HSH (human heat stress): incomplete NWS Heat Index — missing corrections + wrong input temp

[peetmate]

Summary

Review found the HSH (human heat stress) index in hazards_upstream/R/04_indices/calc_HSH.R is an incomplete implementation of the NWS (Rothfusz) Heat Index. The regression core is correct, but it omits the standard correction branches and uses the wrong input temperature — so current Atlas HSH values are biased, materially so in Africa's arid and very-humid zones.

Findings

Correct: all 9 Celsius Rothfusz coefficients (c1..c9) verified against the published NWS regression — exact match.

Issues:

1. Missing all three NWS correction steps:
   • Steadman simple-formula fallback (used when HI < 80 °F) — replaced by a non-standard tmean >= 25 °C gate that returns raw temperature below it.
   • Low-RH adjustment (RH < 13%, 80–112 °F) — index reads high under hot-dry conditions (Sahara/Sahel/Horn).
   • High-RH adjustment (RH > 85%, 80–87 °F) — index reads low under warm-very-humid conditions (Congo basin, Gulf of Guinea coast).
2. Wrong input temperature: feeds daily mean Tavg = (tasmax+tasmin)/2 into the Heat Index. HI is a daytime apparent-temperature metric defined on dry-bulb air temp — should use Tmax. Current code understates the dangerous daytime peak.
3. Non-standard gate: 25 °C (~77 °F) instead of the NWS 80 °F switch.

Options

• (a) Fix the NWS Heat Index in place: switch Tavg→Tmax, add the Steadman fallback + low/high-RH corrections (cleanest done in °F then converted). Keeps the familiar metric, makes it correct.
• (b) Switch to a state-of-the-art human-heat metric: WBGT (ISO 7243) or UTCI. CDH climate-toolkit currently uses xclim humidex for human heat.
• (c) Converge on CDH/xclim human-heat output as part of the hazards merge.

Data note (WBGT/UTCI feasibility)

NEX-GDDP-CMIP6 inputs we currently process: tasmax, tasmin, hurs, rsds, pr — no sfcWind.

• Simplified (shade) WBGT is computable now from tasmax+hurs.
• Full outdoor WBGT and UTCI need wind (natural wet-bulb + globe/mean-radiant temp); sfcWind is a standard NEX-GDDP-CMIP6 variable but is not currently downloaded — addable via the download config.

Ask

@bjyberg — could you review which direction the Atlas should take for human heat stress (fix NWS HI vs humidex vs WBGT/UTCI), given CDH's human-heat method work? Tagging for method sign-off before we change Atlas values.

Filed from a methods review of hazards_upstream/R/04_indices (Adaptation Atlas hazards producer).

[bjyberg]

Given that this will likely be a short term patch until we calculate cmip7 next year for the CDH, I'd suggest we go the easier routes rather than downloading new variables to swap to humidex or another index. Zia's team calculated their own human heat stress for their notebook and they did use wet-bulb. If we can get their data, that could be an easy drop in. Otherwise I'd say we stick with the current hsh method for now, re-run the fix, and upgrade it when the cdh provides a better version.

If we are already re-running it, we could use it as an excuse to start rebuilding the hazard pipeline in python with xclim and xarray to prep for cmip 7. Then when we want to use a wet-bulb based index, we can just swap in the algorithm