Flux everywhere, all of the time?

Abstract:

Vast networks of eddy covariance flux measurements have been made across the globe in the recent decades. The global network – FLUXNET, which contains 700+ historical and active flux measurement sites, provides a comprehensive framework in studying the terrestrial carbon and water cycles. Today, the latest synthesis database contains 2200+ years of data from 400+ sites. While decadal records are becoming available in a subset of the FLUXNET sites (~15%), the majority (>50%) still possess less than five years of data. Thus, it has always been a dilemma for data users to choose between spatial and temporal coverage, or to simply assume exchangeable between space and time. Here, we examine the spatial and temporal representativeness of the FLUXNET sites and the newest synthesis database. Briefly, the general site information is compiled from all available FLUXNET database. The long-term (1981-2013) and global gridded data of climate, land use, and plant phenology are obtained from the CRU, SYNMAP, and GIMMS database. The spatial and temporal representativeness of the site location and operation duration is then examined against the global and long-term database. Potentially, FLUXNET sites have covered the major environmental gradients in most of the biomes. Yet, the data availability is highly imbalanced among ecoregions in terms of the sites existed or the data provided by the existing sites. Arguably, as least five years of records are required to adequately capture the long-term interannual variability of climatic and phenological conditions at most site locations. That means, extra caution is needed to make inference beyond the spatial or temporal extents of current database. At last, a simple Bayesian hierarchical model is developed and fitted against a subset of flux data from multiple sites with similar plant function types and long-term records. The diagnostic model intends to test the hypothesis of exchangeability between space and time in modeling fluxes.