How long the chalk remembers the rain

The question

When it rains, when does the groundwater rise? And if a dry winter is followed by a wet spring, does the aquifer forget?

Every account of a drought turns on this and almost none of them quantify it. Rainfall returning to normal is reported as a drought ending, which is true of the weather and can be false of the water supply — a chalk aquifer at the bottom of its range takes more than a wet month to refill, and the borehole is where that shows. Answering it needs a rainfall record and a groundwater record long enough that the answer is about the aquifer rather than about the particular years that were sampled, on the same time axis, over the same place.

The mechanism this rests on

What the record shows

Rainfall and groundwater level

England rainfall above Chilgrove House, aligned at month resolution over 1836-01 to 2026-05. Separate vertical scales: the two units are not comparable and no height here may be read against another.

Show relevant events (4 events)Shading marks when something was recorded, not what caused the line.

What the shaded periods are (4 events)

A shaded period marks when something was recorded, not what caused the shape of the line. Where the two coincide, they coincided.

Documented record — A documented event, compiled from cited sources after it happened. The dates are as good as the record, which for older events can mean the month rather than the day.

Event	Type	Period	Standing	Authority
The 1976 drought	Drought declaration	1975-05-01 – 1976-08-31	Documented record	Compiled record
Yorkshire drought	Drought declaration	1995-04-01 – 1996-03-31	Documented record	Compiled record
The 2022 drought	Drought declaration	2022-03-01 – 2022-10-31	Documented record	Compiled record
European drought	Drought declaration	2022-05-01 – 2022-10-31	Documented record	Compiled record

Not shown on this chart

Cantareira water crisis, São Paulo — This event’s area does not contain this chart’s subject. Sharing a country is not sharing an event, so it is not drawn.
Cape Town Day Zero — This event’s area does not contain this chart’s subject. Sharing a country is not sharing an event, so it is not drawn.
Hurricane Katrina water and sanitation collapse — This event’s area does not contain this chart’s subject. Sharing a country is not sharing an event, so it is not drawn.

View the data behind this chart

Decade means over the aligned month axis, not the values themselves — a table of every period would be longer than this page. The values are in the download and the API.

Decade	Monthly precipitation (mm)	Groundwater level (m)	Periods held
1830s	69.0	47.20	48
1840s	66.9	47.84	120
1850s	62.7	45.19	120
1860s	68.0	48.78	120
1870s	72.0	50.67	120
1880s	68.4	49.69	120
1890s	64.1	47.05	120
1900s	65.5	47.19	120
1910s	71.6	51.02	120
1920s	70.4	50.16	120
1930s	70.1	49.12	120
1940s	65.7	47.10	120
1950s	69.3	48.92	120
1960s	71.2	51.08	120
1970s	66.0	46.69	120
1980s	70.4	49.89	120
1990s	69.0	47.09	120
2000s	74.4	49.05	120
2010s	72.1	49.63	120
2020s	76.6	51.93	77

Sources: Met Office — HadUK-Grid gridded climate observations (Open Government Licence v3.0); Environment Agency — Hydrology Data Explorer API (Open Government Licence v3.0). Aligned by coarsening the finer series to month resolution using each metric’s declared temporal rule; no value was interpolated.

The top panel is England’s monthly rainfall; the bottom is the water level in the borehole. Read the timing and the shape, never one height against the other — the two are in different units on separate scales, and that is the whole reason they are not on one plot.
The rainfall panel is noise at this resolution, as rainfall is. The groundwater panel is smooth, because the aquifer integrates months of recharge into one level. That difference in texture is the finding before any statistic is computed.
The correlation, in the panel below the chart, is computed on departures from each series’ own seasonal cycle rather than on the raw lines. Both series follow the calendar hard, and a coefficient between the raw series would be mostly the shared year.
The lag at which that correlation is strongest is the aquifer’s response time as this record measures it. Compare it against the multi-year droughts in the event overlay: the deepest levels do not sit under the driest single months.

The association, measured

Computed on each series' departures from its own seasonal cycle, because both series follow the calendar hard and a coefficient between the raw lines would mostly be the shared year.

Lag tested	Correlation	Overlapping pairs
Same period	0.301	2,192
1 month later	0.436	2,191
2 months later	0.334	2,190
3 months later	0.262	2,189
6 months later	0.118	2,186
12 months later	0.003	2,180

Strongest term available for this result: lagged-association. Method signal-lagged-association v1.0.0. Hydrionis has no code path that promotes a correlation to a cause; a causal claim requires external evidence, cited.

Everything that has to be said about this

A groundwater level is measured against a datum, and levels at different boreholes are not on the same datum and must not be read as one series. Recharge is strongly seasonal and largely confined to winter in temperate climates, so summer rain can be substantial and still move the level not at all. Abstraction moves the level too, and is not in either line.
These two quantities are not in the same unit and are on separate panels for that reason. They share only the time axis: the apparent size of any response depends on the scaling each panel was given, so read the timing and the shape, not the heights against one another.
Expected timing: Months, and in a deep or confined aquifer years.
1 of 2 panels have gaps in the common period; a gap is drawn as a break, never bridged.
Not every panel is a measurement. Modelled and aggregated values are shown with their basis and must not be read as gauge readings.
An association, not a mechanism and never a cause. Both series are usually strongly seasonal, and two seasonal series correlate whether or not either affects the other, so a correlation computed on raw values mostly measures the shared calendar. The lag at which r peaks is a property of these two records over this period, not a physical travel time. Refused below 30 shared periods.
Computed over 2,191 overlapping month periods between 1836-01-01 and 2026-05-01.
Each series had its own seasonal cycle removed before correlating, so this is a relationship between departures from normal, not between the raw series.
The two subjects are related as: Chilgrove House lies within England. That is a stated spatial fact, not evidence that one drains to the other.
Chilgrove House lies within England.
A shaded period marks when something was recorded, not what caused the shape of the line. Where the two coincide, they coincided.

What would make this wrong

Where the figures come from

Every number on this page was computed from the observations Hydrionis holds at the moment the page was produced, using methods declared in the methodology register. None of them is typed into this page, so none of them can disagree with the data behind it. The underlying series are in the dataset catalogue and reachable through the API.