The benefits of correlated observation errors for small scales
Abstract
In operational data assimilation, observation errors are generally assumed to be uncorrelated, though some observations, such as satellite data, have correlated errors. We show that, if observation‐error correlations are correctly accounted for, an observing instrument with spatially correlated errors is better able to resolve small scales than an instrument with the same error variance and uncorrelated errors. We explore the disadvantages of falsely assuming uncorrelated observation errors, investigating two methods of compensating for such mis‐specification by either observation‐error inflation or data thinning. We identify scenarios in which correctly specifying the covariance reduces small‐scale error by over 99%.
Document Details
- Document Type
- Pub Defense Publication
- Publication Date
- Jul 28, 2015
- Source ID
- 10.1002/qj.2582
Entities
People
- Craig H Bishop
- Sabrina Rainwater
- William F. Campbell
Organizations
- National Academies of Sciences, Engineering, and Medicine
- Office of Naval Research
- United States Naval Research Laboratory