GloGAS v5 was calibrated using 5379 in-situ discharge gauge stations with at least 365x4 daily measurements from 01/01/1980 to 31/12/2023. Furthermore, reservoir information including inflow, outflow, volume in-situ time series for 360 reserviors as well as climate, geometry, and use attributes for all the 1486 reservoirs included in the modelling domain were used to tune the parameters regulating reservoir release.
Model calibration required to estimate up to14 parameters for each catchment: this objective was achieved in four sequential steps. First, 2 reservoir parameters were calibrated using SCE-UA (where in-situ observed time series were available) and estimated using a random forest regressor estimator (for the remainder of the reservoirs included in GloFASv5 set-up). Second, NSGA-II within DEAP framework was used to optimize up to12 parameters regulating snow melt, water infiltration into the soil, surface water flow, groundwater flow, lakes dynamic of 5379 inter-catchments for which discharge data were available (gauged catchments), covering 51.6% of the semi-global domain. Third, a parameter learning regionalization approach was implemented to estimate up to 12 parameters for 10,916 ungauged catchments, covering 43.0% of the semi-global domain. A minimum drainage area of 500 km2 was used for the above explained steps of the calibration. Lastly, the parameters of the remainder 5.4% area (catchments with upstream area smaller than 500 km2) were estimated using nearest neighbour interpolation.
JDKGE, a revised version of KGE' proposed by Ficchì et al. (2026)1 was used as objective function. JDKGE has the purpose to optimize both low and high flows.
The combined calibration approach delivered 12 parameter maps with quasi-global extent and 2 tables indicating reservoir parameters.
This page summarises GloFAS v5 calibration hydrological skill. Specifically, calibration performance is evaluated using all the available in-situ discharge data for each calibrated station: duration and coverage
Overview
The hydrological performance of GloFAS v5 is expressed by the modified Kling-Gupta Efficiency (KGE', Gupta et al., 20092, Kling et al, 20123); KGE' entails three components: correlation, bias, variability (a detailed description is available in this page).
Albeit JDKGE was used as objective function for model calibration (as explained here), this page presents the outcomes in terms of KGE’ to allow a straightforward comparison with other models and with GloFASv4 evaluation pages. Evaluation of low flow performance is presented by showing the outcomes in terms of JSD, an estimate of the Jensen-Shannon Divergence, as indicated by Ficchì et al. (2026)1.
Figure 1 shows the empirical cumulative distribution function of KGE' values, as well as the KGE' histogram distribution for the 5379 calibration stations. KGE' values were computed using all the available observations for each calibration point.
The median KGE' is 0.716, with the calibrated parameters leading to higher accuracy than the mean flow benchmark for 99.61% of the gauged catchments (KGE’ > -0.41, Knoben et al., 20193).
Figure 1 – KGE' histogram (blue bars) distribution and empirical cumulative distribution function(red line) for all the 5379 calibration stations of GloFAS v5 and all the available data. The green line indicates the optimal performance.
Figure 2 presents the results of GloFAS v5 for the 5379 calibration points in terms of KGE' components: linear correlation between observations and simulations, bias, and a measure of the flow variability error (a detailed description is available in this page).
Median values of correlation, bias, and variability are 0.753, 0.991, and 0.961, respectively.
|
|
|
|---|
Figure 2 – KGE' components (correlation, bias, variability) histogram distribution (light blue bars) and empirical cumulative distribution function (red lines) for all the 5379 calibration stations of GloFAS v5 and all the available data. The green lines show the optimal performance.
Figure 3 shows the results of GloFAS v5 for the 5379 calibration points in terms of JSD. This evaluation metric allows assessing model capability to represent low flow conditions. While JSD optimal value is 0, the work of Ficchì et al. (2026)1 explains that values lower than 0.1 generally indicate adequate representation of low flows. In GloFASv5 calibration, approximately 80% of the 5379 calibration stations achieved JSD smaller than 0.1.
Figure 3 – JSD histogram distribution (pink bars) and empirical cumulative distribution function (magenta line) for all the 5379 calibration stations of GloFAS v5 and all the available data. The optimal value of JSD metric is 0 (green line); values lower than 0.1 (blue dashed line) indicate adequate low flow representation.
Spatial analysis
Figure 4 shows the spatial distribution of KGE' values for the calibration stations. KGE' values > 0.7 are shown in light blue and blue. KGE’ values < -0.41 (model performance lower than the mean flow benchmark) are shown in black.
KGE' is generally uniformly distributed across the domain, with higher performance (light blue and blue) in large parts of North and South America, Central Europe, and Asia. Calibrated catchments with high performances are also found in Africa and Oceania. The lowest performances (black) are often concentrated in arid areas, catchments with strongly regulated rivers, or areas affected by inaccuracies in the forcing dataset.
Figure 4 – Spatial distribution of the hydrological performance (KGE') of GloFAS v5 across the domain for the 5379 calibration stations (evaluated using all the available data).
A low score during evaluation of OS LISFLOOD model calibration is not necessarily an indicator of a decreased forecast performance of the global flood awareness system. GloFAS forecasts are compared to model derived thresholds (Thielen et al., 20094; Bartholmes et al., 20095), this comparison eliminates systematic bias. In some calibration stations, the systematic bias leads to an overall lower score in hydrological performance. Nevertheless, correlation is a desired quality in hydrological performance as it represents the timing of flood peaks. Given the mathematical structure of KGE', all stations where KGE'>=0.7 have correlation >=0.7 (Gupta et al., 20092). Conversely, some of the stations with KGE'< 0.7 can have correlation>= 0.7; but associated to larger bias or variability inaccuracies.
Figure 4 shows a combination of the spatial distribution of GloFAS KGE' and correlation. Stations with KGE'<0.7 and Correlation> 0.7 are highlighted in white. Out of 2471 stations with KGE'<0.7, 540 have Correlation > 0.7.
Figure 5 – Spatial distribution of the hydrological performance (KGE') of GloFAS v5 across the domain combined with correlation: stations with KGE'<0.7 and correlation>0.7 are highlighted in white.
Figures 6, 7, 8 present the spatial distribution of GloFAS v5 hydrological performance across the quasi-global domain in terms of KGE' components: correlation, mean bias and variability bias.
Figure 6 – Spatial distribution of correlation of GloFAS v5 at all 5379 calibration stations (evaluated using all the available data).
Figure 7 – Spatial distribution of bias of GloFAS v5 at all 5379 calibration stations (evaluated using all the available data).
Figure 8 – Spatial distribution of variability of GloFAS v5 at all 5379 calibration stations (evaluated using all the available data).
Figure 9 shows the spatial distribution of JSD performance metric: values lower than 0.1 indicate generally adequate performance in low flow modelling.
Figure 9 – Spatial distribution of JSD of GloFAS v5 at all 5379 calibration stations (evaluated using all the available data).
Comparison of GloFASv5 against GloFASv4: overview
This section presents a comparison between GloFASv5 and GloFASv4 performance for the representation of historical discharge time series.
First, the comparison was performed using all the 5379 calibrated stations in GloFASv5. Clearly, not all the stations used in GloFAS v5 calibration had been used for GloFASv4 calibration. Furthermore, even for stations used in both calibrations, the length and the coverage of discharge observation time series could be different. To enable a more comprehensive assessment, evaluation of both GloFAS v5 and GloFAS v4 made use of the entire observation period available to GloFAS v5.
Figure 10 allows comparing the empirical cumulative distribution functions of KGE', correlation, JSD: the red line represents GloFASv5 performance, the black line represents GloFASv4 performance, the green line the optimal target value.
GloFAS v5 median values of KGE' and Correlation values an improvement of +0.21 and +0.04 compared to the median values of GloFAS v4.
JSD metric was not included in GloFAS v4 calibration: its use in GloFAS v5 calibration led to an increase of +43% in the number of catchments having JSD < 0.1 indicating significantly improved performance in low flow modelling.
|
|
|
|---|
While the plots above highlight the overall comparison between GloFASv5 and GloFASv4, Figure 11 presents a similar comparison, but restricted to a subset of 1563 stations, having the same location in GloFAS v5 and GloFAS v4. When considering this subset of stations, GloFAS v5 median values of KGE' and Correlation show an improvement of +0.07 and +0.005 compared to the median values of GloFAS v4. As highlighted in the description of placeholder - GloFAS v5 calibration data, additional GloFAS v5 calibration points were often located upstream of pre-existing GloFAS v4 calibration points. While GloFAS v5 and GloFAS v4 performance were often similar at common points, especially when located on the major rivers, differences must be highlighted in the catchments hydrological behaviour. In GloFAS v5, the increased fragmentation of large basins allowed parameter calibration closer to the areas where runoff generation occurs, with more sound parameter values, and improved model performances in several points of the basins, rather than just the locations on the major rivers. Model performances in the latter locations were influenced by the reduced degree of freedom: the increased spatial fragmentation reduced the magnitude and occurrence of compensation effects (compared to GloFAS v4).
|
|
|---|
Figure 11 - Empirical cumulative distribution function of KGE' (left) and Correlation (right) for GloFAS v5 (red lines) and GloFAS v4 (black lines), evaluated using 1563 stations with common location in both the versions, and all the available observations.
Comparison of GloFASv5 against GloFASv4: spatial analysis
Figure 12 presents the difference between GloFAS v5 and GloFAS v4 KGE' values, for 5379 stations and all the available observations. Improvements are represented in green and yelllow, substantially similar values with +- 0.05 are in white, degradations are represented in pink and red.
Figure 12 - Spatial distribution of the difference KGE' GloFASv5 - KGE' GloFASv4
Figure 13 presents the difference between GloFAS v5 and GloFAS v4 Correlation values, for 5379 stations and all the available observations. Improvements are represented in green and yelllow, substantially similar values with +- 0.05 are in white, degradations are represented in pink and red.
Figure 13 - Spatial distribution of the difference Correlation GloFASv5 -Correlation GloFASv4
References
1 Ficchì, A., Bavera, D., Grimaldi, S., Moschini, F., Pistocchi, A., Russo, C., Salamon, P., and Toreti, A.: Improving low and high flow simulations at once: An enhanced metric for hydrological model calibration, EGUsphere [preprint], https://doi.org/10.5194/egusphere-2026-43, 2026.
2 Gupta, H. V., Kling, H., Yilmaz, K. K., & Martinez, G. F. (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://www.sciencedirect.com/science/article/pii/S0022169409004843?via%3Dihub
3 Kling, H., Fuchs, M., Paulin, M. (2012). Runoff conditions in the upper Danube basin under an ensemble of climate change scenarios. Journal of hydrology, 424-425, 264-277. https://doi.org/10.1016/j.jhydrol.2012.01.011