From site to map - an analysis of predictability and uncertainty when scaling-up results from process-oriented models to regional maps 

There is a great interest in generalizing models on data input both on models that require less input data but also models that require data of fewer observations of those of national soil and forest inventories. The scale issues have been recognized since the last two decades due to the development of computers. Errors and uncertainty in model predictions arises from the five sources; Field measurements, Errors in input boundary conditions, input variables, Errors in the estimation of effective model parameter values, Model structural errors and Up-scaling models to predict on a larger area than the model was assumed to predict on. Sensitivity analysis is made to assess these sources. More often calibration is made by less subjective and more automatic methods, with the principle of minimization of an objective function through the modification of in-put parameters. The acceptance of a model for its intended use and how the model meets specified performance requirements, validation, is also important for the accuracy of the model predictions. When up-scaling models there are three different scales involved within applied research. These are observation, model and policy scale and between these scales there are several possibilities of up-scaling. There are three main reasons why models are scale specific. First, different processes are dominant at different scales, since models tend to focus on the most important processes and ignore the least ones. Second the availability of data at different scales. Data is often available at small scales. Third reason is the scale of the models variables. The base fundamentals when an up-scaling environmental model is suggested to the ecological hierarchy theory were different processes appears on different scales and affects each other. Model simplification is the most commonly way of up-scaling environmental research models since the models used in this context are often complex. This method can be reached by the two different approaches meta-modelling and lumped conceptual modelling. When estimating the overall uncertainty of the models predictions, an overall, an input and a model structure uncertainty analysis should be made. An important task is to identify the sources of uncertainty that contributed most to the overall uncertainty in the output variable. Since model predictions are uncertain, it is highly recommended that the future research with environmental models take the scale issues and the uncertainty issues into concern. For more accurate and reliable national predictions it is necessary to up-scale observed data and/or results from environmental models and make uncertainty analysis on the model and its outputs. There is also a need of presenting the models predictions together with the uncertainties of the model results in order to facilitate policy decisions. The overall objective of this PhD project is to investigate the uncertainties in the environmental models predictions that arise when up-scaling model predictions to regional and national level.