Ilystics/us-construction-weather-delay-days

US Construction Weather Delay Days by Metro (2016–2025)

How many workdays construction sites actually lose to weather, month by month, for 45 US metropolitan areas — derived from ten years of NOAA daily station observations run through a soil-specific drying model.

This is not a rainfall dataset. Rainfall is the input; the output is unworkable workdays, which is the unit construction schedules and contracts are actually written in. The distinction matters more than it sounds: in 26 of the 45 metros here, the wettest month is not the worst month.

Published by Ilystics. Landing page with per-metro breakdowns: https://ilystics.com/weather-delay-days

Files

weather_delay_days_monthly.csv — 540 rows (45 metros × 12 months)

Column	Description
metro_slug	Stable identifier, e.g. raleigh-durham-nc
metro, state	Display name and USPS state code
lat, lng	Metro centre; the NOAA station picker resolves from here
soil	Representative soil for the metro: clay, silt or sand (see caveats)
base_drying_days	Days to regain trafficability after meaningful rain: clay 3.0, silt 2.0, sand 0
baseline_start_year, baseline_end_year	Observation window (2016–2025)
month, month_name	Calendar month, 1–12
avg_precip_inches	Ten-year mean monthly precipitation
lost_workdays_mean	Ten-year mean of unworkable workdays. The headline figure.
lost_workdays_median	Middle year of the ten; above the mean implies dry-year skew
abnormal_threshold	Mean + 1 standard deviation (see caveat below)
drying_days_per_rain_day	Additional unworkable days each rain day generates
workday_coverage_pct	Share of workdays with a station reading

weather_delay_days_by_year.csv — 5,400 rows

The per-year detail behind every mean above (45 × 12 × 10). Published so the distribution can be inspected rather than taken on trust — a monthly average conceals whether ten years were alike or wildly different.

Method

A day counts as lost when it is a Monday–Friday on which any of the following holds:

1. Precipitation exceeded 0.1 inch, which resets a drying clock
2. The ground had not finished drying from earlier rain — duration set by soil type (clay 3.0 days, silt 2.0, mixed 1.5, sand/rock 0 — only clay, silt and sand occur in this dataset), adjusted for temperature, slope, exposure and antecedent moisture
3. The ground was frozen (max temp < 32°F), with the clock reset on thaw

Weekends are excluded. Days with no station reading are excluded rather than assumed workable.

Full methodology: https://ilystics.com/methodology

Important caveats

Soil is one representative type per metro. Each metro is assigned a single governing soil applied to the whole market — clay (19 metros), silt (16) or sand (10). It is not a spatial average and not a per-site lookup. Real soils vary substantially within any metro, and because the drying clock runs 3.0 days on clay against 0 on sand, site-level results can differ considerably from the metro figure. For a specific location, resolve the actual NRCS map unit rather than using the metro row.

Concretely: this dataset lists Raleigh-Durham as clay with a 3.0-day clock, while an NRCS lookup for ZIP 27606 inside that same metro returns the Beltline-Urban land-Cecil complex and a 1.5-day clock. Both are correct at their own resolution. The metro row is a planning figure, not a site figure.

abnormal_threshold is a convention, not a standard. It is the ten-year mean plus one standard deviation. It is useful for operationalising the contractual phrase "abnormal" or "unusually severe", which appears undefined in AIA A201, FAR 52.249-10 and ConsensusDocs 200 alike. No court, board or contract has adopted this definition. The per-year file is published precisely so anyone can argue for a different line.

Scope is soil-dependent site work only — earthwork, grading, excavation, foundations and underground utilities. Enclosed and vertical construction, roofing, and paving laydown are not modelled.

Do not sum the months into an annual total. The monthly baselines are independent. Adding them assumes a site never improves over a year, when stone haul roads, stabilised pads and ditching go in over the first months of a job. A twelve-month sum produces a number no site contractor would recognise.

8 of 540 rows have an empty drying_days_per_rain_day. All are desert or Mediterranean-climate metros in their dry season (Phoenix in May, Las Vegas in July, Los Angeles and Sacramento in summer) where there were no rain workdays to divide by. Zero lost workdays, no divisor — the blank is correct.

Findings worth checking

• The wettest month is not the worst month in 26 of 45 metros. Boston is the clearest case: July averages 4.28" and loses 8 workdays, while January averages 3.69" and loses 16. Less rain, double the loss. Winter rain arrives spread out onto cold ground and each event collects its own drying tail; summer storms cluster and their tails overlap.
• Worst single month: Minneapolis in January, ~20 lost workdays.
• Widest seasonal swing: Portland, 18 lost workdays in January against 1 in July.

Source and licence

Derived from NOAA National Centers for Environmental Information, Global Historical Climatology Network – Daily (GHCND) — US Government public domain station observations.

Released under CC BY 4.0. Attribution: Ilystics LLC, https://ilystics.com

Model version v2.3.0. Machine-readable per-metro JSON is also served at https://ilystics.com/data/weather-delay-days/{metro_slug}.json.