Evaluating and Benchmarking Land Surface Models

Observations relevant to the assessment of modelled latent and sensible heat are available at a wide range of spatial scales (from the footprint area of flux towers to the large pixels of GRACE satellite products) and time steps (e.g. sub-daily to monthly). Thus an evaluation system should be flexible enough to process model output and observations at whatever time steps and spatial resolutions they are available for. NASA’s LVT (Land Verification Toolkit, Kumar et al., 2012 Geosci. Model Dev.), developed as part of the LIS suite, has this flexibility. It also provides a wide range of analytical metrics with in-built assessment of the uncertainties of the metrics (e.g. 95% CIs). This LVT system is being used to develop a comprehensive benchmarking suite for the whole of the terrestrial water cycle. The initial stages are to establish an evaluation system for momentum, heat, moisture and carbon fluxes using observational data from the FLUXNET2015 database. This will subsequently be expanded to include other aspects of the terrestrial land surface, including LST, soil moisture, ground water, snow cover, vegetation information (such as LAI) and ultimately streamflow. In addition to including all physical aspects pertaining to the water cycle, the system will be extended into a benchmarking suite (following the concepts defined by the PLUMBER paper) rather than just evaluation. The initial development of the system has been applied to the JULES (Joint UK Land Environment Simulator) land surface model. Results demonstrate both the performance of JULES along with the potential of such a system for accelerating future targeted developments within LSMs.