• We aim to estimate, at every point, the frequent floods of biological and morphogenic interest.
• The construction of a reference dataset primarily reveals a heterogeneity in the exploitable measurement periods – a recurring challenge in hydrology.
• We propose a spatial interpolation method, TREK, which accounts for the observation period available at each station.
• Our focus is on estimating, throughout the hydrographic network, the frequent floods of biological and morphogenic interest.

Only a few studies have focused on frequent floods regimes at ungauged locations. Most of works have put their efforts on extreme flood events (return periods of 10 years or more) needed for solving many engineering issues in flood risk management. However, high flows regime is not confined to extremes values. A good understanding of frequent floods is required in a wide array of topics like hydroecology and hydromorphomology. Frequent floods provide many functions, maintaining and rejuvenating ecological habitats and influencing the geomorphology of the streambed, so their distribution must be also known.The main objective of this work is to characterise the frequent floods from a statistical point of view (with a return period between 1 and 5 years) in France. Forming the dataset is a preliminary crucial step to derive both robust and reliable statistics. The selection relies on different criteria, for example related to the quality of discharge measurements, the length of records, the self-assessment of people in charge, and finally on an analysis of extreme values extracted from time series (stationarity, shape of the distributions…).A comprehensive description of frequent floods regimes (intensity, duration and frequency) is required. It is achieved by applying the flow-duration–frequency (QdF) model which takes into account the temporal dynamics of floods. This approach is analogous to the intensity-duration–frequency (IdF) model commonly used for extreme rainfall analysis. At gauged locations, the QdF model can be summarised with only three parameters: the position and scale parameters of the exponential distribution fitted to the samples of instantaneous peak floods and a parameter homogeneous to a decay time computed from observed data.Different regionalisation methods were applied for estimating these three QdF parameters at ungauged locations. Regionalisation methods rely on the concept of transferring hydrological information from a site of measurement to ungauged sites. However these approaches require simultaneous records to avoid that the map is spoiled by temporal variability rather than display truly spatial patterns. Regional empirical formulas were derived but the constraints discussed above lead to discard 30% of the dataset.Time-REferenced data Kriging method (TREK) has been developed to overcome this issue. This alogrithm was developped in order to account the temporal support over which the variable of interest has been calculated, in addition to its spatial support. This approach aims at reducing the loss of data caused by the selection of a common reference period of records required to build a reliable dataset. The performances of each method have been assessed by cross-validation and a combination of best features is finally selected to map the frequent flow features over France.

Funding

100% French ministry for agriculture.

Cite the thesis

Delphine Porcheron. Caractérisation des régimes de crues fréquentes en France - un regard géostatistique. Environnement et Société. Université Grenoble Alpes, 2018. Français. ⟨NNT : 2018GREAU032⟩. ⟨tel-01968715⟩

Access manuscript on HAL thèses