Premium
On generation of an assembly of images in ionospheric tomography
Author(s) -
Andreeva E. S.,
Franke S. J.,
Yeh K. C.,
Kunitsyn V. E.,
Nesterov I. A.
Publication year - 2001
Publication title -
radio science
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.371
H-Index - 84
eISSN - 1944-799X
pISSN - 0048-6604
DOI - 10.1029/2000rs002437
Subject(s) - root mean square , standard deviation , projection (relational algebra) , mean squared error , tomography , inversion (geology) , image (mathematics) , mathematics , computer science , phase (matter) , square root , algorithm , computer vision , remote sensing , statistics , geology , geometry , optics , physics , paleontology , quantum mechanics , structural basin
Radio tomography experiments have demonstrated the promising potential of applying tomographic methods in imaging various ionospheric structures. In actual implementation of image reconstructions one is faced with many choices, which include the following: whether to use the total phase, relative phase, or Doppler as the projection data, how to approximate the projection operator, what inversion algorithm to employ, and the choice of how to include the ancillary data and constraints on the constructed image. Each choice results in an image compatible with the given or measured projection data, yet each choice results in an image different from that of the others, with its own attendant artifacts and distortions. Collectively, the images produced by all the possible choices comprise an assembly of images. In this simulation study of one ionospheric model, 113 members of such an assembly are generated. All images look similar in gross features with a root‐mean‐square deviation not more than 29% from the mean. As expected, the largest deviation occurs near the region of highest gradients. By averaging all of the images in the assembly we show that the mean image is superior because of its smallest root‐mean‐square deviation from the true image. This conclusion, drawn on the simulation study of one model, may in fact have a general applicability, and we discuss why this may be so.