Fig. 1
A comparison of the SOM representation for unit 15 (top left) with the pattern from averaging the correlation maps that are associated with unit 15 (middle left). The base points for all the correlation maps that contribute to the average pattern are shown by the green outline squares, the filled squares correspond to the base points of the correlation maps shown in the right column: light pink is the base point of the correlation map with the largest QE for unit 15, light blue is the base point for the correlation map with the minimum QE for unit 15 and yellow is the base point for the correlation map with the median QE for unit 15. Bottom left shows the difference between the SOM representation and the average correlation map pattern. Three of the correlation maps that are associated with this unit are shown on the right: top, the correlation map the greatest distance away from the unit pattern; middle, the correlation map the shortest distance from the unit; and bottom, the correlation map the median distance from the unit. The base point for each of the correlation maps on the right is indicated with a coloured square on the correlation map itself and on the average pattern from the correlation maps on the left. See Fig. 2 for the location of unit 15 on the SOM.
Fig. 2
SOM trained using NCEP data. The size of the black hexagon superimposed on each unit is proportional to the number of hits that unit received for the NCEP data. The colour of each cluster shows the frequency of occurrence of that cluster, the darker the blue, the more frequently it occurred in NCEP. The cluster numbers are written on each cluster. Unit outlined by a green dotted line (bottom left corner) is unit 15, which is shown in Fig. 1.
Fig. 3
Schematic showing which regions of the SOM correspond to which global teleconnection patterns. Where the teleconnection has no obvious established name the clusters are labelled with a very brief description of the pattern. Acronyms are listed in Appendix C.
Fig. 4
NAO cluster patterns. Top row shows NCEP patterns, with the base points for each pattern indicated by the yellow contour. The title details the cluster number and the frequency of occurrence for that pattern. The middle row shows the same, but for FORTE. The bottom row shows the FORTE patterns minus the NCEP pattern; the green contour shows where the NCEP and FORTE base points agree, the red contour shows where FORTE has include base points that NCEP does not and the yellow contour shows where NCEP has base points that FORTE does not reproduce. The title details the RMSE between the NCEP and FORTE cluster patterns and a ratio of the proportion of NCEP base points FORTE is in agreement with compared to the number of base points FORTE has in the wrong location as a proportion of the total number of NCEP base points for that pattern.
Fig. 5
Same as Fig. 4, except showing the PNA and NPO patterns.
Fig. 6
Same as Fig. 4, except showing the African Monsoon and ENSO.
Fig. 7
Same as top row of Fig. 4 except NCEP cluster patterns for the Indian Monsoon. FORTE provides no hits for these clusters, so there are no corresponding FORTE patterns.
Table 1
The cluster numbers of the teleconnections identified by the SOM, with the associated frequencies of occurrence for NCEP and FORTE
Fig. 8
SOM trained using NCEP data, but superimposed with FORTE hits. The size of the black hexagons is proportional to the number of hits FORTE had on that unit and are scaled the same on all figures (so hexagons of the same size on each figure represent the same number of hits.) The colour of each cluster shows the frequency of occurrence of that cluster, the darker the blue, the more frequently it occurred in FORTE. The cluster numbers are written on each cluster.
Fig. 9
NCEP trained SOM showing both NCEP and FORTE hits as black and grey hexagons, respectively. The colour of the cluster shows the percentage change in frequency between NCEP and FORTE. Red shows overestimation by FORTE, while blue shows underestimation. The cluster numbers are written on each cluster.
Fig. 10
NCEP trained SOM showing both NCEP and FORTE hits as black and grey hexagons, respectively. The colour of the cluster shows the RMSE between the NCEP and FORTE patterns for that cluster. The darker the blue the higher the RMSE. Clusters in pink have an undefined RMSE as there are no FORTE correlation maps associated with them to use to calculate an RMSE. The cluster numbers are written on each cluster.
Fig. A1
Schematic of the SOM process, using current data as an example. Actual data is shown in red, SOM data is shown in dark blue. Green arrows show the iterative loop. Each step is described in more detail in the text.
Fig. A2
SOM units for CM SOM (a) and RD SOM (b) for scenario 1. Unit number is shown in parenthesise above each unit along with percentage frequency of hits for that unit. Red is positive, blue is negative, white is zero.
Fig. A3
As for Fig. A2 using the results from scenario 2.
Fig. A4
First two EOFs for the correlation maps (a) and the raw data (b) under scenario 1. EOF number is shown in parenthesise above each unit along with percentage variance accounted for in each EOF. Red is positive, blue is negative, white is zero.