Adaptive Observation

The adaptive observation problem addresses how to route sensor-laden vehicles over the near future, feasibly (that is, respecting the dynamic constraints on the vehicle motion), in order to obtain the most valuable measurements which minimizes a relevant measure of uncertainty of the forecast. We have developed a novel method to solve this problem in a rigorous manner, quantifying the spread of the Ensemble members at the forecast time (in accordance with the the usual measures provided by the spread of the ensemble members at the forecast time in the EnKF approach).

We now take essentially (but not exactly, there are some devilish details) the dual of the HEnS approach, still using an Ensemble-based method to quantify uncertainty, but now looking at batches of time in the near future, instead of the recent past, to solve the ``adaptive observation'' problem.

Both algorithms incorporate an ensemble of adjoint analyses, and were somewhat difficult to derive and to implement in an efficient/extensible/readily-parallelizable manner. In hindsight, however, the pair of hybrid methods that were developed in this study is rather fundamental and elegant, and quite natural for this class of problems. I'd be happy to share with you further details if you are interested (actually, I'll be coming out to LANL to visit Chuck and his team later this summer, perhaps we could meet then if you are interested).