LAUSR

Hydrological classifications are aimed at simplifying spatiotemporal variability of flow regimes and, secondly, at supporting environmental flow management. The objective of this study was to perform classification of natural flow regimes in Poland using an inductive approach based on a set of hydrological metrics (HMs) and to develop a model for prediction of class membership based on a set of environmental variables (EVs). A set of 147 gauges with unmodified flow regimes was identified, and for each gauge, values of 73 HMs and 28 EVs were computed. Classification was performed using k-means and k-medoids techniques, based on 4 principal components explaining 73.4% of variability in HMs. Out of 7 distinguished classes, 4 (P1–P4) were spread across the Polish Plain, 1 (U5) was restricted to uplands, and 2 (M6 and M7) to mountains. The between-class differences in HMs and EVs were generally high, although classes P1 and P3 were not easily distinguishable. Mean predictive accuracy of the developed random forest model was 79%, which is high compared to other studies of this type. The lowest accuracies (0 and 50%) were achieved by 2 classes with the lowest counts. Variables representing diverse aspects: hydrography, climate, topography, and geology had the highest importance in the random forest model. Future research can benefit from the database of selected gauges with computed HMs, EVs, and assigned classes, freely available through a long-lasting data repository. With this study, the first step towards application of the ELOHA framework for environmental flow management at regional scale has been achieved.