Comparing Land Surface Models with Satellite and in Situ Observations at High Latitudes
Quegan, Shaun1; Kantzas, Euripides1; Lomas, Mark2
1University of Sheffield, UNITED KINGDOM; 2Unjversity of Sheffield, UNITED KINGDOM

Northern high latitudes are predicted to be the part of the planet suffering the largest temperature rises under climate change, with associated changes in precipitation patterns. This will have major effects on the land carbon and hydrological cycles at these latitudes, through the consequences for land cover, phenology and length of growing season, fire regimes, snow cover, runoff into the Arctic Ocean, evapotranspiration, permafrost and greenhouse gas fluxes. Much of our current understanding of these changes is through the application of land surface models that are driven by climate or operate in a coupled mode that can represent land-atmosphere interactions and feedbacks. Such an approach is essential since many of processes interact with each other and are inter-dependent, and models are capable of representing how the overall system behaves and will evolve. Unfortunately, most, if not all, of the models available for applying this full system approach are poorly constrained by data, and we have only a rudimentary knowledge of their biases and uncertainties, and how well they represent the many contributing processes. Hence the accuracy in their calculations and predictions for high latitudes cannot at present be adequately quantified. Until recent times, the disconnection of models from data was understandable, as models were mainly developed at a time when large-scale satellite datasets covering long time periods were comparatively rare and not necessarily of good quality. The current situation is quite different, with a wide range of sensors providing global scale estimates of key land variables relevant to carbon and water calculations. In particular, at high latitudes we now have satellites time series with temporal coverage of more than a decade to more than 30 years in the following variables: burnt area, land cover, snow covered area, start and end dates of the snow season, snow water equivalent, surface water and periods of freeze-thaw. We also have new amalgamated datasets from the ESA Climate Change Initiative activity on soil moisture and a recent pan-boreal dataset of forest biomass (Santoro et al., 2010). There are also very important in situ datasets on basin scale runoff and permafrost. This combined dataset provides an excellent opportunity to test land surface models and to assess (a) whether they are consistent with data and (b) what the consequences are for carbon and water balance calculations. This is the topic of this paper, and will deal with calculations from several state-of-the-art models, including the carbon-water-energy model ORCHIDEE from the Laboratoire des Sciences du Climat et l'Environnement, the Lund-Potsdam-Jena (LPJ) model, the Joint UK Land Environment Simulator (JULES), the NCAR Community Land Model (CLM4) and the Sheffield Dynamic Global Vegetation Model (SDGVM). This paper will discuss the representations of processes and land variables in models, and will demonstrate the following: a. There are major discrepancies between all the models considered and satellite observations of burnt area. The data show that fire is highly variable in space and time, but the models exhibit little such variability. There are also marked differences between the models regarding the carbon emissions per unit of area burned, which are related to how the fuel load is defined in the models and the size of the fuel-related pools (litter and biomass). b. The timings of the snow season in the models agree well with satellite observations, but in general the models predict much greater snow covered area and snow water equivalent than is observed by satellites. There are also significant differences between different models and between different models and the Globsnow dataset. The match between Globsnow and the models is better in Eurasia than in N. America; there are marked disagreements in the latter region, but these seem to be connected to anomalies in the climate data driving the models. c. There are significant discrepancies between the models and data as regards runoff, and marked disagreement between the models. As for snow, at least part of the reason is differences in the driving climate data. d. There are large differences between different land cover maps for the high latitude region, but the effect on net carbon exchange and the runoff calculated by the models is relatively small. In fact, it has its largest impact on emissions due to fire, with variations of around 10% caused by land cover differences. The paper will also discuss the use of satellite–derived biomass and vegetation activity in models and as constraints on the modelling process. References Santoro, M., Beer, C., Cartus, O., Schmullius, C., Shvidenko, A., McCallum, I., Wegmuller, U., Wiesmann, A. (2010). Retrieval of growing stock volume in boreal forest using hyper-temporal series of Envisat ASAR ScanSAR backscatter measurements, Rem. Sens. Env., doi:10.1016/j.rse.2010.09.018.