01
National streamflow LSTM
A single recurrent neural network trained on CAMELS-ES that simulates daily streamflow at any point of peninsular Spain — gauged or ungauged.
A KAJO research initiative · Map app coming soon
The story behind
every Spanish river.
CAMELS-ES is an open large-sample dataset of Spanish catchments — and the ground truth behind a deep-learning model that simulates daily streamflow anywhere on the peninsula. This page tells the story; the interactive map app will live behind it.
/ Background
Why camels,
why beavers?
CAMELS — Catchment Attributes and Meteorology for Large-sample Studies — is the backbone dataset of modern data-driven hydrology. Versions exist for the US, Chile, Brazil, Australia and the UK. This work introduces one for Spain.
The beavers are the LSTM networks: tireless builders that learn how rainfall, soil and terrain shape rivers, day after day. Together they make every Spanish stream observable — even ungauged ones.
Master's thesis, Universidad de Alcalá (2022–2023) · Jesús Casado-Rodríguez · published in Ingeniería del Agua (2026).
/ The dataset
CAMELS-ES,
a national catchment archive.
Daily forcings and observed streamflow paired with hundreds of static descriptors per catchment — the first large-sample hydrological benchmark for Spain, distributed openly via Zenodo.
Coverage
Peninsular Spain
Time series
Daily meteorology + streamflow
Static attributes
Geomorphology, soil, land use, vegetation
Initiative
Aligned with global CARAVAN
/ The models
Two LSTMs,
one peninsula.
02
LISFLOOD-OS emulator
A second LSTM trained to reproduce the LISFLOOD-OS hydrological model used by EFAS, with the same forcings, static maps and calibrated parameters as input.
/ Coming next
The interactive map
lives behind this page.
Browse Spanish catchments, inspect simulated streamflow time series and compare CAMELS-ES observations with both LSTM models — directly on the map.
Map application — in development
/ Cite this work
Casado-Rodríguez, J., Ramos-Gomes, G., & Salamon, P. (2026). Simulación del caudal en España utilizando redes neuronales Long Short-Term Memory. Ingeniería del Agua, 30(1), 63–78.
kajoservices.com