Seasonal forecasts of runoff and river discharge in

Seasonal forecasts of runoff and river discharge in South America: skill and postprocessing Wouter Greuell and Ronald W. A. Hutjes Wageningen University and Research EGU, 2020, session HS 4. 4: Predictability, data assimilation, pre/postprocessing, and verification in deterministic and ensemble hydrological forecasting Friday, May 8, 10: 45 – 12: 30 Wouter Greuell WUR

THE FORECAST SYSTEM: TWO TYPES OF SIMULATIONS (HINDCASTS) Atmosphere and ocean model, SEAS 5 (ECMWF) Forcing: precipitation 2 m temperature, wind, humidity and incoming radiation Hydrological model VIC River discharge Wouter Greuell WUR

HINDCASTS - Forcing from SEAS 5 (ECMWF) hindcasts - Period 1981 -2015 - Start each month (so 12 x 35 starting dates) - 25 members (so 12 x 35 x 25 = 10500 runs) - Duration of runs: 7 months - Hydrological model: VIC - Resolution: 0. 5 degrees; domain South America - Initial conditions from reference simulation (see next slide) Wouter Greuell WUR

IN ADDITION TO HINDCASTS: REFERENCE SIMULATION Is the best possible simulation of the hydrological system during the period of the hindcasts: hydrological simulation forced with meteorological observations (WFDEI – precipitation data from GPCC) Two purposes of this simulation - create initial states for hindcasts (snow, soil moisture) - create pseudo-observations of hydrological variables for skill assessment Wouter Greuell WUR

Wouter Greuell WUR EVALUATION OF FULL HINDCASTS Summary of evaluation: fraction of all (12) target and 6 lead months (first in excluded) with significant skill Example of evaluation: target month November with initialization on September 1 (lead time: 2 months) Evaluation • Of runoff, i. e. streamflow in local rivers • Observations: pseudoobservations, i. e. runoff from reference simulation • Metric: correlation coefficient between observation and ensemble mean of the forecasts • Significant skill: skill exceeding the 95% confidence level Corr. Coeff. Fraction Threshold of significance

3 SETS OF HINDCASTS GOAL: ISOLATE SOURCES OF (DISCRIMINATION) SKILL Note: this scheme represents all hindcasts for with initialization on April 1. Scheme is similar for initialization in other calendar months! Name set Full Init Meteo Includes skill due to. . . Forcing and initial conditions Initial conditions only Forcing year i SEAS 5 initialized on Apr. 1, year i Each year the same selection (25 time series) of SEAS 5 initialized on Apr. 1, year i Initial conditions year i From reference simulation on Apr. 1, year i Mean (of all 35 years) of init. conditions on Apr. 1 Wouter Greuell WUR

SOURCES OF SKILL IN RUNOFF: Spatial distribution Full: all skill North Amazonia Init: initial conditions Meteo: forcing Precipitation Fraction North Chile Conclusions: discrimination skill is due to both forcing and initial conditions but overall forcing is the largest source forcing dominates in parts of the tropics and south-east South America; initial conditions dominate in large parts of Argentina there is more skill in runoff in the Meteo-hindcasts than in the precipitation forecasts Wouter Greuell WUR

EVALUATION NORTH AMAZONIA (outline of region previous slide) Threshold of sign. skill Mean annual cycle in reference simulation precipitation net precipitation runoff baseflow surface runoff evap. transp. Technical description of graphs: • Each curve represents the skill (corr. coeff. ) as a function of lead time of the hindcasts initialized in a specific calendar month • Hence, the x-coordinate of the dots corresponds to target month; thick dots represent the first lead month • Black lines connect all points for specific lead times Conclusions: • In this wet, tropical region, precipitation forecasts have significant skill at all lead times, except in target month April • Except for the first lead month, skill in runoff is almost exclusively due to skill in the forcing • Skill in runoff tends to be larger than skill in precipitation Wouter Greuell WUR

EVALUATION NORTH CHILE (outline of region slide before previous slide) Threshold of sign. skill Technical description of graphs: see previous slide Mean annual cycle in reference simulation precipitation net precipitation runoff baseflow surface runoff evap. transp. Conclusions: • In this dry, relatively cold region, precipitation forecasts have little skill but tends to be slightly higher in the wet season (JJA) • Skill in runoff is considerable • And almost exclusively due to initial conditions • Comparison of the Full with the Init hindcasts leads to two conclusions: 1) in JJA skill in the forcing contributes to skill in the Full hindcasts • 2) In the remaining, dry months the hindcasts of forcing (in Full) lead to a degradation of skill Wouter Greuell WUR

GENERAL CONCLUSION Earlier on we performed similar work on seasonal hydrological forecasts for Europe (see references below). In Europe precipitation forecasts possess hardly any discrimination skill beyond the first lead month. As a result, significant skill in forecasts of runoff and river discharge is mostly due to initial conditions, mainly those of soil moisture but with a smaller contribution by those of snow. The contrast with South America is large. In South America, the forecasted forcing (SEAS 5) of the hydrological model (precipitation is the most important) has significant skill in large parts of the continent, for many target months and even at long lead times. As a result, forcing is the largest source of skill at continental scale. At smaller scale, forcing dominates in parts of the tropics and south-east South America while initial conditions dominate in large parts of Argentina. References: Greuell, W. , Franssen, W. H. , Biemans, H. , & Hutjes, R. W. (2018). Seasonal streamflow forecasts for Europe–Part I: Hindcast verification with pseudo-and real observations. Hydrology and Earth System Sciences, 22(6), 34533472. Greuell, W. , Franssen, W. H. , & Hutjes, R. W. (2019). Seasonal streamflow forecasts for Europe-Part 2: Sources of. Greuell Wouter WUR skill. Hydrology and Earth System Sciences, 23(1), 371 -391.