Quantifying and Classifying Non‐Stationary Turbulence Under Sub‐Mesoscale Disturbances

Abstract Non‐stationary turbulence induced by sub‐mesoscale disturbances can substantially bias flux estimates, yet is largely overlooked in existing evaluation frameworks. Using turbulence measurements over the heterogeneous Loess Plateau, we found that classical stationarity tests classify >60% of records containing clear submesoscale disturbances as stationary, highlighting critical limitations over complex terrain. To address this limitation, a separation method for Sub‐Mesoscale and Turbulent motions (SMT) is applied to synthetic non‐stationary time series, accurately reconstructing prescribed sub‐mesoscale structures and demonstrating high fidelity. Then, we develop multi‐variable non‐stationarity indices based on the SMT‐derived components. Controlled sensitivity experiments with synthetic data demonstrate that these indices track sub‐mesoscale intensity clearly and monotonically, providing a more objective diagnostic than traditional metrics. Anchored by thresholds derived from these experiments, we propose a nine‐grade classification framework. This framework offers a robust diagnostic tool for flux evaluation over complex surfaces, with implications for weather forecasting, carbon cycle assessments, and climate model evaluation.