Multi-site evaluation of the JULES land surface model using global and local data

Abstract. Changes in atmospheric carbon dioxide and water vapour change the energy balance of the atmosphere and thus climate. One important influence on these greenhouse gases is the land surface. Land Surface Models (LSMs) represent the interaction between the atmosphere and terrestrial biosphere in Global Climate Models (GCMs). As LSMs become more advanced, there is a need to test their accuracy. Uncertainty from LSMs contributes towards uncertainty in carbon cycle simulations and thus uncertainty in future climate change. In this study, we evaluate the ability of the JULES LSM to simulate photosynthesis using local and global datasets at 12 FLUXNET sites. Model parameters include site-specific (local) values for each flux tower site and the default parameters used in the Hadley Centre Global Environmental Model (HadGEM) climate model. Firstly, we compare Gross Primary Productivity (GPP) estimates from driving JULES with data derived from local site measurements with driving JULES with data derived from global parameter and atmospheric reanalysis (on scales of 100 km or so). We find that when using local data, a negative bias is introduced into model simulations with yearly GPP underestimated by 16% on average compared to observations while when using global data, model performance decreases further with yearly GPP underestimated by 30% on average. Secondly, we drive the model using global meteorological data and local parameters and find that global data can be used in place of FLUXNET data with only a 7% reduction in total annual simulated GPP. Thirdly, we compare the global meteorological datasets, WFDEI and PRINCETON, to local data and find that the WATCH dataset more closely matches the local meteorological measurements (FLUXNET). Finally, we compare the results from forcing JULES with the remote sensing product MODIS Leaf Area Index (LAI). JULES was modified to accept MODIS LAI at daily timesteps. We show that forcing the model with daily satellite LAI results in only small improvements in predicted GPP at a small number of sites compared to using the default phenology model.