Meteo Data

TODO IN PROGRESS

Calendar explorer

Every month of the record on one grid

Twenty-one years by twelve months, one tile per month, coloured by the metric selected and badged with what was notable in that month. The same grid is drawn at three resolutions: months, meteorological seasons, and a raster of every day of the record, which is the only one of the three that does not cut a spell in half at a boundary. Open a month for its statistics against the calendar-month normal, its rank among the same month of every other year, and a calendar of its days; open a day for that day’s statistics, its flags and its diurnal course.

The page aggregates and compares the exported products and corrects nothing. A month too sparsely measured to support a claim carries no badge and is ranked against nothing. A normal drawn from the whole record is a period average rather than a climate that held still, so the trend of each calendar month is published under it, with the estimator, the coverage rule and the significance test the slope rests on. The file is standalone and works offline.

Pages for the individual variables

Every meteo parameter has its own page, carrying its columns, units, coverage, flag codes and known limitations, and linking the notebooks that produced it. This page stays general: the conventions shared by all products, what they contain, and how they are built. A limitation belongs to a variable and is documented on that variable’s page.

The list below is the full set of parameters. Pages are added as they are written, so an entry without a link is a page not yet written rather than a variable missing from the dataset.

Each parameter also has an interactive dashboard: one self-contained page summarising the exported product year by year, with coverage and provenance, seasonality, distributions, extremes, trends, and the comparison against the reference station where one exists. Every chart carries a table view of the same numbers, and the file is standalone and works offline.

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

A dashboard describes the column a user should analyse. Where a product exports both a measured and a _HOMOGENIZED column, the dashboard uses the homogenised one throughout and names it at the top.

The meteo products

Ten notebooks turn the screened tower and soil measurements into one file per variable, written as parquet and CSV. Each file holds its value columns plus a provenance flag saying, half hour by half hour, whether the number was measured, corrected, reconstructed or modelled.

One notebook produces one variable. They are numbered 01 to 10 in the order in which they may read one another — 02 and 03 read 01, and 07 reads 01, 02 and 04 — and those numbers are used as shorthand throughout this documentation and inside the notebooks themselves. In the repository they are the folder workflow/10_METEO/30_PRODUCTS/; on this site they are rendered under Notebooks, and each is listed with its number and variable under How they are built below.

Two conventions apply to every product.

Timestamps are TIMESTAMP_MID, local time (UTC+1), on a continuous 30-minute index with no daylight saving. The label sits in the middle of the averaging period, so 09:15 covers 09:00 to 09:30.

The provenance flag distinguishes a measurement from an estimate and should be applied before analysis. PA is the only product without one. Neither PA nor LW_IN is gap-filled and both retain their gaps; the LW_IN flag names the instrument and calibration behind each value rather than marking a fill.

Table 2: Meteo variables currently available. 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	2004-2025	100 %	ISFILLED
TA_T1_47_1_gfXG	°C	2004-2025	100 %	ISFILLED + SOURCE
TA_T1_47_1_HOMOGENIZED_gfXG	°C	2004-2025	100 %	ISFILLED + SOURCE
PPFD_IN_T1_47_1_gfXG	µmol m-2 s-1	2004-2025	100 %	ISFILLED
RH_T1_47_1	%	2004-2025	100 %	MISSING + SOURCE
PA_T1_47_1	kPa	2005-2025	98.9 %	none, gaps left in
LW_IN_T1_47_1	W m-2	2005-2025	98.6 %	SOURCE, gaps left in
VPD_T1_47_1	kPa	2004-2025	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

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 recoverable.

_HOMOGENIZED removes a step that a hardware change left in the raw values, so the eras either side become comparable. It does not make either era more accurate: what it removes is the change in the instrument’s bias, not the bias itself. TA carries one because its sensor and acquisition system were replaced together in January 2016, PREC one because its acquisition system changed in 2018, SWC one per depth because the soil profile was replaced in 2020. Most are a rescaling of the earlier era onto the level of the later one; TA also corrects the difference between the two sensors’ radiation shields, which acts only in daylight.

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

Known limitations

Every product carries limitations that are specific to it, and they are documented on that variable’s own page under Known limitations: which dates the record is not homogeneous across, which columns are estimates rather than measurements, and what a trend or a period comparison computed over that series contains besides the weather. Table 1 links all ten pages.

Two limitations are properties of a reference rather than of a product, and recur on several of those pages: the MeteoSwiss Lägern station changed level in October 2010, when its instrumentation was rebuilt, and 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 pulls the quality-screened half-hourly record from the database, joins the older mst screening to the newer diive one with combine_first, converts timestamps, applies the site’s shared 2012 repairs where the variable sat on that logger, and writes out after a set of checks. 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 alone every MeteoSwiss station within 21 km. The soil products have no external reference and are checked against the other depths of their own profile.

Each notebook documents its own decisions and the evidence behind them, and each is named for the variable it produces: 01 SW_IN, 02 TA, 03 PPFD_IN, 04 RH, 05 PA, 06 LW_IN, 07 VPD, 08 PREC, 09 SWC, 10 TS. A handful of unnumbered notebooks beside them export nothing and settle one question each, such as RADIATION_SENSOR_CONTINUITY; the variable pages link the ones that bear on them.