Size-resolved particle-decay measurements in Rayleigh–Bénard turbulence within the Pi Convection-Cloud Chamber

This dataset underpins the paper <i>“Observed deviation from Stokes’ Law in the dry deposition of heavy particles in Rayleigh–Bénard turbulence.”</i> It contains every raw time-series measurement and all derived decay constants used to test whether small, heavy particles settle in buoyancy-driven turbulence according to Stokes' Law.<b>Scientific context</b>Dry deposition governs how long aerosols remain airborne, influencing air-quality, weather and climate. Theory predicts that, for very low-inertia particles (St ≪ 1), the settling and deposition velocity scales with the square of the particle diameter (Stokes settling), even in turbulent flows. We performed laboratory tests of this assumption in anisotropic, buoyancy-driven turbulence generated by Rayleigh–Bénard (RB) convection. Contrary to Stokesian expectations, we find an approximately linear diameter–dependence.<b>Experimental apparatus</b><b>Facility:</b> 3.14 m³ Pi Convection-Cloud Chamber at Michigan Technological University.<b>Flow:</b> Dry RB turbulence driven by ΔT = 10 K and 20 K (Ra ≈ 1–2 × 10⁹). The large-scale circulation period (~60 s) is our characteristic mixing time τₘ.<b>Particles:</b>DEHS oil droplets, 1–10 µm, ρ = 912 kg m⁻³, generated continuously with a Palas MAG-3000.Solid, hollow, and ultra-hollow glass microspheres, 1–38 µm, ρ = 600–2500 kg m⁻³, injected in pulses with a compressed-air “air-cannon.”Resulting Kolmogorov-scale Stokes numbers 3 × 10⁻⁵ – 0.1.<b>Optical particle counters (OPCs):</b>Palas WELAS 2000 (up to 128 bins, 0.2–100 µm, 5 L min⁻¹)Alphasense OPC-N3 (24 bins, 0.35–40 µm, 0.28 L min⁻¹)<br>Counters were mounted at top and bottom ports; most data sampled at 1 Hz, a subset at ~1 min cadence.<b>Trials:</b> 68 experiments (65 usable), July 2023 and May/June 2025, yielding 1,322 fitted decay constants across 83 size–density classes.<b>Methodology &amp; processing workflow</b>Inject particles.Record number concentration decay for each size bin for 30–120 min.Store raw data in a pandas MultiIndex DataFrame with levels<br><code>particle-counter type [welas/opc] → location (top/bottom) → date → trial</code><br>(see README for indexing examples).Apply centered rolling averages and censor start/end segments that deviate from log-linearity.Perform ordinary-least-squares fits to ln C(t) to obtain decay time-constant τ and regression diagnostics (r-value, points-used, etc.).Aggregate τ values by size-density class and compare against Stokes theory; all figure generation is scripted in <b>plots.ipynb</b>.Reproducibility notes<b>Reproducibility notes</b>Running <code>plots.ipynb</code> end-to-end in Google Colab reproduces all figures in ~5 minutes on a free GPU/CPU runtime. All graphical operations use publicly available libraries; no proprietary software is required.<b>File</b><b>Format</b><b>Contents</b><b>Typical loading command</b><b>particle_decay_data.pkl</b>PickleRaw concentration time-series for every trial (MultiIndex DataFrame, units # cm⁻³)<code>pd.read_pickle()</code><b>metadata.pkl</b>PickleTrial-level, bin-by-bin metadata and all fitted parameters (1,322 rows)<code>pd.read_pickle()</code><b>plots.ipynb</b>Jupyter NotebookRecreates every figure from the paper (Colab-ready)Open in Colab or Jupyter; requires <i>pandas ≥ 2.2, numpy ≥ 1.26, scipy ≥ 1.13, matplotlib ≥ 3.9</i><b>README.pdf</b>PDFDetailed file schema, example queries, software dependencies, and Colab setup instructions—<br>