cadrev/lstm-flood-prediction ? reverse-engineered prompt

Build me a simple notebook project that explores using an LSTM to predict river water levels from past measurements.

I want to be able to load the included dataset, train the model on historical water level data only, and then compare the predicted levels against the actual trend so I can see whether the network does a decent job following the pattern. Keep it focused on the flood prediction experiment, with clear steps for cleaning the data, preparing sequences, training the model, and showing the results in a plot like the sample image. If you need to look up the current best way to do this in notebook form, go ahead.