My hydrology training is dated, and it recommended using Log Pearson Type III, for stations with long term record to help estimate skew. I have more recently learned the Gumbel distribution can be very helpful when data includes or is dealing with extremes, such as a very low frequency event(s) included in 30-50 years data. We had such an event and one of the USFS scientists presented how he used Gumbel distribution to more effectively fit data projections when extreme value such as the millennial storm value with flood peak is included. In this instance, it was the nearby long term rainfall records of several stations and Doppler rainfall coverage over time that verified the extreme nature of the event.
Most hydrologists used the Generalized Extreme Value (GEV) distribution, as it generally provides a good fit of most flood dat. This distribution also has some theoretical foundation for block maxima data (such as annual floods). Of course, the Gumbel distribution would be a special case of the GEV distribution. You can generally find GEV codes in R. We have recently compared the performance of the GEV against the Peak Over Threshold (POT) approach, which again has a strong theoretical foundation for extreme data, and the GEV performed equally well and much simpler to apply for hydrologist.