BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Computer\, how likely is it that I need my coat tomorrow?:  How ne
 ural networks can be used for both probabilistic weather forecasting and p
 ost-processing of NWP models - Mariana Clare (European Centre for Medium-R
 ange Weather Forecasts (ECMWF))
DTSTART:20220921T133000Z
DTEND:20220921T135000Z
UID:TALK179084@talks.cam.ac.uk
DESCRIPTION:The success of machine learning techniques over the years\, an
 d in particular neural networks\, has opened up a new avenue of research f
 or weather forecasting. However neural networks suffer as decision-making 
 tools because they lack the ability to express uncertainty. Here we show h
 ow this problem can be alleviated by transforming continuous data to categ
 orical data. Specifically\, we use neural networks to easily generate prob
 abilistic data-driven forecasts of geopotential at the 500hPa level and th
 e temperature at the 850hPa level\, using the WeatherBench dataset (a proc
 essed version of the ERA5 reanalysis dataset regridded onto a coarse resol
 ution). Furthermore\, by using a combination of variable importance analys
 is and ensemble modelling\, we show that our data-driven neural network ap
 proach can achieve better results than both some more complex neural netwo
 rks and some simple NWP models. However\, our approach is not more accurat
 e than the existing operational ECMWF IFS model. Therefore\, in the second
  part of this talk\, we present ongoing work illustrating how neural netwo
 rks can be used for post-processing to improve predictions from NWP models
 . In particular\, we show how the relatively new technique of Bayesian Neu
 ral Networks may help to improve ensemble generation and uncertainty quant
 ification of NWP models.
LOCATION:No Room Required
END:VEVENT
END:VCALENDAR