network.net/)
- Using PlanetDove satellite data for the growing season at multiple sites
- Calculate NDVI time-series
- Explore trends in NDVI and compare with indices from the phenological cameras
Requirements:
- Experience in scripting using R
- Experience with satellite data or time-series analysis helpful
- Interest in grasslands dynamics
Contact:
Development of a Small Lightweight Handheld Spectrometer „STELLA“ after NASA Landsat program initiative.
Call for Master's Thesis: Development of a Small Lightweight Handheld Spectrometer „STELLA“ after NASA Landsat program initiative
Introduction
The ongoing miniaturization of technologies has facilitated significant advances in recent years, opening up new horizons in various scientific disciplines of non-invasive sampling and electromagnetic sensing. In this context, the development of do-it-yourself small lightweight handheld spectrometers is gaining increasing importance. These portable spectrometers offer the possibility to conduct precise optical measurements on-site, enabling a wide range of applications from environmental to vegetation probing, analysis and monitoring.
Background
This research project aims to create the STELLA (Science and Technology Education for Land/Life Assessment) handheld spectrometer that addresses current technological challenges while meeting standards of accuracy, precision and repeatability. The device will be built after a manual provided by the NASA Landsat Science Team https://landsat.gsfc.nasa.gov/stella/ and offers various stages of technological readiness. The work includes wiring, soldering, 3D printing and programming of the device. Required parts and devices will be provided via the institute resp. Fachgebiet of Prof. Birgit Kleinschmit.
Objectives of the Master's Thesis
Conceptualization and Design
- Reproduce a compact and lightweight spectrometer that meets the requirements of mobility and precision
- Create an integration and evaluation plan
- Further information: https://landsat.gsfc.nasa.gov/stella/stella-two/
Technological Implementation
- Integrating and soldering electronic circuit board, sensors, detectors and housing
- Module evaluation to ensure accurate environmental and spectral measurements in the laboratory
- Compare Stella spectra versus industrial grade ASD handheld spectrometer, and UAV-based multi and hyperspectral image spectra
Field test
- Conducting practical tests and validation experiments in selected areas
- In a field campaign encompassing grassland areas of Schorfheide Chorin in the project First 2.0
- Airborne test conducted using in-house UAV system
Calibration and Validation
- Developing an efficient calibration strategy to ensure the reliability and accuracy of measurements.
Profile of the Applicant:
- Student in the advanced stages of a Master's program in Environmental planning, Remote Sensing, Geology, Geosciences, Geophysics, Physics, Electrical Engineering, Optics, or a related discipline
- A drive to experiment with open-source electronics platforms, software and code
- Knowledge in areas such as optics, spectroscopy, signal processing, and/or measurement technology is advantageous.
- Experience in R, Python is advantageous but not required.
- Independent working style, creativity, and an interest in interdisciplinary research to work on current issues of remote sensing.
Contact
Interested students are invited to submit their application documents via email to michael.förster(at)tu-berlin.de or robert.jackisch(at)tu-berlin.de. For further information or questions, Dr. Robert Jackisch is available for inquiries.
Trend and breakpoint detection in Sentinel-2 time series at compensation sites
(Please feel free to contact the contact persons also if you would like to write a Bachelor thesis on this topic)
Background:
One supervisory task of the German federal railway service (Eisenbahnbundesamt) is the inspection of compensation measures (Ausgleichs- und Ersatzmaßnahmen) according to the German federal mitigation regulation (Eingriffsregelung). Instead of expensive and time consuming on-site controls, those monitoring tasks could be potentially covered by time series from the freely available satellite data of the European earth observation program Copernicus. For instance, the Multi-spectral sensors Sentinel-2A and Sentinel-2B are highly sensitive to chlorophyll content and color of vegetation surfaces. Because the Sentinel-2 mission provides images since 2014 with an approximate 5-day temporal resolution, we assume that the 10(-60)m pixel resolved data gives a good baseline for many automated monitoring tasks in the control of nature conservation measures. Potential research questions of the thesis could be: Are long-term trends from afforestation sites measurable? Are breaks from single maintenance measures observable and measurable?
Objectives of this thesis are:
- acquisition of pre-processed Sentinel-2 L2A time series data for multiple test sites
- developing algorithms for the automated detection of trends and breakpoints in S2 backscatter time series (e.g. bfast, trendr, green-brown)
- apply the algorithms on the object-level based on multiple compensation sites
Requirements:
Experience in scripting using R (preferred) or Python
Interest in optical remote sensing and Sentinel-2 MSI data
Expertise in landscape planning, impact assessment and mitigation planning is beneficial
Partners: German Centre for Rail Traffic Research (DZSF) from the Federal Railway Authority (EBA)
Contact:
Christian Schulz (christian.schulz.1@tu-berlin.de)
Birgit Kleinschmit (birgit.kleinschmit@tu-berlin.de)
Anomaly detection in Sentinel-1 time series at decommissioned railway tracks
(Please feel free to contact the contact persons also if you would like to write a Bachelor thesis on this topic)
Background:
One supervisory task of the German federal railway service (Eisenbahnbundesamt) is the inspection of decommissioned rail way tracks to give the permission for further planning decisions. Instead of expensive and time consuming on-site controls, those monitoring tasks could be potentially covered by time series from the freely available satellite data of the European earth observation program Copernicus. For instance, the C-band Synthetic Aperture Radar (SAR) sensors Sentinel-1A and Sentinel-1B are highly sensitive to metal objects. Because the Sentinel-1 mission provides images with an approximate 6-day temporal resolution and is independent from cloud cover, we assume that the 10 m pixel resolved data gives a good baseline for many automated monitoring tasks. Potential research questions of the thesis could be: Can there anomalies be detected during the time frame of the past two years? And at which day and at which exact regions were disturbances in the time series observed?
Objectives of this thesis are:
- acquisition of pre-processed Sentinel-1 time series data for a test site
- developing algorithms for the automated detection of disturbances in S1 backscatter time series (e.g. ts outlier detection, anomaly detection, threshold setting)
- apply the algorithms on the pixel-level based on the example of the test site
Requirements:
Experience in scripting using R (preferred) or Python
Strong interest in radar remote sensing and Sentinel-1 backscatter data
Interest in working with Google Earth Engine is beneficial
Partners: German Centre for Rail Traffic Research (DZSF) from the Federal Railway Authority (EBA)
Contact:
Christian Schulz (christian.schulz.1@tu-berlin.de)
Birgit Kleinschmit (birgit.kleinschmit@tu-berlin.de)
Estimating the relation between hydrological drought and vegetation indices for forest drought events
Background:
The summer 2018 was the warmest summer in Germany since the start of systematic records in 1881. At the same time, it was as dry as rarely ever before. As a result, the climatic water balance (CWB) was as low as just experienced once in the weather records. This extreme year was followed by a very dry and warm year 2019. Both years in combination are estimated as unprecedented in the last 250 years. In response, almost all important tree species almost everywhere in Central Europe showed severe signs of drought stress in 2018 and started to die in large areas in 2019 and 2020. Although the relation between the calamities and the forest species are rather clearly understood, the relation between hydrological deficits and the forest dieback is not understood fully.
Therefore, this thesis aims for estimating the relation between hydrological drought monitoring platforms and vegetation indices derived by remote sensing using data driven regression algorithms.
Possible questions within the overall aim can be for example:
- Which sensor type (eg. SAR, multispectral) leads to best results?
- Are there different relations between tree species and regions in Germany?
- Are there Phenological seasons when a relation is better to detect?
The topic is related to the research project ‘FirSt 2.0’ which aims to monitor forest damages with remote sensing techniques.
Objectives:
Understand relations between hydrological drought and vegetation drought for forests in GermanyRequirements:
Expertise in optical remote sensing
Expertise in GIS and google earth engine is helpful.
Partner:
Luftbild Umwelt Planung GmbH, Thünen Institute of Forest Ecosystems, State Forst Administrations of different federal states, Bavarian Forest National Park
Contact:
Michael Förster (michael.foerster@tu-berlin.de)
Vegetation index vs individual bands/ In which circumstance, predicting plant traits with vegetation indices is more accurate than using a combination of separate remote sensing bands.
Background:
Vegetation index derived from remote sensing bands is widely used to estimate (empirically) plant traits such as leaf area index (LAI), chlorophyll content and leaf water content. However, in which circumstance an index performs better than modelling individual bands is not entirely clear. This study aims to simulate satellite time-series images based on generated plant traits using Radiative Transfer Model (RTM) such as PROSAIL and SCOPE. These scenarios will vary according to the resolution (radiometric, temporal and spatial) of most common satellites such as Sentinel II and Landsat in different locations (i.e. latitudes). Based on these simulated databases, linear regressions' accuracy to predict plant traits using the most famous index (e.g. NDVI) will be assessed against its separate bands (and interactions). We aim to answer the question, which latitude range, resolutions, type of relationship (linear or non-linear) vegetation indices are better preditors than individual bands?
Objectives:
- Simulate spectral time-series images based on plant trait landscape scenarios in different locations according to the resolutions of the main current satellites,
- Modelling the scenarios and predict plant traits using widely accepted vegetation index as predictors against individual bands,
- Assess which circumstances the use of vegetation indices are indicated to predict plant traits instead of individual bands or wavelengths.
Requirements:
Basic remote sensing / familiar or open to learning R language
Contact:
Prof. Dr. Birgit Kleinschmit
Dr. Alby D. Rocha
Selecting the right place for terrestrial soil moisture measurements with remote sensing
Background:
The availability of images collected by Remote Sensing (RS) devices (e.g. satellites and drones) has rapidly increased in the last years. However, the field measurements to model and validate RS products is still restricted to a very limited number of sampling locations. Soil moisture content (SMC) is an essential variable for ecological and agricultural science that can be retrieved by RS data, but sufficient and well distributed samples with ground truth measurements are needed. The align of field samples with RS image will depend on the spatial resolution and extent, while the representativeness will be affected by the spatial autocorrelation (heterogeneity and patterns) and relations with the modelled covariates. When a timeseries of images is available the temporal patters are also an important factor to take consideration. The sampling strategies such as spatial sampling and space-time designs for monitoring can improve model accuracy while reduce the sample size.
Objectives of this thesis are:
- Using Sentinel 1 and 2 timeseries and drone images previously captured, define optimal sample locations to install SMC sensors in the field.
- Assess the difference in model accuracy (e.g. machine learning regressions) for different spatial sampling and space-time designs using a simulated field.
- Evaluate the use of different remote sensing index (covariate) to define sample locations in the lack of SMC preliminary information.
The project is related to an interdisciplinary and transregional research project “Cosmic Sense” that aims to investigate how and where soil moisture changes due to precipitation, evapotranspiration and deep percolation.
Requirements:
Experience in scripting using R (preferred), MATLAB or Python
Interest in working with satellite timeseries and modelling (regressions/machine Learning)
Partners:
Institute of Environmental Science and Geography, University of Potsdam
Contact:
a.duarterocha@tu-berlin.de & michael.foerster@tu-berlin.de
Analysing boreal forest structure to detect, annotate and analyse fuel loads and discriminate between boreal tree species
Background:
The Earth’s boreal forest is subject to transformation linked to global warming. A fire regime change as current visible phenomenon is one of such transformations leading to prolonged and more severe forest fires. 2021 marked a record fire year for the Siberian boreal forest, especially in the eastern Republic of Sakha (Yakutia). An expedition of the Alfred Wegener Institute (AWI) Potsdam together with the TU Berlin acquired a comprehensive LiDAR dataset was using unoccupied aerial vehicles (UAVs) in 2021. Airborne LiDAR scanning is a known technique in forestry, but UAV-based LiDAR and terrestrial LiDAR are recent technological additions that bridge the scale between ground-based forest mensuration and satellite/airborne measurements. However, vast amounts of 3D point data pose a challenge for processing algorithms, while at the same time the often applied conversation into 2D raster for analysis leads to an unfavourable loss of information.
Here, the use of machine learning algorithms from openly accessible code libraries (e.g. PyTorch, Tensorflow) is proposed to be used on selected LiDAR forest plots of Yakutia. This thesis aims to analyse forest structure to detect, annotate and analyse fuel loads (dead wood, snags) and discriminate between boreal tree species.
Objectives of this thesis are:
- Identify and evaluate open-access machine learning/deep learning segmentation tools for forest 3D data (e.g. PointNet++, Silvi-Net, Forest Structural Complexity Tool), in the scale from UAV to terrestrial LiDAR
- Train an artificial neural network using a manually segmented and annotated training data set
- Classify billions of LiDAR points to assess fuel loads aiming to evaluate forest fire risk
The project is related to the research project “BioS” that aims to monitor and model the influence of wildfire on boreal forest of Central Yakutia.
Requirements:
Experience in scripting using Python (preferred) or R
Experience working with 3D point cloud data, especially LiDAR and structure-from-motion photogrammetry
Expertise in working with deep learning techniques is helpful
Partners:
Alfred Wegener Institute Potsdam: Polar Terrestrial Environmental Systems Research Group
Contact:
Robert.jackisch@tu-berlin.de & Michael.foerster@tu-berlin.de & Stefan.kruse@awi.de & birgit.heim@awi.de
Thermal data from UAV and Landsat: Can tree water deficit and drought stress be predicted nationwide with Landsat data?
During the FirSt 2.0 project, primarily concerned with the derivation of forest damage in Germany, a method was developed and tested at the Thünen Institute that makes it possible to detect drought stress in trees using thermal data. Initially, numerous UAV data were collected at a small permanent observation plot near Britz in Brandenburg. Together with other input parameters, a model was trained and then transferred to a larger area (Schorfheide near Chorin). The results are promising. However, as the FirSt 2.0 research project aims to derive drought stress in forest areas for the whole of Germany, possibilities are now being thought to transfer the methodology from UAV to satellite data. At present, only data from Landsat satellites can be considered for this purpose, as they are available free of charge and contain one or two thermal channels.
The following questions are to be answered within the scope of a master thesis:
1) How do the temperature signals of the already recorded UAV images differ from the Landsat data? Both the different resolution (centimeter range vs. 100m pixels), the different sensitivity of the sensors as well as possibly the temporal offset of the recordings shall be considered.
2) If the derivation of temperature based on Landsat satellite data is comparable to drone temperature measurements considering the coarser resolution (upscaling), it shall be evaluated if the satellite data can be used as an input data for a Germany-wide model.
Provided that Landsat data are suitable for application of the model, it can be applied nationwide.
The drought stress maps produced could then be made available to a wider user community in the free forest damage portal ForestWatch or the Thünen-Landatlas.
Contact:
Tanja Sanders (tanja.sanders(at)thuenen.de)
Michael Förster (michael.foerster(at)tu-berlin.de)
Masterarbeiten (laufend oder abgeschlossen)
Laufend
Katharina Horn
Forest fire detection in Brandenburg with XAI
Jaey Vallapurackal
Spatial Distributional Energy Justice of Wind Energy Deployment in Germany
Elisabeth Köhler
Evaluating the use of remote sensing to detect valuable habitats under PV systems
Ramona Bauer und Nicola Kist (gemeinsame Arbeit)
Biodiversity and environmental justice in Berlin: Linking the availability of urban biodiversity with socio economic factors
Nermin Varol
Multifunctional green infrastructure planning: Revealing trade-offs & synergies and the hotspot areas in Berlin
Audrey Adisi
Assessment of fuel loads in Siberian Larch forest using on UAV-based LiDAR data and 3D segmentation
Qiuandra Taylor
Remote Sensing methods to restore trees and livelihoods in Kenya
Pia Kräft
Estimating the relation between hydrological drought monitoring platforms and vegetation indices derived by remote sensing.
Shirin Akter Mony
Integrating IceSat-2 discrete Lidar canopy information with UAV-based Lidar ground truth for boreal forest systems (working title)
Ramsha Tauqeer
Algorithms, tools and workflows to process and analyse Lidar-based information with focus on forested ecosystems (working title)
Abgeschlossen
2023
Igor Mishurov
Impacts of recent droughts on plant vitality in Berlin – a satellite time series analysis
Pia Bettancourt
Once there was a trail: how we lost the possibility of walking through the territory, the case of Chile.
2022
Aanjali Fernando
Mapping the invasive species Alianthus altissima in the Urban Environment with satellite Data
Sebastian Lehmler
Modelling green volume using Sentinel-1, -2 and PALSAR-2 satellite data and different machine learning algorithms for urban and semi-urban areas in Germany
Ahuvit Trumper
Evaluation of the temporal development of ash decline with hyperspectral imagery in Demmin, Germany
Kazi Sajjad Hossain
Time series analysis of the deciduous forest degradation in the Berlin-Brandenburg region based on Sentinel-2 data
Frederic Sorbe
Maxent modeling to forecast the future distribution of Acacia dealbata and Ulex europaeus in Chile under changing climate conditions.
Bráullio Nunes de Souza
Mapping expansion of agricultural frontier in Brazil: the case of coffee production in the Cerrado Mineiro Region.
Andrew Rasmussen
Methodische Ansätze für Schwerpunkträume zum Artenschutz in der Windenergieplanung.
2021
Hyunjae Kim
Impacts of land cover on Evapotranspiration in Berlin
Bronwyn L. Dyson
Testing the PlanktoScope: can low-cost flow imaging provide reliable and cost-effective phytoplankton estimates?
Florencia Arias
Analysis of the greenhouse gas emissions from Lake Dagow (Brandenburg, Germany) using the floating chamber method and eddy covariance.
Kayynat Shafiq
Identification of burn severity after the forest fires in Truenbreitzen using remote sensing techniques.
Leon-Friedrich Thomas
Object-based classification of individual tree vitality using LiDAR and hyperspectral data.
2020
Anjes Bloch
Rapid Assessment of forest storm damages using PlanetScope and Sentinel-2 satellite imagery.
Natalie Arnold
Investigation on the vitality of selected urban trees in parks in Berlin and Brandenburg
Irina Stockmann (Uni Potsdam)
Biomasseschätzung in einem temperierten Mischwald auf der Basis von Höheninformationen und multitemporalen RapidEye Daten.
2019
Balázs Lajos Dienes
Intra-annual and inter-generic variability of transmissivity of solar radiation through crowns of urban trees and its impact on solar radiation on building facades – a modelling approach
Jill Leonie Wabra (geb. Wagner)
Maximum Entropy Modelling and Classification of Pinus radiata Invasion in Central Chile
Kathrin Frank
Explaining occurrence of different wild bee species groups with terrain indices, climate parameters, land-cover structure, and spectral indices.
Isabel Blanke
How does precipitation influence the backscatter signal of grasslands?
Gina Maskell
How useful is Twitter data for the delineation of flood extent maps? A comparison of satellite-derived and Twitter-derived flood assessments
2018
Sophie Renner
Urban blue infrastructure patterns – a review of European cities.
Sara Muznik
Assessing the quality of urban green spaces in European cities. Case study Berlin.
Alexandra Rios
Modelling of Invasive Plant Species in Central Chile - Spreading patterns of Ulex europaeus and Acacia dealbata.
Liubov Shirotova
Estimating biophysical parameters of larix gmelinii using rapideye satellite imagery. The case study of Khatanga, central Siberia, Russia.
Rayk Albrecht
„Forest Cover Mapping and Multi-Temporal Analysis of Tugai Forests in Central Asia using Sentinel-2 and Landsat data.
2017
Roman Wolf
Linear Spectral Unmixing of Sentinel2 time series for detecting invasion of Pinus radiata in the Maule Region in Chile.
Ann-Kathrin Holtgrave
Estimation of Soil Moisture in Vegetation Covered-Floodplains with Sentinel-1 SAR Data Using Support Vector Regression
Sara Knuth
Does urban structure trigger the spread of canine distemper virus in Berlin red foxes?
2016
Alexandra Singleton
Assessing drivers of vegetation response in the lower reaches of the Tarim River, China using boosted regression trees.
Niklas Moran
An Automatic Tool for EUNIS Habitat Classification Using Ontologies and SEaTH.
Nicolas Specklin
Assessing desertification processes by remote sensing data in the Safi Region, Marocco.
2015
Cornelius Jopke
Change Detection analysis for meadow monitoring using Rapid Eye satellite data.
Nele Steimetz (in Kooperation mit der Uni Potsdam)
Machine learning regression algorithms for biophysical parameter retrieval for Alder and Beech forests in North-East Germany
Sabine Koch (in Kooperation mit dem Institut für Geodäsie)
Storage and Querying of CityGML Models in BaseX - Evaluation of the usage of a native XML database system for 3D city models
Christine Schwarzer
Klassifikation von Offenlandvegetation in einem Natura 2000 Gebiet – der Effekt von hoher räumlicher, spektraler und temporaler Auflösung
Lisa Heinsch
Agent-based modeling of human decisions on residential land uses in Berlin
Kyle Pipkins
A Comparison of Feature Selection Methods for Multitemporal Tree Species Classification
Christian Schulz
Quantifying land cover and landscape diversity changes at the caatinga (2001-
2012) – Landscape pattern analysis with modis land cover products
Eva Tsimeka
Tree species classification using intra-annual Rapid-Eye time series A case study: in Rhineland Palatinate, Germany
Kathrin Ward (in Kooperation mit HU Berlin)
How do urban structure influence the urban heat island effect? - A review of European cities.
2014
Pierre-Adrien Dugord
Assessing the influence of land use Patterns on urban climate and citydwellers vulnerability toward heat stress
Guilherme Henrique Braga Klaussner
Assessing Urban Environmental Justice in two subprefectures of São Paulo, Brazil – a GIS-based synoptic analysis
2013
Adina Tillack
Estimation of Seasonal Leaf Area Index in an Alluvial Forest Using High Resolution Satellite-based Vegetation Indices
2012
Marlen Diederitz (in Kooperation mit HU Berlin)
Szenarienbasierte Analyse der Versorgung mit Naherholungsflächen für die Stadtregion Berlin 2030
2011
Iftikhar Ali
Investigating the Potential of TerraSAR-X time series for the detection of swath events in NATURA 2000 grassland habitats
Theresa Garske
When is a city green? - Eine GIS-basierte Methode zur Ermittlung der städtischen Grünflächen-Versorgung für Erholungszwecke
Masterarbeit Garske Zusammenfassung
Rolf Breitschaft
Potentiale von Geographischen Informationssystemen in der Immobilienwirtschaft
