WP leader: Wageningen University, Gerard Heuvelink
We will validate and analyse uncertainty of the e-SOTER output by comparing it with independent validation data for both 1:1 million-scale windows in Europe and for two 1:250 000-scale pilot areas resulting in a validation strategy and accuracy assessment of e-SOTER products. The work will consist of:
- the design of a validation strategy based on statistical sampling and inference methods. Basic design principles include precise definition of the objective, specification of the quality measure (width of confidence intervals, accuracy of bias estimate, power in hypothesis testing), inventory of constraints and prior information, assessment of anticipated operational costs and set-up of a detailed field and lab protocol. The validation strategy must be able to incorporate various types of existing data.
- the collection of independent validation data for both 1:1 million-scale test windows in Europe and for two of the four 1:250 000-scale pilots using the generic sampling design procedure worked out above.
- the comparison of the e-SOTER data with the independent validation data. A comparison will be made between the accuracy obtained with products of WP1/2 with that of WP3.
- identification of those input data that have large uncertainties and for which the SOTER methodologies are sensitive. The uncertainty in these inputs is assessed and quantified by probability distribution functions. The assessment is done by identifying the sources of uncertainty and consulting e.g. instrument precision specifications, GIS meta-information on data-quality, replication measurements, expert elicitation, purity indices for large and small-scale soil maps, quantified spatial interpolation errors. The input uncertainty assessment must also consider cross-correlations and spatial autocorrelation.
- analysis of the uncertainty propagation by applying Monte Carlo simulation for selected test windows (minimum two). The contribution of individual uncertain inputs to the final uncertainty is also assessed, thus identifying the weakest links in the SOTER modeling chain. The propagated uncertainty in the SOTER products is also compared to the uncertainty that resulted from the validation analysis (Task 4.3), thus providing insight to the contribution of model structural and parameter errors.