A new approach is developed for the automatic (computer based) exploration
and analysis of hydrological data, particularly focused on the identificati
on of shifting relationships among hydrological variables. The methodology
developed is applicable to many hydrological problems, such as identificati
on of multiple stage-discharge relationships in a river section, data homog
eneity analysis, analysis of temporal consistency of hydrological data, det
ection of outliers, and determination of shifts and trends in hydrological
time series. Such problems are examined here as particular applications of
the single methodology developed. A general mathematical representation of
the data exploration problem is initially proposed, based on set theory con
siderations. Several statistical tests, as well as auxiliary information of
physical conditions, are systematically combined to form an objective func
tion to be optimised. This objective function represents the performance of
a solution, (where a solution is a specific partitioning of a data set int
o subperiods), in a manner, that holds in each subperiod a single relations
hip among data values. It is shown that an exhaustive search of all candida
te solutions is intractable. Therefore, a heuristic algorithm is proposed,
which emulates the exploratory data analysis of the human expert. This algo
rithm encodes a number of search strategies in a pattern directed computer
program and results in an automatic determination of a satisfactory solutio
n.