GloFAS version 4 forecast skill (for details on the modelling aspects of v4 please see: GloFAS v4.0) is evaluated using the Continuous Ranked Probability Skill Score (CRPSS) for ECMWF-ENS 20-year reforecasts against both persistence and climatology benchmark forecasts with respect to GloFAS-ERA5 river discharge reanalysis (also known as forced simulations (sfo)) as proxy observations at n=4544 GloFAS diagnostic river points. Further details on the evaluation methodology can be found here: GloFAS forecast skill.
Ensemble forecast skill with lead time
Overall, GloFAS version 4 is skilful at the global scale for the majority of catchments across all lead times analysed (Figure 1). The median CRPSS against persistence decays exponentially as a function of lead time reaching a minimum of 0.6 by day 21. The CRPSS at day 1 is 0.96 (interquartile range of 0.90 to 0.99), day 3 = 0.84 (0.74, 0.92), day 5 = 0.76 (0.66, 0.85), day 10 = 0.65 (0.57, 0.74). The CRPSS against climatology begins higher than persistence and decays continuously towards day 30. At day 15 is 0.36 (0.20, 0.58), day 20 = 0.24 (0.09, 0.45), day 25 = 0.15 (0.02, 0.35), day 30 = 0.09 (0, 0.27).
Figure 1: Skill of GloFAS v4 with global median Continuous Ranked Probability Skill Score (CRPSS) for reforecasts against persistence (blue line) and climatology (red line) benchmarks from 1- to 30-day lead times with respect to GloFAS-ERA5 river discharge reanalysis at 4544 diagnostic river points. The interquartile range of CRPSS values at each lead time are shown by semi-transparent bands.
Spatial distribution of skill
Short to medium lead time (river discharge persistence as benchmark)
At short lead times (1 and 3 days), GloFAS is skilful compared to the persistence benchmark forecast in all catchments (Figure 2). Regions with the lowest skill against persistence include the Amazon basin, central North America and North Africa. By day 10, the strength of skill has decreased, but 99 % of catchments remain skilful (i.e. CRPSS > 0). Regions with the highest skill (CRPSS >= 0.8) include South America, especially the Amazon basin, the US, southern Africa, central Asia, and eastern Australia. By day 10, the strength of skill has decreased, but 99 % of catchments remain skilful (i.e. CRPSS > 0).
Figure 2: Skill of GloFAS v4 with global median Continuous Ranked Probability Skill Score (CRPSS) for reforecasts against the persistence benchmark for short to medium lead times (1- (a), 3- (b), 5- (c), and 10-days (d)) with respect to GloFAS-ERA5 river discharge reanalysis at 4544 diagnostic river points. Optimum value of CRPSS is 1. Blue (red) dots show catchments with positive (negative) skill.
Extended lead time (river discharge climatology as benchmark)
For extended lead times shown in Figure 3, GloFAS is skilful compared to the climatology benchmark forecast in 97 % of catchments by day 15, reducing to 93 %, 87 % and 81 % by lead times 20-, 25- and 30-days, respectively. The regions of lowest skill against climatology at the extended-range lead time include Northwest Canada and Alaska, North and central Africa, and the Northwest coast of South America.
Figure 3: Skill of GloFAS v4 with global median Continuous Ranked Probability Skill Score (CRPSS) for reforecasts against the climatology benchmark for extended lead times (15- (a), 20- (b), 25- (c), and 30-days (d)) with respect to GloFAS-ERA5 river discharge reanalysis at 4544 diagnostic river points. Optimum value of CRPSS is 1. Blue (red) dots show catchments with positive (negative) skill.