Original paper
Localized variational blending for nowcasting purposes.
Atencia, Aitor; Kann, Alexander; Wang, Yong; Meier, Florian

Meteorologische Zeitschrift Vol. 29 No. 3 (2020), p. 247 - 261
50 references
published: Oct 16, 2020
published online: Apr 21, 2020
manuscript accepted: Mar 4, 2020
manuscript revision received: Feb 6, 2020
manuscript revision requested: Jan 15, 2014
manuscript received: Oct 4, 2019
Open Access (paper may be downloaded free of charge)
Abstract
Mesoscale models' improvement of recent years (like spin-up reduction, assimilation techniques, and assimilation of new observation) and increased computational resources justify a rapid update cycle (1 hour). Despite all these improvements, precipitation forecasts provided by these models are not able to beat the observation-based Lagrangian extrapolation nowcasting for the first forecast steps. In this paper, these two forecasting sources are merged by a new blending technique, more complex than the regular global weights. It takes advantage of a variational technique commonly used in building the analysis in data assimilation cycles. This methodology allows to keep the spatial correlation of the errors and to merge the forecast with locally different weights. The results show an improvement over the original sources of forecast in terms of deterministic, dichotomic and spatial scores.
Keywords
Variational • Blending • Precipitation • Seamless