Physics-Constrained Probabilistic Tomography of 0–2 km Eddy Dissipation Rate Fields from Heterogeneous Ground-Based Atmospheric Observations

Low-altitude aviation requires spatially resolved turbulence information, but routine observing networks do not directly measure dense three-dimensional eddy dissipation rate (EDR) fields. We formulate EDR retrieval as physics-constrained probabilistic tomography of EDR (PCT-EDR) and construct an author-curated private multi-source low-altitude observation dataset designated SORA2025 to support the retrieval and benchmark evaluation. The observation operator partitions Doppler spectral-width variance into turbulent, beam, within-volume shear, hydrometeor, instrument, and residual components, then converts the corrected turbulent contribution into posterior distributions of log10ϵ and EDR on a 0–2 km grid. Wind-profiling and S/X-band Doppler radars supply vertical and horizontal constraints, while microwave radiometers and automatic weather stations provide stability and near-surface context. Twelve cases from four campaign dates characterize the retrieved posterior fields. A separate frozen processed benchmark contains 240 tower-sonic and unmanned aerial vehicle (UAV) windows from four later dates. Recalculation from the supplied processed package gives a root mean square error (RMSE) of 0.238 in log10ϵ, a Spearman correlation of 0.812, and an area under the receiver operating characteristic curve (AUC) of 0.913 for the stored prefit-calibrated product at EDR>0.10m2/3s−1. The nominal 90% interval covers 84.2% of the processed targets, indicating mild under-dispersion.