ClimateCast Documentation

1Geography

Each ZIP code is represented by a single latitude and longitude used for all climate data lookups. We use a centroid hierarchy to pick the most representative point:

1. Primary source: HUD population-weighted centroid. The U.S. Department of Housing and Urban Development publishes pre-computed centroids that reflect where residents are concentrated within each ZIP code, rather than the geometric center of the ZIP's boundary.
2. Fallback: ZCTA geometric centroid. When HUD does not provide a centroid for a given 5-digit ZIP code, we fall back to the geometric center of the Census Bureau's ZIP Code Tabulation Area (ZCTA) polygon from TIGER/Line 2023.
3. Land-snap. If the chosen point falls in water (common for coastal ZIPs), we shift it inland to the nearest land pixel using a CONUS land mask derived from WorldClim.

2Current Monthly Climate

Source: WorldClim historical monthly weather (CRU-TS 4.09 downscaled with WorldClim 2.1), averaged over the 2005–2024 reference period, 2.5 arc-minute resolution (~4.6 km per pixel).

Method: Year-by-year monthly rasters over 2005–2024 are averaged per calendar month, then point-sampled at the representative latitude and longitude selected for each ZIP. No area-averaging is performed; each ZIP gets the value of the single grid cell containing its representative point.

3Extreme Heat Days and Wet-Bulb Days (Current)

These metrics require daily temperature distributions and cannot be derived from WorldClim monthly means alone. We compute them from NOAA's Global Historical Climatology Network Daily (GHCNd) dataset.

Station Selection

The assigned station is recorded in:

Heat Day Thresholds

For each month, ClimateCast counts the average annual days where TMAX exceeds the threshold below.

Wet-Bulb Temperature Calculation

Wet-bulb temperature (Tw) uses the Stull (2011) empirical approximation:

Tw = T · atan(0.151977 · (RH + 8.313659)^0.5)
   + atan(T + RH)
   - atan(RH − 1.676331)
   + 0.00391838 · RH^1.5 · atan(0.023101 · RH)
   − 4.686035

Where T is dry-bulb temperature in °C and RH is relative humidity in %.

Relative humidity is derived from TMAX and dew-point using the Magnus formula. When dew-point is missing, ClimateCast uses TMIN as a dew-point proxy (a conservative approach that can bias wet-bulb counts high in arid climates).

Frost Dates

4Future Monthly Climate

Source: WorldClim CMIP6 downscaled projections, 10 arc-minute resolution (~18 km per pixel on the UC Davis mirror), point-sampled at the same representative location as historical data (nearest finite cell search handles occasional coastal nodata on the coarse grid).

Scenarios and Time Periods

ClimateCast exposes three SSP pathways:

Each scenario includes four 20-year windows:

Multi-Model Mean

Future values are averaged across eight WorldClim CMIP6 GCMs (same set for every emissions pathway):

1. ACCESS-CM2
2. BCC-CSM2-MR
3. CNRM-CM6-1
4. CNRM-ESM2-1
5. CanESM5
6. IPSL-CM6A-LR
7. MIROC6
8. MRI-ESM2-0

GFDL-ESM4 (the ninth GCM on WorldClim) is excluded because the UC Davis download mirror lacks complete coverage for the SSPs we use.

Scenario keys are low, mid, and high; period keys are 2040, 2060, 2080, and 2100 (each maps to the twenty-year windows above). All monthly.* arrays are 12 elements (Jan–Dec). In the paths below, <tier> is one of low, mid, or high, and <window> is one of 2040, 2060, 2080, or 2100.

Delta Method for Future Heat and Wet-Bulb Days

ClimateCast uses a mean-shift (delta) approach:

1. Compute Δtmax = CMIP6_future_tmax − WorldClim_historical_tmax
2. Shift the historical daily TMAX distribution by Δtmax
3. Recount threshold exceedances and recompute wet-bulb days

This method does not model shifts in variance/tails, only the mean.

5Climate analog (reference locations)

Shows which existing places have a baseline climate most similar to this ZIP's projected future — one closest match from the combined U.S. + international catalog. This is an illustrative similarity in climate-variable space, not a forecast that a ZIP "becomes" another city.

How it works

The app compares each reference location's WorldClim / CRU-TS 4.09 (2005–2024) baseline against this ZIP's projected monthly climate (CMIP6 multi-model mean for the selected scenario and period) across seven dimensions:

Distance calculation

All reference features are z-scored using means and standard deviations from the full reference set. The projected vector is z-scored with the same parameters. Distance is weighted Euclidean in z-space with weights: mean high 2.2, mean low 2.2, aridity index 1.9, temperature seasonality 0.85, precipitation phase 0.7 each (cos and sin), and annual precipitation 0.8. Temperature is weighted most, followed by aridity; annual precipitation is down-weighted because it overlaps heavily with the aridity and precipitation-phase dimensions.

The single closest row from the full combined reference set is reported. The user's own location is excluded from candidates by ZIP and by normalized place name + state (same city at a different catalog ZIP).

When it's omitted

The section is hidden when the reference dataset fails to load, monthly futures are incomplete, or centroid latitude is missing. Centroid longitude is only needed for the map arcs.

Downloads (same origin as the site; use “Save link as…” if your browser opens JSON instead of downloading):

6Drought Risk

Drought risk is derived, not pass-through from an external risk tier dataset. It uses an Aridity Index (AI):

AI = MAP / MAE

Where:

Classification thresholds:

7Risk Level Binning

ClimateCast reports eight risk categories:

• Six county-level scores from ProPublica/Rhodium
• One storm intensity category from EPA CREAT
• One derived drought category

ProPublica / Rhodium Group Risks

Categories: heat, wet_bulb, farm_crop_yields, sea_level_rise, very_large_fires, economic_damages.

ProPublica publishes county-level scores on a 1–10 scale. ClimateCast bins those scores:

Storm Intensity (EPA CREAT)

Category: storm_intensity

Binning thresholds:

Null Reason Values

8Demographics

Three demographic fields are included as pass-through values from named sources.

On city and county pages (see City and County Pages), median household income and total population are pulled directly from ACS at the Census place and county summary levels rather than aggregated from ZIP values, so they are the official figures for the whole place or county.

9ZIP-to-County Assignment

Each ZIP is assigned to a single county for county-level risk joins.

Representative ZIP per county

Each county also has one representative ZIP, used for the county page's climate data (see City and County Pages) and as the climate source for that county.

Selection order (first match wins; later steps break ties):

10City and County Pages

Alongside ZIP reports, ClimateCast publishes city and county pages. Searching for a city opens its city page, and clicking a county in the county-risk map opens that county page.

Geography

How values are sourced

Income and population are read directly from ACS for the whole place or county, so they are not a single ZIP's numbers nor an average of ZIP medians. Climate, life expectancy, and plant hardiness use one representative ZIP (see Representative ZIP per county for how the county ZIP is chosen). For cities, the pipeline first picks the highest-population allowed ZCTA whose centroid falls inside the Census place; if none qualify, it uses the highest-population ZCTA that intersects the place polygon.

Search routing

Search routes by the selected result's place type: a city or locality opens the matching city page, an administrative area opens the county page, and a ZIP code or street address opens the ZIP page. The city search index lists only places with a published city/{geoid}.json (representative ZIP must exist in the location dataset). When a city can't be matched to that index, search shows an error instead of silently falling back to a ZIP page.

11Known Limitations

12Data Sources

• WorldClim ↗ — Historical climate normals and CMIP6 downscaled future projections.
• NOAA GHCNd ↗ — Daily station data used for heat-day, wet-bulb, and frost metrics.
• ProPublica Climate Risk ↗ — County-level climate risk indicators sourced from ProPublica/Rhodium.
• EPA CREAT ↗ — Legacy tool reference for the Climate Resilience Evaluation & Awareness Tool (CREAT).
• CREAT ArcGIS Data (CRWU) ↗ — Primary downloadable storm projection dataset used because the original CREAT app is no longer active.
• U.S. Census ACS ↗ — Median household income by ZCTA.
• CDC USALEEP ↗ — Small-area life expectancy estimates.
• USDA PHZM ↗ — Plant hardiness zone reference values.
• HUD ZIP Crosswalk ↗ — ZIP-to-county assignment using residential address weights.