Methodology
Every timing recommendation on this site comes from a number we can point at. This page says where that number comes from, how wrong it can be, and when we decline to give you one.
Where the soil temperature comes from
Primary source: measured. The U.S. Climate Reference Network (USCRN, NOAA/NCEI) runs research-grade stations that record soil temperature at 5 cm depth. We pull one station per region every day. The data lands with about a one-day delay, which does not matter for a seasonal decision. When a page says measured, a real probe in real soil produced that reading.
Fallback: estimated. Stations go offline. When ours has no fresh reading, we estimate soil temperature from air temperature reported by the National Weather Service (api.weather.gov), using a 12-day rolling average of daily means plus a per-region correction. Pages label this estimated, always.
The temperature windows
Each timing call on this site maps to a band on the soil thermometer, not to a month. The bands differ in width — and that width is why some calls require a measured reading while others tolerate an estimate.
- Cool-season seeding — 50–65°F (15° wide; estimate OK)
- Crabgrass pre-emergent — 52–58°F (6° wide; measured only)
- Warm-season seeding — above 65°F
How the fallback model was calibrated
We took three years of daily USCRN data from twelve stations — one per region — and compared measured 5 cm soil temperature against rolling averages of air temperature. Three things came out of it, and all three changed the site:
- A 12-day window fits best. Our first guess was 5 days. It was worse. Windows between 9 and 13 days are effectively tied.
- Soil runs warmer than the air, and by how much is regional. The correction ranges from −1.12°F in the Upper Midwest to +7.03°F in the Pacific Northwest. A single national number would be several degrees off at the edges.
- The model is accurate to about ±2.6°F (mean absolute error), with a 90th percentile around 5°F, across the months that matter for lawn decisions.
Per-region correction
| Region | USCRN station | Correction (°F) |
|---|---|---|
| Upper Midwest | MN_Sandstone_6_W | -1.12 |
| Great Lakes | OH_Wooster_3_SSE | +1.29 |
| Northeast | NY_Ithaca_13_E | +1.19 |
| Mid-Atlantic | VA_Charlottesville_2_SSE | +0.44 |
| Transition Zone | KY_Versailles_3_NNW | +1.51 |
| Southeast | GA_Watkinsville_5_SSE | +3.85 |
| Florida & Gulf | FL_Sebring_23_SSE | +4.98 |
| South Central | TX_Austin_33_NW | +1.39 |
| Southern Plains | KS_Manhattan_6_SSW | -1.13 |
| Northern Plains | IA_Des_Moines_17_E | +0.28 |
| Mountain West | CO_Nunn_7_NNE | +2.12 |
| Pacific Northwest | OR_Corvallis_10_SSW | +7.03 |
Accuracy by length of history
| Days of data in the average | Mean absolute error (°F) |
|---|---|
| 1 | 6.2 |
| 3 | 3.2 |
| 7 | 2.7 |
| 12 | 2.6 |
When we refuse to answer
- Fewer than 7 days of history. A one-day estimate looks exactly like a normal reading and can be 6°F off. The page says we are still collecting instead of showing it.
- Crabgrass pre-emergent on estimated data. The useful application range is roughly 52–58°F — six degrees wide. Our estimate carries ±2.6°F of error, which is noise at that resolution. So we only make that call from a measured reading. Seeding cool-season grass, by contrast, has a 15°F range, where the estimate holds up fine.
- Winter. The model's bias jumps in midwinter, when snow insulates the ground. We calibrated on the months we actually give advice in (March–May, August–November) and do not make lawn calls in January.
Why regions and not states
Lawn advice follows climate and grass type, not state lines. Twelve regions, each with a real station and its own guide, beats fifty near-identical state pages — for you and for search engines alike.
Sources
Calibration run: August 17, 2026. Re-run annually.