How the River Pang reacts to Rain

How the River Pang reacts to Rain Last generated: 24 August 2026, 07:11

Testing whether our Natural Flood Management (NFM) work has changed it

Since 2019, the Pang Valley Flood Forum and partners have built a series of Natural Flood Management (NFM) schemes - leaky dams and similar "slow the flow" measures - on tributary streams feeding the River Pang. See the full list and photos on our Projects page.

In 2021, PVFF carried out a statistical analysis to test whether these schemes were reducing how much the river rises after rain (the original PVFF NFM Final Report - Statistical Analysis v1.0), but with only 16 usable rain events since the first scheme, it couldn't draw a conclusion either way. It's now several years further on, so this analysis re-runs that method with everything recorded since - generated automatically from the underlying rain and river gauge data. The schemes feed two separate river gauges, so the results are split across two pages, one per gauge:

The conclusions below are based on data up to the "Last generated" date shown above, and the dataset is still relatively small - a few years' worth of rain events, not decades. As more events are recorded, this page and the two gauge pages update automatically, so the analysis and its conclusions can be sharpened, and may change, as more data is gathered.

Tidmarsh / Bourne NFM

Has the Bourne NFM scheme (38 leaky dams, completed Dec 2019) changed how the Tidmarsh gauge responds to rain?

Can't reliably test this at Tidmarsh

See the full assessment →

Bucklebury / Elmwood + Redhill Copse NFM

Has the combined set of Elmwood (Sep 2020) and Redhill Copse (May 2023) schemes, and other upstream changes, made a difference to the river response to rain at the Bucklebury gauge?

The river is rising less at large rain events

See the full assessment →

All the interventions

Scheme	Stream	Affects gauge	Completed
Bourne NFM - 38 leaky dams	River Bourne	Tidmarsh	20 Dec 2019
Elmwood NFM - 24 leaky dams, log barriers & "dragon's teeth"	Elmwood Stream	Bucklebury	25 Sep 2020
Redhill Copse NFM - 13 leaky dams	Osgood's Gully	Bucklebury (same gauge as Elmwood)	1-31 May 2023

Bourne feeds into the Pang above Tidmarsh, and both Elmwood and Redhill Copse feed in above Bucklebury, so this splits into two separate questions: has Tidmarsh's response to rain changed since Bourne, and has Bucklebury's changed since the combined set of Elmwood and Redhill Copse. The Bucklebury Waven Field scheme (2025) manages village surface water and isn't expected to affect the river gauge, so isn't tested here.

What this does and doesn't tell us

This is a before/after comparison of the one river that exists, using its own monitored record across time. Anything else that changed over the same ~7 years (rainfall patterns, land use elsewhere in the catchment, sensor drift) could show up alongside an apparent "scheme effect" here - a known example for Bucklebury: alterations to Bucklebury Common's tree and vegetation cover (upstream of Elmwood) since 31 May 2023, alongside the NFM works - additional leaky dams were built there with the intention of offsetting that reduced cover. Where a result shows the river rising more or less since a scheme was built, it may not be possible to attribute that to the NFM works alone - only to the combined set of changes in the catchment over that period.
The Bucklebury page's comparison mixes together the effects of two schemes (Elmwood, then Redhill Copse) by design - that's the question it's answering: have the NFM works, taken together, made a difference - not either scheme in isolation.
Where a result isn't statistically significant, that means the data can't currently rule out "no effect" - it does not mean the scheme has no effect, just that we can't yet be confident either way.
Thatcham's weather station data has known problems: its feed has occasionally produced obviously-wrong single readings (e.g. 220mm+ of "rain" in one 15-minute interval), which this analysis filters out, and PVFF isn't fully confident the station's software has always converted local time to GMT/UTC consistently across its history (the operator may have changed the underlying software without realising that mattered). A check of Thatcham's rain timing against Yattendon's across every year in the dataset didn't show a clear, consistent sign of a timestamp problem, but it's inherently a noisy check and can't rule one out for certain. Both pages show the fit quality for each of the 3 gauges on pre-scheme data, so which gauge was picked (and how it compares to the others) is visible before trusting any result - but the "after" comparison itself only ever uses the one gauge picked from the "before" data, not a re-check across all three.
River levels are measured by radar gauges, accurate to 1mm - the limitation isn't measurement precision, it's that they record water level, not flow volume (there's no simple, accurate way to convert a level rise in cm to a volume in m³/s for these gauges), so "rise in cm" is what this analysis works with throughout, the same limitation the original 2021 report had.
Above ~74cm, the Bucklebury gauge under-reads: the river overflows upstream of the gauge there, so the raw reading no longer reflects the true level - this is why the original 2021 report excluded events peaking above 70cm entirely. The Bucklebury page now corrects for this (a linear adjustment above the critical level, calibrated against one independent point - the 2014 event's citizen-measured village depth against the simultaneous raw gauge reading) so the largest events on record can be included rather than silently dropped. The very largest, 19-21 July 2007, still falls outside the Bucklebury page's "before" comparison for an unrelated reason - it predates the 2011 flood alleviation channel, and the "before" period there is restricted to after that channel was built (see that page's Methods). Treat the overflow correction itself as the best currently-available correction, not a precisely-instrumented one - see PVFF's recorded modelling assumptions for the full parameter set.
Events are excluded from both pages where the West Berkshire Groundwater Scheme was running - EA drought-relief pumping into the Pang above Bucklebury that raises Bucklebury, Tidmarsh and Pangbourne readings independent of rainfall (see the note on the dashboard when it's active). The underlying record of when this has run only covers one known episode (24 Oct-14 Nov 2022), so this may not catch every pumping period across the full dataset.
This treats each scheme as a single before/after step change, but the schemes themselves aren't static once built - leaky dams degrade over time, and streams can erode around or under them, changing how effective they are years after completion. A "post-scheme" result averages over however well-maintained the structures were across the whole after period, so it doesn't distinguish a scheme working as designed from one whose effect has faded since it was built - maintenance records aren't currently part of this analysis.

As more rain events accumulate, both pages update automatically to sharpen these estimates - unlike the 2021 report, nothing here needs to be redone by hand.

How it works

Rain events

A "rain event" is a spell of rain preceded and followed by at least 8 dry hours (so separate downpours aren't run together). Events totalling less than 5mm are ignored as noise. For each event we record its length, total rainfall, peak 15-minute intensity, and how much rain fell in the preceding 7 days (a proxy for how saturated the ground already was). A "hard frost in the days before" flag was also tried, since the 2021 report noted several outlier events followed cold snaps - but no version of it (tested at several lookback windows) could be reliably estimated from the data, so it was dropped; see the Bucklebury page's methods for why.

The model, and when we don't trust it

One multi-linear regression predicts the gauge rise from those event characteristics plus the month of the year, fitted separately on each of three rain gauges (Yattendon, Thatcham, and a blend of the two) using only pre-scheme ("before") events. We tested all three; whichever gauge's model fits that pre-scheme data best by cross-validated R² (a fairer measure of real predictive skill than in-sample R², since it's tested on data the model wasn't fitted on) is the only one used for everything that follows.

If none of the three reaches a cross-validated R² of 50%, the before-model isn't trustworthy enough to build an after-comparison on - that page reports no verdict at all, rather than showing a comparison against a prediction we don't trust. This varies genuinely by catchment: a gauge with more, or faster-reacting, upstream inflows is inherently harder to predict from rainfall alone than a simpler one - it isn't a flaw in the method.

Testing the effect

Rather than just counting how many post-scheme events came in higher or lower than the pre-scheme model predicts, the primary result on each page pools all before-and-after events into one regression with a "post-scheme" flag (and its interaction with rain amount, so the effect isn't forced to be a flat shift) - giving a direct significance test of the scheme's effect. This supersedes PVFF's original 2021 report's method (a simpler model, gauge picked by in-sample R², no formal significance test) entirely; the numbers here aren't comparable to that report's.