How reliable would be a hydrological model, trained/verified with gridded reanalysis datasets (e.g NCEP/NCAR, ECMWF) instead of gauge based data, specially in terms of future streamflow projections. Thanks
Any model regardless of how much it was calibrated or validated will not be 100% accurate, but an appropriately calibrated/validated model can provide useful information.
Skilful and reliable precipitation data are essential for seasonal hydrologic forecasting and generation of hydrological data.Although output from dynamic downscaling methods is used for hydrological application, the existence of systematic errors indynamically downscaled data adversely affects the skill of hydrologic forecasting. This study evaluates the precipitation dataderived by dynamically downscaling the global atmospheric reanalysis data by propagating them through three hydrologicalmodels. Hydrological models are calibrated for many watersheds located across any area that is minimallyaffected by human intervention.
If you had observed dataset to calibrate your model, why do you need a model output?
I guess you don't and you rely on a globally validated dataset. This validated model output may not give you local scale specific accurate results. Therefore, the error from the historical phase (calibration) will propagate to the future projections. The result could be misleading. Please, use the observed dataset. A model remains a model. It is not accurate in its entirety.