View in EDS

Calibration of TIGGE ensemble precipitation forecasts using Bayesian model averaging for a semi-arid river basin.

Saved in:

Bibliographic Details
Title:	Calibration of TIGGE ensemble precipitation forecasts using Bayesian model averaging for a semi-arid river basin.
Authors:	Yadav, Rashmi¹ (AUTHOR) rashmiy276@gmail.com, Yadav, Sanjaykumar M.¹ (AUTHOR)
Source:	Acta Geophysica. Apr2026, Vol. 74 Issue 2, p1-13. 13p.
Abstract:	The ensemble precipitation forecasts exhibit uncertainty due to the perturbations in the initial condition. A forecast is considered accurate if the predicted field and the true state of the system agree substantially. Raw ensemble precipitation forecasts frequently include systematic biases and spread deficits, as well as coarse spatial resolutions. It makes them unsuitable for driving hydrological models for ensemble streamflow forecasts. To eliminate biases and fix dispersion errors in raw ensemble precipitation forecasts, statistical postprocessing is required. The Bayesian model averaging (BMA) is employed grid-wise for the postprocessing of the short-range ensemble precipitation forecasts by the European Centre for Medium-Range Weather Forecasts (ECMWF), The International Grand Global Ensemble (TIGGE) for the Vishwamitri River catchment. In BMA, the component distributions are mixtures of gamma distributions and point masses at zero instead of the Gaussian distribution. The average Brier score (BS) of raw and postprocessed ensemble precipitation showed a significant improvement after the postprocessing of 1–3-day lead time. The verification of the postprocessed ensemble shows the reduction of the area under the curve (AUC) of receiver operator characteristics (ROC) plots with the increase of the lead time. Thus, the postprocessing method BMA calibrates the ensemble precipitation forecasts with the observed precipitation and reduces the systematic biases. The calibrated forecasts will be useful in generating the flood forecasts at the lead time of up to 3 days. [ABSTRACT FROM AUTHOR]
Database:	Energy & Power Source

Description
Abstract:	The ensemble precipitation forecasts exhibit uncertainty due to the perturbations in the initial condition. A forecast is considered accurate if the predicted field and the true state of the system agree substantially. Raw ensemble precipitation forecasts frequently include systematic biases and spread deficits, as well as coarse spatial resolutions. It makes them unsuitable for driving hydrological models for ensemble streamflow forecasts. To eliminate biases and fix dispersion errors in raw ensemble precipitation forecasts, statistical postprocessing is required. The Bayesian model averaging (BMA) is employed grid-wise for the postprocessing of the short-range ensemble precipitation forecasts by the European Centre for Medium-Range Weather Forecasts (ECMWF), The International Grand Global Ensemble (TIGGE) for the Vishwamitri River catchment. In BMA, the component distributions are mixtures of gamma distributions and point masses at zero instead of the Gaussian distribution. The average Brier score (BS) of raw and postprocessed ensemble precipitation showed a significant improvement after the postprocessing of 1–3-day lead time. The verification of the postprocessed ensemble shows the reduction of the area under the curve (AUC) of receiver operator characteristics (ROC) plots with the increase of the lead time. Thus, the postprocessing method BMA calibrates the ensemble precipitation forecasts with the observed precipitation and reduces the systematic biases. The calibrated forecasts will be useful in generating the flood forecasts at the lead time of up to 3 days. [ABSTRACT FROM AUTHOR]
ISSN:	18956572
DOI:	10.1007/s11600-026-01851-2