Soil water content at the forest floor

Half-hourly volumetric soil water content at five depths of the FF1 soil profile at CH-LAE, 2004-2025. The file is 09_METEO_SWC_FF1_2004-2025 (parquet and CSV): 373,728 records on a continuous 30-minute middle-timestamp index (named TIMESTAMP_MIDDLE in the file) in local time (UTC+1, no daylight saving), from 2004-09-07 00:15 to 2025-12-31 23:45. Values are in % VWC. The index begins with the soil record rather than on 1 January, because there is no soil moisture at this plot before September 2004.

The interactive dashboards summarise one depth each on a single page — coverage and provenance, seasonality, distributions, extremes and trends, with a table view behind every chart. They are standalone and work offline: 0.05 / 0.1 / 0.2 / 0.3 / 0.5 m.

Method, evidence and checks:

09_METEO_SWC_FF1 — builds the product, splices the sensor generations, and tests the older era for internal breaks.
The screening notebooks of the current profile, one per depth, which correct the raw high-resolution record before this product reads it: 0.05 / 0.1 / 0.2 / 0.3 / 0.5 m, and the profile download they share as a cross-check reference.
Meteo_Product_Chain — where this product sits in the chain.

Which column to use

No column is homogeneous across April 2020, when the whole soil profile was replaced. The step between the two profiles is +7.6 to +10.7 % VWC depending on depth, and it is an instrument change, not weather.

Inside one profile, use SWC_FF1_<d>_1, which is what the probe recorded.
Across April 2020, use SWC_FF1_<d>_1_HOMOGENIZED — but read The 2020 profile replacement below first, because that column rests on climatology alone and is a derived estimate rather than a measurement.
At 0.5 m there is no homogenised column, because that depth has only one profile.

Nothing in this product is gap-filled, and coverage varies strongly by depth, from 80.8 % at 0.05 m to 96.0 % at 0.2 m over the full period. A missing record is NaN, never 0.

Columns

Fourteen columns: one measured value, one homogenised twin and one flag per depth, except at 0.5 m, which has no homogenised twin. They are grouped by depth in the file, so a value and its provenance stay adjacent. <d> is one of 0.05, 0.1, 0.2, 0.3, 0.5.

Table 1: Columns of 09_METEO_SWC_FF1_2004-2025.

column	unit	description
`SWC_FF1_<d>_1`	% VWC	Soil water content as the probe recorded it, screened. Not homogeneous across April 2020.
`SWC_FF1_<d>_1_HOMOGENIZED`	% VWC	The same series with the pre-2020 era shifted onto the level of the current profile. A derived estimate. Absent at 0.5 m.
`FLAG_SWC_FF1_<d>_1_SOURCE`	-	Which sensor generation measured the value.

A value is present exactly where its flag is greater than 0, and the homogenised column is present exactly where the measured one is.

`FLAG_SWC_FF1_<d>_1_SOURCE`

The codes name a sensor generation, not a gap-filling model: any code above 0 means this depth measured this value. Filter on > 0 for measurements.

Table 2: Source codes and their record counts per depth.

code	sensor generation	0.05 m	0.1 m	0.2 m	0.3 m	0.5 m
0	no measurement: a gap, or before this depth’s record begins	71,670	36,422	14,865	50,638	275,140
1	EC-20 profile, screened with the MeteoScreeningTool	207,596	238,767	260,311	239,424	0
2	TEROS 12 `FF1` profile, installed 19 March 2020	94,462	98,539	98,552	69,152	98,588
3	TEROS 12 replacement probe, 40 cm downslope	0	0	0	14,514	0

Code 1 runs from 2004 to each probe’s own death, between December 2018 and March 2020. Code 2 begins on 2020-04-10 at every depth. Code 3 exists at 0.3 m only, from 2025-03-04. The whole of 2004-2020 is code 0 at 0.5 m, because the older profile had no probe at that depth — no instrument, rather than missing data.

Coverage

Coverage differs by depth more than for any other meteo variable, because each probe failed on its own schedule and was not replaced until the whole profile was.

Table 3: Coverage and gap structure since each depth’s own record begins. A gap is measured as the interval between the values surrounding it.

depth	record begins	coverage	gaps	gaps > 30 d	longest gap	ends	range	mean
0.05 m	2004-09-07	80.8 %	100	6	695 d	2015-06-03	7.7-39.7	23.5
0.1 m	2004-09-07	90.3 %	105	3	480 d	2020-04-10	0.8-41.7	22.9
0.2 m	2004-09-07	96.0 %	108	2	92 d	2009-06-11	7.3-39.0	23.6
0.3 m	2004-09-07	86.5 %	70	6	328 d	2025-03-04	3.3-36.4	21.1
0.5 m	2020-04-10	98.2 %	31	0	14 d	2024-04-23	18.9-31.7	24.1

Means and ranges mix the two profiles and are given for orientation only. Three of the five longest gaps are hardware rather than downtime: 0.1 m waits 480 days between its old probe dying and the new profile starting, 0.3 m waits 328 days for its replacement in 2025, and the 695 days at 0.05 m are the unexplained absence discussed under Known limitations.

The 2020 profile replacement

The five TEROS 12 probes of the current profile were installed on 19 March 2020 and began reporting on 10 April. The EC-20 probes they replaced had failed one by one over the preceding year and a half, so the two generations never overlap at any depth: the interval between them runs from 22 days at 0.2 m to 480 days at 0.1 m. That interval is a dead probe, not an outage, and it is not filled.

Because nothing measured through the break — no second profile, no weather service, and not the other depths, which changed on the same day — the step cannot be calibrated the way the precipitation step at 2018 can. The _HOMOGENIZED columns shift the older era onto the current level using additive calendar-month offsets fitted on climatology alone, which assumes the two periods had the same monthly soil-moisture climate. Use them for continuity and trend work that has to cross 2020. Do not read a homogenised pre-2020 value as an estimate of what the soil held at the time.

The four homogenised columns are also not on one reference period: each is shifted onto its own depth’s current-profile era, and at 0.3 m that era ends at the April 2024 probe failure while elsewhere it runs to the end of 2025. Within a depth this changes nothing; differencing between depths across 2020 carries the difference.

The 2015 break in the older era

The older era is not one level either, and this matters for anyone using it. On 25 June 2015, between 08:15 and 09:15, all four depths drop by about 3.9 % VWC at once and stay down. The maintenance record has people working on the forest-floor logger’s wiring that morning, and three weeks earlier it records the plot’s power supply being replaced and the soil-moisture probes’ connectors being found loose. These probes were read as a voltage scaled by a fixed multiplier, so a poor connection in that path shifts the reported level directly.

The break is not corrected, because removing it alone would leave a larger problem in place. Underneath it, both the wettest and the driest conditions of each year rise steadily at every depth through 2005-2015, by +0.5 to +1.2 % VWC per year — five to nine % VWC in total — while precipitation over the same decade trends slightly downward. The driest half-hour of a year is a property of the soil, so a record whose dry end nearly doubles is not tracking the weather. It is consistent with the site inventory, which flags three of the four old probes as defective. A drift has no date at which a correction could be applied and no reference to fit against, so it is documented rather than removed, as for the tower pyranometer on the SW_IN page and the PAR sensor on the PPFD_IN page.

The practical consequence: a level comparison inside the pre-2020 era across June 2015 is not a comparison of measurements, and a trend fitted through that era is dominated by instrument drift rather than by climate.

No break at the January 2016 site renewal

The January 2016 tower renewal moved air temperature, relative humidity and air pressure, and a program change that June left much of the longwave record reading low. It did not affect this profile, which is on the forest-floor datalogger rather than the tower one: the tower program contains no soil-moisture instruction at all, the profile lost no records while the tower logger was replaced, and neither the level nor any individual half-hour moves at the date. The notebook asserts all three, so the question is settled rather than open.

Known limitations

Depths are not replicates. Each is a single probe at one point in one profile. They share a logger, a power supply and an SDI-12 bus, so they fail together; but they do not measure the same soil, and their disagreement is information rather than error.
0.3 m carries three sensor generations and steps twice. The original probe stopped communicating in April 2024 and, in failing, took the whole profile’s SDI-12 port down, which is why every depth has an April 2024 gap. Its last value in this file is 2024-04-09 16:45, and the 328-day gap that follows ends when a replacement went in on 4 March 2025 40 cm down the slope, in different soil, reading about 3 % VWC high relative to its neighbours. Code 3 marks it; the measured column keeps the step and the homogenised column removes it.
The pre-2020 era was screened once, with the deprecated MeteoScreeningTool, and never re-screened. Values outside the physical range are removed here — failed sensor reads, which that hardware stores as a fixed negative constant rather than as a gap — but a probe that stopped responding to rainfall while still reporting plausible values would not be caught. The per-depth screening notebooks run that test for the current profile only.
The two largest gaps in the older era were power failures, documented in the site record: 10 March to 11 June 2009 and 6 March to 28 May 2010, when a fuse slipped out and the battery-powered forest-floor logger lost power. The forest floor had no site record at all before 2006, so the first two years of the soil record are unadjudicable in principle.
The nearly two-year absence at 0.05 m (July 2013 to June 2015) has no explanation anywhere in the site record, and unlike the 2009 and 2010 gaps it affects one depth only. It is why 0.05 m has the lowest coverage and the longest gap of the five depths.
Sub-daily timestamps in August 2012 carry the same uncertainty as the tower variables. A logger clock error that month is corrected in the tower products. Whether the forest-floor logger shared it is not recorded, and the fault is undetectable in soil moisture, which has no diurnal cycle to align against. Nothing is applied.
FLUXNET’s SWC_F_MDS_1..4 are not this product. They splice the same two profiles with the 2020 step passed straight through, and flat-fill some stretches at a constant value. The source flag exists so that this product cannot be used the same way by accident.