Machine learning models in hydrology often achieve excellent performance in the environments in which they were trained – but can they truly generalize beyond these conditions?
To what extent are these models transferable across different catchments, spatial scales, and monitoring conditions? Can data-driven hydrological models reliably predict extreme events in ungauged or data-scarce catchments?
This PhD project addresses a fundamental challenge in contemporary hydrology: understanding the generalizability and transferability of machine learning hydrological models beyond the environments in which they were developed.
Research objectives
The PhD project focuses on fundamental research into the limits, generalizability, and transferability of machine learning-based flood prediction across different physiographic and hydroclimatic conditions. The main objective is to understand how the availability of hydrological information, flood-generating mechanisms, and environmental characteristics affect prediction reliability, uncertainty, and achievable forecast lead time.
The research will investigate the conditions under which models developed in well-instrumented experimental catchments can be transferred to other environments, particularly data-scarce and ungauged mountain and headwater catchments.
Particular attention will be paid to changes in flood predictability under climate variability, climate change, and environmental transformations, including the ability of models to capture extreme events and extrapolate beyond the conditions represented in their training data.
Research focus
The research will use long-term, high-frequency hydrological observations and data from sensor networks operated by the Faculty of Science, Charles University, in experimental mountain catchments. These will be complemented by hydrometeorological data, remote sensing observations, catchment physiographic characteristics, and large-scale open hydrological datasets (e.g., CAMELS and LamaH).
Machine learning methods will be used to investigate relationships between the availability and characteristics of hydrological information, catchment physiography, flood-generating mechanisms, and flood predictability. The research will focus on identifying the causes and conditions of prediction performance degradation and assessing the transferability of the resulting knowledge across catchments, spatial and temporal scales, and changing hydroclimatic conditions.
Key research questions:
- How do the availability, quality, and spatial distribution of hydrological observations influence the predictability of extreme events and its limits?
- To what extent are models and identified predictive relationships transferable between catchments with different physiographic characteristics?
- How do predictability and model transferability vary across temporal and spatial scales?
- How do climate variability, climate change, and environmental transformations affect the stability of predictive relationships and the ability of models to predict extreme events?
The research will address the following tasks:
- Development and evaluation of reference ML models for rainfall–runoff response and flood prediction, including assessment of their performance across different flood types, forecast lead times, and data availability conditions.
- Analysis of the information value of hydrological observations, focusing on the effects of data quantity, type, quality, and the spatial configuration of sensor networks, including data representativeness, redundancy, and complementarity.
- Identification of hydrological limits to predictability in relation to flood-generating mechanisms, event magnitude and dynamics, antecedent conditions, and forecast lead time, including analysis of model failures during extreme and unusual hydrological events.
- Assessment of the stability of predictive relationships in relation to catchment physiography, hydrological regime, climate variability, and environmental change, including the ability of models to extrapolate beyond the range of training conditions.
- Analysis of model generalizability and transferability across catchments with contrasting environmental conditions, spatial scales, and hydroclimatic regimes, leading to general insights applicable to hydrological prediction and monitoring in data-scarce and ungauged catchments.
The aim of the project is not merely to develop effective predictive models, but primarily to advance our understanding of the hydrological principles and limitations underlying data-driven flood prediction and to establish generalizable knowledge about its reliability and transferability under changing climatic and environmental conditions.
The results will be published in relevant peer-reviewed international journals and presented at international scientific conferences.
Candidate profile
We welcome applicants with an academic background in fields such as:
- Hydrology or physical geography
- Earth sciences or environmental sciences
- Geoinformatics and remote sensing
- Data science, statistics, or related quantitative disciplines
The ideal candidate has a strong interest in the research topic, scientific curiosity, and motivation to address interdisciplinary challenges at the intersection of hydrology and data science. We particularly value analytical thinking, a willingness to explore new methodological approaches, and the ability to critically interpret research findings.
Programming skills (particularly Python) and previous experience with machine learning, statistical modelling, or hydrological data analysis are advantageous but not required. Candidates are expected to be motivated to develop these skills during their doctoral studies.
The ability to work independently, take initiative, and develop original research approaches is equally important as the ability to collaborate effectively within a research team.
Research environment
The position is hosted by the Hydrology Research Group, Faculty of Science, Charles University (Prague) — an internationally active group working on experimental catchments, sensor networks, hydrological extremes, and data-driven modelling. The PhD is conducted in collaboration with the T. G. Masaryk Water Research Institute (Prague), linking fundamental research with applied water management.
What we offer
- a fully funded doctoral position,
- access to unique experimental hydrological datasets,
- publication-oriented research with international visibility,
- conference participation and international collaboration,
- a friendly and vibrant research environment in Prague.
Prospective candidates are encouraged to contact the supervisor before submitting a formal application.
Supervisor agreement
Before submitting a formal PhD application, candidates must obtain the supervisor’s consent.
To request this consent, applicants must contact the supervisor and submit the required materials (CV and motivation letter) via the designated form no later than November 30, 2026. The form requires login using a Google account.
All submitted materials will be carefully reviewed by early April. Applicants will be informed about the results shortly thereafter. Candidates who best meet the criteria will be invited for an interview in mid-April.
Following a successful interview, selected candidates will be eligible to submit a formal PhD application through the Charles University Study Information System (SIS).
The official deadline for submitting the formal application via the Charles University Study Information System (SIS) is December 15, 2026.
Binding information and official application via SIS
(link will be activated in October, 2026)
Important dates
- Deadline for requesting supervisor consent (CV + motivation letter via Google form): November 30, 2026
- Group interviews for shortlisting candiates: December 2-11, 2026
- Official application deadline (via Charles University Study Information System): December 15, 2026
- Admission interview: February 1, 2027
- Expected start date: April 1, 2027
Important notice
The advertised PhD position is subject to securing funding for the associated research project.
The Faculty of Science, Charles University, reserves the right not to open the position if the required funding is not awarded.
