A comprehensive benchmarking system for evaluating global vegetation models

We present a benchmark system for global vegetation models. This system provides a quantitative evaluation of multiple simulated vegetation properties, including primary production; seasonal net ecosystem production; vegetation cover; composition and height; fire regime; and runoff. The benchmarks are derived from remotely sensed gridded datasets and site-based observations. The datasets allow comparisons of annual average conditions and seasonal and inter-annual variability, and they allow the impact of spatial and temporal biases in means and variability to be assessed separately. Specifically designed metrics quantify model performance for each process, and are compared to scores based on the temporal or spatial mean value of the observations and a "random" model produced by bootstrap resampling of the observations. The benchmark system is applied to three models: a simple light-use efficiency and water-balance model (the Simple Diagnostic Biosphere Model: SDBM), the Lund-Potsdam-Jena (LPJ) and Land Processes and eXchanges (LPX) dynamic global vegetation models (DGVMs). In general, the SDBM performs better than either of the DGVMs. It reproduces independent measurements of net primary production (NPP) but underestimates the amplitude of the observed CO2 seasonal cycle. The two DGVMs show little difference for most benchmarks (including the inter-annual variability in the growth rate and seasonal cycle of atmospheric CO2), but LPX represents burnt fraction demonstrably more accurately. Benchmarking also identified several weaknesses common to both DGVMs. The benchmarking system provides a quantitative approach for evaluating how adequately processes are represented in a model, identifying errors and biases, tracking improvements in performance through model development, and discriminating among models. Adoption of such a system would do much to improve confidence in terrestrial model predictions of climate change impacts and feedbacks.

The development of a data-driven application benchmarking approach to performance modelling

Performance modelling is a useful tool in the lifeycle of high performance scientific software, such as weather and climate models, especially as a means of ensuring efficient use of available computing resources. In particular, sufficiently accurate performance prediction could reduce the effort and experimental computer time required when porting and optimising a climate model to a new machine. In this paper, traditional techniques are used to predict the computation time of a simple shallow water model which is illustrative of the computation (and communication) involved in climate models. These models are compared with real execution data gathered on AMD Opteron-based systems, including several phases of the U.K. academic community HPC resource, HECToR. Some success is had in relating source code to achieved performance for the K10 series of Opterons, but the method is found to be inadequate for the next-generation Interlagos processor. The experience leads to the investigation of a data-driven application benchmarking approach to performance modelling. Results for an early version of the approach are presented using the shallow model as an example.