Please enable JavaScript.
Coggle requires JavaScript to display documents.
Weather forecasting: Numerical weather prediction (Storm Doris (2017)…
Weather forecasting: Numerical weather prediction
Storm Doris (2017)
Underwent explosive
cyclogenesis
Heavy snow to Scotland during rush hour caused disruption- closure of m80
Yellow/ amber warnings for wind, snow and rain
Strong winds/ gusts (gust of 94mph)
Numerical Weather prediction was very good- predicted future
Great storm of 1987
one in 200 year storm
First gale warnings on morning of storm
2235 winds of force 10 forecasted
assistance from military an option
Forecast has gained one day per decade
How NWP works
Analytical solution exists for trivial cases ono- NWP models, supercomputers
Challenges
Initialisation
Don't know initial state of atm for weather forecasting w/ high accuracy
Measurements have errors & limited coverage
Data assimilation
techniques generate analysis fields by combining
observations & short term predictions
from NWP models
Predictability
Limited predictability for chaotic systems- small differences in initial conditions -> large diffs in final results
Ensemble predictions
Atm is chaotic system- weather forecasts are sensitive to initial conditions
uncertanties in initial conditions -> large prediction errors
Idea: estimate uncertainties through computing an ensemble of predictions with slightly diff initial conditions
Spread allows inferring the robustness of forecast
Discretisation
Need to approximate atm with values on discrete grid-
1.5km grid-spacing in short range UK forecast
20km grid spacing for global forecast models, 100km for climate models
State of art models have >10^7 grid points
Vertical levels over idealised mountain follow orography near surface and become constant
Many small-scale processes can't be resolved & must be parametrised (radiation, cloud physics, convection, turbulence, evaporation, soil moisture...)
Computing
UK Met Office Unified model (UM)
MetOffice UK currently uses model chain that links global long-term simulations w/ 1.5km resolution forecasts over UK
Computer resources put limits on:
No. of ensemble members
The domain
The resolution
Met office runs 3 versions of UK at diff resolutions over dif domains
UKV: 36 hour forecast 8x/day
Euro4: 60 hour forecast twice/day
Global: 48 hour forecast twice/day
European Centre for Medium-Range Weather Forecasting (ECMWF) runs 15-day global forecasts
Physical laws form a system of coupled equations
Historical names
Vilhelm Bjerknes (1904)
Weather prediction as initial value problem based on physical equations
Lewis Richardson (1922)
numerical integration of equations of motion
Manual computation of pressure change over England based on equations: 145hPa in 6 hours
History
First 'supercomputer' for weather forecasting
First NWP with ENIAC
Highly simplified weather equations
24 hour forecast took 24 hours to compute
NWP now
Computational time: <1 sec
Reproduction with mobile phone (java script)
Despite exponential increases, c
omputer time is still limiting factor in operational NWP
Timeline towards modern NWP
1979: Establishment of ECMWF- now 32 member states
1992: First ensemble prediction system at ECMWF
1966: First usage of full physical equations
1995: Development of non-hydrostatic regional models
1958: first generation of initial conditions w/ objective analysis
1954/5: first operation NWP in US and swededn