9  Performance indicators

The indicators calculated by the Comparator object are based on observed and simulated values. Since the Comparator object is able to compare variables of type flow, height, altitude, power and True/False, the descriptions below only refer to these 2 variables.

9.1 Nash coefficient

The Nash-Sutcliffe criteria (Nash and Sutcliffe 1970) is used to assess the predictive power of hydrological models (Ajami et al. 2004; Schaefli et al. 2005; Jordan 2007; Viviroli et al. 2009; García Hernández 2011). It is defined as presented in Equation 9.1.

\[ Nash = 1 - \frac{\sum_{t = t_{i}}^{t_{f}}{(X_{sim,t} - X_{ref,t})^{2}}}{\sum_{t = t_{i}}^{t_{f}}{(X_{ref,t} - {\overline{X}_{ref}})^{2}}} \tag{9.1}\]

with \(Nash\): Nash-Sutcliffe model efficiency coefficient [-]; \(X_{sim,t}\): simulated variable (discharge [L3/T] or height [L]) at time \(t\); \(X_{ref,t}\) : observed variable (discharge [L3/T] or height [L]) at time \(t\); \(\overline{X}_{ref}\): average observed variable (discharge [L3/T] or height [L]) for the considered period.

It varies from -∞ to 1, with 1 representing the best performance of the model and zero the same performance than assuming the average of all the observations at each time step.

9.2 Nash coefficient for logarithm values

The Nash-Sutcliffe coefficient for logarithm flow values (\(Nash-ln\)) is used to assess the hydrological models performance for low values (Krause, Boyle, and Bäse 2005; Nóbrega et al. 2011). It is defined as presented in Equation 9.2.

\[ \text{Nash-ln} = 1 - \frac{\sum_{t = t_{i}}^{t_{f}}{(ln(X_{sim,t}) - ln(X_{ref,t}))^{2}}}{\sum_{t = t_{i}}^{t_{f}}{(ln(X_{ref,t}) - ln({\overline{X}_{ref}}))^{2}}} \tag{9.2}\]

with \(Nash-ln\): Nash-Sutcliffe coefficient for log values [-].

It varies from -∞ to 1, with 1 representing the best performance of the model.

9.3 Pearson Correlation Coefficient

The Pearson correlation coefficient shows the covariability of the simulated and observed values without penalizing for bias (Aghakouchak and Habib 2010; Wang et al. 2011). It is defined as presented in Equation 9.3.

\[ Pearson = \frac{\sum_{t = t_{i}}^{t_{f}}{(X_{sim,t} - {\overline{X}_{sim}}) \cdot (X_{ref,t} - {\overline{X}_{ref}})}}{\sqrt{\sum_{t = t_{i}}^{t_{f}}{(X_{sim,t} - {\overline{X}_{sim}})^{2}} \cdot \sum_{t = t_{i}}^{t_{f}}{(X_{ref,t} - {\overline{X}_{ref}})^{2}}}} \tag{9.3}\]

with \(Pearson\): Pearson Correlation Coefficient [-]; \({\overline{X}_{sim}}\): average simulated variable (discharge [L3/T] or height [L]) for the considered period.

It varies from -1 to 1, with 1 representing the best performance of the model.

9.4 Kling-Gupta Efficiency

The Kling-Gupta efficiency (Gupta et al. 2009) provides an indicator which facilitates the global analysis based on different components (correlation, bias and variability) for hydrological modelling issues.

Kling, Fuchs, and Paulin (2012) proposed a revised version of this indicator, to ensure that the bias and variability ratios are not cross-correlated. This update is proposed as indicator in RS MINERVE (Equation 9.4):

\[ KGE' = 1 - \sqrt{(r - 1)^{2} + (\beta - 1)^{2} + (\gamma - 1)^{2}} \tag{9.4}\]

\[ \beta = \frac{\mu_{s}}{\mu_{o}} \tag{9.5}\]

\[ \gamma = \frac{{CV}_{s}}{{CV}_{o}} = \frac{\sigma_{s}/\mu_{s}}{\sigma_{o}/\mu_{o}} \tag{9.6}\]

with \(KGE'\): modified KGE-statistic [-]; \(r\): correlation coefficient between simulated and reference values [-]; \(\beta\): bias ratio [-]; \(\gamma\): variability ratio [-]; \(\mu\): mean discharge [L3/T]; \(CV\): coefficient of variation [-]; \(\sigma\): standard deviation of discharge [L3/T]; the indices \(s\) and \(o\) indicate respectively simulated and observed discharge values.

It varies from 0 to 1, with 1 representing the best performance.

9.5 Bias Score

The Bias Score (\(BS\)) is a symmetric estimation of the match between the average simulation and average observation (Wang et al. 2011). It is defined as presented in Equation 9.7.

\[ BS = 1 - \bigg( max \Big(\frac{{\overline{X}}_{sim}}{{\overline{X}}_{ref}};\frac{{\overline{X}}_{ref}}{{\overline{X}}_{sim}} \Big) - 1 \bigg)^2 \tag{9.7}\]

with \(BS\): Bias Score [-].

It varies from -∞ to 1, with 1 representing the best performance of the model.

9.6 Relative Root Mean Square Error

The Relative Root Mean Square Error (\(RRMSE\)) is defined as the RMSE normalized to the mean of the observed values (Feyen et al. 2000; El-Nasr et al. 2005; Heppner et al. 2006) and is presented in Equation 9.8.

\[ RRMSE = \frac{\sqrt{\frac{\sum_{t = t_{i}}^{t_{f}}{(X_{sim,t} - X_{ref,t})^{2}}}{n}}}{{\overline{X}}_{ref}} \tag{9.8}\]

with \(RRMSE\): relative \(RMSE\) [-]; \(n\): number of values [-].

It varies from 0 to +∞. The smaller \(RRMSE\), the better the model performance is.

9.7 Relative Volume Bias

The Relative Volume Bias (\(RVB\)), sometimes called differently, corresponds in this case to the relative error between the simulated and the observed volumes during the studied period (Ajami et al. 2004; Schaefli et al. 2005; Moriasi et al. 2007; Aghakouchak and Habib 2010) according to Equation 9.9. This indicator is envisaged for the comparison between observed and simulated discharges.

\[ RVB = \frac{\sum_{t = t_{i}}^{t_{f}}{(X_{sim,t} - X_{ref,t})}}{\sum_{t = t_{i}}^{t_{f}}{(X_{ref,t})}} \tag{9.9}\]

with \(RVB\): relative volume bias between forecast and observation for the considered period [-]; \(X\) usually corresponding to the discharge variable.

The \(RVB\) varies from -1 to +∞. An index near to zero indicates a good performance of the simulation. Negative values are returned when simulated variable is, in average, smaller than the average of the observed one (deficit model), while positive values mean the opposite (overage model).

9.8 Normalized Peak Error

The Normalized Peak Error (\(NPE\)) indicates the relative error between the simulated and the observed maximum values (Masmoudi and Habaieb 1993; Sun et al. 2000; Ajami et al. 2004; Gabellani et al. 2007). It is computed according to Equation 9.10 to Equation 9.12.

\[ NPE = \frac{S_{\max} - R_{\max}}{R_{\max}} \tag{9.10}\]

\[ S_{\max} = \overset{t_{f}}{\underset{t = t_{i}}{\vee}}Q_{sim,t} \tag{9.11}\]

\[ R_{\max} = \overset{t_{f}}{\underset{t = t_{i}}{\vee}}Q_{ref,t} \tag{9.12}\]

with \(NPE\): relative error between simulated and observed peak value [-]; \(S_{max}\): maximum simulated value (discharge [L3/T] or height [L]) for the studied period; \(R_{max}\) : maximum observed value (discharge [L3/T] or height [L]) for the studied period.

The NPE varies from -1 to +∞. Negative values are returned when maximum simulated value is below the observed one, while positive values mean the opposite. Values near to zero indicate a good performance of simulated peaks regarding observed ones.

Warning

The indicator is computed over the entire simulation period and the absolute maximum of the simulated and the observed peaks are considered! This indicator should therefore be used with care when simulating over long periods of time.

9.9 Peirce Skill Score

The Peirce Skill Score (\(PSS\)) indicates the performance of the model to reproduce the overrun of a threshold (Peirce 1884; Manzato 2007). Based on a contingency table definging the number of cases where the simulation and the obseration exceed or not the threshold, the \(PSS\) is computed according to Equation 9.13.

\[ PSS = \frac{ad - bc}{(a + c)(b + d)} \tag{9.13}\]

with \(a\): the number of cases when both simulation and observation exceed the threshold defined in the Comparator (event); \(b\): the number of cases when the simulation exceeds the threshold but not the obseration (false); \(c\): the number of cases when the observation exceeds the threshold but not the simulation (miss); \(d\): the number of cases when both simulation and obseration are below the threshold (no event).

Note

If the denominerator equals 0 (division by 0), a value of 0 is returned for the PSS.

9.10 Overall Accuracy

The Overall Accurary (\(OA\)) indicates the performance of the model to reproduce the overrun of a threshold (Parajka and Blöschl 2008). Based on the same contingency table as the Pierce Skill Score, the \(OA\) is computed according to Equation 9.14.

\[ OA = \frac{a + d}{a + b + \ c + d} \tag{9.14}\]

References

Aghakouchak, Amir, and Emad H. Habib. 2010. “Application of a Conceptual Hydrologic Model in Teaching Hydrologic Processes.” International Journal of Engineering Education 26: 963–73.
Ajami, Newsha K., Hoshin Gupta, Thorsten Wagener, and Soroosh Sorooshian. 2004. “Calibration of a Semi-Distributed Hydrologic Model for Streamflow Estimation Along a River System.” Journal of Hydrology 298 (1-4): 112–35. https://doi.org/10.1016/j.jhydrol.2004.03.033.
El-Nasr, Ahmed Abu, Jeffrey G. Arnold, Jan Feyen, and Jean Berlamont. 2005. “Modelling the Hydrology of a Catchment Using a Distributed and a Semi-Distributed Model.” Hydrological Processes 19 (3): 573–87. https://doi.org/10.1002/hyp.5610.
Feyen, L., R. Vázquez, K. Christiaens, O. Sels, and J. Feyen. 2000. “Application of a Distributed Physically-Based Hydrological Model to a Medium Size Catchment.” Hydrology and Earth System Sciences 4 (1): 47–63. https://doi.org/10.5194/hess-4-47-2000.
Gabellani, S., G. Boni, L. Ferraris, J. von Hardenberg, and A. Provenzale. 2007. “Propagation of Uncertainty from Rainfall to Runoff: A Case Study with a Stochastic Rainfall Generator.” Advances in Water Resources 30 (10): 2061–71. https://doi.org/10.1016/j.advwatres.2006.11.015.
García Hernández, Javier. 2011. “Flood Management in a Complex River Basin with a Real-Time Decision Support System Based on Hydrological Forecasts.” {PhD} {Thesis} {N}°5093, Lausanne, Switzerland: Ecole Polytechnique Fédérale de Lausanne, EPFL.
Gupta, Hoshin V., Harald Kling, Koray K. Yilmaz, and Guillermo F. Martinez. 2009. “Decomposition of the Mean Squared Error and NSE Performance Criteria: Implications for Improving Hydrological Modelling.” Journal of Hydrology 377 (1-2): 80–91. https://doi.org/10.1016/j.jhydrol.2009.08.003.
Heppner, Christopher S., Qihua Ran, Joel E. VanderKwaak, and Keith Loague. 2006. “Adding Sediment Transport to the Integrated Hydrology Model (InHM): Development and Testing.” Advances in Water Resources 29 (6): 930–43. https://doi.org/10.1016/j.advwatres.2005.08.003.
Jordan, F. 2007. “Modèle de Prévision Et de Gestion Des Crues - Optimisation Des Opérations Des Aménagements Hydroélectriques à Accumulation Pour La Réduction Des Débits de Crue.” {PhD} {Thesis} {N}°3711, Lausanne, Switzerland: Ecole Polytechnique Fédérale de Lausanne, EPFL.
Kling, Harald, Martin Fuchs, and Maria Paulin. 2012. “Runoff Conditions in the Upper Danube Basin Under an Ensemble of Climate Change Scenarios.” Journal of Hydrology 424-425 (March): 264–77. https://doi.org/10.1016/j.jhydrol.2012.01.011.
Krause, P., D. P. Boyle, and F. Bäse. 2005. “Comparison of Different Efficiency Criteria for Hydrological Model Assessment.” Advances in Geosciences 5 (December): 89–97. https://doi.org/10.5194/adgeo-5-89-2005.
Manzato, Agostino. 2007. “A Note On the Maximum Peirce Skill Score.” Weather and Forecasting 22 (5): 1148–54. https://doi.org/10.1175/WAF1041.1.
Masmoudi, M., and H. Habaieb. 1993. “The Performance of Some Real-Time Statistical Flood Forecasting Models Seen Through Multicriteria Analysis.” Water Resources Management 7 (1): 57–67. https://doi.org/10.1007/BF00872242.
Moriasi, D. N., J. G. Arnold, M. W. Van Liew, R. L. Bingner, R. D. Harmel, and T. L. Veith. 2007. “Model Evaluation Guidelines for Systematic Quantification of Accuracy in Watershed Simulations.” Transactions of the ASABE 50 (3): 885–900. https://doi.org/10.13031/2013.23153.
Nash, J. E., and J. V. Sutcliffe. 1970. “River Flow Forecasting Through Conceptual Models Part IA Discussion of Principles.” Journal of Hydrology 10 (3): 282–90. https://doi.org/10.1016/0022-1694(70)90255-6.
Nóbrega, M. T., W. Collischonn, C. E. M. Tucci, and A. R. Paz. 2011. “Uncertainty in Climate Change Impacts on Water Resources in the Rio Grande Basin, Brazil.” Hydrology and Earth System Sciences 15 (2): 585–95. https://doi.org/10.5194/hess-15-585-2011.
Parajka, J., and G. Blöschl. 2008. “Spatio-Temporal Combination of MODIS Images - Potential for Snow Cover Mapping: SPATIO-TEMPORAL COMBINATION OF MODIS IMAGES.” Water Resources Research 44 (3). https://doi.org/10.1029/2007WR006204.
Peirce, C. S. 1884. “The Numerical Measure of the Success of Predictions.” Science ns-4 (93): 453–54. https://doi.org/10.1126/science.ns-4.93.453.b.
Schaefli, B., B. Hingray, M. Niggli, and A. Musy. 2005. “A Conceptual Glacio-Hydrological Model for High Mountainous Catchments.” Hydrology and Earth System Sciences 9 (1/2): 95–109. https://doi.org/10.5194/hess-9-95-2005.
Sun, X., R. G. Mein, T. D. Keenan, and J. F. Elliott. 2000. “Flood Estimation Using Radar and Raingauge Data.” Journal of Hydrology 239 (1-4): 4–18. https://doi.org/10.1016/S0022-1694(00)00350-4.
Viviroli, Daniel, Heidi Mittelbach, Joachim Gurtz, and Rolf Weingartner. 2009. “Continuous Simulation for Flood Estimation in Ungauged Mesoscale Catchments of SwitzerlandPart II: Parameter Regionalisation and Flood Estimation Results.” Journal of Hydrology 377 (1-2): 208–25. https://doi.org/10.1016/j.jhydrol.2009.08.022.
Wang, Q. J., T. C. Pagano, S. L. Zhou, H. A. P. Hapuarachchi, L. Zhang, and D. E. Robertson. 2011. “Monthly Versus Daily Water Balance Models in Simulating Monthly Runoff.” Journal of Hydrology 404 (3-4): 166–75. https://doi.org/10.1016/j.jhydrol.2011.04.027.