Does rainfall drive WWTP sludge pumping?

From the QuickBooks ledger, June 2025 through March 2026: 29 sludge-removal transactions totaling roughly $107,000. Of those, 16 are categorized “emergency” and 4 “routine.” Emergency rate ≈ $323/kgal; routine rate ≈ $194/kgal. The FY 25-26 adopted sewer budget assumed $5,500 for the full year; the projected actual is $60,837 — a 10× overrun, with a mid-year amendment of $5K already approved.

The operator’s memos on the emergency invoices repeatedly describe the pumps as “EQ tank pump-down” or “sewer plant pump” — language consistent with drawing down the equalization tank when the plant can’t process inflow fast enough. If that’s the mechanism, the trigger should be antecedent rainfall.

For each pump-out date, we compute cumulative rainfall over the three days before the event and compare against the same statistic for every non-pump-out day in the same period. If pump-outs were rainwater-driven, pump-out days would sit in the wetter tail of the rainfall distribution.

A: No — and the difference is statistically significant in the opposite direction. Pump-out days had a 3-day-prior median of 0.00″ vs 0.08″ on typical days (Mann–Whitney U test, p ≈ 0.02). Wider 7-day and 14-day windows give the same qualitative answer with smaller effect size. Whatever is triggering pump-outs, it is not recent rainfall.

The most plausible mechanism for the dry-day clustering is operator scheduling: trucks need access roads dry, and routine metrics (MLSS readings, sludge consistency) are easier to measure outside storm events. So the EQ tank is drawn down at operator convenience, not in response to hydraulic pressure.

Weakly. To find out which rainfall window best predicts plant flow, the heatmap below tests every (start, end) cumulative window between days[d − start] and day[d − end] and colors each cell by its Pearson correlation with flow on day d. Red Hook Commons has a hot bottom row — a slow-infiltration response consistent with groundwater leaking into the collection system over many days. Village WWTP is flat across the whole grid: no window of rainfall predicts its flow at any lag.

Picking each plant’s strongest cell from the heatmap and plotting daily flow against rainfall over that window quantifies the slope — extra MGD of flow per inch of rainfall. The Commons line is real but shallow; the Village line is statistically indistinguishable from flat.

Even accepting the Commons response, a typical 1″ storm spread over two weeks adds only a few percent to base flow. The histogram below puts that in context: it shows every day of measured plant flow against each plant’s SPDES design capacity (0.050 MGD at the Village WWTP via sub-outfall 01A; 0.025MGD at Red Hook Commons via sub-outfall 01B). Days that fall to the right of each plant’s dashed capacity line are days the plant ran above its permitted monthly-average design flow — the operational condition that would justify drawing down the EQ tank.

So rainfall does affect flow at one of the two plants, by a small amount — and capacity excursions are rare even at the smaller Commons plant. Neither is in a regime that would explain a 10× budget overrun on pump-out volume.

The DEC monthly DMR forms include a column for the operator’s gauge readings of daily rainfall. Anyone looking at the forms would naturally use that column for any rainfall analysis. We did, initially. The first pass of this analysis showed an apparent weak rain-flow signal at Village WWTP, which seemed to support the rainwater-pump hypothesis.

Once we cross-checked the operator’s readings against NOAA observations from a CoCoRaHS station 1.4 miles from the village, the operator-reported series turned out to be unreliable: 11% lower in total rainfall, ~5% of days disagree by more than half an inch, and many days have a major storm recorded by NOAA that the operator’s column shows as 0.00″. The apparent Village fast-inflow signal was an artifact of those inconsistencies.

Switching to NOAA as the source of truth removed the apparent signal and revealed the dry-day clustering above. A side effect of this finding: the operator’s rainfall column on the DMR is itself unreliable enough to be of concern to NYSDEC, independent of the pump-out cost question.

We cannot tell from the data the village currently has. The QuickBooks memos describe pumps but not their causes. We can’t reliably split events between the Village WWTP and Red Hook Commons. Daily flow data on the DEC forms is too coarse to identify process upsets in real time. The candidates we can’t rule in or out:

• Equipment failures or maintenance backlog — pump-outs cluster in time (4 events in August, 4 in December, 4 in January), which fits an equipment-issue pattern more than a steady-state operating pattern.
• Biological process upsets — sudden MLSS drops or WAS-rate changes around event clusters would point to the activated-sludge tank failing, with the EQ tank used as a hydraulic buffer to keep influent off the bug population.
• Operator scheduling thresholds — the EQ tank may be drawn down at a fixed level the operator chose, regardless of what was filling it. If so, the cost is a function of operator choice, not plant condition.
• Base-load drift — village sewer base flow may be growing fast enough to fill the EQ tank more often, with neither weather nor equipment as the proximate cause.

To answer “what is driving the pump-outs” with the rigor the budget overrun warrants, the village would need to publish, per pump-out event:

• The plant pumped (Village WWTP / Commons / both).
• The operator’s stated cause (process upset, equipment failure, scheduled maintenance, hydraulic load).
• The EQ tank level at the time of the call and after the pump.
• The volume actually hauled (vs the volume billed, since three of the larger emergency invoices use rate-divided volume rather than measured volume).

None of this requires special equipment. It does require the operator to keep a log and the village to publish it.