Results from the international benchmarking project, PLUMBER (PALS Land sUrface Model Benchmarking Evaluation pRoject) (Invited Presentation)

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner
Tuesday, 4 February 2014: 3:45 PM
Room C210 (The Georgia World Congress Center )
Martin Best, Met Office, Exeter, United Kingdom; and G. Abramowitz, H. Johnson, M. Ek, P. A. Dirmeyer, Z. Guo, B. Pak, L. Stevens, M. Decker, G. Balsamo, B. J. J. M. van den Hurk, J. A. Santanello Jr., C. D. Peters-Lidard, S. Kumar, A. J. Pitman, A. A. Boone, H. Kim, and T. Oki

There have been many international land surface model comparison studies over the last couple of decades. These have been valuable for identifying areas for development for the poorer performing models, but does not enable us to discover if our models are “good enough”. Although “good enough” may be a subjective choice that is application specific, it does at least enable us to identify if our current development plans are appropriate for delivering improvements to the parts that are not fit for purpose at present.

The land surface benchmarking project, PLUBMER, uses the PALS web based system for the analysis of land models (http://www.pals.unsw.edu.au/pals/Welcome.action) and is a comparison project that does not compare the performance of each model against another, but against some standard metrics. These metrics have been chosen to be a reference crop Penman-Monteith model, a Manabe bucket model and three forms of statistical regression. Initial results from the project have shown that the models have similar results relative to the benchmarks. All models perform better than the Penman-Monteith model and the Manabe bucket model, but struggle to outperform the regressions, especially for the sensible heat flux.

In this presentation the initial results from the comparison will be presented, along with further analysis of the performance of the models and the regressions for the sensible heat flux. An attempt will be made to conclude why the regressions perform so well for this flux and how the models could be improved. In addition, analysis will also be undertaken for a handful of sites where the models' performance is poor. This performance will be compared to the benchmarks to identify if this is a generic model problem, or a limitation in terms of our physical understanding of these sites.