Talks & Teaching

In this talk, I explored the potential of large-scale, multi-species animal movement tracking for Earth observation and opportunities in combining remote sensing and ecological data. The talk was given as part of the Geoinformatics Forum at ifgi at the invitation of Prof. Dr. Edzer Pebesma and Johannes Heisig. From the abstract: Land use/land cover (LU/LC) classifiers are models that transform environmental data into interpretable, quantifiable information. Conventional LU/LC classification approaches rely on human-defined classes and human-collected ground truth data to train a classifier. Animals, like humans, perceive, interpret and differentiate their environment, resulting in distinctive movement behavior governed by the animals’ needs and the surrounding conditions. In this study, we utilize the animal’s unique view on the landscape to classify environmental data: We propose a Deep Learning approach that learns cluster-friendly representations from animal-environment interactions, generating behavior-derived ground truth data for animal land use (ALU) classification from environmental data. The classifier’s applicability was measured within 153 areas of interest (AOI) within Europe, covering an area of 53,167.55 km2. Our approach achieved high accuracies in both representation learning (0.881) and ALU classification (0.806), showing that, overall, classes defined by clustered representations of animal-environment interactiond were well-differentiable in the environmental data. With an average ground truth spatial density of one sample per 0.51 km2 every three years, the presented approach exceeded sampling densities of state-of-the-art human labelling campaigns such as the European Land Use/Cover Area frame Survey (LUCAS) with 15 km2 per sample/three years. In comparison to human LU/LC, thematic resolution of ALU varied per land surface type, with higher thematic resolution in natural habitats such as grasslands and lower thematic resolution in urban areas. Our study demonstrates how animals can be employed as environmental informants in remote sensing research by using them as landscape interpreters to generate ground truth for land use classifications.

Duration: 30 minutes

In this talk, I presented work on deriving snow depth data using UAV LiDAR remote sensing on Svalbard to investigate snow-ecosystem interactions at fine spatial scales. The talk was given as part of the SvalbardMonitoring workshop organized by Dr. Hans Tømmervik (NINA) which took place in Longyearbyen, Svalbard, from 20th to 23rd of October 2025. From the abstract: Snow is an essential climate variable crucial to Arctic ecosystem functioning. Its high albedo and seasonal variability drive thermal gradients that globally affect atmospheric temperature. Snow also has strong local impacts: Acting as an insulating layer, snow depth regulates subnivean temperature and active layer thickness, protects vegetation from icing, and impacts water and soil nutrient availability during melt. It thereby affects plant growth dynamics and phenology, influencing resource availability for herbivorous wildlife, such as reindeer. At the same time, snow depth can constrain forage accessibility for herbivores and increase the energetic costs of moving and feeding during winter. However, local ecological effects of snow are still not fully understood, largely owing to a lack of snow depth data at ecologically relevant spatio-temporal scales. Such are challenging to cover adequately through traditional in-situ snow depth measurements. In this study, we investigate the potential of high resolution, spatially continuous UAV-derived snow depth data to advance Arctic ecological research. Using a fixed-wing, LiDAR-enabled Uncrewed Aerial Vehicle (UAV), we derive snow depth as the difference between digital surface models recorded under snow-covered and snow-free conditions, with a spatial resolution of 2.5 cm per pixel, for an area of 216.5 ha across two study sites in high-arctic Svalbard. We validate the model with 340 in-situ snow depth measurements temporally synchronous to UAV acquisitions, showing a high accuracy (R-squared = 0.93). Using a semi-variogram and autocorrelation function thresholding approach, we analyse at which spatial scales snow depth varies to quantify the folding frequency, i.e. the minimum sampling distance to measure snow depth with, at or below 2.5 metres. In addition, we demonstrate that UAV-derived snow depth shows no spatially stable correlation with standard proxies (e.g., terrain ruggedness, hillshade, elevation, distance to coast) conventionally used in Arctic ecology research when spatially continuous snow data are unavailable. Our findings suggest that UAV-derived, spatially continuous snow depth data at sub-decimeter spatial resolutions have a high potential to contribute to a better understanding of the impact of snow depth on Arctic ecosystem dynamics.

Duration: 30 minutes

In this guest lecture, I introduced satellite and UAV remote sensing, geo-analysis and spatial modelling for population ecology research to Arctic Population Ecology BSc. students, together with my colleague Sebastian Buchelt. The guest lecture was held at the invitation of Assoc. Prof. Dr. Larissa Beumer (UNIS) as part of a six-week international BSc. course on Arctic Population Ecology. Students were introduced into UAV (drone) remote sensing and its applicability for detecting, counting and mapping animals to estimate animal populations, such as Svalbard reindeer. Students were presented different sensing and mapping methods, UAV platforms as well as the effects of flight parameters such as flight altitude or flight speed on the quality and resolution of the acquired data. Students learned how UAV remote sensing data can be used to describe the environment for being able to model relationships between animal occurrences and environmental covariates. In addition, they learned how UAV missions are planned, which regulations apply and how collected images are converted into geo-spatial maps (orthomosaics) using photogrammetric methods such as structure-from-motion (SfM). Outside, in the tundra, students were shown hands-on how to use UAVs to map an area of interest using RGB and thermal sensors, as well as how to find, identify and classify reindeer from the air. They also learned how to use differential GNSS to measure positions of objects, making use of Real-Time Kinematics (RTK), and how to use this technology for high-accuracy UAV mapping. Back in the lab, the group was shown how to analyse selection decisions made by reindeer from the acquired environmental data and to model reindeer selection/habitat use. Students learned to work with spatial point and spatial image data and were introduced to spatial statistics, e.g. point pattern analysis for quantifying the distribution of reindeer using methods such as Clark Evan‘s test, Ripley’s reduced second moment function K or kernel density estimations. In addition, students learned how to fit Generalized Linear Models (GLMs) on UAV-derived remote sensing data for modelling selection and predicting habitat use. In the end, the students were able to fit their own selection models and to therefrom create maps that showed areas of higher and lower selection probability within the experiment area.

Duration: 3 sessions (full days, 3-days block)

In this guest lecture, my colleague Sebastian Buchelt and myself taught UAV (drone) remote sensing of the cryosphere to remote sensning MSc. students. The guest lecture was held at the invitation of Assoc. Prof. Dr. Eero Rinne (UNIS) as part of a multi-week international MSc. course on Remote Sensing of the Cryosphere. The course combined classroom sessions on flight regulations, data georeferencing and processing, and platform and sensor types with practical training in mission preparation and planning. On the second day, students were taught how to acquire high-resolution UAV imagery at the Longyearbreen, a glacier nearby Longyearbyen, to study surface temperature variations, melt channel features, and snow cover. On the third day, students processed the acquired data into orthomosaics and digital elevation models (DEMs), gaining first-hand experience with the challenges of handling large UAV datasets. The exercise demonstrated how UAV-derived spectral and thermal data can be applied to improve our understanding of cryospheric processes.

Duration: 3 sessions (full days, 3-days block)

This course introduces UAV/UAS (drone) remote sensing and field methods to ecologists. They get to know field campaigning (sampling methods, routing, positioning etc.), in-situ data sampling (parameters, field devices such as spectrometers, soil moisture probes etc.) and UAS (drone) imagery acquisition (platforms, sensors, flight planning, licensing, training). The course prepares them for a field day at a research site where data are collected under real-world research conditions. Afterwards, they learn to handle, process and analyse the recorded data and turn them into interpretable information. During the course, students create a personal guide book that they can use later in their careers to successfully conduct UAV research.

Duration: 10 sessions (full days, 2-weeks block)

This methodological course introduces Deep Learning with a practical focus on how to use it for image processing in Earth observation. Students get to know the principles behind the design of Neural Networks and the training thereof using Deep Learning. They learn about loss, backpropagation, optimization, activation functions & vanishing gradients, over- & underfitting, regularization, augmentation, convolutions, layers of state vs. stateless layers, sequential and non-sequential network designs for image processing tasks such as classification and segmentation etc. The course is taught Pyhton and R, mainly using keras and tensorflow.

Duration: 12 sessions (1.5 hours each)

This modelling course introduces concepts and methods to Earth observation students who want to learn to work with animal movement trajectories, a special type of spatio-temporal data, and integrate such with Earth observation analyses. The course aims to find data-driven answers to questions such as: Why do animals move through the landscape the way they do? How are they impacted by their environments? And: Which environmental conditions are tied to what kind of of movement behavior? While the course introduces background knowledge on topics such as movement theory, the effects of discretely observing continuous processes (e.g. sampling rate, autocorrelation, bias etc.), scale-dependencies/matching, tracking approaches and location error, its practical focus lays on methods to handle and analyze tracking data (e.g. using geometric & variance component analyses, behavioral segmentation, area-metrics such as home ranges, remote-sensing driven trajectory analysis incl. corridor analysis & habitat analysis) as well as tracking data in combination with remotely sensed environmental data (e.g. through resource utilization modelling, resource selection modelling and step selection modelling). This leads students to eventually be able to independently spot patterns in movement data, make connections to environmental conditions, and, finally, jointly model movement tracking and remotely sensed environmental data.

Duration: 4 sessions (4 hours each)

In this course, ecology students are introduced to remote sensing and geo-analysis for ecological research applications and learn to independently use remote sensing in their own research fields. In the first week, students are introduced to remote sensing and geo-spatial data and methods and learn to use QGIS and R for geo-spatial analysis. In the second week, they apply remote sensing and geo-analysis to answer their own real-world ecological research questions. At the end of the course, they present their remote-sensing-enabled findings.

Duration: 10 sessions (full days, 2-weeks block)

In this guest lecture, Arctic winter ecology MSc. students were introduced to satellite and UAV remote sensing and geo-analysis for ecological research applications in the Arctic. The guest lecture was held at the invitation of Assoc.Prof. Dr. Simone Lang (UNIS) as part of a six-week international MSc. course on Arctic winter ecology. On the first day, students were introduced to remote sensing research applications, physical principles of remote sensing and fundamental methods for remote sensing data analysis. In an hands-on session, they learned to handle, visualize and analyze remote sensing data from both Earth observation satellites and Unoccupied Aerial Vehicles (UAVs) using geo-spatial software, including QGIS and R. Afterwards, students were introduced to UAV remote sensing sensors and platforms as well as mission planning. The day finished with the design of a real-world field campaign for the next day, including in-situ snow data collection and UAV thermal and LiDAR data collection using two drones. On the second day, students learned in the field how to in-situ collecting snow parameters,, including snow depth, snow wetness, snow temperatures, snow hardness etc. To geo-locate in-situ measurements with high accuracy, students learned to use real-time kinematic (RTK) corrected GPS. Afterwards, two UAV systems were flown to map the research area, equipped with a thermal and a LiDAR sensor. Later-on, students had the chance to experiment with the data they had collected themselves, e.g. by validating UAV-derived snow depth or snow temperature data against their own measurements. The data collected during the course are of high potential value for Arctic ecology research, as they can be used to quantify the impact of snow properties on vegetation phenology and resource development in spring and summer.

Duration: 2 sessions (full days, 2-days block)

Interactive visualizations using shiny for R in combination with leaflet, plotly, ggplot2 and LaTeX. Basics of web design using html, css and javascript. Basics on Declarative Programming and Lazy Execution. Introduction into client-server communication using reactive contexts and event observation. Introduction into basic UI design such as dashboard UIs, interactive maps, interactive plots etc.

Duration: 3 sessions (1.5 hours each)

This methodological course introduces Deep Learning with a practical focus on how to use it for image processing in Earth observation. Students get to know the principles behind the design of Neural Networks and the training thereof using Deep Learning. They learn about loss, backpropagation, optimization, activation functions & vanishing gradients, over- & underfitting, regularization, augmentation, convolutions, layers of state vs. stateless layers, sequential and non-sequential network designs for image processing tasks such as classification and segmentation etc. The course is taught Pyhton and R, mainly using keras and tensorflow.

Duration: 12 sessions (1.5 hours each)

Remote sensing applications; physical principles (electromagnetic radiation, absorption, emission, reflectance, optics, spectral information); spatial raster & vector data types; coordinate reference systems & projections; QGIS & R; R for spatial & remote sensing data analysis; spectral indices; basic modelling (classifiers, regressions); generating experience-driven ground truth using digitization; supervised classifications; (statistical) validation; accuracy assessment; scale of data vs. scale of observation targets; resolutions.

Duration: 10 sessions (full days, 2-weeks block)

This modelling course introduces concepts and methods to Earth observation students who want to learn to work with animal movement trajectories, a special type of spatio-temporal data, and integrate such with Earth observation analyses. The course aims to find data-driven answers to questions such as: Why do animals move through the landscape the way they do? How are they impacted by their environments? And: Which environmental conditions are tied to what kind of of movement behavior? While the course introduces background knowledge on topics such as movement theory, the effects of discretely observing continuous processes (e.g. sampling rate, autocorrelation, bias etc.), scale-dependencies/matching, tracking approaches and location error, its practical focus lays on methods to handle and analyze tracking data (e.g. using geometric & variance component analyses, behavioral segmentation, area-metrics such as home ranges, remote-sensing driven trajectory analysis incl. corridor analysis & habitat analysis) as well as tracking data in combination with remotely sensed environmental data (e.g. through resource utilization modelling, resource selection modelling and step selection modelling). This leads students to eventually be able to independently spot patterns in movement data, make connections to environmental conditions, and, finally, jointly model movement tracking and remotely sensed environmental data.

Duration: 5 sessions (full days, 1-week block)

Basics of R & QGIS; version control using git; R in comparison to other languages (interpreter vs. compiler, memory management etc.); programming paradigms; procedural vs. object-oriented vs. functional programming; types/modes, structures, indexing; implicit vs. explicit type conversion; control flow constructs & vectorization; functions; package building; data visualization; statistics; spatial data analysis; image processing; classification models etc.

Duration: 12 sessions (1.5 hours each)

A practical guide towards building custom GEE processing pipelines natively in R using rgee.

Duration: 1 session (1.5 hours)

This methodological course introduces Deep Learning with a practical focus on how to use it for image processing in Earth observation. Students get to know the principles behind the design of Neural Networks and the training thereof using Deep Learning. They learn about loss, backpropagation, optimization, activation functions & vanishing gradients, over- & underfitting, regularization, augmentation, convolutions, layers of state vs. stateless layers, sequential and non-sequential network designs for image processing tasks such as classification and segmentation etc. The course is taught in R and Pyhton, mainly using keras and tensorflow.

Duration: 12 sessions (1.5 hours each)

This course introduces remote sensing field methods to ecologists. They are get to know field campaigning (sampling methods, routing, positioning etc.), in-situ data sampling (parameters, field devices such as spectrometers, soil moisture probes etc.) and UAS (drone) imagery acquisition (platforms, sensors, flight planning, licensing, training). This prepares them for a field day at a research site where data are collected under real-world research conditions. Afterwards, they learn to handle, process and analyse the recorded data and turn them into interpretable information.

Duration: 10 sessions (full days, 2-weeks block)

This modelling course introduces concepts and methods to Earth observation students who want to learn to work with animal movement trajectories, a special type of spatio-temporal data, and integrate such with Earth observation analyses. The course aims to find data-driven answers to questions such as: Why do animals move through the landscape the way they do? How are they impacted by their environments? And: Which environmental conditions are tied to what kind of of movement behavior? While the course introduces background knowledge on topics such as movement theory, the effects of discretely observing continuous processes (e.g. sampling rate, autocorrelation, bias etc.), scale-dependencies/matching, tracking approaches and location error, its practical focus lays on methods to handle and analyze tracking data (e.g. using geometric & variance component analyses, behavioral segmentation, area-metrics such as home ranges, remote-sensing driven trajectory analysis incl. corridor analysis & habitat analysis) as well as tracking data in combination with remotely sensed environmental data (e.g. through resource utilization modelling, resource selection modelling and step selection modelling). This leads students to eventually be able to independently spot patterns in movement data, make connections to environmental conditions, and, finally, jointly model movement tracking and remotely sensed environmental data.

Duration: 5 sessions (full days, 1-week block)

Remote sensing applications; physical principles (electromagnetic radiation, absorption, emission, reflectance, optics, spectral information); spatial raster & vector data types; coordinate reference systems & projections; QGIS & R; R for spatial & remote sensing data analysis; spectral indices; basic modelling (classifiers, regressions); generating experience-driven ground truth using digitization; supervised classifications; (statistical) validation; accuracy assessment; scale of data vs. scale of observation targets; resolutions.

Duration: 10 sessions (full days, 2-weeks block)

Basics of R & QGIS; version control using git; R in comparison to other languages (interpreter vs. compiler, memory management etc.); programming paradigms; procedural vs. object-oriented vs. functional programming; types/modes, structures, indexing; implicit vs. explicit type conversion; control flow constructs & vectorization; functions; package building; data visualization; statistics; spatial data analysis; image processing; classification models etc.

Duration: 12 sessions (1.5 hours each)

Remote sensing of forests, vulnerabilities & risks, biodiversity, wildlife ecology, natural resources, fire & burnt areas, coasts, diseases & health, agriculture, soil, land cover & land use, human settlements, policy etc.

Duration: 12 sessions (1.5 hours each)

Duration: 1 session (2.5 hours)

Duration: 2 sessions (full days)

This course introduces remote sensing field methods to ecologists. They are get to know field campaigning (sampling methods, routing, positioning etc.), in-situ data sampling (parameters, field devices such as spectrometers, soil moisture probes etc.) and UAS (drone) imagery acquisition (platforms, sensors, flight planning, licensing, training). This prepares them for a field day at a research site where data are collected under real-world research conditions. Afterwards, they learn to handle, process and analyse the recorded data and turn them into interpretable information.

Duration: 10 sessions (full days, 2-week block)

This methodological course introduces Deep Learning with a practical focus on how to use it for image processing in Earth observation. Students get to know the principles behind the design of Neural Networks and the training thereof using Deep Learning. They learn about loss, backpropagation, optimization, activation functions & vanishing gradients, over- & underfitting, regularization, augmentation, convolutions, layers of state vs. stateless layers, sequential and non-sequential network designs for image processing tasks such as classification and segmentation etc. The course is taught in R and Pyhton, mainly using keras and tensorflow.

Duration: 12 sessions (1.5 hours each)

(Interactive) visualizations using LaTeX, leaflet, plotly, ggplot2, shiny, basics of web design using html, css, javascript (including frameworks such as bootstrap), static site generation frameworks such as hugo, jekyll etc.

Duration: 5 sessions (1.5 hours each)

This modelling course introduces concepts and methods to Earth observation students who want to learn to work with animal movement trajectories, a special type of spatio-temporal data, and integrate such with Earth observation analyses. The course aims to find data-driven answers to questions such as: Why do animals move through the landscape the way they do? How are they impacted by their environments? And: Which environmental conditions are tied to what kind of of movement behavior? While the course introduces background knowledge on topics such as movement theory, the effects of discretely observing continuous processes (e.g. sampling rate, autocorrelation, bias etc.), scale-dependencies/matching, tracking approaches and location error, its practical focus lays on methods to handle and analyze tracking data (e.g. using geometric & variance component analyses, behavioral segmentation, area-metrics such as home ranges, remote-sensing driven trajectory analysis incl. corridor analysis & habitat analysis) as well as tracking data in combination with remotely sensed environmental data (e.g. through resource utilization modelling, resource selection modelling and step selection modelling). This leads students to eventually be able to independently spot patterns in movement data, make connections to environmental conditions, and, finally, jointly model movement tracking and remotely sensed environmental data.

Duration: 5 sessions (full days, 1-week block)

Remote sensing applications; physical principles (electromagnetic radiation, absorption, emission, reflectance, optics, spectral information); spatial raster & vector data types; coordinate reference systems & projections; QGIS & R; R for spatial & remote sensing data analysis; spectral indices; basic modelling (classifiers, regressions); generating experience-driven ground truth using digitization; supervised classifications; (statistical) validation; accuracy assessment; scale of data vs. scale of observation targets; resolutions.

Duration: 10 sessions (full days, 2-weeks block)

Basics of R & QGIS; version control using git; R in comparison to other languages (interpreter vs. compiler, memory management etc.); programming paradigms; procedural vs. object-oriented vs. functional programming; types/modes, structures, indexing; implicit vs. explicit type conversion; control flow constructs & vectorization; functions; package building; data visualization; statistics; spatial data analysis; image processing; classification models etc.

Duration: 12 sessions (1.5 hours each)

Remote sensing of forests, vulnerabilities & risks, biodiversity, wildlife ecology, natural resources, fire & burnt areas, coasts, diseases & health, agriculture, soil, land cover & land use, human settlements, policy etc.

Duration: 12 sessions (1.5 hours each)

This methodological course introduces Deep Learning with a practical focus on how to use it for image processing in Earth observation. Students get to know the principles behind the design of Neural Networks and the training thereof using Deep Learning. They learn about loss, backpropagation, optimization, activation functions & vanishing gradients, over- & underfitting, regularization, augmentation, convolutions, layers of state vs. stateless layers, sequential and non-sequential network designs for image processing tasks such as classification and segmentation etc. The course is taught in R and Pyhton, mainly using keras and tensorflow.

Duration: 12 sessions (1.5 hours each)

Spectrometric field sampling, hyperspectral image processing, spectral unmixing in R & QGIS etc.

Duration: 2 sessions (full days, 2-days block)

(Interactive) visualizations using LaTeX, leaflet, plotly, ggplot2, shiny, basics of web design using html, css, javascript (including frameworks such as bootstrap), static site generation frameworks such as hugo, jekyll etc.

Duration: 5 sessions (1.5 hours each)

This modelling course introduces concepts and methods to Earth observation students who want to learn to work with animal movement trajectories, a special type of spatio-temporal data, and integrate such with Earth observation analyses. The course aims to find data-driven answers to questions such as: Why do animals move through the landscape the way they do? How are they impacted by their environments? And: Which environmental conditions are tied to what kind of of movement behavior? While the course introduces background knowledge on topics such as movement theory, the effects of discretely observing continuous processes (e.g. sampling rate, autocorrelation, bias etc.), scale-dependencies/matching, tracking approaches and location error, its practical focus lays on methods to handle and analyze tracking data (e.g. using geometric & variance component analyses, behavioral segmentation, area-metrics such as home ranges, remote-sensing driven trajectory analysis incl. corridor analysis & habitat analysis) as well as tracking data in combination with remotely sensed environmental data (e.g. through resource utilization modelling, resource selection modelling and step selection modelling). This leads students to eventually be able to independently spot patterns in movement data, make connections to environmental conditions, and, finally, jointly model movement tracking and remotely sensed environmental data.

Duration: 5 sessions (full days, 1-week block)

Basics of R & QGIS; version control using git; R in comparison to other languages (interpreter vs. compiler, memory management etc.); programming paradigms; procedural vs. object-oriented vs. functional programming; types/modes, structures, indexing; implicit vs. explicit type conversion; control flow constructs & vectorization; functions; package building; data visualization; statistics; spatial data analysis; image processing; classification models etc.

Duration: 12 sessions (1.5 hours each)

Remote sensing of forests, vulnerabilities & risks, biodiversity, wildlife ecology, natural resources, fire & burnt areas, coasts, diseases & health, agriculture, soil, land cover & land use, human settlements, policy etc.

Duration: 12 sessions (1.5 hours each)

introduction into machine learning, deep neural networks, computer vision, discriminative modelling; (web-)APIs, web protocols, machine-to-machine communication; development environments, code maintenance, unit testing, continuous integration, version control etc.

Duration: 12 sessions (1.5 hours each)

Spectrometric field sampling, hyperspectral image processing, spectral unmixing in R & QGIS etc.

Duration: 2 sessions (full days, 2-days block)

(Interactive) visualizations using LaTeX, leaflet, plotly, ggplot2, shiny, basics of web design using html, css, javascript (including frameworks such as bootstrap), static site generation frameworks such as hugo, jekyll etc.

Duration: 5 sessions (1.5 hours each)

Duration: 1 session (2.5 hours)

Duration: 1 session (2.5 hours)

introduction into machine learning, deep neural networks, computer vision, discriminative modelling; (web-)APIs, web protocols, machine-to-machine communication; development environments, code maintenance, unit testing, continuous integration, version control etc.

Duration: 12 sessions (1.5 hours each)

Title of the talk: Potentials of integrating animal movement tracking data with remote sensing.

Duration: 1 session (2.5 hours)

Duration: 1 session (2.5 hours)