Examination of the sources of uncertainty in land-atmosphere model results for boreal ecosystems

This dissertation examines of the sources of uncertainty in land-atmosphere model results in boreal ecosystems. The research is done as part of the Boreal Ecosystem-Atmosphere Study (BOREAS). The relative importance of differences in model parameterizations, uncertainty in meteorological forcing data, and uncertainty in soil/vegetation data in the uncertainty in model results is examined. As a first step, a land-atmosphere model is improved to represent processes typical for boreal landscapes. The model is thoroughly validated using point scale surface flux, soil temperature and soil moisture observations. It is found that the accuracy performance of the model is essentially the same as the accuracy of the input and validation data. As a second step the model is applied over two smaller study areas inside the BOREAS region. Land cover classifications determined by two different sensors at different resolutions are used to examine the uncertainty in model outputs caused by misclassifications and different representations of spatial variability in land cover data. Further, weather forecast data from the European Center for Medium-Range Weather Forecasts (ECMWF) Numerical Weather Prediction (NWP) model, as well as surface observations, are used as meteorological model forcing data, and the results obtained using the two different data sources are analyzed. The intercomparison between obtained results allows the assessment of the uncertainty in model results caused by uncertainty in meteorological forcing data. As a final step, the model is applied over a large domain inside the BOREAS region. Again, both surface observations and weather forecast data are used as meteorological model forcing data. Model results are compared to results from the ECMWF NWP model. This intercomparison allows the examination of the uncertainty in model results caused by both uncertainty in meteorological forcing data and differences in model parameterizations. It is found that, over a simulation period of approximately 200 days, uncertainty in land cover classification causes a limited uncertainty in the individual components of the energy and water balances. These uncertainties have been found to be originated by misclassifications, while the effect of spatial resolution is secondary. The precipitation, incoming solar and longwave radiation, and, during the winter, the air temperature, are the forcing variables causing the highest uncertainties in model results. The parameterization of a moss layer has been found to be the model improvement causing the highest differences in model results. Overall, both differences in model parameterizations and uncertainties in meteorological forcing data have the highest impact on model results, with the uncertainty in land cover classification only having a secondary effect on model results.