3.2. Hydrologic classification for non-perennial Mediterranean rivers and streams
The optimized SOM with minor quantization and topographic errors included a grid of 3 x 8 nodes. Fig. 3 shows maps gradually distributed according to the five selected hydrological indices, thus the gauging stations with similar characteristics are located into the same node and remain close to other neighbourhood nodes in a two-dimensional heatmap. As a consequence of applying the cluster classification, the maps show two boundaries that divide the maps into three zones, grouping the stations into three clusters (I, II and III). For example, gauging stations with high values of number of days per year without flow are shown in yellow and located in the lower part of the map, while low values are represented in blue in the upper part of the map. The same occurs with the annual percentage of months without flow and coefficient of variation of Julian date of annual start of zero flow. In contrast, gauging station with low values of the mean daily annual flow are located in the upper area of the map.
(Here Fig. 3)
With regard to cluster classification, Fig. 4 represents the groups obtained by training a SOM with the five selected variables whereas table 2 shows the main characteristics of each cluster. Cluster I is composed of gauging stations with high values of number of days per year without flow (231.45 ± 97.07 days/year, see Fig. 4 and Table 2), annual percentage of months without flow (76.87 ± 23.52%), CV of Julian date of annual start of zero flow (2.23 ± 0.92), and annual rise rate (3.69 ± 6.19 l/day). It also included rivers with low values of mean daily annual flow (0.53 ± 0.70 l/day). Cluster II is an intermediate group between the other two clusters. It displayed mean values between 144.14 ± 54.40 days/year for the number of days per year without flow and 47.32 ± 15.23% for the annual percentage of months without flow. It also exhibited values of 0.72 ± 0.65 l/day for mean daily annual flow, 0.63 ± 0.27 for CV of Julian date of the annual start of zero flow, and 2.00 ± 1.98 l/day to annual rise rate. In contrast, cluster III included rivers with low values of number of days per year without flow (24.57 ± 25.68 days/year), annual percentage of months without flow (11.25 ± 10.09%), Julian date of annual zero flow (0.25 ± 0.17), and annual rise rate (1.35 ± 1.72 l/day), but with high values of mean daily annual flow (2.06 ± 3.80 l/day).
(Here Table 2)
(Here Fig. 4)