A fuzzy classification of the hydrodynamic forcings of the Rhone River plume: An application in case of accidental release of radionuclides

Hydrological classification by clustering approach of time-integrated samples at the outlet of the Rhône River: Application to Δ14C-POC

“…The reduction of dimensionality provides a framework to visualize the Rhône hydrology. The essential of the information being on the first components (Husson et al, 2010), PCA is usually used as a pre-processing tool before clustering analysis (Delaval et al, 2021).…”

“…One approach is to run different simulations with different {C,m} pairs and to check the efficiency of clustering with a quality criteria (Ramze Rezaee et al, 1998). River hydrology related to clustering studies have a number of clusters falling in the interval [2,8] (Delaval et al, 2021;Javed et al, 2021;Ouyang et al, 2010). Our number of ∆ 14 C-POC observations being limited (n=54), we are expecting 8 clusters at most (Chaimontree et al, 2010) and we will set the test interval for the number of clusters to [2,8].…”

“…Unfortunately, such data was not available from all the tributaries and for the entire period of study (starting since 2011), while clustering using water discharge was already commonly used (Delaval et al, 2021). Finally, samples in the Cluster 5 seem to be a mix between the Cluster 4 (low-water level cluster) and the flood-related clusters (1, 2 and 3).…”

“…Multivariate analysis and clustering to characterize catchment seasonal hydrology could be linked with time-integrated sample. The Principal Component Analysis (PCA) coupled to Clustering Analysis (CA) has been widely used in Earth and Environmental sciences (e.g., Delaval et al, 2021, Raux et al, 2011. It is also an usual method for data mining when it comes to identify groups and classify individuals .…”

“…However, many algorithms exist and present different results and convergence speeds (Jain et al, 1999). Fuzzy c-mean algorithms are usually considered as reliable algorithms in environmental sciences (Delaval et al, 2021, Kim et al, 2011. Their success is due to their simple geometric criteria and their robustness without any prior information on the cluster structure such as possible overlap or sphericity (Kamel and Selim, 1994).…”