Investigating sources of uncertainty associated with the JULES land surface model

Abstract

The land surface is a key component of the climate system and exchanges energy, water and carbon with the overlying atmosphere. It is the location of the terrestrial carbon sink and changes in the land surface can impact weather and climate at various time and spatial scales. It's ability to act as a source or a sink can influence atmospheric CO2 concentrations. Both models and observations have shown the reduced ability of the land surface to absorb increased anthropogenic CO2 emissions with results from the Coupled Climate{Carbon Cycle Model Intercomparison Project (C4MIP) and phase 5 of the Coupled Model Intercomparison Project (CMIP5) have shown that the terrestrial carbon cycle is a major source of model uncertainty. Land surface models (LSMs) represent the interaction between the biosphere and atmosphere in earth system models (ESMs) and are important for simulating the terrestrial carbon cycle. In the context of land surface modelling, uncertainty arises from an incomplete understanding of land surface processes and the inability to model these processes correctly. As LSMs become more advanced, there is a need to understand their accuracy. In this thesis, the ability of the Joint UK Land Environment Simulator (JULES), the land surface scheme of the UK Met Office Unified Model, to simulate Gross Primary Productivity (GPP) fluxes is evaluated at various spatial scales (point, regional and global) in order to identify and quantify sources of uncertainty in the model. This thesis has three main objectives. Firstly, JULES is evaluated at the point scale across a range of biomes and climatic conditions using local (site-specific), global and satellite datasets. It was found that JULES is biased with total annual GPP underestimated by 16% and 30% across all sites compared to observations when using local and global data, respectively. The model's phenology module was tested by comparing results from simulations using the default phenology model to those forced with leaf area index (LAI) from the MODIS sensor. Model parameters were found to be a minor source of uncertainty compared to the meteorological driving data at the point scale as was the default phenology module in JULES. Secondly, in addition to evaluating simulated GPP fluxes at the point scale, the ability of JULES to simulate GPP at the global and regional scale for 2000{2010 was investigated with being able to simulate interannual variability and simulated global GPP estimates were found to be greater than the observation-based estimates, FLUXNET-MTE and MODIS, by 8% and 25 %, respectively. At the regional scale, differences in GPP between JULES, FLUXNET-MTE and MODIS were observed mostly in the tropics and this was the reason for differences at the global scale. Simulating tropical GPP was found to be a major source of uncertainty in JULES. JULES was found to be insensitive to spatial resolution and when driven with the PRINCETON meteorological dataset, differences between model simulations driven using WFDEI-GPCC and PRINCETON occurred in the tropics (at 5°N{5°S) and extratropics (at 30°N{60°N). Finally, the response of JULES to changes in climate (surface air temperature, precipitation, atmospheric CO2 concentrations) was explored at the global and regional scale. Simulated GPP was found to have greater sensitivity to changes in precipitation and CO2 concentrations than air temperature at the global scale while LAI was sensitive only to changes in temperature and insensitive to changes in precipitation and CO2 concentrations. It was found that model sensitivity to climate at the global scale was determined by its behaviour at the regional scale.