Level-3.3: USTAR filtering

Level 3.3 flags the NEE records of low turbulence. A record with a friction velocity USTAR below the threshold gets flag 2, by day and by night; unlike ONEFlux, the first record above the threshold is kept. The filter is run at three constant thresholds, which give three USTAR scenarios: CUT_16, CUT_50 and CUT_84. The flags enter the overall quality flag (QCF) of each scenario.

Method, evidence and checks:

• 01_FPPC_IRGA72_NEE, Step 5: the filter, the records per scenario and the gaps left to fill.
• diive: USTAR filter: how the filter works.

Thresholds and records

Table 1: USTAR thresholds and NEE records per scenario, as a share of all 188,448 half-hours of FP2026.1. Records already removed by Levels 2 to 3.2 (51,004, 27 %) are not counted as removed by the USTAR filter.

Scenario	Threshold (m s-1)	Removed by the USTAR filter	Records for gap-filling
CUT_16	0.272	31,447 (17 %)	92,412 (49 %)
CUT_50	0.304	37,102 (20 %)	86,757 (46 %)
CUT_84	0.340	43,610 (23 %)	80,249 (43 %)

CUT_50 is the best estimate. CUT_16 and CUT_84 bound it: the lower threshold keeps more records of low turbulence, the higher one removes more. Their difference gives the USTAR uncertainty of Level 4.1.

The same three thresholds are used for every year. They are kept on purpose, although the thresholds that FLUXNET uses for CH-Lae are much higher.

Columns

For each scenario <CUT>:

Table 2: Level-3.3 columns of NEE.

Column	Content
FLAG_L3.3_<CUT>_NEE_L3.1_USTAR_TEST	USTAR filter: 0 kept, 2 removed
FLAG_L3.3_<CUT>_NEE_L3.1_QCF	Overall quality flag after all tests of Levels 2 to 3.3
NEE_L3.1_L3.3_<CUT>_QCF	NEE with quality flag 0 or 1, the input of the gap-filling
NEE_L3.1_L3.3_<CUT>_QCF0	NEE with quality flag 0 only

Latent and sensible heat flux

LE (and therefore ET) and H are not filtered for USTAR, following Pastorello et al. (2020): it is not established that the advection that biases the CO2 flux at low turbulence also affects the energy fluxes, which are small at night when advection is largest. Their Level-3.3 columns carry CUT_NONE in the name, for example LE_L3.1_L3.3_CUT_NONE_QCF.

Threshold detection

The thresholds were detected with the R package REddyProc (usControlUstarEst settings, 100 bootstrap runs), following Pastorello et al. (2020). The detection used nighttime NEE after Level 3.2 and was run separately for the two gas analysers. The 16th, 50th and 84th percentiles of the bootstrap results give the three scenarios.

Table 3: USTAR thresholds (m s-1) detected for the two analysers, and the thresholds used. For each scenario the higher of the two is used. Defaults in brackets.

Analyser	Period	Nighttime records used	REddyProc settings changed from default	CUT_16	CUT_50	CUT_84
LI-7200 (IRGA72)	2016-2024	QCF 0 and 1	ustPlateauFwd = 9 (10), ustPlateauBack = 5 (6), UstarClasses = 30 (20)	0.271620	0.301811	0.333478
LI-7500 (IRGA75)	2005-2017, 2019	QCF 0	ustPlateauFwd = 8 (10), ustPlateauBack = 4 (6), UstarClasses = 30 (20)	0.268384	0.303628	0.339684
Used				0.271620	0.303628	0.339684

The LI-7200 detection included records of QCF 1, because with QCF 0 only the plateau of nighttime NEE against USTAR was not detected as well.

Notebooks that prepared the detection input:

• IRGA72: 11_IRGA72_FluxProcessingChain_L3.2_NEE_QCF11_2016-2024, 13_IRGA72_PrepareForREddyProc
• IRGA75: 01_IRGA75_FluxProcessingChain_L3.2_NEE_QCF10_2005-2019, 03_IRGA75_PrepareForREddyProc

The IRGA75 fluxes themselves are in progress for FP2026.2 and are not part of FP2026.1.