Meteo Data

Thirteen meteo products from the CH-LAE tower and forest-floor soil profile, each a half-hourly file with its value columns and provenance flags. Products 01 to 07 run to 30 September 2026; products 08 to 13 still end on 31 December 2025. This page covers what all products share. Each variable has its own page with its columns, units, coverage, flag codes and known limitations.

Pages for the individual variables

An entry without a working link is a page or dashboard not yet written, not a variable missing from the dataset.

Table 1: The per-variable pages and dashboards.

parameter	measures	page	dashboard
`SW_IN`	incoming shortwave radiation	Incoming shortwave radiation at 47 m	`SW_IN`
`TA`	air temperature at 47 m	Air temperature at 47 m	`TA`
`PPFD_IN`	photosynthetic photon flux density	Photosynthetic photon flux density at 47 m	`PPFD_IN`
`RH`	relative humidity	Relative humidity at 47 m	`RH`
`PA`	air pressure	Air pressure at 47 m	`PA`
`LW_IN`	incoming longwave radiation	Incoming longwave radiation at 47 m	`LW_IN`
`VPD`	vapour pressure deficit	Vapour pressure deficit at 47 m	`VPD`
`PREC`	precipitation	Precipitation at 47 m	`PREC`
`SWC`	soil water content, five depths	Soil water content at the forest floor	0.05 / 0.1 / 0.2 / 0.3 / 0.5 m
`TS`	soil temperature, seven depths	Soil temperature at the forest floor	0.05 / 0.1 / 0.15 / 0.2 / 0.3 / 0.5 / 0.6 m
`G`	soil heat flux, three plates	Soil heat flux at the forest floor	`G`
`SW_OUT`	outgoing shortwave radiation	Outgoing shortwave radiation at 47 m	`SW_OUT`
`LW_OUT`	outgoing longwave radiation	Outgoing longwave radiation at 47 m	`LW_OUT`

A dashboard is one standalone page per exported series: coverage and provenance, seasonality, distributions, extremes, trends and, where one exists, the comparison against the reference station, with a table view behind every chart. It works offline. Where a product exports both a measured and a _HOMOGENIZED column, the dashboard uses the homogenised one and names it at the top.

The merged file and FP2026.1

Notebook 99 joins all thirteen products onto one half-hourly index. The file is 99_METEO_MERGED_2004-2026.09 (parquet and CSV): 398,832 records from 2004-01-01 00:15 to 2026-09-30 23:45, with 61 columns, 31 data series and 30 flag columns. The values are exactly what each product notebook exported; the merge computes and corrects nothing.

01 to 07 (SW_IN, TA, PPFD_IN, RH, PA, LW_IN, VPD) run to 30 September 2026.
08 to 13 (PREC, SWC, TS, G, SW_OUT, LW_OUT) still end on 31 December 2025. Their columns are NaN from January to September 2026 until their notebooks are extended.
For TS and G the merged file carries only the reconciled series, one per depth and one plate mean. The individual channels and plates stay in the product files.

The file 99_METEO_MERGED_2004-2026.09_INFO.md, written beside the data, lists every column with its unit, range and flag codes, and the reason each series has gaps. Read the column list there; this page does not repeat it.

FP2026.1 carries all 61 meteo columns of the merged file, cut to the FP2026.1 period, 1 January 2016 to 30 September 2026. The six products that end in 2025 are therefore empty in FP2026.1 for January to September 2026. See Dataset versions.

Calendar explorer

Every month of the record on one grid

One tile per month, coloured by the selected metric and marked with what was notable in that month. The grid is drawn at three resolutions: months, meteorological seasons, and single days. The daily raster is the only one of the three that does not split a spell at a boundary. Open a month for its statistics against the calendar-month normal, its rank among the same month of the other years, and a calendar of its days; open a day for its statistics, flags and diurnal course.

The explorer aggregates and compares the exported products and corrects nothing. A month with too few measured records carries no badge and is not ranked. The normal is a period average over the whole record, not a stationary climate, so the trend of each calendar month is shown with it, together with the estimator, the coverage rule and the significance test. The file is standalone and works offline.

The meteo products

Each product notebook writes one file per variable as parquet and CSV. The notebooks are numbered 01 to 13 in the order in which they may read one another: 02 and 03 read 01, 04 reads 02, and 07 reads 02 and 04. These numbers are used as shorthand throughout the documentation. In the repository the notebooks are in workflow/10_METEO/30_PRODUCTS/; on this site they are under Notebooks and are linked under How they are built.

Two conventions apply to every product.

Timestamps are TIMESTAMP_MIDDLE, local standard time (UTC+01:00), on a continuous 30-minute index with no daylight saving. The timestamp marks the middle of the averaging period, so 09:15 covers 09:00 to 09:30.

Apply the provenance flag before analysis. It says, record by record, whether a value was measured, corrected, reconstructed or modelled. PA is the only product without a flag. PA, LW_IN, SWC, G, SW_OUT and LW_OUT are not gap-filled, so their gaps are NaN.

Table 2: Meteo variables in the product files. Coverage is the share of that file’s own period carrying a value; flag columns are defined at every record. <d> is the depth in metres.

variable	unit	period	coverage	provenance flag
`SW_IN_T1_47_1_gfXG`	W m^-2	2004-Sep 2026	100 %	`ISFILLED`
`TA_T1_47_1_gfXG`	°C	2004-Sep 2026	100 %	`ISFILLED` + `SOURCE`
`TA_T1_47_1_HOMOGENIZED_gfXG`	°C	2004-Sep 2026	100 %	`ISFILLED` + `SOURCE`
`PPFD_IN_T1_47_1_gfXG`	µmol m^-2 s^-1	2004-Sep 2026	100 %	`ISFILLED`
`RH_T1_47_1`	%	2004-Sep 2026	100 %	`MISSING` + `SOURCE`
`RH_T1_47_1_HOMOGENIZED`	%	2004-Sep 2026	100 %	`MISSING` + `SOURCE`
`PA_T1_47_1`	kPa	2005-Sep 2026	98.9 %	none, gaps left in
`LW_IN_T1_47_1`	W m^-2	2005-Sep 2026	98.7 %	`SOURCE`, gaps left in
`VPD_T1_47_1`	kPa	2004-Sep 2026	100 %	`ISFILLED` + `SOURCE`
`VPD_T1_47_1_HOMOGENIZED`	kPa	2004-Sep 2026	100 %	`ISFILLED` + `SOURCE`
`PREC_TOT_T1_47_1`	mm (30 min)	2004-2025	99.4 %	`SOURCE` + `ISFILLED`
`PREC_TOT_T1_47_1_HOMOGENIZED`	mm (30 min)	2004-2025	99.4 %	`SOURCE` + `ISFILLED`
`SWC_FF1_<d>_1`, depths 0.05 / 0.1 / 0.2 / 0.3 / 0.5 m	% VWC	2004-2025	26-96 %	`SOURCE`
`SWC_FF1_<d>_1_HOMOGENIZED`, the four depths above 0.5 m	% VWC	2004-2025	81-96 %	`SOURCE`
`TS_FF1_<d>_HOMOGENIZED_GAPFILLED`, depths 0.05 / 0.1 / 0.15 / 0.2 / 0.3 / 0.5 / 0.6 m	°C	2004-2025	27-94 %	`METHOD` + `SUSPECT`
eighteen individual `TS_FF1_` / `TS_PRF_FF1_` channels	°C	2004-2025	21-72 %	none, raw as screened
`G_FF1_0.05_<n>`, three plates	W m^-2	2004-2025	73-96 %	`SOURCE`
`G_FF1_0.05_HOMOGENIZED`, the plate mean	W m^-2	2004-2025	95.6 %	`NPLATES`
`SW_OUT_T1_47_1`	W m^-2	2005-2025	98.8 %	`INSTRUMENT` + `SOURCE`
`LW_OUT_T1_47_1`	W m^-2	2005-2025	98.1 %	`SOURCE`, gaps left in

Flag names are given by their suffix; the full column is FLAG_<variable>_<suffix>.

The derived columns

A _HOMOGENIZED or _GAPFILLED column sits beside the measured one rather than replacing it, so the measured record remains available.

_HOMOGENIZED removes a step that a hardware change left in the values, so the eras on either side can be compared. It does not make either era more accurate: it removes the change in the instrument’s bias, not the bias itself. TA and RH carry one because their probe and acquisition system were replaced together in January 2016 and the probe again in June 2026, and VPD because it is computed from them. PREC carries one because its acquisition system changed in 2018, and SWC one per depth because the soil profile was replaced in 2020. Most rescale the earlier era onto the level of the later one. TA also corrects the difference between the radiation shields of the probes, which acts only in daylight.

_GAPFILLED means a model produced some of the values. For TS the model is a regression on the other depths of the profile, and the METHOD flag says which values are measured and which are modelled.

Known limitations

The limitations of each product are on that variable’s page under Known limitations: the dates across which the record is not homogeneous, the columns that are estimates rather than measurements, and what a trend or period comparison over the series contains besides the weather. Table 1 links every page.

Two limitations belong to a reference rather than to a product and recur on several pages. The MeteoSwiss Lägern station changed level in October 2010, when its instrumentation was rebuilt. The NABEL series on the same tower stop being independent records in mid-2018. Neither changes a value in any file, but both change what a comparison against those series means.

How they are built

Every product reads the quality-screened half-hourly record from the database, joins the older mst screening to the newer diive one, applies the site’s shared 2012 repairs where the variable was on the affected logger, and is checked before it is written. Independent references do the validation and, where a product is filled, much of the filling: MeteoSwiss Lägern 2.5 km away, NABEL at 49 m on the same tower until 2018, the MeteoSwiss gauge at Ehrendingen for rain, and for PREC every MeteoSwiss station within 21 km. The soil products have no external reference and are checked against the other depths of their own profile. Meteo Product Chain shows which product reads which.

Each notebook documents its own decisions and the evidence behind them: 01 SW_IN, 02 TA, 03 PPFD_IN, 04 RH, 05 PA, 06 LW_IN, 07 VPD, 08 PREC, 09 SWC, 10 TS, 11 G, 12 SW_OUT, 13 LW_OUT, and 99, the merge. Unnumbered notebooks beside them export nothing and each settle one question, such as RADIATION_SENSOR_CONTINUITY; the variable pages link the ones that bear on them.