Data and code for "Frenetic Accessibility as a Cloud-Regime Transition Diagnostic"
Description
Reproducibility deposit for the manuscript "Frenetic Accessibility as a Cloud-Regime Transition Diagnostic." Contains the Python analysis and figure-generation code, intermediate .npz caches (coarse-grained regime-label time series and per-region transition summaries for four months), machine-readable toy-model output tables, and a frozen, checksummed manifest pinning the exact public ISCCP-Basic HGG satellite granules used (992 granules, 3.02 GB; the raw NetCDF is not included). The caches reproduce the toy-model results and the satellite cross-regime, obstruction, and overdispersion analyses without the multi-gigabyte download. The satellite input is the NOAA ISCCP-Basic H-Series Climate Data Record, obtained anonymously from the NOAA Open Data Dissemination S3 bucket noaa-cdr-cloud-properties-isccp-pds. See README.md for per-script documentation, region boxes, and download instructions; SHA256SUMS.txt lists checksums for every file.
Files
Steps to reproduce
Environment: Python 3 with numpy, scipy, matplotlib (netCDF4 only for the raw-data pipeline). Figure scripts set matplotlib pdf.fonttype=42. Unzip deposit_payload.zip in place so that code/, cache/, data_manifest/, and toy_outputs/ sit beside README.md. Optionally check integrity: sha256sum -c SHA256SUMS.txt. A. Reproduce from caches (no 3 GB download). The cache/*.npz files hold the regime-label time series and per-region summaries for the four analysed months, so the survey, obstruction, and overdispersion results run directly: - python3 code/verify_r7.py and python3 code/verify_r8.py — referee-round checks (WLS overdispersion, trend decomposition, small-count-bias bootstrap; spatiotemporal 5-day block bootstrap, leave-one-out, ratio/pair confidence intervals). Expected: spatial-only WLS chi-square ~ 155; a_BC slope ~ +0.27; bootstrap bias <= 0.011. - python3 code/obstruction_search.py — four-state {clear, low, mid, high} scan with the mid-excision contraction check. - python3 code/occ_residual.py — overdispersion via the spatiotemporal block bootstrap (5-day default). B. Toy model (constants only, no data). - python3 code/make_toy_csv.py toy_outputs/ regenerates every table in toy_outputs/ by importing the three toy modules (mep_alignment_decomp.py, channel_loss_demo.py, level25_loss.py); no numbers are hand-entered. Expected: Fisher weight B->C ~ 7.1%; obstruction gap ~ +0.561; stationary laws pi1 ~ (0.167, 0.186, 0.647) and pi2 ~ (0.044, 0.718, 0.239). - Figure generators: channel_loss_demo.py, level25_loss.py, fw_quasipotential.py, make_alignment_fig.py. C. Full pipeline from raw granules (optional). Download the exact 992 granules (3.02 GB; Jan and Jul of 2015 and 2016) pinned in data_manifest/ISCCP_granule_manifest.csv, with no AWS account needed: aws s3 cp --no-sign-request --recursive s3://noaa-cdr-cloud-properties-isccp-pds/data/isccp-basic/hgg/<YYYYMM>/ ./isccp/<YYYYMM>/ or: wget -i data_manifest/ISCCP_granule_urls.txt. Verify files against the md5_etag column. Then run python3 code/isccp_frenetic_pipeline.py (download -> regime labels -> transition diagnostics) for a chosen region/month; the eight region boxes are tabulated in README.md. This regenerates the caches used in section A. All region definitions, per-script descriptions, and full data provenance are given in README.md.