Studying the hypothetical case of a trend superimposed on a random sta
tionary variable, we highlight the strong influence of possible non-st
ationarities on exceedance probability. After a general outline, the s
ubject is analytically developed using the Gumbel distribution, emphas
izing the quick increase of the exceedance probability over time in th
e presence of weak rising trends, and its sensitive underestimation wh
ere the non-stationarity goes unnoticed or is considered negligible. F
inally the work is applied to hydrological series of rainfall and rive
r flow.