@article{fdi:010070270,
  title = {{P}artitioning into hazard subregions for regional peaks-over-threshold modeling of heavy precipitation},
  author = {{C}arreau, {J}ulie and {N}aveau, {P}. and {N}eppel, {L}.},
  editor = {},
  language = {{ENG}},
  abstract = {{T}he {F}rench {M}editerranean is subject to intense precipitation events occurring mostly in autumn. {T}hese can potentially cause flash floods, the main natural danger in the area. {T}he distribution of these events follows specific spatial patterns, i.e., some sites are more likely to be affected than others. {T}he peaks-over-threshold approach consists in modeling extremes, such as heavy precipitation, by the generalized {P}areto ({GP}) distribution. {T}he shape parameter of the {GP} controls the probability of extreme events and can be related to the hazard level of a given site. {W}hen interpolating across a region, the shape parameter should reproduce the observed spatial patterns of the probability of heavy precipitation. {H}owever, the shape parameter estimators have high uncertainty which might hide the underlying spatial variability. {A}s a compromise, we choose to let the shape parameter vary in a moderate fashion. {M}ore precisely, we assume that the region of interest can be partitioned into subregions with constant hazard level. {W}e formalize the model as a conditional mixture of {GP} distributions. {W}e develop a two-step inference strategy based on probability weighted moments and put forward a cross-validation procedure to select the number of subregions. {A} synthetic data study reveals that the inference strategy is consistent and not very sensitive to the selected number of subregions. {A}n application on daily precipitation data from the {F}rench {M}editerranean shows that the conditional mixture of {GP}s outperforms two interpolation approaches (with constant or smoothly varying shape parameter).},
  keywords = {{ZONE} {MEDITERRANEENNE} ; generalized {P}areto distribution ; probability weighted moment ; spatial interpolation ; regional frequency analysis ; {F}rench {M}editerranean ; {FRANCE}},
  booktitle = {},
  journal = {{W}ater {R}esources {R}esearch},
  volume = {53},
  numero = {5},
  pages = {4407--4426},
  ISSN = {0043-1397},
  year = {2017},
  DOI = {10.1002/2017wr020758},
  URL = {https://www.documentation.ird.fr/hor/fdi:010070270},
}