Model evaluation using a community benchmarking system for land surface models

Evaluation of atmosphere, ocean, sea ice, and land surface models is an important step in identifying deficiencies in Earth system models and developing improved estimates of future change. For the land surface and carbon cycle, the design of an open-source system has been an important objective of the International Land Model Benchmarking (ILAMB) project. Here we evaluated CMIP5 and CLM models using a benchmarking system that enables users to specify models, data sets, and scoring systems so that results can be tailored to specific model intercomparison projects. Our scoring system used information from four different aspects of global datasets, including climatological mean spatial patterns, seasonal cycle dynamics, interannual variability, and long-term trends. Variable-to-variable comparisons enable investigation of the mechanistic underpinnings of model behavior, and allow for some control of biases in model drivers. Graphics modules allow users to evaluate model performance at local, regional, and global scales. Use of modular structures makes it relatively easy for users to add new variables, diagnostic metrics, benchmarking datasets, or model simulations. Diagnostic results are automatically organized into HTML files, so users can conveniently share results with colleagues. We used this system to evaluate atmospheric carbon dioxide, burned area, global biomass and soil carbon stocks, net ecosystem exchange, gross primary production, ecosystem respiration, terrestrial water storage, evapotranspiration, and surface radiation from CMIP5 historical and ESM historical simulations. We found that the multi-model mean often performed better than many of the individual models for most variables. We plan to publicly release a stable version of the software during fall of 2014 that has land surface, carbon cycle, hydrology, radiation and energy cycle components.