PRISM offset from a nearby station assumes the spatial pattern of mean minimum temperature predicts the spatial pattern of last-frost date.
Promising · suboptimal if ignored → advise
How it works
PRISM models orography, temperature inversions, slope aspect, and coastal proximity at 800m. It provides monthly mean/min/max temperature. It does not provide frost exceedance probabilities; those exist only at stations. Combining them is an inference.
How much it matters: Offset mean-absolute last-frost error ~4-7 days within 80 km; skill over a bare nearest-station lookup +1.4 days at 10-20 km rising to +4.0 days at 40-80 km (190-station leave-one-out held-out fit, engine/climate_calibrate.py, 2026-07-11).
The evidence
Evidence: confirmed against 2 sources in the research-based literature.
the 2 sources
Daly et al. (2008), 'Physiographically Sensitive Mapping of Climatological Temperature and Precipitation across the Conterminous United States', Int. J. Climatol. 28: 2031-2064, prism.oregonstate.edu/pubs/link/2008_daly-etal_ijoc.pdf (read 2026-07-11): PRISM produces '1971-2000 mean monthly precipitation and minimum and maximum temperature ... 30-arcsec (~800-m) grids'; station weighting considers 'location, elevation, coastal proximity, topographic facet orientation, vertical atmospheric layer, topographic position, and orographic effectiveness of the terrain'; the western US, 'characterized by ... large elevation gradients, rain shadows, inversions, cold air drainage, and coastal effects, showed the greatest improvement.' Confirms PRISM provides mean monthly temperature fields, NOT frost-date exceedance probabilities -- so applying the offset to last-frost date remains an inference (below).
engine/climate_calibrate.py (held-out-station leave-one-out fit, 2026-07-11, ISSUES#6): over 190 station pairs spanning flat plains and mountain terrain, predicting a station's last-frost date from its nearest neighbour via the PRISM offset reduces mean-absolute error vs a bare nearest-station lookup beyond ~10 km (skill +1.4 days at 10-20 km, +2.5 at 20-40 km, +4.0 at 40-80 km; offset MAE stays <= 7 days out to 80 km). Offset MAE ~doubles (7.6 -> 12.6 days) once local PRISM tmin std >= 1.0 C. Quantifies the assumption; the provenance distance threshold (80 km), terrain downgrade (1.0 C), and on-site cap (0.5 C) in docs/CLIMATE.md §4 fall out of it.
Change history
Every change to this rule is one commit. This is what the project's version history records — the date and the reason — not a summary of it.
- drop TWO letter grades from the calibrator pointer (D-083, O66)
- corpus: another rule here's pointer prose said "unverified"; the rule is verified (BACKLOG O3)
- promote to verified — PRISM-offset last-frost inference
- quantify effect_size from the held-out-station calibration (ISSUES#6)
- cite Daly et al. 2008 (read) for PRISM methodology; mechanism confirmed
Cite this page
Milpa Gardens. "PRISM offset from a nearby station assumes the spatial pattern of mean minimum temperature predicts the spatial pattern of last-frost date." Evidence grade: Promising. https://milpa.garden/rules/prism-offset-from-nearby-station-assumes-spatial/
A garden-planning corpus: every claim states its evidence grade and links its sources on the page. Cite it with the grade, and add the date you read it.