conditions and longer-term climate change.
Weather forecasting and climate simulation are both
founded on solving large systems of partial differential
equations that describe the flows in the atmosphere.
Figure 3 shows the primary equations. Equation (1)
is the Navier-Stokes equation that describes the flow
of incompressible fluids. Equation ( 2) accounts for the
conservation of momentum in a particular atmospheric
“box.” Equation ( 3) handles atmospheric heating, and
finally Equation ( 4) is the meteorological form of the
equation of state for an ideal gas.
Next step: Divide up the atmosphere into grids, use
height as the vertical dimension, and model the conditions
in each cube. Since each weather event must be at least
three grid points in size to be recorded, a smaller grid
gives greater accuracy.
Different resolutions suit different applications.
The UM global model uses a 40-km grid, so it can’t
account for events such as small depressions or large
thunderstorms. On the other hand, over the UK the
UKV model uses 1.5-km resolution and seventy vertical
height levels, and can reportedly model individual
The UM dates back to 1990. Over the years, the model
has undergone continual refinement. Originally written
in Fortran 77, the current UM uses Fortran 90; the next
major revision will be written in Fortran 2003, the first
version to support object-oriented programming and
interoperability with the popular C language.
Many other countries use derivatives of the UK model,
but not the US: The NOAA uses the Global Forecast
System (GFS). The system received a major upgrade in
2016 with the selection of a new dynamic core.
Making a more accurate prediction of extreme weather
events and gaining a better understanding of the
consequences of climate change require bigger computers,
and Met Offices worldwide are upgrading their systems.
For example, their new Cray XC40 will allow the
Australian BOM weather model to move from a 4-km to
a 1.5-km grid.
Unlike many supercomputers on the TOP500 list, the
UK machine doesn’t rely on graphical processing units
(GPUs) from Nvidia or others for its processing power.
One reason is the difficulty of porting the existing
code to these new massively parallel platforms; doing so
requires a different approach to low-level programming,
a different approach for different hardware platforms,
and research on the best optimization strategies. UK
researchers are working to identify which current
algorithms don’t work very well at high levels of
parallelism, as well as those that are functional but must
be implemented differently.
Despite the great increase in computing power, many
important physical processes like tropical convection
and the effect of clouds are still not fully accounted for
in even the newest numerical models. This situation
will continue for the foreseeable future and is a major
source of forecasting errors. Including these small
effects by adding a random element to the model to
account for uncertainty, stochastic modeling is another
area of active research.
As a final note, the British care about their weather
so much that they have an expression, defined as “the
act of getting a prediction catastrophically wrong:”
A “Michael Fish moment.” It’s named after the
unfortunate TV weatherman who, in October 1987, told
BBC viewers not to worry about rumors of a hurricane
a few hours before the worst storm in three centuries
killed nineteen people. They even played a clip of the
moment during the London Summer Olympics opening
ceremony in 2012.
You can’t make this stuff up. ECN
Figure 3: The fundamental equations for modeling the weather. (Source: The Guardian)