# Data & content license The **software** in this repository (build pipeline, scripts, front-end code) is licensed under the **MIT License** — see [`LICENSE`](LICENSE). The **data, map output, and written documentation** that this project *produces* — the composite risk scores, the within-county percentile ranks, and the layers computed here (impervious surface, energy burden, land-surface temperature, WBGT) — are licensed under: **Creative Commons Attribution 4.0 International (CC BY 4.0)** You may share and adapt this material, including commercially, **as long as you give appropriate credit.** Suggested attribution: > San Diego Heat Risk Explorer — Southern California Extreme Heat Research Hub & > San Diego Regional Climate Collaborative. Licensed CC BY 4.0. ## Third-party source data retains its own terms This tool incorporates data from external sources. **CC BY 4.0 applies only to this project's own contributions, not to the underlying third-party data**, each of which keeps its original license. Most sources are U.S. federal public-domain works (Census/ACS, CDC PLACES & SVI, USGS Landsat, NOAA HRRR, NLCD, DOE LEAD), but several have their own conditions: | Source | Terms | Handling | |---|---|---| | [Healthy Places Index (HPI)](https://www.healthyplacesindex.org/terms-and-conditions) | **Non-commercial** + attribution; mark if modified | **Excluded from the "Download Data" export.** Shown on the map under HPI's terms with attribution; obtain the data from HPI directly. | | [CalHeatScore (CalEPA / OEHHA)](https://calheatscore.calepa.ca.gov/) | **No published license for the per-ZCTA effect estimates** | **Excluded from the export** (2026-07-27 audit). See below. | | [SANDAG / SanGIS](https://www.sangis.org/) | Raw parcel/land-base layers need a SanGIS sublicense | Only a derived per-tract green-space percentage is used; raw parcels are not redistributed. The land-use input came from the open [SANDAG Open Data portal](https://opendata.sandag.org/). | | [MesoWest / Synoptic](https://synopticdata.com/tcs/) | Non-commercial; summary statistics may be published only if the underlying data "cannot be decompiled, reverse engineered, or otherwise extracted" | Derived per-tract temperature statistics only, each a blend of six stations with no station exceeding 40% of the value. See below. | | [Romitti et al. 2022](https://doi.org/10.7910/DVN/HWFVP6) | CC BY article; data deposited on Harvard Dataverse | Open. Cited to the Dataverse DOI (the redistributable artifact), not the article page. | | [UCLA Heat Maps](https://uclaheatmaps.org) | Public research product, aggregated from California HCAI emergency-department records | Display-only map layer; **not included in the export.** | ## Provenance audit, 2026-07-27 Every column reaching the "Download Data" export was traced to its source file and that file's origin checked. Findings: **Verified open** — CDC PLACES, CDC/ATSDR SVI, ACS (B25034 / B25003 / C24030 / B11007), Census LACE 2023, USGS Landsat, NOAA HRRR, NLCD/MRLC, DOE LEAD, TIGER/Line, LOCA2, SANDAG Open Data, and Romitti et al. via Harvard Dataverse. All are U.S. federal public-domain works, state open-data releases, or CC BY. **Withdrawn from the export** — `heat_hosp_day` and `heat_hosp_night`. Their input (`data/hospitalizations/joined_rel.csv`) arrived as an emailed attachment; it carries no license, has no public download URL, and its column schema (`max99_2`, `hi95_SNR`, …) is analyst-internal rather than a published product. CalHeatScore's public release is daily scores plus a REST API — the per-ZCTA effect estimates are not part of it. Absent a demonstrable right to redistribute, they are now treated like HPI: on the map, out of the download. Request them from `CalHeatScore@oehha.ca.gov`. ### Resolved: single-station recoverability in the MesoWest columns Synoptic permits publishing summary statistics *provided the underlying observations cannot be extracted from them*. The audit found two places where our tract aggregates failed that test. Both are fixed. **What was wrong.** Of the 1,093 contributing stations, 511 are citizen networks (Tempest 228, CWOP 226, WXM 57) and 414 professional (SDG&E mesonet 220, SGXWFO 110). CWOP is public domain; Tempest, WXM, SDG&E and SC-Edison are privately operated. - **Daytime** (`mean_tmax`, `p85_tmax`, `p99_tmax`, `max_tmax`): the pipeline pooled stations into their containing tract, ran IDW from those ~212 tract centroids, then overwrote every station-containing tract with the pooled direct computation. 131 of those 212 tracts held exactly **one** station, so the published value *was* that station's statistic — at least 14 of them on the private SDG&E mesonet. - **Nighttime** (`mean_tmin_overnight`, `p90_tmin_overnight`, `no_relief_fraction`): one-hop IDW (k=6) normally blends six stations, but in 35 tracts a single station carried >90% of the inverse-distance weight (one centroid sits 14 m from its station). 24 of those were private networks. **What changed** (both in the `heat-risk-sd` repo): 1. `analyze_extreme_heat.py` now writes `station_heat_metrics.csv` — per-station heat statistics for the 503 stations with at least a year of record. A stub for this existed but returned an empty DataFrame, which is why interpolation ran from tract centroids in the first place. 2. `map_tract_heat_risk.py` interpolates in **one hop from those 503 station points** and no longer overwrites station-containing tracts. Control points went from 212 to 503, so this also raises resolution where stations are densest — the urban cores. 3. Both `map_tract_heat_risk.py` and `compute_nighttime_tract_metrics.py` cap any single station at **40%** of a tract's interpolated value, redistributing the excess across the remaining neighbours. Verified: 0 of 737 tracts exceed the cap on either path, and every tract now blends six stations. The cap is enforced by `tests/test_idw_weight_cap.py` in that repo. Effect on the published values: daytime `mean_tmax` rose 0.48 °C on average (Spearman 0.83 against the old field), `max_tmax` 1.59 °C (Spearman 0.68 — the old field was a station maximum passed straight through, which the blend now smooths). Nighttime shifted by −0.03 to +0.05 °C on average. Because HPI is non-commercial and CC BY permits commercial use, HPI-derived columns are **omitted from the bulk data download** to keep the downloadable dataset CC BY-clean. They remain visible on the map, used under HPI's own terms with attribution. If you redistribute the raw inputs (not the normal "view/use the map" path), check each source's terms. The exported data dictionary lists every withheld variable with `in_download = no` and the reason, so an absent column is explained rather than mysterious. Exclusions are driven by a single table — `RESTRICTED_EXPORT_SOURCES` in `scripts/build_diy_pca_viewer.py` — and enforced by `tests/test_export_licensing.py`. ### Why HPI is excluded but MesoWest/Synoptic is not Both sources are non-commercial, so the differing treatment is deliberate. The HPI columns on the map are HPI's *published product* — their percentile values, passed through unchanged; redistributing them is redistributing HPI. The MesoWest columns (`mean_tmax`, `p85_tmax`, `p99_tmax`, `max_tmax`, `mean_tmin_overnight`, `p90_tmin_overnight`, `no_relief_fraction`) are new statistics this project computes from raw station observations and interpolates to tracts, which is the case Synoptic's terms expressly allow publishing. The same reasoning covers the derived SanGIS green-space percentage. An earlier version of this file claimed the observations "cannot be recovered from the aggregates," and extended the reasoning to the CalHeatScore variables. The 2026-07-27 audit found both claims wrong as written: the aggregates *were* invertible in ~166 single-station tracts, and the CalHeatScore input has no license at all. The CalHeatScore variables have been withdrawn from the export, and the interpolation was rebuilt so the claim now holds — every tract value is a six-station blend with a 40% per-station ceiling. ### Attribution travels with the download Every row of the CSV and every feature of the GeoJSON produced by "Download Data" carries three constant columns — `citation`, `license` (CC BY 4.0), and `sources_url` — so a file that leaves this site still states its terms and links back to the per-variable source table. These are stamped from `EXPORT_META` in `scripts/build_diy_pca_viewer.py`. See the full source list in the [User Guide](userguide/index.html).