NeuralHydrology

NeuralHydrology is the PyTorch framework from Frederik Kratzert and collaborators that made LSTM-based rainfall-runoff modelling reproducible. It handles the full experimental loop — CAMELS-style dataset loading, training, hyperparameter configs, multi-basin regional models, fine-tuning and evaluation — and underlies a large share of the published work showing deep learning beating conceptual models at ungauged and gauged prediction alike.

For practitioners it is the most direct path to a state-of-the-art data-driven baseline: if you have forcing and discharge data, a regional LSTM is a few config files away. Research-grade software, but unusually well documented for it.