Join our workshop during the 11th Scientific Assembly of the IAHS
in Montpellier (June 2022)!
Objectives
For decades, hydrologists have been using time series of rainfall, air temperature, water level and streamflow data to feed, calibrate and evaluate statistical and deterministic hydrological models. These models have many applications in hydrological engineering, such as estimating the mean annual streamflow of an ungauged catchment, flood frequency analysis at gauged and un-gauged sites, extrapolation of streamflow response in changing climate (e.g. extreme floods and droughts), etc. Unavoidably, the temporal series used within these methods are characterized by epistemic and aleatory errors (e.g. incorrect transcription of a handwritten value) and/or systematic inconstancies (e.g. change of the gauging location). These errors raise the issue of the (dis)informative content of the data used in hydrology and specifically in rainfall-runoff modeling: are all periods equally informative, are there seasonal trends? If some time series or periods can be considered as disinformative, how is it possible to detect them and what effects might they have on model identification and inference? Are there criteria or methods (quantitative and/or qualitative) that might be used to classify data in this way given different sources of uncertainty, independent of the model being used? Might the classification be different for catchments with more or less baseflow for example? And if we can avoid using these series or periods in model identification, will this have a significant impact on the performance of the considered models and thus on their robustness?
We propose to organize a collaborative workshop during 2022 IASH General Assembly to bring together a group of researchers around the following question:
Can we quantify the (dis)information content of the uncertain hydro-meteorological data series used in hydrology?
Participants will have to propose innovative solutions to identify informative or disinformative data taking account of sources of data uncertainty and to show how this classification might be used to produce more robust hydrological calculations and models. The participants are free to develop methods based on series analysis (e.g. analysis of runoff coefficients), statistical tests, rainfall-runoff modeling or machine learning. Contributions are welcomed in various domains: flood frequency analysis, regionalization, rainfall-runoff modeling (both lumped and semi-distributed), hydro-meteorological forecasting, modelling of climate change impact on water resources, etc.
Potential road maps for the development of methods to identify informative or disinformative data are given through the following types of questions:
- If some periods of data are disinformative in model identification, how can we validate or invalidate different potential model structures?
- If some periods of data are disinformative in model identification how can we assess predictability in flood frequency analyses or simulations of future flow regime on a catchment?
- If some periods of data are disinformative for model identification, how can we gain insight into the importance of different sources of uncertainty in the modelling process?
To answer one or more of these questions, participants are invited to test their methods on “their” catchment dataset or on catchment datasets available online (e.g. CAMELS), and to discuss the obtained results in terms of catchment characteristics (flashiness, baseflow or snow influence, etc.)
The outputs of this workshop may be published in a dedicated Hydrological Sciences Journal special issue!
Program
Workshop n°8: When are hydrological data informative (or not)? Testing for information content in the face of sources of uncertainty.
- Convener: Pierre Brigode | Co-conveners: Vazken Andréassian, Stacey Archfield, Keith Beven, Louise Crochemore, Alexander Gelfan, Julien Lerat & Ralf Merz.
- Orals: Wed, 01 Jun, 08:30–15:00 | Rooms Sully 1 and Sully 2
- Discussion and posters: Attendance Wed, 01 Jun, 08:30–15:00 | Rooms Sully 1 and Sully 2
- Abstracts are available here
Orals (last update: 29/05/2022)
- 08:30 - 08:45: Workshop introduction. P. Brigode & V. Andréassian.
- 08:45 - 09:00: Is model calibration compensation for input errors? F. Anwar & A. Bárdossy.
- 09:00 - 09:15: On parameter instability in model building. P. Astagneau, F. Bourgin, V. Andréassian & C. Perrin.
- 09:15 - 09:30: Recalculation of historical streamflow series. Impact assessment and valorization. A. Belleville & D. Sevrez.
- 09:30 - 09:45: Dealing with uncertainty of rainfall-runoff parameters in a stochastic simulation method for extreme flood estimation: milking the most of a calibration dataset. E. Paquet.
- 09:45 - 10:00: Which range of discharge data is most informative in the calibration of a rainfall-runoff model? M. Saadi, C. Furusho-Percot & S. Kollet.
Coffee break
- 10:30 - 10:45: Hunting after shoals of red herrings - cataloguing disinformative data and their causes. I. Westerberg & J. Seibert.
- 10:45 - 11:00: Impact of dubious data in streamflow time series on the efficiency and the parameters of rainfall-runoff models. C. Thébault, C. Perrin, V. Andréassian, G. Thirel & S. Legrand.
- 11:00 - 11:15: Value of historical information for flood frequency estimation: case study on the upper Rhine River. M. Lang, B. Renard, J. Le Coz & M. Darienzo.
- 11:15 - 11:30: Presentation of the posters and of the Hydrological Science Journal Special Issue associated with the workshop. P. Brigode.
- 11:30 - 12:00: Discussions.
Lunch break
- 13:30 - 14:00: Information and disinformation in data – can we do better as a community? K. Beven.
- 14:00 - 15:00: Poster tour.
Posters
- Choosing acceptable parameter sets when calibrating a continental rainfall-runoff model (E-HYPE v.4). C. Brendel, R. Capell, J. Musuuza, K. Isberg & B. Arheimer.
- Why calibrate and average over time if we can directly estimate parameters and their temporal evolution from data? M. Hrachowitz
- Errors and uncertainties in streamflow data. J. Le Coz, B. Renard, M. Darienzo, I. Horner, F. Branger & M. Lang.
- Use of data assimilation to improve rainfall-runoff model structure for climate change projections. J. Lerat, F. Chiew, Zheng & D. Robertson.
- Use of transfer entropy between streamflow and forcing time series for the identification of dominant rainfall-runoff dynamics and for the assessment of hydrological similarity. M. Neri, Coulibaly & E. Toth.
- Evaluating the value of crowd-based observations for hydrological modelling. J. Seibert, F. Schwarzenbach, S. Blanco, M. Scheller, W. Ze & I. van Meerveld.
- Can quantitative hydrological models benefit from water quality measurements? Revisiting the baseflow separation issue. G. Tallec, J. Tunqui Neira, V. Andréassian & J.-M. Mouchel.
- Can observed information be transferred from gauged to ungauged catchments? S. Tian, J. Lerat, Renzullo, & Pipunic.
- On snow undercatch by raingages: looking for the missing information. V. Andréassian, Gevorgyan, Misakyan & Azizyan.
- Customizing large-scale hydrological models for local applications: Is it a data information extraction challenge? I. Pechlivanidis, & J. Musuuza.
- Critical approach on the trend analyses of groundwater levels time series: study case of Telangana state, India. A. Selles, A. Paswan, S. Ferrant, B. Dewandel & J-C. Marechal.
- Do hydrological model errors spread across time and timescales? (withdrawn) P. Royer-Gaspard.
- On bad and good neighbors in a hydrological regionalization perspective. (withdrawn) P. Brigode, F. Bourgin, V. Andréassian, C. Perrin & L. Oudin.
- Using snow data to improve realism of a semi-distributed hydrological model. (withdrawn) D. Ruelland.
- Efficiently exploiting the information content of in-situ soil moisture measurements with machine learning. (withdrawn) R. Orth, S. O.