Sources of variation in hydrological classifications: Time scale, flow series origin and classification procedure

- Hydrological classifications based on different data and procedures have been compared.
- Classifications of monthly data accounted for the critical hydrological information.
- Classifications of gauged series outperformed those based on simulated series.
- Data quality and origin are more critical than classification procedure.
- Expert-driven procedures led to the least effective classifications.

SummaryClassification of flow regimes in water management and hydroecological research has grown significantly in recent years. However, depending on available data and the procedures applied, there may be several credible classifications for a specific catchment. In this study, three inductive classifications derived from different initial flow data and one expert-driven classification were defined. The hydrological interpretation, statistical performance and spatial correspondence of these classifications were compared. Daily Gauged Classification (DC) was derived from daily flow data while Monthly Gauged Classification (MC) and Monthly Modeled Classification (MMC) were derived from monthly flow series, using gauged and modeled flow data, respectively. Expert-Driven Classification (EDC) was based on a Spanish nationwide hydrological classification, which is being used in the current River Basin Management Plans. The results showed that MC accounted for much of the critical hydrological information variability comprised within the DC. However, it also presented limitations regarding the inability to represent important hydroecological attributes, especially those related to droughts and high flow events. In addition, DC and MC presented an equivalent performance more than 60% of the time and obtained a mean ARI value of 0.4, indicating a similar classification structure. DC and MC outperformed MMC 100% and more than 50% of the times when they were compared by means of the classification strength and ANOVA, respectively. MMC also showed low correspondence with these classifications (ARI = 0.20). Thus, the use of modeled flow series should be limited to poorly gauged areas. Finally, the significantly reduced performance and the uneven distribution of classes found in EDC questions its application for different management objectives. This study shows that the selection of the most suitable approach according to the available data has significant implications for the classification uses. Therefore, caution is recommended, especially if classifications are to be use in a normative manner.