Evaluation of rainfall networks using entropy. I. Theoretical development

This paper, the first in a series of two, develops an entropy-based approach for evaluating rainfall networks. Space and time dependencies between raingages are examined by autocovariance and cross-covariance matrices. Multivariate distributions, associated with different dependencies, are obtained using the principle of maximum entropy (POME). Formulas for entropy (uncertainty in data of one raingage), joint entropy (uncertainty in data of two or more raingages) and transinformation (common information content among two or more raingages) are derived for each distribution, based on normal data. The decision whether to keep or eliminate a raingage depends entirely on reduction or gain of information at that raingage. The lines of equal information (isoinformation contours) are defined by considering two raingages (bivariate case) and many raingages (multivariate case).