NEON's approach to uncertainty estimation for sensor-based measurements

Background/Question/Methods

The Fundamental Instrument Unit (FIU) of the National Ecological Observatory Network (NEON) is responsible for generating QA/QC protocols and uncertainty estimates for all of NEON’s level-one (i.e., calibrated and temporally averaged) in-situ, sensor-based measurements.  Identifying and quantifying uncertainties are the keystones of deriving statistical interpretations about mean quantities and variance structure; both are needed to construct NEON’s higher level data products.  These data products will be publically available, and as such, a goal of NEON is to ensure all sources of uncertainty are identified and if possible, quantified in a traceable manner.  Here, our approach for identifying and quantifying uncertainties of these level-one data products is presented.  It emulates the approach taken by the International Organization for Standardization (ISO), while leaving room for scientific creativity, that which can be justified.  Specific examples are given to exhibit the strengths, limitations, and plausibility of estimating uncertainty (in a standardized fashion) for those data products produced by the FIU. 

Results/Conclusions

Although sources of uncertainties can be identified, some cannot currently be quantified, especially at a high level of confidence.  Much work is still needed to better understand the uncertainties of various ecological processes (e.g., fine-root turnover) and their associated sampling methods (e.g., use of minirhizotrons).  Additionally, efforts should be aimed toward minimizing specific assumptions that uncertainty analyses comprise, such as inferring human induced errors are negligible.  This is especially important for NEON’s Observatory, as many ecological measurements will be human-based.  As time progresses and NEON data are analyzed, a better understanding of sensor specific (and human induced) uncertainties may arise, thus making it possible to quantify previously unquantifiable uncertainties. 

The presented approach serves as a standardized method by which ecological data uncertainties can be quantified.  It is our hope that current and future ecological networks will adapt this method, thereby strengthening ecological datasets and promoting interoperability among ecological networks.