Optimizing the Use of Redundant and Similar Sensors in Automated Gap-Filling of Meteorological Measurements at the Network Level

Redundant and similar sensors are ideal for gap-filling missing meteorological measurements. However, this is more complex than it might seem when implementing an automated gap-filling procedure for bottom-up measurement networks such as AmeriFlux and FLUXNET because there are varying levels of sensor redundancy/similarity across the network yet the procedure must be applied consistently across all sites. Thus, automatable decisions must be made about what constitutes a redundant sensor, and whether and how to use similar measurements that are confounded by different measurement theory and/or spatial variability.

Network-level processing by AmeriFlux gap-fills meteorological measurements with a combination of marginal distribution sampling, mean diurnal variation and downscaled atmospheric reanalysis. It does not currently incorporate redundant or similar measurements, except in cases where the site PI has performed this step prior to data submission to the network. By incorporating in-situ knowledge of the site, such as one sensor being more reliable than another, site PI aggregations are often the most accurate, but are nearly impossible to apply consistently and in a reproducible form by the networks.

Here we test the performance of a gap-filling framework that incorporates a sequence of automated decisions and processing steps to leverage redundant and similar sensor measurements when available. This framework builds on top of the current AmeriFlux/FLUXNET pipeline and is applied consistently across sites and data products. Preliminary results indicate that the tested framework improves gap-filling performance by a significant margin compared to a procedure that does not incorporate redundant/similar measurements.