Proposal for a new customization process for a data-based water quality index using a random forest approach

As the water quality index (WQI) represents water quality, it is crucial to customize the WQI for a specific purpose. In this study, to better represent water quality data using WQI, a random forest (RF) approach was used to derive the parameter weight and calculate the WQI according to the watershed and its use. Eight parameters (water temperature, dissolved oxygen, pH, electrical conductivity, suspended solids, total nitrogen, total phosphorus, and total organic carbon) were evaluated using a total of 220,103 data points collected from 900 monitoring sites throughout South Korea between 2011 and 2020. The estimation of parameter weights, key elements in developing the WQI model, was performed through the variable importance estimation method that can be derived from the RF model. The parameter weights were derived based on various spatiotemporal datasets, and it was confirmed that the spatiotemporal differences in weights according to data characteristics represented the regional and seasonal water quality characteristics. Consequently, a customized WQI representing water quality characteristics could be calculated using data-based weights, and it is expected that a data-based customized WQI could be developed to better match the previous WQI to the purpose and target source. [Display omitted] •A new customization of water quality index (WQI) was proposed.•A random forest algorithm was used to develop the customization.•Water quality spatiotemporal characteristics can be evaluated by parameter weight.•Customized WQI effectively reflects spatiotemporal characteristics of water quality.•The customized WQI can be adjusted to reflect the actual water quality grade.