Regionalising the available water capacity from readily available data

In this study the available water capacity (AWC) is predicted for a 10 km2 catchment in Northern Bavaria, Germany using only readily available data. Therefore classification and regression trees (CART), a feed-forward multilayer perceptron (MLP) and kriging are applied. In this context we propose an alternative procedure for fitting the covariance function. The different regionalisation model performances are compared and the influence of the sample size on the model performance is investigated systematically. Unfortunately the three regionalisation approaches provide AWC maps containing only limited information indicating that the information content of the data is not sufficient for regionalisation. We determine that the model predictions are unlikely to be significantly improved by the addition of more data points. Other possibilities regarding the improvement of the AWC predictions are discussed in detail.