Elsevier

Agricultural and Forest Meteorology

Volume 265, 15 February 2019, Pages 145-151
Agricultural and Forest Meteorology

Effects of the Gill-Solent WindMaster-Pro “w-boost” firmware bug on eddy covariance fluxes and some simple recovery strategies

https://doi.org/10.1016/j.agrformet.2018.11.010Get rights and content

Highlights

  • Eddy covariance fluxes from some Gill sonic anemometers are affect by a firmware bug.

  • The bug can be exactly compensated by reprocessing raw data files.

  • Approximate compensation can be done by applying a fixed multiplier to existing data.

  • The bug also affects some angle of attack corrections.

Abstract

In late 2015 and early 2016, work done by the AmeriFlux Management Project Technical Team (amerilfux.lbl.gov) helped to uncover an issue with Gill WindMaster and WindMaster Pro sonic anemometers used by many researchers for eddy covariance flux measurements. Gill has addressed this issue and has since sent out a notice that the vertical wind speed component (a critical piece of all eddy covariance fluxes) was being erroneously computed and reported. The problem (known as the “w-boost” bug) resulted in positive (upward) wind speeds being under-reported by 16.6% and negative (downward) wind speeds being under-reported by 28.9%. This has the potential to cause similar under estimates in fluxes derived from measurements using these instruments. Additionally, the bug affects corrections for angle of attack as derived by Nakai and Shimoyama, rendering them invalid. While the manufacturer has offered a firmware upgrade for existing instruments that will fix this issue, many existing data sets have been affected by it and are currently in use by the scientific community.

To address the issue of affected data, currently in use, we analyzed multi-year and short-term data sets from a variety of ecosystems to assess methods of correcting existing flux data. We found that simple multiplicative correction factors (∼1.18) may be used to remove most of the “w-boost” bias from fluxes in existing data sets that do not include angle of attack corrections.

Introduction

The importance of sonic anemometry to the eddy covariance technique is well known. It is at the heart of this method which is widely used to estimate the exchange of matter and energy between the land surface and the atmosphere. The accuracy and precision of vertical wind speed measurements directly affects the quality of calculated fluxes (Baldocchi et al., 1988; Lee et al., 2004; Aubinet et al., 2012). In the last 40 years, much effort has been expended to refine and improve sonic anemometry with the goal of reducing biases and uncertainties in eddy covariance fluxes (Wyngaard and Zhang, 1985; Nakai et al., 2006; Kochendorfer et al., 2012; Nakai and Shimoyama, 2012; Frank et al., 2013; Horst et al., 2015). Many of these works have focused on topics related to instrument geometry and transducer-induced flow distortions, and they all share the common assumption that the tested instruments correctly report wind velocity components as defined by the instrument specifications. This assumption is made based on the fact that all manufacturers test and calibrate instruments (individually or by model representatives) in wind tunnels.

Field researchers also rely on this assumption, but at typical installations, there is usually no opportunity to verify it by comparison with a second instrument. In this respect, the AmeriFlux Management Project (AMP) Technical Team (ameriflux.lbl.gov/tech) is in a unique position. One of the main functions of the Tech Team is to maintain and improve data quality from network sites through an on-going inter−COmparison campaign (ameriflux.lbl.gov/site-visits). During these exercises, Tech Team members deploy a well vetted, portable eddy covariance flux system adjacent to a client system for an observation period of about 10 days. Results from both systems are then compared and differences are assessed. In late 2015 and early 2016, the Tech Team identified anomalous wind velocity behavior during several flux site comparisons, all of which involved the same model and vintage Gill sonic anemometer (WindMaster and WindMaster Pro). A Gill distributor in the U.S. (LI−COR Biosciences, Lincoln, NE) and the manufacturer (Gill Solent, Lymington, UK), investigated this and discovered that between 2006 and 2015, this specific sonic anemometer model suffered from a firmware error that resulted in an asymmetric under estimation of vertical wind speeds. This flaw caused a 16.6% under estimation of positive (or upward) vertical wind speeds and a 28.9% under estimation of negative (or downward) vertical wind speeds and became known as the “w-boost” bug. Gill further confirmed that this issue was isolated to the vertical wind velocity component, did not affect the horizontal velocity components or the sonic temperature measurements, and was only present in the WindMaster and WindMaster Pro models manufactured between 2006 and 2015. This error has since been fixed in new units and Gill offers several remedies for owners of affected instruments (Gill-Solent website, 2016).

The manufacturer has identified specific firmware versions and serial number blocks which allow identification of potentially affected instruments (see the Appendix). This method however, is not an infallible indicator of whether or not the “w-boost” bug has propagated into public data sets derived from them. In the simplest case, the data-originator may be unaware of the “w-boost” bug, or may not have an opportunity to modify their sonic anemometer or their post-processing work flow. This will result in publicly available flux data that is affected by the “w-boost” bug and is associated with metadata necessary to detect the situation. In other cases, the effects may have been corrected in the post-processing work flow, but non-updated metadata (instrument serial number or firmware version) could indicate that the data still contains the “w-boost” bias.

Aside from these direct effects, the “w-boost” bug can be manifested in other flux corrections derived from affected instruments. One notable case concerns the application of the transducer flow distortion (“angle of attack” or AoA) corrections calculated by Nakai and Shimoyama (2012). These algorithms were derived from data produced by affected sonic anemometers, and did not include compensation for the (then undiscovered) “w-boost” bug. The resulting AoA corrections effectively mix unaffected u and v wind speeds with affected w ones. Because of this, application of the Nakai and Shimoyama (2012) AoA correction will produce erroneous results in all cases, and is not recommended. Unfortunately, this algorithm has been incorporated into widely used, raw data processing streams (e.g., LiCor’s EddyPro), and it is likely that data sets with this secondary error have been made available to the public. Also, because the essential geometry of the Windmaster/Windmaster Pro sonic anemometers is shared with the Gill R3 and R3-50 models, it is possible that some data-originators have applied the Nakai and Shimoyama (2012) AoA correction to these instruments as well, creating yet another class of affected data. In all, we recognize five potential cases of data that are directly or indirectly affected by the “w-boost” bug:

  • 1

    Fluxes from affected Windmaster and Windmaster Pro sonic anemometers where no corrections have been applied.

  • 2

    Fluxes from affected Windmaster and Windmaster Pro sonic anemometers where only the “w-boost” bug correction has been applied.

  • 3

    Fluxes from affected Windmaster and Windmaster Pro sonic anemometers where only the erroneous Nakai and Shimoyama (2012) AoA correction has been applied.

  • 4

    Fluxes from affected Windmaster and Windmaster Pro sonic anemometers where both the Nakai and Shimoyama (2012) AoA and the “w-boost” corrections have been applied

  • 5

    Fluxes from unaffected (R3, R3-50, or repaired or older WindMaster and WindMaster Pro) sonic anemometers where the Nakai and Shimoyama (2012) AoA correction has been applied.

It appears that data-originators may have unknowingly yet systematically calculated and shared under estimated values for eddy covariance fluxes derived from both affected and unaffected instruments and that researchers have been using these biased data sets for other studies. Two challenges therefore emerge regarding historical observations: first, the development and implementation of simple and practical correction strategies and second, a method to unambiguously identify affected data sets.

The first challenge is itself twofold. First a method of correcting fluxes for the “w-boost” bug must be developed and second, a method of removing the erroneous AoA correction is needed.

Because of the nature of the eddy covariance process, it is not obvious that a simple multiplicative factor will adequately correct existing fluxes for the “w-boost” bug, or what that factor might be. The ideal solution of course is to correct and re-process raw instrument data (LI-COR website, 2016). In practice, this solution could be difficult and inconvenient for many data providers to implement, due to its labor-intensive nature and will often be impossible for data users and archive managers since the raw data streams are not usually available. A more feasible approach would be the development and use of a simple mathematical transform that employs only variables commonly available to end-users which can be applied directly to affected fluxes.

Regarding the indirect effects of the “w-boost” bug that are propagated through the Nakai and Shimoyama (2012) AoA corrections, we expect the two to be independent, but without access to the original high-speed data used to calculate the AoA corrections, we cannot exactly determine the magnitude of the effect that the “w-boost” bug had on them, nor can we modify the algorithms. We can, however attempt to develop a simple strategy to effectively remove the AoA correction from existing data sets similar to the one outlined above.

A similar approach could be taken to address the second issue (identification of affected data sets). Commonly available variables could be compared to expected values for flux sites to determine if an affected sonic anemometer was present. Alternatively, and perhaps more reliably, efforts could be made (by data users or network data curators) to reach out to data-originators and determine directly, whether an affected instrument was used to produce the data set in question and whether or not corrections have been applied.

The AmeriFlux Tech Team launched an effort to explore these options and determine which if any are practical for wider use. Our unique position allowed us to obtain long-term and short-term, raw data sets from eddy covariance flux sites in a variety of ecosystems. From an analysis of exactly corrected fluxes with respect to uncorrected fluxes, we have developed a simple correction strategy for removal of the “w-boost” bug that can be used by the flux community.

Section snippets

Methods and study sites

From the beginning, we adopted the restriction that any correction scheme must rely only on variables that are commonly available from the large data archives (e.g., AmeriFlux, FluxNet, ARM, etc.). This precluded any method that would rely on original, high-speed data which might only be available from the data originator and would not be commonly available from public archives. For our purposes, these data included fluxes, mean wind velocity components, and simple statistical moments (e.g.,

Results

Our first task was to determine if a simple, linear transform could be used to compensate for the “w-boost” bug and/or AoA correction in existing flux data sets. We arbitrarily set the criterion that if linear regressions between corrected and uncorrected fluxes showed little scatter about regression lines (R2 values greater than 0.9), the relationship would be considered linear. Fig. 1 shows comparisons between “w-boost” corrected (W) and uncorrected (U) fluxes for one site-year of data (2015)

Discussion

Considering the direct effects of the “w-boost” bug, there clearly exists a bias in eddy covariance fluxes obtained from affected sonic anemometers. While random uncertainties in fluxes are often between 10% and 15% (Billesbach, 2011; Richardson et al., 2012); they will tend to cancel out in long term averages. The “w-boost” bug, however, produces a bias (15% to 18%) and has the potential to significantly impact annual budgets of carbon, water and energy. Ideally, all affected raw data should

Conclusions and Recommendations

There is a firmware bug in many Gill-Solent WindMaster and WindMaster Pro sonic anemometers (manufactured between 2006 and October 2015) that can directly cause under estimation of fluxes by about 18% and can also introduce biases indirectly through an angle-of-attack (AoA) correction (Nakai and Shimoyama, 2012). The direct effect can be exactly compensated in the raw, high-frequency data by multiplying all positive (or upward) wind speeds by 1.166 and all negative (or downward) wind speeds by

Acknowledgment

Funding for the AmeriFlux Management Project was provided by the U.S. Department of Energy’s Office of Science under Contract No. DE-AC02-05CH11231.

References (14)

There are more references available in the full text version of this article.

Cited by (5)

  • Potential and challenges of wind measurements using met-masts in complex topography for bridge design: Part I – Integral flow characteristics

    2021, Journal of Wind Engineering and Industrial Aerodynamics
    Citation Excerpt :

    It should be noted that the ratio σw/σu ​≈ ​0.4 recorded on the Hardanger bridge ​(Fenerci and Øiseth, 2018b) is slightly lower than reported in the other Norwegian fjords. It is unsure whether it is due to flow-distortion by the deck, sensor calibration error ​(Gill Instruments, 2016; ​Billesbach et al., 2019), or simply peculiar flow characteristics at the bridge location. The ratio σv/σu is expected to range from 0.7 to 0.9 in flat terrain (Solari and Piccardo, 2001).

  • Comparison of gas analyzers for eddy covariance: Effects of analyzer type and spectral corrections on fluxes

    2019, Agricultural and Forest Meteorology
    Citation Excerpt :

    Unless otherwise specified, default software settings were used (e.g., block averaging, despiking thresholds, 2-D coordinate rotation, etc.). No angle of attack corrections (Nakai and Shimoyama, 2012; Nakai et al., 2006) were applied to the Gill sonic anemometer since it does not apply to the R3-50 (Billesbach et al., 2019). Temporal lags between the sonic anemometer and the gas analyzer, whether due to digital signal processing delays or physical (separation of sensors or residence time in inlet, were calculated for each averaging period using the covariance maximization procedure.

View full text