*2.4. Analytical Methods*

The self-organizing map (SOM) is an unsupervised neural network as machine learning, and it is commonly known as a powerful tool for pattern recognition from complex data [38]. In ecological research, the SOM has recently been considered as a more appropriate multivariate analysis than other conventional statistical approaches [39]. The SOM is robust and suitable in providing comprehensive views on highly complex and multi-dimensional data through reducing the data dimension. The efficiency of SOMs in information extraction was demonstrated across different hierarchical levels of life from molecules to ecosystems [40]. Several studies showed that the SOM was robust enough to capture the nonlinear pattern of an ecosystem [39,41,42]. For these reasons, the SOM has been extensively applied to pattern recognition in various ecological domains including benthic macroinvertebrates [43,44], plankton communities [45–48], dissolved organic matters [49], fish assemblages [50,51], and biomanipulation assessment [52,53].

In the SOM analysis, a total of 33 variables were used including six physicochemical parameters, 27 dominant plankton populations (10 from the eDNA, and 17 from the CMI samples). In selecting the number of variables, we only included the plankton communities, of which abundance was greater than 5% of the total abundance. That is, otherwise, the variables would contain too many zero values which could lead to topological biases in the SOM visualization. The SOM size was determined by the rule of 5 sample size [54]. The SOM model was developed using MATLAB 6.1 (The MathWorks Inc., Natick, MA, USA) and the SOM Toolbox (Helsinki University of Technology, Espoo, Finland).

For assessment of richness and diversity, the former simply equals to number of species, while the latter is based on the Shannon–Weaver index (H = − *pi* ln *pi*, *pi* indicates a fraction of *i*th species) [55]. In calculating those biological indices, we excluded the taxonomical groups from eDNA samples, such as bacteria, mammals, reptiles, terrestrial plants, and amphibians, because the comparison between two different methods should be done at the same level of analytical resolution.
