Why hydrological predictions should be evaluated using information theory