Heatmaps for the comparative assessment of the forecasting methods according to the median values of the mNSE metric and the condition stated on Table 9.
Comparison of stochastic and machine learning methods for multi-step ahead forecasting of hydrological processes: Supplementary material Selected figures for the qualitative comparison of the forecasting methods
Georgia A Papacharalampous, Hristos Tyralis, Demetris Koutsoyiannis
SE_1a
SE_1b
SE_2a
SE_2b
Figure S1. Heatmaps for the comparative assessment of the forecasting methods on the level of each simulation experiment according to the median values of the metrics and the conditions listed on Table 9 (part 1). The simulation experiments are named according to Table S1. The darker the color the better the forecasts. The missing values of the metrics Pr, r2 και KGE for the Naive and SES forecasting methods are not taken into consideration during the comparative assessment and are imprinted with white color. We also apply clustering analysis on the forecasting methods based on their performance. 1
SE_3a
SE_3b
SE_4a
SE_4b
Figure S2. Heatmaps for the comparative assessment of the forecasting methods on the level of each simulation experiment according to the median values of the metrics and the conditions listed on Table 9 (part 2). The simulation experiments are named according to Table S1. The darker the color the better the forecasts. The missing values of the metrics Pr, r2 και KGE for the Naive and SES forecasting methods are not taken into consideration during the comparative assessment and are imprinted with white color. We also apply clustering analysis on the forecasting methods based on their performance.
2
SE_5a
SE_5b
SE_6a
SE_6b
Figure S3. Heatmaps for the comparative assessment of the forecasting methods on the level of each simulation experiment according to the median values of the metrics and the conditions listed on Table 9 (part 3). The simulation experiments are named according to Table S1. The darker the color the better the forecasts. The missing values of the metrics Pr, r2 και KGE for the Naive and SES forecasting methods are not taken into consideration during the comparative assessment and are imprinted with white color. We also apply clustering analysis on the forecasting methods based on their performance. 3
SE_7a
SE_7b
SE_8a
SE_8b
Figure S4. Heatmaps for the comparative assessment of the forecasting methods on the level of each simulation experiment according to the median values of the metrics and the conditions listed on Table 9 (part 4). The simulation experiments are named according to Table S1. The darker the color the better the forecasts. The missing values of the metrics Pr, r2 και KGE for the Naive and SES forecasting methods are not taken into consideration during the comparative assessment and are imprinted with white color. We also apply clustering analysis on the forecasting methods based on their performance. 4
SE_9a
SE_9b
SE_10a
SE_10b
Figure S5. Heatmaps for the comparative assessment of the forecasting methods on the level of each simulation experiment according to the median values of the metrics and the conditions listed on Table 9 (part 5). The simulation experiments are named according to Table S1. The darker the color the better the forecasts. The missing values of the metrics Pr, r2 και KGE for the Naive and SES forecasting methods are not taken into consideration during the comparative assessment and are imprinted with white color. We also apply clustering analysis on the forecasting methods based on their performance.
5
SE_11a
SE_11b
SE_12a
SE_12b
Figure S6. Heatmaps for the comparative assessment of the forecasting methods on the level of each simulation experiment according to the median values of the metrics and the conditions listed on Table 9 (part 6). The simulation experiments are named according to Table S1. The darker the color the better the forecasts. The missing values of the metrics Pr, r2 και KGE for the Naive and SES forecasting methods are not taken into consideration during the comparative assessment and are imprinted with white color. We also apply clustering analysis on the forecasting methods based on their performance.
6
Figure S7. Heatmaps for the comparative assessment of the forecasting methods according to the median values of the MAE metric and the condition stated on Table 9. The simulation experiments are named according to Table S1. The darker the color the better the forecasts. We also apply clustering analysis on the forecasting methods based on their performance.
7
Figure S8. Heatmaps for the comparative assessment of the forecasting methods according to the median values of the MAPE metric and the condition stated on Table 9. The simulation experiments are named according to Table S1. The darker the color the better the forecasts. We also apply clustering analysis on the forecasting methods based on their performance.
8
Figure S9. Heatmaps for the comparative assessment of the forecasting methods according to the median values of the RMSE metric and the condition stated on Table 9. The simulation experiments are named according to Table S1. The darker the color the better the forecasts. We also apply clustering analysis on the forecasting methods based on their performance.
9
Figure S10. Heatmaps for the comparative assessment of the forecasting methods according to the median values of the NSE metric and the condition stated on Table 9. The simulation experiments are named according to Table S1. The darker the color the better the forecasts. We also apply clustering analysis on the forecasting methods based on their performance.
10
Figure S11. Heatmaps for the comparative assessment of the forecasting methods according to the median values of the mNSE metric and the condition stated on Table 9. The simulation experiments are named according to Table S1. The darker the color the better the forecasts. We also apply clustering analysis on the forecasting methods based on their performance.
11
Figure S12. Heatmaps for the comparative assessment of the forecasting methods according to the median values of the rNSE metric and the condition stated on Table 9. The simulation experiments are named according to Table S1. The darker the color the better the forecasts. We also apply clustering analysis on the forecasting methods based on their performance.
12
Figure S13. Heatmaps for the comparative assessment of the forecasting methods according to the median values of the cp metric and the condition stated on Table 9. The simulation experiments are named according to Table S1. The darker the color the better the forecasts. We also apply clustering analysis on the forecasting methods based on their performance.
13
Figure S14. Heatmaps for the comparative assessment of the forecasting methods according to the median values of the ME metric and the condition stated on Table 9. The simulation experiments are named according to Table S1. The darker the color the better the forecasts. We also apply clustering analysis on the forecasting methods based on their performance.
14
Figure S15. Heatmaps for the comparative assessment of the forecasting methods according to the median values of the MPE metric and the condition stated on Table 9. The simulation experiments are named according to Table S1. The darker the color the better the forecasts. We also apply clustering analysis on the forecasting methods based on their performance.
15
Figure S16. Heatmaps for the comparative assessment of the forecasting methods according to the median values of the PBIAS metric and the condition stated on Table 9. The simulation experiments are named according to Table S1. The darker the color the better the forecasts. We also apply clustering analysis on the forecasting methods based on their performance.
16
Figure S17. Heatmaps for the comparative assessment of the forecasting methods according to the median values of the VE metric and the condition stated on Table 9. The simulation experiments are named according to Table S1. The darker the color the better the forecasts. We also apply clustering analysis on the forecasting methods based on their performance.
17
Figure S18. Heatmaps for the comparative assessment of the forecasting methods according to the median values of the rSD metric and the condition stated on Table 9. The simulation experiments are named according to Table S1. The darker the color the better the forecasts. We also apply clustering analysis on the forecasting methods based on their performance.
18
Figure S19. Heatmaps for the comparative assessment of the forecasting methods according to the median values of the Pr metric and the condition stated on Table 9. The simulation experiments are named according to Table S1. The darker the color the better the forecasts. We also apply clustering analysis on the forecasting methods based on their performance. We note that the Pr metric cannot be calculated for the Naive and SES forecasting methods.
19
Figure S20. Heatmaps for the comparative assessment of the forecasting methods according to the median values of the r2 metric and the condition stated on Table 9. The simulation experiments are named according to Table S1. The darker the color the better the forecasts. We also apply clustering analysis on the forecasting methods based on their performance. We note that the r2 metric cannot be calculated for the Naive and SES forecasting methods.
20
Figure S21. Heatmaps for the comparative assessment of the forecasting methods according to the median values of the d metric and the condition stated on Table 9. The simulation experiments are named according to Table S1. The darker the color the better the forecasts. We also apply clustering analysis on the forecasting methods based on their performance.
21
Figure S22. Heatmaps for the comparative assessment of the forecasting methods according to the median values of the md metric and the condition stated on Table 9. The simulation experiments are named according to Table S1. The darker the color the better the forecasts. We also apply clustering analysis on the forecasting methods based on their performance.
22
Figure S23. Heatmaps for the comparative assessment of the forecasting methods according to the median values of the rd metric and the condition stated on Table 9. The simulation experiments are named according to Table S1. The darker the color the better the forecasts. We also apply clustering analysis on the forecasting methods based on their performance.
23
Figure S24. Heatmaps for the comparative assessment of the forecasting methods according to the median values of the KGE metric and the condition stated on Table 9. The simulation experiments are named according to Table S1. The darker the color the better the forecasts. We also apply clustering analysis on the forecasting methods based on their performance. We note that the KGE metric cannot be calculated for the Naive and SES forecasting methods.
