When we chose the criteria that should be used to automatically assess the quality of the MT curves, we tried to make the subjective assessment of the handler operator coincide with the result of the automatic evaluation. As a result, we stopped at three criteria:
1) Estimate for confidence intervals of the impedance modulus,
2) Estimation by dispersion ratio,
3) Estimation by variance (variation) of intermediate realizations.
Criteria calculation:
Criterion 1: for each frequency the value (var|Z| / |Z|) is calculated for the mean curve of the impedance amplitude, where var|Z| + |Z| = |Z|max. The assessment is then taken as a geometric mean of these values.
If the loaded data contains not only an averaged curve, but also intermediate implementations, then before calculating the estimates, the average curve is recalculated from them using the following algorithm. To calculate the average curve at each frequency, a set of amplitude, phase, and residual values is formed from individual realizations. From each such set, a subset of values between the first and third quartiles (the 25th and 75th percentiles, respectively) is selected from the values of the impedance modulus. This subset is called "auto-edited". Averaging the subset yields the value of the average record.
Criterion 2: the mean curve of the impedance amplitude is interpolated with a spline, which is then transformed into a phase curve phs' via the corresponding dispersion relation (Weidelt relation). For each frequency within the middle part of the measured period range (the 1/3 of a frequency axis decade is omitted at both left and right ends of the curve) the value |phs — phs'| mod Pi is calculated. The estimation is then taken as an arithmetic mean of these values.
Criterion 3: the phase valuation val is picked for each frequency and impedance estimation value. The standard deviation for each set is calculated as sqrt (sum ((val — avg)^2)/n), where avg — is the arithmetic mean of all phase values, and n — is the number of phase values at a given frequency. The estimation is then taken as a geometric mean of these values. The algorithm employs all available values regardless of whether they were used for the mean value calculation or not.
Data quality in a field site is assessed on a five-grade basis and represented by means of the following color scale:
5 — excellent quality
4 — good quality
3 — satisfactory quality
2 — unsatisfactory quality
1 — bad quality
Each of three criteria is assessed according to the obtained values as follows:
Criteria 1: "1" > 1.0 ⩾ "2" > 0.2 ⩾ "3" > 0.1 ⩾ "4" > 0.05 ⩾ "5" > 0
Criteria 2: "1" > 1.0 ⩾ "2" > 0.2 ⩾ "3" > 0.1 ⩾ "4" > 0.05 ⩾ "5" > 0
Criteria 3: "1" > 1.0 ⩾ "2" > 0.2 ⩾ "3" > 0.1 ⩾ "4" > 0.05 ⩾ "5" > 0
Each or the criteria is calculated independently for the principal components Zxy and Zyx and the minor value between these two calculations is then used for the criteria assessment.
Summative estimation represents an average value of the criteria above, weighted in a certain manner:
e = (e1 + 2 * e2 + 2 * e3) / 5
If any of the necessary estimations is missing, it is assigned the value 5.
If the "apply period constraints" mode is active, then the assessment is performed only for the chosen frequency range.
If the "frequency range penalty" mode is active, then the summative estimation is multiplied by the overlapping coefficient between the data range [0;1] (for the active point) and the chosen frequency range.