Spatiotemporal-SOM analysis code and data

The self-organizing map (SOM) is a neural network-based classification method with unsupervised learning (Kohonen, 1982). The excellent clustering ability of SOM is valuable because of its noise tolerance and nonlinearity characteristics. This repository provides several Matlab script files that utilize the SOM algorithm for spatiotemporal analysis.

The spatiotemporal SOM analysis is applied to explore the spatiotemporal variability in water quality in Hong Kong marine water areas. SOMs are applied at both spatial and temporal domains for the multivariate marine water quality observations, and principal component analysis (PCA) is used to cluster SOM neurons and component planes. The analysing results demonstrate that applying a SOM at a spatial domain allowed us to obtain spatial clusters of monitoring stations and to investigate relationships among the parameters, while applying a SOM at a temporal domain allowed us to obtain temporal variability in observational time series. Overall, the combined use of SOM and PCA provides a holistic view of the complex multivariate water quality data.

Spatiotemporal-SOM is developed by a candidate Ph.D. student Jiang Yu at College of Engineering, Peking University. The SOM Toolbox 2.0 developed by Vatanen et al. is employed to provide the SOM algorithm.