Effects of measurement uncertainties of meteorological data on estimates of site water balance components

- We developed a robust approach for uncertainty assessment of input data.
- Simulation results were evaluated on measured evapotranspiration.
- The uncertainty of eddy-covariance measurements was tackled.
- Empirical and process based models are similarly affected by input uncertainties.
- Uncertainties of meteorological input and of parameters have a similar importance.

SummaryNumerical water balance models are widely used in ecological and hydro sciences. However, their application is related to specific problems and uncertainties. The reliability of model prediction depends on (i) model concept, (ii) parameters, (iii) uncertainty of input data, and (iv) uncertainty of reference data. How model concept (i) and parameters (ii) effect the model's performance is an often treated problem. However, the effects of (iii) and (iv) are typically ignored or only barely treated in context of regionalisation and generalisation. In this study, the actual measurement uncertainties of input and reference data are the main focus. Furthermore, the evaluation of model results is analysed with regard to uncertainties of reference data. A special feature is the use of evapotranspiration (measured via the eddy covariance) instead of runoff for evaluation of simulation results. It is shown that seemingly small uncertainties of measurements can create significant uncertainties in simulation results depending on the temporal scale of investigation. As an example, the uncertainty of measurements of daily global radiation sum up to an uncertainty of 250 MJ (equivalent to 100 mm) on an annual scale, which causes an uncertainty of 40 mm in simulated grass-reverence evapotranspiration. Summarised and generalised, the measurement uncertainties of all input data create an uncertainty on average of around 5% in the simulated annual evapotranspiration and of around 10% in the simulated annual seepage. However, the effects can be significantly higher in years with extreme events and can reach up to 15%. It is demonstrated that uncertainties of individual variables are not simply superposed but interact in a complex way. Thereby, it has become apparent that the effects of measurement uncertainties on model results are similar for complex and for simple models.