CAM reanalysis (site-level) bias correction tool

Use case

Currently to perform a site-level CLM-DART simulation we use the CAM reanalysis global gridded data to provide the ensemble met forcing. This has been shown to work for certain site level applications, for example the assimilation of LAI at a New Mexico site (Fox et al., 2018; doi.org/10.1029/2018MS001362). However, when assimilating in site level eddy covariance flux data, the original CAM reanalysis could be introducing enough boundary condition error (due to site to grid spatial mismatch) which negatively impacts the assimilation. This was observed when an OSSE simulation was performed successfully for a site level CLM-DART assimilation for synthetic flux tower data, however, when real flux data was used for a similar setup the RMSE and bias statistics did not improve significantly (Fox, Hoar et al., unpublished).

I propose to use the tower meteorology to perform a bias correction upon the CAM reanalysis ensemble met forcing. Not only will this make the CAM reanalysis more accurate, but will improve the hour-to-hour variation in met forcing (radiation, temp) which is highly influential upon observed tower fluxes (e.g. GPP, ER, latent heat flux). By applying the bias correction as a pre-processing step this should drastically improve the open-loop simulation, and also increase the likelihood for improved RMSE and bias statistics during assimilation. The bias correction tool will follow a scaling approach used for snow cover assimilation work (Zhang et al., 2015; PhD Dissertation).

Is your feature request related to a problem?

Intended to address challenge of assimilating site level flux tower data. Applicable to 2nd SIF model inter-comparison project; or widely applicable for any site level (NEON, Ameriflux) work.

Describe your preferred solution

The preferred solution is following the Zhang et al., scaling approach. Preliminary results show this approach is successful for improving open loop site level simulation runs where the bias correction tool has been applied to the 'adjusted CAM6'. Note the improved representation of LAI in the open loop simulation below:

Describe any alternatives you have considered

As mentioned above, the original CAM reanalysis can work for certain site level applications for slowly varying properties such as LAI, biomass and soil carbon, but does not work for faster varying tower fluxes.

The AR1 approach has also been used to generate a site level met ensemble (Reichle et al., 2002; Reichle & Koster 2003) and applied to OSSE site level soil moisture assimilation within GEOS-DART (Dibia,Reichle,Anderson et al. (in revision)). It is unclear how the AR1 approach will compare to the CAM bias correction approach. The advantage of the CAM bias correction approach is that it still maintains the physical relationships between the atmospheric properties as represented by CAM. As far as I know this is an open, scientific question how AR will compare to CAM bias correction.

Yet another approach to generate site level met ensemble is to perform a single column CAM-DART simulation using site level met as observations. The posterior of this single column CAM-DART simulation could be used directly as the met ensemble. I don't have experience performing single-column CAM simulations.

NCAR / DART