Abstract. Objective criteria for catchment classification are identified by the scientific community among the key research topics for improving the interpretation and representation of the spatiotemporal variability of streamflow. A promising approach to catchment classification makes use of unsupervised neural networks (Self Organising Maps, SOM's), which organise input data through non-linear techniques depending on the intrinsic similarity of the data themselves. Our study considers ~300 Italian catchments scattered nationwide, for which several descriptors of the streamflow regime and geomorphoclimatic characteristics are available. We qualitatively and quantitatively compare in the context of PUB (Prediction in Ungauged Basins) a reference classification, RC, with four alternative classifications, AC's. RC was identified by using indices of the streamflow regime as input to SOM, whereas AC's were identified on the basis of catchment descriptors that can be derived for ungauged basins. One AC directly adopts the available catchment descriptors as input to SOM. The remaining AC's are identified by applying SOM to two sets of derived variables obtained by applying Principal Component Analysis (PCA, second AC) and Canonical Correlation Analysis (CCA, third and fourth ACs) to the available catchment descriptors. First, we measure the similarity between each AC and RC. Second, we use AC's and RC to regionalize several streamflow indices and we compare AC's with RC in terms of accuracy of streamflow prediction. In particular, we perform an extensive cross-validation to quantify nationwide the accuracy of predictions in ungauged basins of mean annual runoff, mean annual flood, and flood quantiles associated with given exceedance probabilities. Results of the study show that CCA can significantly improve the effectiveness of SOM classifications for the PUB problem.
Data-driven catchment classification: application to the PUB problem
