Leakin-run Analysis
2025-10-08
John Dinnall
Site Description
Catchment: Leakin Run is an urban headwater stream in Baltimore City, Maryland (39°17’55.88”N, 76°41’44.72”W). It drains a short reach of approximately 0.15–0.25 km along a residential corridor before feeding into a tributary of Dead Run within the Gwynns Falls watershed. The catchment transitions from urban forest and institutional open space into low-density residential land use. A mature riparian canopy provides shade and bank stabilization, although the channel shows localized erosion from surrounding roadway runoff, a pattern typical of urban stream syndrome.
Sensor Deployment: Five sensors were deployed at this site: four in-stream sensors at the Downstream, Midstream, Upstream, and Upstream 2 positions, and one Air sensor mounted as a dry atmospheric reference. The two upstream positions belong to the same channel network and converge near the midstream sensor. All in-stream positions had active flow at the time of deployment.
Goal: To document flow patterns at Leakin Run and assess whether low-cost thermal sensing is a viable approach for detecting surface water presence and flow permanence at each sensor point.
Data Note: The downstream sensor was lost during this observation period. Because of how the processing code is structured, data from the second upstream location (Upstream 2) are recorded in the downstream position for this site. Upstream 2 is not a true downstream location; it is a naturally formed channel with denser riparian cover in its immediate vicinity. In the summary figures below it is labeled “Upstream 2,” but in the four-panel diagnostic plots it appears under the “Downstream” label.
Study Area Map
Figure 1: Study area map of Leakin Run, Baltimore City, Maryland. Sensor points indicate the locations of the in-stream temperature loggers deployed along the channel. The inset map shows the location of Leakin Run relative to all study sites in the Maryland Piedmont.
Figure 2: Daily precipitation recorded at Leakin Run from July 2025 through May 2026.
Figure 3: Stacked bar chart showing the proportion of days classified as wet (blue), dry (red), and uncertain (gray) for each sensor position at Leakin Run across the study period.
Figure 4: Daily dryness state classifications for all sensor positions at Leakin Run from May through August 2026. Each row represents a sensor position (Air, Midstream, Upstream, Upstream 2) and each column represents a single day. Blue indicates wet, red indicates dry, and gray indicates uncertain.
Sensor Images
05/29/26
Figure 5: Sensor positions at deployment in late
spring. The Upstream position is fed by a culvert outlet, while Upstream
2 is a naturally formed channel with more riparian cover in its
immediate vicinity. Both the Midstream and Downstream positions had
active flow.
10/03/26 (Retrieval)
Figure 6: Sensor positions at retrieval in early fall. The Upstream position had a gentler current and Upstream 2 was reduced to a trickle. The Midstream position had moderate flow. The Downstream sensor was missing, but flow was still present at that location.
Thermal Overview Plots
Figure 7: Temperature recorded at 15-minute intervals by all four sensors at Leakin Run from June through August 2026. The Air sensor reflects ambient atmospheric conditions, while the three in-stream sensors reflect channel conditions at each position. Air shows the largest fluctuations and the warmest overall temperatures. Upstream 2 shows the opposite, recording the lowest overall temperatures of any sensor.
Figure 8: Daily mean temperature for all four sensors at Leakin Run from May through August 2026. Averaging to a daily mean smooths short-term fluctuations and reveals seasonal trends. Air ranges from the high 60s to about 80°F, while the in-stream sensors vary within a narrower range of about 5°F.
Case studies
Purpose: To illustrate sensor response during different atmospheric and seasonal conditions. Each case study presents a four-panel diagnostic plot followed by a dryness score table for a single selected day.
4-Panel Plots: A four-panel diagnostic figure showing each sensor’s thermal behavior over a 24-hour period.
Panel 1 shows raw temperature readings.
Panel 2 shows temperature rise above each sensor’s morning minimum.
Panel 3 shows the rate of temperature change per minute.
Panel 4 shows the computed dryness score for each sensor. Scores near 0 indicate wet conditions, scores near 1 indicate dry conditions. Dryness Score table: A summary of each sensor’s thermal classification for the selected day.
Warming score and cooling score reflect thermal response during the morning and afternoon phases respectively.
Daily score is the average of both.
A conflict flag indicates the two phases produced contradictory results.
Dryness state is the final classification — wet, dry, or uncertain.
July 20th (Normal day) Close-up
Figure 9: Four-panel diagnostic plot for July 20th,
2026, a hot summer day with no precipitation and regional air
temperatures of 63–85°F. The shaded Air sensor rose more than any other
sensor, climbing from roughly 65°F to 75°F in the raw panel and about
9°F above its morning minimum. In the rate-of-change panel, the sensor
labeled Downstream (Upstream 2) shows the steepest drop, at
approximately 5 PM.
| Sensor | warming_score | cooling_score | daily_score | conflict_flag | wetness_state |
|---|---|---|---|---|---|
| Air | 1.0000 | 0 | 5e-01 | TRUE | uncertain |
| Downstream | 0.0000 | 1 | 5e-01 | TRUE | uncertain |
| Midstream | 0.0000 | 0 | 0e+00 | FALSE | wet |
| Upstream | 0.0011 | 0 | 6e-04 | FALSE | wet |
Table 1: The Air and Downstream (Upstream 2) sensors both received conflict flags because their warming and cooling scores pointed in opposite directions. For Air, this likely reflects its strong afternoon warming for Upstream 2,further reflecting the brief spike visible in its rate-of-change panel. The Upstream and Midstream sensors were both classified as wet with no conflict flags.
August 5th, 2025 (Precipitated day ) Close-up
Figure 10: Four-panel diagnostic plot for August 5th,
2026, a day with precipitation and air temperatures of 74–88°F. Rain
began at roughly 6:00 AM. In the raw and normalized panels, the
in-stream sensors do not warm steadily like the Air sensor, instead they
alternate between rising and falling throughout monrning to ealry
afternoon.
| Sensor | warming_score | cooling_score | daily_score | conflict_flag | wetness_state |
|---|---|---|---|---|---|
| Air | 0e+00 | 0.0000 | 0.0000 | FALSE | wet |
| Downstream | 8e-04 | 1.0000 | 0.5004 | TRUE | uncertain |
| Midstream | 0e+00 | 0.0000 | 0.0000 | FALSE | wet |
| Upstream | 1e+00 | 0.4994 | 0.7497 | TRUE | dry |
Table 2: The Air and Upstream sensors both received conflict flags. Rainfall disrupted the normal daily pattern of morning warming followed by afternoon cooling, and because the classifier relies on that pattern, the warming and cooling phases produced contradictory scores.
Analysis
This deployment captured Leakin Run during the height of summer baseflow. Over the study period, Leakin Run showed the most consistent surface flow at its midstream location, which was classified as wet on roughly two-thirds of days and dry on only a few. The upstream location was wet on 58.1% of days and dry about 20.9%. Upstream 2 was the most thermally responsive stream sensor, classified as dry 34.9% of the days, approaching the Air sensor’s dry frequency. This is notable because Upstream 2 sits within denser riparian cover and recorded the coolest temperatures of any sensor, both of which would be expected to buffer temperature swings rather than amplify them. Combined with the trickling flow observed at retrieval, this suggests that its dry classifications may reflect very shallow or intermittent surface water rather than solar exposure. All sensors, including Air, showed a substantial share of uncertain days (roughly a quarter to a third), and Air itself was classified as wet 31.4% of the days, likely on overcast or rainy days with small temperature swings. Both points limit confidence in individual daily classifications. Because Leakin Run was monitored only during summer, these results cannot capture seasonal recovery. During this period, the midstream point behaved consistently with perennial flow, while both upstream points showed intermittent behavior.