Analysis Error

No matter how fast computers become, their ability to forecast the weather will always be limited by the accuracy of the instruments used to observe the weather. Conversely, even with perfect instruments, the accuracy of a computer forecast is constrained by the ability to use the observations to construct an initial condition for the model with information at every grid point. The process of taking observations and constructing a set of data at regularly-spaced grid points is called "objective analysis" or "data assimilation".

Instruments always have some inherent inaccuracy. For rawinsondes (instruments attached to weather balloons), the temperature observations are usually accurate to within a few tenths of a degree Celsius, the pressures are accurate to within about a millibar, relative humidity is accurate to about 5%, and wind is accurate to a few meters per second. Because heights are computed from temperatures and pressures by integrating the hydrostatic equation, temperature errors cause errors in heights which get bigger at higher altitudes.

Even if an observation is accurate, it may not be "representative". Rawinsonde stations are typically more than a hundred miles apart, so each observation must represent conditions in the atmosphere over a large area. Suppose there are isolated thunderstorms, and the weather balloon ascends through a thunderstorm. The observations will indicate 100% relative humidity and variable winds, and the computer will have no way of knowing that the air is not saturated with water vapor for 100 miles in all directions. On the other hand, if the balloon ascends in between clouds, the sounding will be relatively dry throughout the troposphere, and the thunderstorm will remain undetected. Either way, the observation will lead to an incorrect analysis even though the observation itself might be perfectly accurate.

Because of these inherent inaccuracies, the model analyses avoid incorporating unusual data from single stations. This minimizes the impact of erroneous data, but also means that many smaller-scale features in the atmosphere are drastically smoothed by the analysis. Things like shallow, intense cold fronts are often mis-analyzed and therefore mis-forecast by the computer models.

There are other sources of data besides rawinsonde observations. The primary source is satellite observations of radiation emitted by the atmosphere at different temperatures. Most of this other data is not readily accessible by the forecaster, so it is impossible to compare the analysis to these observations. Satellite observations tend to be less useful than rawinsonde observations at the same place, but the virtue of satellites is that they can take measurements across the globe at fairly high horizontal resolution.

Another source of data are observations recorded by commercial aircraft and transmitted back to the ground. In what way do you think aircraft data is most beneficial to weather forecasting, based on what you know?