MSc Research Projects

Introductory Materials (Video, Slides, Description)

Representatives:

Dr. Rubén Díaz Sierra, Universidad Nacional de Educación a Distancia (UNED); Dr. Mara Baudena, National Research Council of Italy

Join Teams Channel for Challenge 2 to meet representatives Dr. Rubén Díaz Sierra and Dr. Mara Baudena, and other potential teammates

Background of the project:

The well-known and controversial intermediate disturbance hypothesis (IDH) predicts that intermediate frequency or intensity of perturbations maximize biodiversity, allowing the coexistence of many different species competing for the same resources^1,2. We recently adapted a classical implicit-space model for plant competition and coexistence³ by including stochastic perturbations to model the effects of forest fires on the persistence of ecosystems⁴. More recently, we have developed the mathematical conditions that determine how periodic perturbations alter species coexistence in a two-dimensional impulsive differential system.

Project:

The students are expected to perform simulations exploring the effect of frequency and intensity of (or recovery capacity after) perturbations on species coexistence, and how it is affected by colonization rates and competition hierarchy. Students will also attempt to propose generalizations of the coexistence conditions in higher dimensional systems.

Main questions to be addressed:  

1. Is the IDH observed in simulations for large stochastic models?
2. Can conditions in low dimensional systems be generalized for high dimensional models?

Complex Systems Science aspects:

General properties in population ecology are powerful tools in articulating theoretical discussions and inspiring empirical research. Proving them for general systems with partially random elements is a hard task. The proposed modeling scheme allows up-bottom (simulations) and bottom-up (coexistence conditions) approaches to understand the mechanisms behind the IDH.

Possible societal importance/impact:  

The loss of biodiversity is one of our main current societal challenges. Understanding general drivers of species coexistence is very relevant. In particular, perturbations (including fires) are expected to increase, both in frequency and intensity, with climate and land use change, threatening ecosystems diversity, functions and resilience.

Initial literature:  

1. Huston, M. A general hypothesis of species diversity. American Naturalist 113, 81-101 (1979).
2. Hughes, A. (2010) Disturbance and Diversity: An Ecological Chicken and Egg Problem. Nature Education Knowledge 3(10):48
3. Tilman, D. (1994) Competition and biodiversity in spatially-structured habitats. Ecology, 75, 2–16.
4. Baudena, M., Santana, V.M., Baeza, M.J., Bautista, S., Eppinga, M.B., Hemerik, L., Garcia Mayor, A., Rodriguez, F., Valdecantos, A., Vallejo, V.R., Vasques, A., Rietkerk, M., 2020. Increased aridity drives post‐fire recovery of Mediterranean forests towards open shrublands. New Phytol. 225, 1500–1515. https://doi.org/10.1111/nph.16252. Code: https://github.com/baudenam/FireMed-Baudena-et-al-2019-New-Phytologist
5. A. Tchuinté Tamen, Y. Dumont, J.J. Tewa, S. Bowong, P. Couteron. A minimalistic model of tree–grass interactions using impulsive differential equations and non-linear feedback functions of grass biomass onto fire-induced tree mortality. Math. Comput. Simul., 133 (2017)

Representatives:

Paul Petersik (VineForecast) and Prof. dr. ir. Hendrik Dijkstra

Introduction:

High resolution weather model data contains a vast amount of valuable information. However, for practical applications often just a very small fraction of that data is needed. For small businesses that want to work with this data, saving a vast amount of model data with a high resolution is often too costly and not feasible. Therefore, clever compression and downscaling algorithms for weather data are needed to achieve a lean and efficient approach to work with high-resolution model data.

Aim of the project:

The aim of this project is to build and train machine learning models that take low resolution spatio-temporal variables, e.g. temperature, and high resolution spatial variables, e.g. topography, as input to predict the spatio-temporal variables with a higher resolution.
For instance, one could reduce the needed disk storage by a factor of 100 if a downscaling algorithm is able to recreate high resolution data which comes from a 1 km x 1km grid based on input data which is stored on a coarser 10 km x 10 km grid.

Moreover, it would be very interesting to investigate if this downscaling algorithm has some skill to downscale data even further, i.e. downscale the recreated 1km x 1km data to a resolution of 100m x 100m. This might be indeed possible for some variables such as 2m temperature, because of scale independent relations of the variable with the topographic influence.

Requirements:

- solid programming skills in python (numpy, matplotlib, xarray)
- first experience with machine learning python-packages (scikit-learn, keras)
- first experience in working with weather data (netcdf, grib)

Benefits:

- Working closely together with the team of a growing research startup (recently received research grant from German government)
- Hands-on work with the possibility for a direct implementation of the gained insights
- Possibility to work at the Seedhouse Startup Accelerator in Osnabrück (www.seedhouse.de)

If you are interested and want to get more details about this project feel free and contact Paul Petersik via mail to paul@vineforecast.com. You can find more information about VineForecast on its website: www.vineforecast.com.

Representatives:

Paul Petersik (VineForecast) and Prof. dr. ir. Hendrik Dijkstra

Introduction:

Fungal diseases of grape vine are one of the biggest sources of yield loss in viticulture. To control the disease development, winemakers need to apply fungicides against these fungal diseases. A precise forecast of fungal diseases can support winemakers to apply fungicides with an optimal timing and hence prevent overuse of these chemical agents.

Nowadays, forecast systems mostly use a combination of meteorological (e.g. relative humidity, temperature or leaf wetness) and phenological (e.g. leaf number or maturity of grapes) variables as predictors to estimate the infection risk. On the one hand, certain weather conditions favour the development of fungal disease e.g. the downy mildew disease needs rainy weather and long leaf wetness periods for its development. On the other hand, certain phenological stages are more susceptible to infections than others. For example, grape vines are most susceptible to the powdery mildew disease around bloom.

Current approaches to model grape vine diseases are built around a “mean” approach. Hence, without saying, one assumes that an entire vineyard behaves as an “average” grape vine which is exposed to the vineyard-“averaged” environmental condition.

However, in reality microclimatic conditions can vary on a short distance, especially when the vineyard is located in a complex terrain. Moreover, phenological development can vary quite significantly from shoot to shoot on a single plant as well as between different plants. The influence of these spatial heterogeneities was only examined in very few studies (e.g. Calonnec et al., 2014) and is not yet integrated into disease forecast models.

Aim of the project:

The aim of this project is to investigate how spatial heterogeneities are influencing disease development in a modelling study. For this, an idealized 2-D vineyard model will be used and developed further in this project.

Requirements:

- solid programming skills in python (numpy, matplotlib)
- basic understanding of ordinary and partial differential equations (ODE/PDE)
- basic understanding of numerical integration schemes for ODEs and PDEs
- first experience to work with disease models is a plus

Benefits:

- Working closely together with the team of a growing research startup (recently received research grant from German government)
- Hands-on work with the possibility for a direct implementation of the gained insights
- Possibility to work at the Seedhouse Startup Accelerator in Osnabrück (www.seedhouse.de)

If you are interested and want to get more details about this project feel free and contact Paul Petersik via mail to paul@vineforecast.com. You can find more information about VineForecast on its website: www.vineforecast.com.

References:

Calonnec, Agnes, et al. "Modelling of powdery mildew spread over a spatially heterogeneous growing grapevine." Modelling of powdery mildew spread over a spatially heterogeneous growing grapevine. 105 (2014): 137-148.

Representative:

Dr. Tjebbe Hepkema (NS), Simone Griffioen (NS), Bert de Vries (NS), and Dr. Deb Panja (UU)

Introduction:

The Nederlandse Spoorwegen (NS) is the main railway operator in the Netherlands. Its goal is to make the Netherlands accessible for everybody in a sustainable way. An important aspect of this is to accurately estimate the number of passengers that wish to travel from one place to another. Based on historical data the number of passengers is predicted for every train at different timescales in advance.

Aim of the project:

The project aims at improving the prediction of the number of passengers. Multiple research options are possible. The first is to take an probabilistic approach and quantify the uncertainty of the prediction. Besides the expected number of passengers, the complete distribution is relevant for scheduling the rolling stock. The second option is to find patterns in the dynamics of the number of passengers. This will help to choose the dates on which predictions are based. Besides these two options, others can be discussed.

Requirements:

• Quantitively mature (working on a MSc in math/physics/AI/…),
• Fluent in Python, Julia and/or R,
• Affinity with data, programming and logistic processes,
• At least six months available,
• Proactively asking questions at NS meetings.

Benefits:

• You’ll contribute to sustainable mobility,
• The NS has a lot of data,
• Internship association to meet other NS interns,
• €27,46 per workday.

For more details contact Dr. Tjebbe Hepkema at Tjebbe.hepkema@ns.nl.

Representatives:

Paul Petersik (VineForecast) and Prof. dr. ir. Hendrik Dijkstra

Introduction:

Grapevine phenological models play a crucial role in grapevine disease prediction because most grapevine diseases are favoured by specific phenological stages. For example, vines are most susceptible to the powdery mildew disease around bloom.

Grapevine phenology is often modelled using a cumulative degree days approach (e.g. Molitor et al., 2020). For this, the cumulated daily mean temperature starting from a specific date is used as a predictor for specific phenological stages.

Current approaches employ rather simple statistical models for this, e.g. logistic regression (as in Leolini et al, 2020). These models may have the advantage that they can give insight into the direct influence of the predictor variable onto the target. However, they are likely to underfit the real world data. Using more complex methods, e.g. random forest or neural networks, might lead to better performance which is strongly needed for real world applications.

Furthermore, phenological models are often focused on a specific variety, i.e. Riesling or Müller-Thurgau. For an efficient implementation of a phenological model into an operational system, there is a strong need for a general model that takes the variety directly a predictor variable into account.

Aim of the project:

The aim of this project is to build and train a machine learning model that can predict grape vine phenology for a wide range of varieties. Afterwards, the model performance should be compared to the performance of conventional models.

Requirements:

- solid programming skills in python (numpy, matplotlib)
- first experience with machine learning python-packages (scikit-learn, kears) is a plus
- first experience in working with weather data is a plus

Data:

- Phenological data for the model training is available through the PERPHECLIM data platform (https://data.pheno.fr/)

Benefits:

- Working closely together with the team of a growing research startup (recently received research grant from German government)
- Hands-on work with the possibility for a direct implementation of the gained insights
- Possibility to work at the Seedhouse Startup Accelerator in Osnabrück (www.seedhouse.de)

If you are interested and want to get more details about this project feel free and contact Paul Petersik via mail to paul@vineforecast.com. You can find more information about VineForecast on its website: www.vineforecast.com.

References:

Leolini, Luisa, et al. "Phenological model intercomparison for estimating grapevine budbreak date (Vitis vinifera L.) in Europe." Applied Sciences 10.11 (2020): 3800.

Molitor, Daniel, Helder Fraga, and Jürgen Junk. "UniPhen–a unified high resolution model approach to simulate the phenological development of a broad range of grape cultivars as well as a potential new bioclimatic indicator." Agricultural and Forest Meteorology 291 (2020): 108024.