Manual editing of the diagnostic profiles, Te(R), ne(R), Ti(R) and to remove spurious data points is one of the more important and delicate operations that must be performed when preparing a SNAP input file. There is some danger that a user will substitute eyeball judgement of expected profile smoothness for that of the diagnostic--and therefore delete features from the profile which are physically important.
A classic example of the limitations of SNAP with respect to relatively small-scale structures (5-8 cm), which might be obscured by excessive profile editing, are the flat spots observed in the Thomson scattering measurements of Te(R). Boris Grek has demonstrated that such flat (or inverted) spots exist, in the sense that they appear to be statistically significant. Additionally, they appear on both sides of the profile, at about the same q(r). So there is some reason to believe that there are regions in the plasma which, at least for short time periods, have extremely high values of local heat transport. Because these flat or inverted spots typically encompass only a handful of data points (3-5), a SNAPIN\ user might regard one or two of them as spurious, because they lie more than one error bar outside the local gradient, and therefore delete them from the profile. The crucial information is thereby lost.
However, it is important to recognize that special precautions would have to be taken in the SNAP analysis to ensure that it properly inferred high heat transport in regions where the Te(R) profile is flat. First, since the smallest physical structure that can be resolved in a smoothed profile is comparable to the smoothing radius (which is typically set to 8 cm), SNAP's smoothing would either have to be turned off or else set at a value small compared to the size of the flat spot. Second, the major radius profile Te(R) must be mapped to minor radius using either the outside portion or the inside portion of the profile, but not both (i.e., not both nor slice & stack mapping). (Mapping algorithms are described in section 3.7.6.) Otherwise, small differences in the measured Te at which the flat spots occurred on the inside versus outside parts of the profile (for slice & stack), or small errors in SNAP's calculation of the magnetic geometry (in errors in the calculation of the Shafranov shift), will effectively average the flat spot on the inside with a non-flat spot on the outside, or vice versa, thereby washing out the regions of zero gradient.
The proper way to handle this difficulty would be to analyze the profile twice with SNAP first using profile data from inside the magnetic axis, then using the profile data from outside the magnetic axis. By comparing the two profiles of Te(r) mapped onto minor radius, or the profiles of inferred , one can assess whether the inside and outside parts of the profile show regions of high heat transport on the same flux surface.
By smoothing the data and by using the data from both sides of the diagnostic profile, a user is effectively choosing to ignore small-scale structure.
On the other hand, there are genuine needs for editing the profiles. First, the smoothing algorithm used in SNAP and SNAPIN does not consider the profiles error bars when it performs smoothing. So, for example, an edge Thomson scattering data point which has Te = 15 keV and 20 keV error bars will contribute with equal weight as its neighbors with Te = 1 keV and 0.2 keV error bars. This will surely lead to nonsense. More generally, the error bars quoted for many diagnostics do not (and probably cannot) include all of the systematic errors which may affect the result, particularly in regions of borderline signal/noise.
Some examples of spurious data points:
