A new study by CLEX researchers identifies regions of high and low predictability and will likely help improve land surface model evaluation.

It focuses on observations of the flows, or fluxes, of heat, moisture, and carbon between the land surface and atmosphere. These observations play an important role in determining the processes behind drought, flooding and heatwaves. For this reason, it is vital that we have a good understanding of the variability of these processes at different observational sites.

Until recently, there have been few attempts to quantify the predictability of land surface fluxes. In this case, predictability means the ability of a climate or weather model to accurately estimate the flux observations based on the meteorological variables and the physical characteristics of the site.

To see if there was a way to determine which observation sites were most reliable/predictable for evaluating land surface models, the researchers first looked at observations of energy, water and carbon fluxes and their relation to climatology, geography, vegetation and other factors. The researchers used observations across 155 ground-based sites that were part of the FLUXNET 2015 dataset.

This observational data of all 155 sites was combined then averaged to create a mean “global” estimate of these flows. The researchers then took the same observational data from each site individually.

The two sets of data, local and global, were then used to train statistical models to predict how these flows change at each of the 155 sites. When the models were run, the two sets of simulations (global and local) were compared. The differences between the two projections were then used to develop a predictability metric.

Where local projections were better than global projections for the site, that site was generally considered less predictable.  These differences suggested local characteristics that were unique to the site, potentially indicating sites that could be of particular interest in evaluating global land surface models, since their processes were more unusual.

Sites where local and global projections were closely aligned were classified as more predictable.

The researchers had expected to find consistent patterns in mean precipitation, vegetation type or other metrics that might reveal what made a site less predictable. These patterns did not appear and in particular, there was little evidence that flux behaviour was well discretised by vegetation type – a common assumption in land surface science. The only substantial exception was an indication that drier sites may be less predictable.

It is hoped the results can help land surface modellers make more informed choices of which sites to use for model development and evaluation, and may provide guidance to the FLUXNET community for future flux tower deployments.

 

  • Paper: Haughton, N., Abramowitz, G., De Kauwe, M. G., and Pitman, A. J.: Does predictability of fluxes vary between FLUXNET sites?, Biogeosciences, 15, 4495-4513, https://doi.org/10.5194/bg-15-4495-2018, 2018.