| 26 May 2025
CAMELS-LUX: Highly Resolved Hydro-Meteorological and Atmospheric Data for Physiographically Characterized Catchments around Luxembourg
Abstract. Harmonized large-sample datasets have become a central pillar of hydrological research, particularly in the machinelearning era, where data-based algorithms and machine-learning techniques are gaining increasing importance in daily life. The CAMELS-LUX dataset (Catchment Attributes and MEteorology for Large-sample Studies – LUXembourg) described here covers 56 nested catchments (0.46 km2 – 4256.62 km2) that contribute to the Luxembourgish stream network. While Luxembourg has a relatively homogeneous climate, the physiography varies significantly on a small scale making it a suitable study area for investigating different hydrological processes, such as runoff generation or groundwater recharge. The CAMELS-LUX dataset contains hydrological observations, meteorological data, and atmospheric reanalysis data from 2004–2021. Moreover, comprehensive physiographic catchment characteristics are provided that incorporate geology classes, land use classes, and a range of topographic indices. CAMELS-LUX is distinctive as the first dataset in the CAMELS series that offers data at three different temporal resolutions: daily, hourly, and in a 15-minute time step. Furthermore, CAMELS-LUX includes a series of flash floods in 2016 and 2018 as well as major large floods in 2010 and 2021. The extensive information contained in CAMELS-LUX is instrumental in advancing our understanding of varying discharge behaviour within Luxembourg and beyond. The CAMELS-LUX dataset has been utilized to develop and train a Long-Short-Term-Memory (LSTM) model, that simulates discharge time series, providing a benchmark for subsequent hydrological modelling efforts in the area. The model based on this dataset sufficiently reproduces hydrological rainfall-runoff dependencies and can be applied to simulate discharge in sparsely gauged basins for approximation. The CAMELS-LUX dataset is available on zenodo: https://doi.org/10.5281/zenodo.13846619 (Nijzink et al., 2024).
