Barrier-layer temperature inversion dataset and analysis code for the Bay of Bengal (Argo, GLORYS12V1, ERA5; 2004-2025)

Derived data, analysis code and figure files for a study of barrier-layer temperature inversions in the Bay of Bengal, built from 32,943 Argo profiles collocated with GLORYS12V1 and ERA5 over January 2004 to December 2025. The archive supports the companion paper "Three-dimensional modulation of barrier-layer temperature inversions in the Bay of Bengal: the dominant role of barrier-layer thickness over seasonal and regional forcing". Its central result is that barrier-layer thickness, rather than season or region, sets whether the water column inverts: inversion frequency rises from 3.0% in the thin regime (BLT below 25 m) to 79.5% in the thick regime (BLT above 63 m), the transition is continuous, and it steepens in two places, near 25 m and near 63 m. Bulk Richardson numbers from ERA5 winds have medians of 100 to 750, that is 400 to 3,000 times the critical value of 0.25, so shear instability does not create these inversions; salinity stratification does. Contents - data/argo_master_v2_samecaliber.parquet: the master table, 32,943 profiles by 60 fields, the source of every number in the paper. DATA_DICTIONARY.md describes each field. - data/Fig01 to data/Fig08 and data/FigS1 to data/FigS3: figure-level data subsets, each with a self-contained plotting script, a README and the rendered figure. - code/processing/: nine scripts that build the master table from Argo profiles, GLORYS12V1 and ERA5. - data/_shared/bathy_BoB.nc: 15 arc-second GEBCO_2026 bathymetry subset (2-26 N, 75-102 E), the map base of Figures 1, 3, and 4. - All 11 figures as 300 dpi PNG and vector PDF, stored next to the data and code that produce them. Reproducing the figures The record holds a single file, BoB_BLT_v1.0.0.zip. Extract it and the directory tree described above is what you get, unchanged. Every plotting script reads the file sitting in its own folder and writes its output there, with no absolute paths and no cross-folder dependencies. Running "python fig02.py" inside data/Fig02/ regenerates Figure 2 exactly. The package was tested with Python 3.13 on Windows 11; exact library versions are in requirements.txt. All 11 scripts were re-run in a clean copy of the archive and their output matched the distributed figures pixel for pixel. Note for reuse: two Delta-T definitions The archive deliberately keeps two definitions of the mixed-layer to subsurface temperature difference in separate columns, because they are not interchangeable. delta_T_argo follows the original Argo convention, with the subsurface reference 2 to 5 m below the mixed-layer base. delta_T_argo_v2 and delta_T_glorys_v2 use the paired MLD + 10 m convention applied identically to GLORYS and to Argo. Mixing the two reproduces numbers that appear nowhere in the paper; the inversion rate alone differs substantially between them, and delta_T_argo_v2 is undefined for 6,539 profiles. Section 2 of README.md and data/README.md state which columns belong to which definition and why the two cannot be unified. Provenance and licensing No third-party raw data are redistributed. Argo (doi:10.17882/42182), GLORYS12V1 (doi:10.48670/moi-00021) and ERA5 (doi:10.24381/cds.f17050d7) are cited by DOI in the paper and downloaded directly from their providers. The only gridded product included is the GEBCO_2026 bathymetry subset, which is in the public domain (doi:10.5285/4f68d5c7-45eb-f999-e063-7086abc036fa) and must not be used for navigation. Data are released under CC BY 4.0, code under MIT.