Factors contributing to data impact

Discussion

All contribute.

• The weather itself and time of day affect the number and location of aircraft observations and number, location, and type of usable satellite observations. For instance, IR radiances are not used in cloudy conditions, cloud drift winds don't exist in clear areas, and polar orbiter data swaths vary with time of day.
• Time of day also affects the number of radiosonde and surface observations.
• An observation type with more observations will have a larger number of small contributions yielding a bigger overall impact.
• However, if the observation error assigned by the analysis to that observation type is large, the contributions from each such observation will be very small. The observation error could be large because the instrument is not very accurate or precise or, more typically, because the scale of features sampled differs from model resolution (representativeness incompatibility).
• Even a dense network of perfect observations will have little impact where the first guess is good, because no change is needed. This means if RAOB data usually leads to a good forecast over the CONUS, other data types will show little impact because there isn't much to correct.
• The flow regime itself may affect the quality of the first guess and thus where the largest corrections are indicated.
• Assumptions about the first guess also are crucial.
  • The smaller the first guess error is assumed to be, the less influence all of the data types are allowed.
  • The forecast error covariance functions determine the spreading of observation influence, thereby affecting whether a particular observation type can impact the analysis at many different levels and at some distance from where its observations are concentrated.

The question of how much influence data has on the analysis is a complex issue drawing on many aspects of data assimilation. One more factor is explored in the next question.