Clustering Flood Events from Water Quality Time-Series using Latent Dirichlet Allocation Model

摘要: To improve hydro-chemical modeling and forecasting, there is a need to better understand flood-induced variability in water chemistry the processes controlling it watersheds. In literature, assumptions are often made, for instance, that stream reacts differently rainfall events depending on season; however, methods verify such not well developed. Often, few floods studied at time chemicals used as tracers. Grouping similar from large multivariate datasets using principal component analysis clustering helps explain hydrological processes; these currently have some limits (definition of flood descriptors, linear assumption, instance). Most been context regionalization, focusing more mapping results than understanding processes. this study, we extracted patterns probabilistic Latent Dirichlet Allocation (LDA) model, its first use hydrology, our knowledge. The LDA method allows temporal be considered without having define explanatory factors beforehand or select representative floods. We analyzed dataset long-term observatory (Kervidy-Naizin, western France) containing data four solutes monitored daily 12 years: nitrate, chloride, dissolved organic carbon, sulfate. different were distributed by season. Each pattern can explained seasonal Hydro-meteorological parameters help leading patterns, which increases quality. Thus, appears useful analyzing datasets.