LDI Intellectual Property.
Not for secondary distribution or replication, in part or entirety.
DIGISONDE-4D
SYSTEM MANUAL
VERSION 1.2.11
5-22 SECTION 5 - SYSTEM SOFTWARE
ARTIST-5 Ionogram Qualifiers
5:42. In addition to the ACL value determined by ARTIST-5 by inspecting autoscaling process and its out-
come for anomalies, another level of ionogram qualification is provided by automatic classification of each
ionogram in terms of the level of the ionosphere/ionogram disturbance. A total of six ionogram categories is
defined for the error analysis by assigning the following ionogram qualifiers:
QC = Quiet ionospheric conditions, Confident ARTIST scaling ACL = 1
MC = Moderate Spread F conditions, Confident ARTIST scaling ACL = 1
HC = Heavy Spread F conditions, Confident ARTIST scaling ACL = 1
QL = Quiet ionospheric conditions, Low confidence of ARTIST scaling ACL = 0
ML = Moderate Spread F conditions, Low confidence of ARTIST scaling ACL = 0
HL = Heavy Spread F conditions, Low confidence of ARTIST scaling ACL = 0
5:43. For certain locations such as Jicamarca (Peru), Gakona (Alaska), or Campo Grande (Brazil), we ob-
served another class of ionograms for which it is impossible to derive any meaningful vertical electron density
profile from the ionogram because of extremely heavy spread F conditions. Such class received a separate qual-
ifier EH:
EH = Excessively Heavy Spread F conditions, no autoscaling results attempted
5:44. Using this system of ionogram qualifiers, it is possible to better reflect long-term statistical differences
in the ARTIST accuracy by isolating anomalous records that tend to distort the Gaussian error distribution into
separate categories for which the long-term accuracy is expected to be worse and likely non-Gaussian. By ex-
cluding the outliers from the main categories of the confidently scaled data, their distributions get closer to the
Gaussian.
Long-term Statistical Evaluation of Digisonde-4D Accuracy
5:45. The DPS error analysis focuses on the errors introduced by the automatic scaling and assumes correct
manual scaling. Clearly the instrumental errors in measuring the virtual heights h’ and the plasma density are
small (see below) compared to the errors introduced by the autoscaling. The long-term statistical accuracy of
DPS4D ionogram-derived data is evaluated by calculating error bars for the scaled characteristics and the re-
sulting error boundaries for the EDP. The error bars and boundaries describe the probability that the true value
lies within specified bounds placed around the reported automatically derived values. In this approach, the
probability is fixed at a particular level acceptable to the user application (e.g., 95%, 1σ, etc.), and the bounds
are then determined from statistical comparison of reported (autoscaled) values to the true (manually scaled)
values. Once the bounds are determined from the comparison set of ionograms, they are applied to the rest of
the data. Such method benefits from ionogram classification into subcategories in which data possess statisti-
cally different accuracy. In this case, ARTIST-5 software determines the ionogram subcategory (using the sys-
tem of ionogram qualifiers) and then applies the appropriate set of error measures to describe its accuracy.
Sensitivity Study of Ionogram-Derived Data Accuracy
5:46. A full scale study of the ARTIST data was conducted to determine ionogram subcategories that would
reveal different levels of accuracy. Close to 250,000 manually scaled Digisonde
®
ionograms were involved in
the study that tested dependence of the autoscaled errors on location, season, time of day, level of ionospheric
disturbance, and autoscaling confidence level. The study results demonstrated clear dependence of the accuracy
on the location of the sounder, level of ionospheric disturbance, and ACL. For the purpose of this study, the ac-
curacy was still evaluated for ACL=0 data, and because it indeed appears to be considerably worse, rejecting
ACL=0 data from assimilation makes sense. Alternatively, they still could be assimilated, but with large error
To Next Page