On the predictability of turbulent fluxes from land: PLUMBER2 MIP experimental description and preliminary results

Abstract. Accurate representation of the turbulent exchange of carbon, water, and heat between the land surface and the atmosphere is critical for modelling global energy, water, and carbon cycles, both in future climate projections and weather forecasts. We describe a Model Intercomparison Project (MIP) that compares the surface turbulent heat flux predictions of around 20 different land models provided with in-situ meteorological forcing, evaluated with measured surface fluxes using quality-controlled data from 170 eddy-covariance based flux tower sites. Several out-of-sample empirical model predictions of site fluxes are used as benchmarks to quantify the degree to which land model performance could improve across a broad range of metrics. The performance discrepancy between empirical and mechanistic model predictions also provides a potential pathway to understand sources of model error. Sites with unusual behaviour, complicated processes, poor data quality or uncommon flux magnitude will be more difficult to predict for both mechanistic and empirical models. Results suggest that latent heat flux and net ecosystem exchange of CO2 are better predicted by land models than sensible heat flux, which at least conceptually would appear to have fewer physical processes controlling it. Land models that are implemented in Earth System Models also appear to perform notably better than stand-alone ecosystem (including demographic) models, at least in terms of the fluxes examined here. Flux tower data quality is also explored as an uncertainty source, with the difference between energy-balance corrected versus raw fluxes examined, as well as filtering for low wind speed periods. Land model performance does not appear to improve with energy-balance corrected data, and indeed some results raised questions about whether the correction process itself was appropriate. In both cases results were broadly consistent, with simple out-of-sample empirical models, including linear regression, comfortably outperforming mechanistic land models. The PLUMBER2 approach, and its openly available data, enable precise isolation of the locations and conditions in which model developers can know that a given land model can improve, allowing information pathways and discrete parametrisations in models to be identified and targeted for model development.