© 2013 National Ecological Observatory Network, Inc. ALL RIGHTS RESERVED.
CONNECTING UNCERTAINTY
ESTIMATES AND QA/QC METHODS
Josh Roberti
Derek Smith
Jeff Taylor
Janae Csavina
Dave Durden
Stefan Metzger
Sarah Streett
LTER All Scientists Meeting
1 September 2015
Estes Park, CO
© 2013 National Ecological Observatory Network, Inc. ALL RIGHTS RESERVED.
TRACEABLE – STANDARDIZED –
TRANSPARENT
• In-house calibrations
• Algorithm Theoretical Basis Documents (ATBDs)
http://data.neoninc.org/documents
• Theory of measurement
• Theory of algorithms
• converting from raw, uncalibrated data
• QA/QC
• Temporal averaging (Level 1 data products)
• One- and thirty-minute averages
• Uncertainty estimates
• Measurement uncertainty as well as the uncertainty of our
temporally averaged data products
© 2013 National Ecological Observatory Network, Inc. ALL RIGHTS RESERVED.
NEON’S CVAL
Purpose of having an in-house metrology lab
 Nationwide Sensor Based Observations
 Thousands of sensors through one lab using
same nationally recognized standards.
 Control over quality and internal assessment of
measurement uncertainty.
 terrestrial and aquatic sensors
 gas and isotopic water transfer standards
 standards calibrations
 external labs
 Technology changes
 Compare old to new with same standards to
understand changes in the 30 years of the project.
© 2015 National Ecological Observatory Network, Inc. ALL RIGHTS RESERVED.
NEON is solely sponsored by the National Science Foundation
QA/QC Flags:
• Range – realistic data range
• Persistence – realistic data variability
• Null –missing data points
• Gap –large data gaps
• Step –large changes between
consecutive data points
• De-spiking –spikes in the data set
• Consistency – comparable
measurements
Sensor Flags:
• Identify problem areas, e.g.,
consistency checks
• Assist algorithm refinement and
development
QA/QC
© 2013 National Ecological Observatory Network, Inc. ALL RIGHTS RESERVED.
UNCERTAINTY ESTIMATES
NEON calculates measurement uncertainties according to recommendations
of the Joint Committee for Guides in Metrology (JCGM) 2008.
𝑢 𝑐 𝑦 =
𝑖=1
𝑁
𝜕𝑓
𝜕𝑥𝑖
2
𝑢2 𝑥𝑖
1
2 where ,
𝜕𝑓
𝜕𝑥 𝑖
= partial derivative of y with respect to xi
𝑢 𝑥𝑖 = combined standard uncertainty of xi.
aka QA/QC
© 2015 National Ecological Observatory Network, Inc. ALL RIGHTS RESERVED.
NEON is solely sponsored by the National Science Foundation
Example
1. Convert [Ω] to [°C]
2. Sensor flags
A. Heater (ON/OFF)
B. Turbine (flow > threshold)
3. QA /QC testing
4. Calculate temporally averaged data
product
5. Publish data
1 Hz Tem. (Ω)
• Inform assembly uncertainty to set thresholds for certain QA/QC tests
(e.g., persistence)
• Make confident inferences between signal (environmental phenomenon)-
to-noise
• Promote standardization and interoperability among ecological networks
Traceable uncertainty estimates
© 2015 National Ecological Observatory Network, Inc. ALL RIGHTS RESERVED.
NEON is solely sponsored by the National Science Foundation
Future work
• Data commissioning
• Audits
• Threshold refinement
• Non-analytical uncertainty analyses
• Bayesian – Monte Carlo
Thanks!
jroberti@neoninc.org
Contact info:

Roberti lter asm_september2015_jr_ds_jt_jc

  • 1.
    © 2013 NationalEcological Observatory Network, Inc. ALL RIGHTS RESERVED. CONNECTING UNCERTAINTY ESTIMATES AND QA/QC METHODS Josh Roberti Derek Smith Jeff Taylor Janae Csavina Dave Durden Stefan Metzger Sarah Streett LTER All Scientists Meeting 1 September 2015 Estes Park, CO
  • 2.
    © 2013 NationalEcological Observatory Network, Inc. ALL RIGHTS RESERVED. TRACEABLE – STANDARDIZED – TRANSPARENT • In-house calibrations • Algorithm Theoretical Basis Documents (ATBDs) http://data.neoninc.org/documents • Theory of measurement • Theory of algorithms • converting from raw, uncalibrated data • QA/QC • Temporal averaging (Level 1 data products) • One- and thirty-minute averages • Uncertainty estimates • Measurement uncertainty as well as the uncertainty of our temporally averaged data products
  • 3.
    © 2013 NationalEcological Observatory Network, Inc. ALL RIGHTS RESERVED. NEON’S CVAL Purpose of having an in-house metrology lab  Nationwide Sensor Based Observations  Thousands of sensors through one lab using same nationally recognized standards.  Control over quality and internal assessment of measurement uncertainty.  terrestrial and aquatic sensors  gas and isotopic water transfer standards  standards calibrations  external labs  Technology changes  Compare old to new with same standards to understand changes in the 30 years of the project.
  • 4.
    © 2015 NationalEcological Observatory Network, Inc. ALL RIGHTS RESERVED. NEON is solely sponsored by the National Science Foundation QA/QC Flags: • Range – realistic data range • Persistence – realistic data variability • Null –missing data points • Gap –large data gaps • Step –large changes between consecutive data points • De-spiking –spikes in the data set • Consistency – comparable measurements Sensor Flags: • Identify problem areas, e.g., consistency checks • Assist algorithm refinement and development QA/QC
  • 5.
    © 2013 NationalEcological Observatory Network, Inc. ALL RIGHTS RESERVED. UNCERTAINTY ESTIMATES NEON calculates measurement uncertainties according to recommendations of the Joint Committee for Guides in Metrology (JCGM) 2008. 𝑢 𝑐 𝑦 = 𝑖=1 𝑁 𝜕𝑓 𝜕𝑥𝑖 2 𝑢2 𝑥𝑖 1 2 where , 𝜕𝑓 𝜕𝑥 𝑖 = partial derivative of y with respect to xi 𝑢 𝑥𝑖 = combined standard uncertainty of xi. aka QA/QC
  • 6.
    © 2015 NationalEcological Observatory Network, Inc. ALL RIGHTS RESERVED. NEON is solely sponsored by the National Science Foundation Example 1. Convert [Ω] to [°C] 2. Sensor flags A. Heater (ON/OFF) B. Turbine (flow > threshold) 3. QA /QC testing 4. Calculate temporally averaged data product 5. Publish data 1 Hz Tem. (Ω) • Inform assembly uncertainty to set thresholds for certain QA/QC tests (e.g., persistence) • Make confident inferences between signal (environmental phenomenon)- to-noise • Promote standardization and interoperability among ecological networks Traceable uncertainty estimates
  • 7.
    © 2015 NationalEcological Observatory Network, Inc. ALL RIGHTS RESERVED. NEON is solely sponsored by the National Science Foundation Future work • Data commissioning • Audits • Threshold refinement • Non-analytical uncertainty analyses • Bayesian – Monte Carlo Thanks! jroberti@neoninc.org Contact info:

Editor's Notes

  • #2 (30 seconds) Introduce myself and the FIU:
  • #3 (60 seconds): NEON’s approach to uncertainty – why do we quantify it? How do we quantify it? How are the end-users aware of NEON’s algorithms and procedures, etc. 1. Web-link: Just make sure it’s visible to the audience and also mention that we do have streaming data at current date, and we will have more in the near future. All information can be found on the website shown here. 2. ATBDs – what is found in them… brief explanation will suffice. If the audience wants more information, they can check out the web-link
  • #4 (60 seconds) Provide an overview of Note that as a first approach we are considering all input quantities to be independent of one another. However, we are aware that correlations exist, over time, and after the observatory is built, we will spend time tweaking our approach, e.g., we plan to incorporate correlations, components of drift (sensor and DAS), etc. The main goal of our uncertainty analyses and more importantly, ATBDs, are to provide traceable, standardized approaches while being transparent to the end users (e.g., scientific community, students, politicians). As of current, our approach for quantifying measurement uncertainty comprises analytical derivation.
  • #6 (60 seconds) Provide an overview of Note that as a first approach we are considering all input quantities to be independent of one another. However, we are aware that correlations exist, over time, and after the observatory is built, we will spend time tweaking our approach, e.g., we plan to incorporate correlations, components of drift (sensor and DAS), etc. The main goal of our uncertainty analyses and more importantly, ATBDs, are to provide traceable, standardized approaches while being transparent to the end users (e.g., scientific community, students, politicians). As of current, our approach for quantifying measurement uncertainty comprises analytical derivation.