Functional networks have become a standard tool for the analysis of complex systems, allowing the unveiling of their internal connectivity structure while only requiring the observation of the system’s constituent dynamics. To obtain reliable results, one (often overlooked) prerequisite involves the stationarity of an analyzed time series, without which spurious functional connections may emerge. Here, we show how ordinal patterns and metrics derived from them can be used to assess the effectiveness of detrending methods. We apply this approach to data representing the evolution of delays in major European and US airports, and to synthetic versions of the same, obtaining operational conclusions about how these propagate in the two systems.
Four historical Pisano power-law equations for daily dry-weather solids deposition are audited using transparent numerical and analytical diagnostics. EQ1 is an endogenous surrogate: its slope descriptors depend on prior modeled deposition, whereas EQ4 can be evaluated from standalone network-scale inputs. An independent-uniform exploration of historical input ranges, rather than real-world occurrence probabilities, uses a common Latin hypercube sample of 200,000 combinations and gives median deposited-solids masses of 34.9, 168.0, 164.5, and 159.2 kg day−1 for EQ1–EQ4. The paired EQ1/EQ4 median log-ratio is , so EQ1 is the lower formulation. However, 19.7% of these unconstrained baseline combinations violate slope-descriptor ordering. Additional designs enforce ordering, a field-informed length-to-area envelope, and, separately, linked exponential slope quantiles. EQ1 remains distinct in these constrained tests, but its EQ4 log-ratio changes to , , and , respectively. High global rank correlations do not imply tail interchangeability: baseline EQ1/EQ4 and EQ2/EQ4 top-decile overlaps are 21.6% and 76.2%. Elasticities are invariant model properties, whereas variance shares depend on bounds and dependence: narrowing the length interval makes the EQ1 slope descriptor outweigh length. Under log-uniform marginals, output medians nearly coincide; retaining the four original flow levels also changes the induced distributions. All reported distributions and sensitivity rankings are conditional on these explicitly assumed designs. These findings support a conditional mathematical comparison of a model-dependent surrogate with standalone screening equations; they do not establish physical dominance, predictive equivalence, or design suitability.