Adaptive Observations

Improved initial conditions result in improved forecasts. Improved observations, either through accuracy, number or distribution result in improved initial conditions. One way to improve observations is through using mobile observing platforms to make observations in areas that will maximally benefit resulting forecasts.

Deciding where to place the additional observation(s) is not as straight-forward as simply making observations where we believe errors in our estimates of the state of the atmosphere or ocean to be large. Perhaps those large errors evolve only over unimportant regions, or perhaps small errors in other regions will experience explosive error growth and ultimately have a much larger impact on forecast errors.

But even observing in regions where error growth is large does not guarantee success. Observations are blended with a first guess of the system state (a short-term forecast) through data assimilation. The approach is to move the first guess closer to the true state by using information contained in observations. Even if a targeted observation is made in an area of large uncertainty and/or a dynamically sensitive area, the data assimilation scheme could produce an estimate of the system state that results in a worse forecast than if no targeted observation were made at all! One must account for uncertainty magnitude, uncertainty growth and the details of one's data assimilation scheme when selecting targeted observations.

Faculty involved with adaptive observations are Kerry Emanuel and Sai Ravela