(one or two tips)
≥1 mm per 5 min
falling 06–12h
best predictor AUC
The station's rain gauge is a tipping bucket that reports a running daily total to a thousandth of an inch, every five minutes, since November 2023. Reconstructing rain from it means taking the positive increments of that counter within each local day, discarding the first reading after midnight when it still carries yesterday's total, and adding nothing when the counter dips. Done that way the record holds 2,213 mm across 4,702 wet steps, of which 4,671 are clean (the previous reading was five minutes earlier) and the rest accrued across outages.
Three things fall out. Rain here is mostly drizzle-sized steps punctuated by a small number of very hard ones; it is a night and evening phenomenon with a near-dry late morning; and the outages are not random — the station goes down in storms, so the biggest events are partly hidden and every peak rate on this page is a floor. The last section is a negative result we think is worth publishing: nothing the station sees in the afternoon predicts whether it will rain that night.
Most steps are a tip; a few steps are the storm.
Each row is a bin of five-minute increments. The gauge's smallest increment is 0.004 in = 0.10 mm, so "≤0.25 mm" is one or two tips.
| 5-min step | Share of wet steps | Share of rain |
|---|---|---|
| <=0.25 mm (1-2 tips) | 61.9% | 19.1% |
| 0.26-0.5 mm | 19.9% | 17.6% |
| 0.51-1.0 mm | 9.0% | 16.3% |
| 1-2 mm | 5.5% | 18.4% |
| 2-4 mm | 2.9% | 19.0% |
| >4 mm | 0.8% | 9.6% |
62% of wet steps are ≤0.25 mm and carry 19% of the rain. Steps of 1 mm or more are 9% of the steps and carry 47%.
1 mm in five minutes is 12 mm/h — a proper downpour. Nearly half of everything that has fallen on the station fell at that rate or harder, in about one wet step in eleven. The estuary and the road do not experience the median step; they experience the top decile.
The hardest clean five minutes: 8.5 mm. The hardest clean hour: 35.1 mm.
"Clean" means no missed reading inside the window. The day totals include rain that accrued across outages; the "hidden" table lists the biggest of those, where the true peak rate is unknown and the amount is a floor.
| Ending (local) | mm / 5 min |
|---|---|
| 2025-09-21 16:51 | 8.5 |
| 2026-07-29 03:26 | 8.3 |
| 2025-10-09 14:20 | 6.4 |
| 2026-06-18 23:55 | 6.1 |
| 2024-09-09 21:40 | 5.9 |
| Hour ending | mm / 1 h (gap-free) |
|---|---|
| 2026-06-04 03:00 | 35.1 |
| 2023-11-23 01:20 | 31.8 |
| 2026-06-19 00:40 | 31.4 |
| 2024-10-10 01:35 | 30.4 |
| 2026-06-19 19:00 | 30.1 |
| Local day | mm | Gauge coverage |
|---|---|---|
| 2024-07-21 | 103.2 | 93% |
| 2024-09-26 | 101.8 | 100% |
| 2024-09-27 | 78.1 | 87% |
| 2024-07-30 | 72.2 | 98% |
| 2025-06-14 | 67.0 | 100% |
| 2024-09-09 | 64.9 | 98% |
| 2026-06-19 | 64.4 | 100% |
| 2024-10-10 | 48.0 | 100% |
2024-09-26 and 09-27 are Hurricane John (101.8 + 78.1 mm in two days). 2024-07-21 is the record day and also the most hidden one — see below.
| Gap ending | Gap length | Rain across the gap (floor) |
|---|---|---|
| 2024-07-30 15:06 | 41 min | ≥42.8 |
| 2024-07-21 22:56 | 61 min | ≥42.1 |
| 2024-07-21 21:10 | 35 min | ≥24.3 |
| 2024-09-09 19:09 | 32 min | ≥18.8 |
| 2025-09-21 17:06 | 15 min | ≥16.6 |
| 2025-06-23 00:18 | 38 min | ≥12.2 |
| 2025-06-18 20:38 | 23 min | ≥10.7 |
2024-07-21: 103.2 mm in the day, at least 42.1 mm in a 61-minute outage and 24.3 mm in a 35-minute one. 2024-07-30: at least 42.8 mm across 41 minutes.
Those two gaps almost certainly hold the true hourly records. The clean 1-h record (35.1 mm, ending 2026-06-04 03:00) is what the gauge saw without blinking; the July 2024 storms delivered more than that in less time and the station lost power in the middle of it. Say "at least".
Evening and night. The morning is nearly dry.
| Block (local) | Share of rain (clean steps) | Share of rain (all steps) | Share of wet steps |
|---|---|---|---|
| 00-03h | 20.2% | 18.5% | 15.7% |
| 03-06h | 16.3% | 14.5% | 16.2% |
| 06-09h | 6.6% | 5.8% | 10.4% |
| 09-12h | 1.6% | 1.4% | 3.3% |
| 12-15h | 9.4% | 8.4% | 5.9% |
| 15-18h | 14.1% | 16.4% | 15.0% |
| 18-21h | 17.1% | 18.2% | 18.7% |
| 21-24h | 14.7% | 16.7% | 14.9% |
06–12h carries about 8% of wet-season rain; 18–24h carries about 32%; 00–06h about 36%.
Two peaks: an evening one from convection built over the afternoon (15–21h) and an overnight one (00–06h). The 09–12h block gets 1.6%. If you need a dry three hours in August, the late morning is the bet — which is also when the sea breeze has just arrived (the wind clock).
The station goes down when it storms.
| Measure | Value |
|---|---|
| Gaps ≥15 min in the gauge record | 112 |
| Share of the record's time inside those gaps | 7.8% |
| Share of all rain that accrued across gaps ≥15 min | 8.0% (176 mm in 12 gaps) |
| Share of all rain across any missed reading (interval >7.5 min) | 12.3% |
| Clean 5-min intervals that are wet | 1.68% |
| Gaps ≥15 min that are wet | 10.7% |
| Ratio | 6.4× |
An outage is 6.4 times as likely to contain rain as a normal interval. That is the power grid, not the gauge: lightning and wind take the station down. It means day totals on stormy days are floors, and the peak-rate tables above are conservative in exactly the cases that matter.
No. Not with anything the station measures.
Jun–Oct days with no rain between 12:00 and 18:00 and good coverage overnight: n=332, of which 68 had ≥1 mm between 18:00 and 06:00. Each afternoon (13–17h) reading is scored as a ranking predictor of a wet night (AUC; 0.5 is a coin flip). Pressure is de-tided by removing the mean diurnal cycle for each 5-minute slot.
| Afternoon predictor (13–17h) | AUC | Mean, dry nights | Mean, wet nights |
|---|---|---|---|
| Afternoon dew point | 0.52 | 79.2°F | 79.2°F |
| Afternoon relative humidity | 0.54 | 70.9% | 71.4% |
| Afternoon max temperature | 0.47 | 91.9°F | 91.6°F |
| Afternoon mean solar | 0.51 | 652.3 W/m² | 641.3 W/m² |
| De-tided pressure anomaly | 0.41 | -0.0 inHg | -0.0 inHg |
| Afternoon mean wind | 0.55 | 4.1 kt | 4.4 kt |
| Afternoon max gust | 0.54 | 10.3 mph | 10.9 mph |
AUC 0.41–0.55 across seven predictors. The means on dry and wet nights are indistinguishable.
Wet nights are not hotter, more humid, calmer or lower-pressure in the afternoon than dry ones. With 68 events the confidence band on an AUC is roughly ±0.08, so a weak signal cannot be excluded, but nothing here is usable. The storm signature exists — it just arrives with the rain.
The storm signature, composited
64 onsets: the first wet step after ≥6 dry hours that was followed by ≥5 mm within 3 h. Anomalies relative to the mean of the hour before onset (−60 to −5 min). Mean and median across onsets.
| Minutes from onset | Pressure, mean (mb) | Temperature, mean (°F) | Temperature, median (°F) |
|---|---|---|---|
| -60 | -0.12 | +0.7 | +0.5 |
| -30 | +0.00 | -0.2 | -0.1 |
| -15 | +0.14 | -0.5 | -0.3 |
| +0 | +0.31 | -2.1 | -1.5 |
| +15 | +0.40 | -4.2 | -3.2 |
| +30 | +0.46 | -5.1 | -4.2 |
| +60 | +0.51 | -6.1 | -5.7 |
| +120 | +0.65 | -7.1 | -7.1 |
| +180 | +0.89 | -7.3 | -7.1 |
Pressure is flat until 15 minutes before the first tip, then rises (+0.31 mb at onset, +0.89 mb after three hours — the cold-pool "pressure jump"). Temperature is -2.07 °F at the first tip and -6.11 °F an hour later (median -5.71). A prior pass of this analysis, with a different onset rule, put the jump at +0.65 mb and the temperature drop at −1.2 then −4.3 °F; the shape is the same, the size depends on how you define the onset and the baseline.
One gauge, 0.1 mm resolution. The bucket tips at 0.1 mm (0.004 in) and the counter is reported to 0.001 in, so the smallest steps land at 0.08–0.10 mm and the bins are approximate. Heavy-rain undercatch of tipping buckets is real and unquantified here; the hardest steps are floors for that reason too.
Outages hide the peaks. 8.0% of all rain accrued across gaps ≥15 min and 12.3% across any missed reading, and gaps are 6.4× as likely to be wet as clean intervals. Every 'largest' on this page is the largest the gauge saw; 2024-07-21 and 2024-07-30 almost certainly exceeded the clean hourly record. eventrainin and hourlyrainin from the station are not used (unreliable).
2.9 seasons. November 2023 to September 2026, with the array stuck 2024-08-18 → 08-21 and dead 2026-05-15 → 07-18. Wet-season shares pool three summers; Hurricane John (Sept 2024) alone is 8% of the record.
The null is a null on this station's variables. 68 wet nights is enough to rule out a strong predictor, not a weak one (AUC ±0.08). Satellite, radar, soundings or model convective indices were not tested; the claim is only that the local surface afternoon does not tell you about the local night.
Rain here is a night animal. It arrives as a handful of hard steps inside a lot of drizzle, mostly between dusk and dawn, hardest when the station cannot see it. The dry hours are the late morning. And the afternoon — hot, humid, still — knows nothing about what the night will do.Synthesis · 26 September 2026
Sources and method
Data: on-site Ambient Weather station, dailyrainin every 5 minutes, 2023-11-09 → 2026-09-19. Local day = UTC−6.
- Steps: positive increments of dailyrainin against the running maximum within the local day; a first reading stamped before 00:05 whose value exceeds the next reading is dropped (it carries the previous day); counter dips add 0. Inches ×25.4.
- Clean step: interval to the previous reading ≤7.5 min. Gap step: interval ≥15 min (rain accrued during an outage).
- Gap-free hour: trailing 60-min sum with ≥10 of 12 five-minute slots present and no gap step inside; peaks deduplicated at 2 h.
- Diurnal blocks: Jun–Oct steps by local hour of the reading that recorded them; clean steps for the headline, all steps in the second column.
- Night prediction: Jun–Oct days with ≥30 readings 13–17h, no rain 12–18h, ≥100 readings 18–06h; wet night = ≥1 mm 18:00–06:00; AUC by rank statistic. Dew point from the Magnus formula; pressure anomaly = reading minus the mean for that 5-min slot of the day.
- Storm composite: onset = first wet step after ≥6 dry hours followed by ≥5 mm within 3 h (n=64); 10-min windows at 15-min lags, anomalies vs the mean of −60..−5 min.
Script: scripts/analyze_station_climatology.py → functions/api/_findings_station_climatology.js (export STATION_CLIMATOLOGY, generated 2026-09-27).
Cite as: Conner, A. (2026). "How It Rains at La Saladita — Night Rain, Hard Rain, and a Null." La Saladita Field Guide, lasaladita.com/findings/how-it-rains/, 26 September 2026. Data: on-site station 2023–2026; artifact /api/findings-station-climatology.