Three-Station Rain Temporal Footprint Analysis

Stations: CRBN01 at Weatherly in Carbon County, SHKL01 at Penn State Schuylkill, and BRKS01 at Penn State Berks. Data run through July 26, 2026. All clock-time results are converted from UTC to America/New_York with daylight-saving time handled.

Principal findings

• Summer has the clearest clock-time signal. In the longer Schuylkill–Berks record, 71.3% of JJA rain falls from noon through 11:59 p.m.; only 6.8% falls from 6–11 a.m. The single strongest hour is 6–7 p.m. local, containing about 9.7% of summer rainfall.
• Summer events are shorter and harder. Median JJA event span is 2.73 hours, versus about 4.94–5.60 hours in the other seasons. Median peak 60-minute accumulation is 0.16 in/hr, roughly 1.6–2.3 times the other seasons.
• Late autumn is the quietest event season in this sample. The Schuylkill–Berks average is lowest in October and November, while June, July, February, and December produce the most events per 30 observed days.
• Most storms are regional, but not all. During the common three-station period, 122 of 212 classified regional events (57.5%) reached all three stations, 36 (17.0%) reached exactly two, and 54 (25.5%) reached only one.
• Spatial coherence is seasonal. All three stations participated in 75.0% of autumn and 64.2% of spring events, but only about half of summer and winter events. Summer patchiness is physically plausible; winter must also be treated cautiously because tipping-bucket response to frozen precipitation was not independently verified here.
• Onset differences are usually under two hours, but sometimes much longer. Median absolute pairwise onset lag ranges from 0.73 to 1.37 hours; 90th-percentile lag ranges from 4.22 to 5.90 hours.

Data coverage and QC

The records are strong for minute-scale work: valid rainfall coverage ranges from 99.02% to 99.36%. Unequal station histories are handled two ways: monthly and seasonal rates are divided by valid exposure time, and spatial comparisons use only periods when the relevant stations overlap.

Isolated one-tip clusters are also common: 49 at Carbon, 166 at Schuylkill, and 113 at Berks. They make up roughly 17–22% of all raw six-hour clusters. Some are real tiny showers; others can be delayed bucket completion, dew, meltwater, or mechanical noise. They are stored, but excluded from the principal “how often does it rain?” count.

Why a six-hour event separator

There is no universal rain-event definition. Published gauge studies explicitly note that event boundaries depend on the chosen rainless interval, while six hours is a common hydrologic convention. For these data, three hours fragments long intermittent systems; twelve hours merges too many neighboring systems. Six hours is the defensible middle, provided the results are accompanied by sensitivity tests and a shorter burst definition.

Recommended hierarchy:

• Rain tip: any minute with at least 0.01 inch.
• Trace cluster: a six-hour cluster totaling exactly 0.01 inch. Retain, flag, do not use in the primary frequency count.
• Wet burst: tips separated by no more than 30 minutes. This captures starts, stops, and embedded showers.
• Station event: primary depth at least 0.02 inch, with six observed dry hours required to separate events.
• Regional episode: union of station events across the network, again using six hours.

Monthly footprint

The table below uses the longer Schuylkill–Berks records and averages station-normalized rates. These are sample climatologies, not 30-year normals. Carbon is plotted separately but has only about 21 months and should not be used to proclaim eternal truths about May merely because May behaved dramatically twice.

What time does it rain?

The month–hour heatmap is the most direct answer. The strong summer late-day maximum is unmistakable. Outside summer, the timing is broader and more synoptically controlled, so “usually” becomes a much weaker word.

Intensity, duration, and stop–start behavior

Events are not treated as continuously wet. Every station event is decomposed into 30-minute bursts. A typical event contains two to three bursts, while the 90th percentile reaches five bursts in summer and seven to nine in the other seasons. This is why event span and actual raining time are not the same thing, despite software’s touching faith that a single duration number can represent weather.

Interpretation: summer events are shorter, more intense, and somewhat less intermittent. Winter and spring events occupy longer clock spans and contain more pauses. Autumn has fewer events but a relatively large upper tail of event depth.

Three-station temporal and spatial footprint

The stations are separated by about 21 to 40 miles. During the common period from October 30, 2024 through July 26, 2026, 212 regional episodes passed coverage checks and contained at least one primary station event.

The signed median lags are only a few tens of minutes and change sign among pairs. That is not a stable storm-motion vector. Three gauges can show onset sequencing, but they cannot by themselves distinguish translation, growth, decay, terrain enhancement, and threshold timing. Radar or gridded precipitation should be joined later if the goal becomes storm-track reconstruction.

Limitations

• Schuylkill and Berks provide roughly 5.6 years; Carbon provides only about 1.7 years. This supports exploratory local climatology, not a stable climate normal.
• A 0.01-inch tipping bucket quantizes light rainfall. The first tip can occur long after rain begins, and a partially filled bucket can tip during a later event. Minute start times are therefore measurement times, not perfect physical onset times.
• The analysis uses gauge rainfall as supplied. Instrument heating, snow adapters, maintenance history, and site-level exposure metadata were not available in the archives, so winter spatial differences deserve caution.
• Event matching is threshold-based. A station receiving 0.01 inch is treated as trace-only, while 0.02 inch is primary. Both raw and filtered tables are included so the threshold can be changed without excavating the whole project again.

Reproducible outputs

The package contains the full station event catalogs, month/hour tables, seasonal summaries, regional-event catalog, pairwise footprint table, event-definition sensitivity tables, QC flags, figures, and the Python analysis scripts. The most useful starting files are:

• tables/*_events_6h_timing_qc.csv: station event catalogs after timing QC.
• tables/three_station_regional_events.csv: common-period regional episodes and station participation.
• tables/pairwise_temporal_footprint.csv: event overlap, onset lag, and depth correlation.
• tables/long_pair_pooled_season_hour_timing_qc.csv: local-hour seasonal timing.
• tables/suspect_slow_drain_events.csv: every excluded timing episode, with dates and diagnostic metrics.

Method references

The six-hour convention and its sensitivity are discussed in EPA storm-event work and hydrologic literature. A gauge-comparison study notes that there is no universal event criterion. Tipping-bucket measurements are discrete increments, so low-rate onset and event boundaries are intrinsically quantized.

• U.S. EPA, Pittsburgh green-infrastructure technical assistance report: https://19january2021snapshot.epa.gov/sites/static/files/2015-10/documents/pittsburgh-united-abundant-rainfall-508.pdf
• EPA, Analysis of Storm Event Characteristics for Selected Rainfall Gages: https://nepis.epa.gov/Exe/ZyPURL.cgi?Dockey=P100F7F7.TXT
• Tokay et al., Rain Gauge and Disdrometer Measurements during KAMP: https://journals.ametsoc.org/view/journals/atot/20/11/1520-0426_2003_020_1460_rgadmd_2_0_co_2.xml
• UCAR MetEd, Instrumentation and Measurement of Surface Precipitation: https://www.meted.ucar.edu/instrumentation/precipitation/print.php