Industry-standard streamflow forecasts use state-of-the art weather forecasts to drive hydrological models and predict hydrological extremes (floods and droughts) weeks to months before they occur. However, these traditional approaches are often computationally demanding and have limited skill, which hinders their uptake by end users. Now, with the advent of advanced statistical forecasting techniques alongside a range of Big Data sources (such as Earth Observation from satellite archives), it is possible to start exploring ‘intelligent’ approaches to forecast hydrological extremes.
This project will evaluate the feasibility of skilful long-range statistical forecasting methods based on a range of Global Big Data sources, including climate forecasts, teleconnection indices, and Earth Observation data.
The research will involve accessing climate hindcasts from a range of modelling centres (e.g. ECMWF, NOAA, and NASA), Earth Observation data from satellite archives (e.g. SMOS, Landsat, or Sentinel), and streamflow time series for catchments in different physical and climatic environments around the world. Seasonal streamflow extremes will be forecast using a range of different statistical models in R, including machine learning models.
This research will require a degree of quantitative expertise, understanding of time series analysis, and the ability to code in R. The proposed approach opens the way for more computationally efficient flood and drought forecasting and is of practical interest to a broad range of end users and practitioners. Critical new scientific understanding will come from this studentship, in terms of developing competitive forecasting methodologies, and defining the usability of Big Data approaches for forecasting applications.
The student will: (1) Obtain publicly-available forecasts of a range of climate quantities (e.g. precipitation, temperature, sea surface temperature, atmospheric moisture) to compute climate indices and teleconnection modes over a range of spatial and temporal scales; (2) Prepare associated time series of streamflow and Earth Observation data using hydrometric and satellite archives (e.g. land cover and soil moisture); (3) Develop statistical models (building on existing models) to compute streamflow forecasts over a range of lead times; (4) Compare the skill of the statistical forecasts with that of industry-standard forecasts, using bench-marking tools like Ensemble Streamflow Prediction (ESP); (5) Contribute to the dissemination and operationalisation of these new tools via presentations at national and international conferences, and via discussions with end-users and practitioners. The student will acquire state-of-the-art skills and knowledge of forecasting methodologies, data science and ‘Big Data’ analysis techniques.
Training and Skills
CENTA students are required to complete 45 days training throughout their PhD including a 10 day placement. In the first year, students will be trained as a single cohort on environmental science, research methods and core skills. Throughout the PhD, training will progress from core skills sets to master classes specific to CENTA research themes.
Extensive training will be given to the student in the fundamentals of hydrological forecasting, climate informatics and data science at ECMWF and Loughborough University. The successful student will require strong analytical skills which will be applied through algorithm development within this studentship.
The student will be integrated within an experienced project team and will work alongside existing CENTA and ECMWF PhD students. The student will attend at least one relevant international conference in their first year in order to gain an understanding of the acknowledged state-of-the-art in the area.
Year 1: Comprehensive literature review of long-range streamflow forecasting methodologies; In-depth assessment and acquisition of available data (climate forecasts and Earth Observation); Training to acquire data science and programming skills in R; Initial development of a statistical forecasting algorithm in R, applied to selected test locations/catchments identified with ECMWF.
Year 2: Further development of the statistical forecasting algorithm; evaluation of ancillary predictors and other statistical approaches; Benchmarking with existing methods that are operational at ECMWF; and/or comparison with a standard approach like ESP.
Year 3: Finalising the forecasting algorithm; Discussion and knowledge exchanges with end-users and practitioners; Final analyses of model skill relative to established products; Publication in peer-reviewed journals; Presentation to national and international forecasting communities; Writing up and submission of the thesis.
Partners and collaboration (including CASE)
Fieldwork will include a secondment for the PhD student within ECMWF, working alongside other streamflow forecasters. Support will be provided for the student throughout the project by statisticians and computer scientists at ECMWF and Loughborough University.
For information about this project, please contact Dr Louise Slater (email@example.com). For enquiries about the application process, please contact the School of Social, Political and Geographical Sciences Research (firstname.lastname@example.org).
Please quote CENTA when completing the application form: http://www.lboro.ac.uk/study/apply/research/.