•
Temperature
•
Relative humidity
•
Wind direction (in three dimensions)
•
Wind speed
•
Barometric pressure
All of this data is fed into a computer, which is programmed with one objec-
tive: determining what each of these parameters will be, for each cubic kilome-
ter of air or water on the entire planet, five seconds (5 s) into the future. It’s not
difficult to imagine that a powerful computer, programmed according to all the
laws of physics, ought to be able to figure out a problem like that. This is, sim-
plified terms, what numerical weather prediction is all about.
SMALLER PACKETS, LONGER TIME
Of course, a forecast for the next 5 s is not going to be of much use to anybody.
But if we can forecast the weather 5 s in advance, we’ll get a new set of param-
eters, and we can apply the same computer program to the same parcels of air
again, getting another 5 s forecast. This can be repeated, or iterated, as many
times as we like, ultimately getting forecasts for the next minute, hour, day,
week, month, or year!
Obviously, as this numerical process is extended by repetition, the range of
error increases. Even if we can predict, with 99.99% certainty, the parameters for
every cubic kilometer of air and water for the next 5 s, we can’t expect to get the
same accuracy after 10 iterations (50 s), 100 iterations (500 s or 8.3 min), 1000
iterations (5000 s or 83 min), or 1,000,000 iterations (5,000,000 s or 58 days). The
error, however small, that occurs with each iteration is multiplied by itself again
and again. Over time, the resulting uncertainty can become so great as to render
the forecast meaningless. If you multiply 99.99% by itself 1000 times, you get
90.48%. If you multiply 99.99% by itself 10,000 times, you get 36.79%. If you
multiply 99.99% by itself 100,000 times, you get a number pretty close to 0%.
Yet, with each passing year, computers become more and more powerful, and
the programs they use become better and better. Observing apparatus becomes
more accurate, and can be placed in more and more locations. Imagine obtaining
data, by a combination of observation and interpolation, for each parcel of air and
water measuring just one cubic meter (1 m
3
) in size! Suppose the accuracy is
refined a thousandfold for every observation and every calculation! This might
result in a 10 s prediction certainty of 99.9999999%. When this number is mul-
tiplied by itself 1000 times, you get 99.9999%. If you multiply 99.9999999%
CHAPTER 3 Observation and Forecasting
101