This file was generated on 2025-09-30 by Mohamad Hakam Shams Eddin A GENERAL INFORMATION 1. Title of the dataset: RiverMamba: A State Space Model for Global River Discharge and Flood Forecasting [data set] 2. Brief description of the research project and its aims: This is the dataset used in the RiverMamba paper (see https://arxiv.org/abs/2505.22535). The aim of the RiverMamba project is to develop a deep learning model that is pretrained with long-term reanalysis data and fine-tuned on observations to forecast global river discharge and floods up to 7 days lead time on a 0.05° grid. The dataset includes the necessary data to train and run the model. The dataset includes the following: 1- CPC precipitation data 2- ECMWF-HRES meteorological forecasts 3-ERA5-Land reanalysis data 4- GloFAS reanalysis data 5- GloFAS static data 6- Reforecasts generated by RiverMamba 7- Pretrained RiverMamba models. 3. Author Information A. Project Supervisor (Principal Investigator) Contact Information Name:Jürgen Gall Institution: Institute of Computer Science III, Department of Information Systems and Artificial Intelligence Address: Friedrich-Hirzebruch-Allee 8, 53115 Bonn Email: gall@iai.uni-bonn.de B. In case of questions related to this dataset, please contact: Name: Mohamad Hakam Shams Eddin Institution: Institute of Computer Science III, Department of Information Systems and Artificial Intelligence Address: Friedrich-Hirzebruch-Allee 8, 53115 Bonn Email: shams@iai.uni-bonn.de C. Collaborator: Name: Yikui Zhang Institution: Research Centre Jülich Address: Wilhelm-Johnen-Straße, 52428 Jülich Email: yik.zhang@fz-juelich.de D. Collaborator: Name: Stefan Kollet Institution: Research Centre Jülich Address: Wilhelm-Johnen-Straße, 52428 Jülich Email: s.kollet@fz-juelich.de 4. Date of data collection: 2024-07-01 5. Information about funding sources that supported the collection of the data: This work was supported by the Federal Ministry of Research, Technology, and Space under grant no. 01|S24075A-D RAINA and by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) – SFB 1502/1–2022 – project no. 450058266 within the Collaborative Research Center (CRC) for the project Regional Climate Change: Disentangling the Role of Land Use and Water Management (DETECT). 6. Language of the dataset: English 7. Geographic location of data collection: Global (90◦N-60◦S, 180◦W-180◦E) B DATA | FILE OVERVIEW 1. File List: . ├── CPC_Global │   └── CPC_Global.7z ├── ECMWF_HRES_Global │   ├── │   │   ├── .7z.001 │   │   ├── .7z.002 │   │   └── .7z.003 │   └── hres_statistics_train.json ├── ERA5-Land_Reanalysis_Global │   ├── │   │   ├── .7z.001 │   │   ├── .7z.002 │   │   ├── .7z.003 │   │   └── .7z.004 │   └── ERA5_Land_statistics_train.json ├── GloFAS_Reanalysis_Global │   ├── .7z │   └── GloFAS_statistics_train.json ├── GloFAS_Static │   └── GloFAS_Static.7z ├── Licenses │   ├── HydroATLAS_HydroRIVERS │   │   ├── ClimateClassification.txt │   │   ├── GloRiC_TechDoc_v10.pdf │   │   ├── GLWD_TechDoc_v2_delta.pdf │   │   ├── HydroRIVERS_TechDoc_v10.pdf │   │   ├── HydroSHEDS_TechDoc_v1_4.pdf │   │   └── Overview_NeuralFAS_HydroRIVERS_static.ods │   ├── licence-to-use-copernicus-products.pdf │   ├── license_cpc_data.txt │   ├── license_ecmwf_hres_data.txt │   ├── license_era5_land_data.txt │   ├── license_GloFAS_cems-floods.pdf │   ├── license_lisflood_static_data.txt │   └── license_rivermamba_reforecasts_models.txt ├── Reforecasts │   ├── GRDC_Meta.txt │   ├── LSTM │   │   ├── LSTM_glofas_reanalysis.zip │   │   └── LSTM_grdc_obs.zip │   └── RiverMamba │   ├── RiverMamba_glofas_reanalysis_full_map.zip │   ├── RiverMamba_glofas_reanalysis.zip │   └── RiverMamba_grdc_obs.zip └── RiverMamba_pretrained_models ├── RiverMamba_aifas_grdc_obs │   ├── RiverMamba_aifas_grdc_obs.pth │   └── RiverMamba_aifas_grdc_obs.txt ├── RiverMamba_aifas_reanalysis │   ├── RiverMamba_aifas_reanalysis.pth │   └── RiverMamba_aifas_reanalysis.txt └── RiverMamba_full_map_reanalysis ├── RiverMamba_full_map_reanalysis.pth └── RiverMamba_full_map_reanalysis.txt 2. Are there multiple versions of the dataset? No C SHARING/ACCESS INFORMATION 1. Was data derived from another source? Yes - ERA5-Land Reanalysis: ERA5-Land hourly data from 1950 to present. Copernicus Climate Change Service (C3S) Climate Data Store (CDS) - CPC: CPC Global Unified Gauge-Based Analysis of Daily Precipitation data provided by the NOAA PSL, Boulder, Colorado, USA, from their website at https://psl.noaa.gov - ECMWF-HRES: the data is based on data and products from the archive of the European Centre for Medium-Range Weather Forecasts (ECMWF). - GloFAS Reanalysis: Joint Research Center, Copernicus Emergency Management Service (2019): River discharge and related historical data from the Global Flood Awareness System. Early Warning Data Store (EWDS). DOI: 10.24381/cds.a4fdd6b9 - Static data: - LISFLOOD: LISFLOOD static and parameter maps for GloFAS. European Commission, Joint Research Centre (JRC) [Dataset] PID: http://data.europa.eu/89h/68050d73-9c06-499c-a441-dc5053cb0c86 - HydroRIVERS and HydroATLAS: data were provided from https://www.hydrosheds.org/products/hydrorivers 2. Licenses/restrictions placed on the data: This dataset processing for RiverMamba: Creative Commons Attribution CC BY 4.0: https://creativecommons.org/licenses/by/4.0/ Licesnes of the data: - CPC data: There is no usage restrictions - ECMWF-HRES: The Creative Commons Attribution 4.0 International (CC BY 4.0) - ERA5-Land: The Creative Commons Attribution 4.0 International (CC BY 4.0) - Rivermamba reforecasts and pretrained models: The Creative Commons Attribution 4.0 International (CC BY 4.0) - GloFAS reanalysis: CEMS-FLOODS datasets licence (https://ewds.climate.copernicus.eu/datasets/cems-glofas-historical?tab=overview) - Static data: - LISTFLOOD static data: The Creative Commons Attribution 4.0 International (CC BY 4.0) - HydroATLAS and HydroRIVERS: HydroSHEDS core products license (https://www.hydrosheds.org/products/hydrorivers). The climate data included in this data package are climate indices derived from CRU TS v3.23 data, which is made available here under the Open Database License (ODbL) (https://opendatacommons.org/licenses/odbl/1.0/). See the folder "Licenses" for more details about the licenses of the data 3. Links to publications that cite or use the data: see paper: RiverMamba: A State Space Model for Global River Discharge and Flood Forecasting, NeurIPS 2025. link: https://arxiv.org/abs/2505.22535 D METHODOLOGICAL INFORMATION 1. Description of methods used for collection/generation of data: - CPC data: This observational precipitation estimates are obtained from the National Oceanic and Atmospheric Administration (NOAA), Climate Prediction Center (CPC). The CPC precipitation product is accumulated daily and provided globally at 0.5◦ × 0.5◦. Operational CPC data can be obtained from https://psl.noaa.gov/data/gridded/data.cpc.globalprecip.html - ECMWF-HRES: We use the deterministic forecast of the ECMWF Integrated Forecast System (IFS) High Resolution (HRES) atmospheric model. The HRES data were obtained from the ECMWF Archive Catalogue https://www.ecmwf.int/en/forecasts/dataset/operational-archive. We use HRES up to 7 days lead time and once per day at 00:00 UTC. The data does not include any forecast for nowcasting at the time of the forecasts (analysis step). The processed data include 2 instantaneous and 5 daily accumulated variables at the surface level forecasts - ERA5-Land: Data are obtained from the Copernicus Climate Change Service (C3S) Climate Data Store (CDS) https://doi.org/10.24381/cds.e2161bac. We processed 14 instantaneous state variables at 00:00 UTC and 18 daily accumulated state variables (00:00 UTC previous day to 00:00 UTC current day) - Rivermamba reforecasts: These are the medium-range forecasts of river discahrge as produced by RiverMamba deep learning model. The reforecasts files are generated by the software https://github.com/HakamShams/RiverMamba_code v1.0.0. - RiverMamba pretrained models: These are the pretrained models generated by the software https://github.com/HakamShams/RiverMamba_code v1.0.0. - GloFAS reanalysis: The reanalysis is generated by coupling surface and subsurface runoff from the ERA5 reanalysis, produced by the H-TESSEL and surface model with the LISFLOOD hydrological and river routing model. The daily GloFAS reanalysis discharge data represents the mean value between 00:00 UTC previous day and 00:00 UTC current day. The dataset is publicly available on Climate Data Store and Early Warning Data Store (EWDS) https://doi.org/10.24381/cds.a4fdd6b9 - LISFLOOD static data: The LISFLOOD static maps are obtained from the Joint Research Centre Data Catalogue http://data.europa.eu/89h/68050d73-9c06-499c-a441-dc5053cb0c86. This includes 96 time-invariant variables from 7 different categories - HydroATLAS and HydroRIVERS: We obtained static river attributes from https://www.hydrosheds.org/products/hydrorivers See paper for more details (RiverMamba: A State Space Model for Global River Discharge and Flood Forecasting, NeurIPS 2025) 2. Methods for processing the data: - CPC data: Data are mapped onto the GloFAS domain (regular latitude-longitude grid) using nearest point algorithm which preserves the original coarse grid structure but refines the resolution. We did not do any modification for the CPC time zones. - ECMWF-HRES: Data are processed from 2010 to 2024. Data before 2010 were replaced by ERA5-Land in the RiverMamba framework. To match the resolution of the target GloFAS domain, we regridded HRES to 0.05◦ × 0.05◦ regular latitude-longitude grid. - ERA5-Land: Data are provided originally at 0.1◦ × 0.1◦. We mapped the data onto the GloFAS regular latitude and longitude (Plate Carrée projection) using bilinear mapping. - Rivermamba reforecasts: The reforecasts files are generated by the software https://github.com/HakamShams/RiverMamba_code v1.0.0. - RiverMamba pretrained models: These are the pretrained models generated by the software https://github.com/HakamShams/RiverMamba_code v1.0.0. - GloFAS reanalysis: we don't do any preprocessing for this dataset - LISFLOOD static data: The maps are provided at the same resolution as GloFAS at 3 arcmin and covering the globe (90◦N-60◦S, 180◦W-180◦E). We excluded the lakes, reservoirs and some static water demand maps. We don't do any further preprocessing for this dataset - HydroATLAS and HydroRIVERS: The original data are stored as shape files. To map them onto the GloFAS domain, we first extract the coordinates of the rivers and then project them with the grid points on the WGS-84 ellipsoid. Then, for each GloFAS grid point, depending on the attribute type, we either average the attributes or take the most frequent attribute within a radius of 5 km. If no attributes were found, we increase the radius to 12 km, and 24 km, respectively. We processed 299 river feature attributes overall. The global data has 6,221,926 points along the x dimension. This represents points on land at 0.05° resolution. Points on oceans where removed and the image is flattened to generate the output points on land. See paper for more details (RiverMamba: A State Space Model for Global River Discharge and Flood Forecasting, NeurIPS 2025) 3. Instrument- and/or software-specific information needed to interpret the data: For analysing and processing of the data, see the code on GitHub: https://github.com/HakamShams/RiverMamba_code v1.0.0 The software is written in Python programming language 4. People involved in sample collection, processing, analysis and/or submission: Mohamad Hakam Shamd Eddin (shams@iai.uni-bonn.de) ---------------------------------------------------------------------- E DATA-SPECIFIC INFORMATION FOR GloFAS Reanalysis: 1. Variable list including full names and definitions of column headings for tabular data: - acc_rod24 | runoff water equivalent | kg/m^2 | surface and subsurface | accumulated - dis24 | river discharge in the last 24 hours | m^3/s | surface | averaged over 24 hours - sd | snow depth water equivalent | kg/m^2 | surface | instantaneous - swi | soil wetness index | - | root zone | instantaneous 2. Missing data codes/symbols: NaN. E DATA-SPECIFIC INFORMATION FOR ERA5-Land Reanalysis: 1. Variable list including full names and definitions of column headings for tabular data: - d2m | 2m dewpoint temperature | K | 2m | instantaneous - e | total evaporation | m of water equivalent | surface | accumulated - es | snow evaporation | m of water equivalent | surface | accumulated - evabs | evaporation from bare soil | m of water equivalent | surface | accumulated - evaow | evaporation from open water | m of water | surface | accumulated | surfaces excluding oceans | equivalent - evatc | evaporation from the top of | m of water equivalent | surface | accumulated | canopy - evavt | evaporation from vegetation | m of water equivalent | surface | accumulated | transpiration - lai_hv | leaf area index | m^2/m^2 | 2m | instantaneous | high vegetation - lai_lv | leaf area index | m^2/m^2 | 2m | instantaneous | low vegetation - pev | potential evaporation | m | 2m | accumulated - sf | Snowfall | m of water equivalent | surface | accumulated - skt | skin temperature | K | surface | instantaneous - slhf | surface latent heat flux | J/m^2 | surface | accumulated - smlt | snowmelt | m of water equivalent | surface | accumulated - sp | surface pressure | Pa | surface | instantaneous - src | skin reservoir content | m of water equivalent | surface | instantaneous - sro | surface runoff | m | surface | accumulated - sshf | surface sensible heat flux | J/m^2 | surface | accumulated - ssr | surface net solar radiation | J/m^2 | surface | accumulated - ssrd | surface solar radiation | J/m^2 | surface | accumulate | downwards - ssro | subsurface runoff | m | subsurface | accumulated - stl1 | soil temperature | K | soil layer (0 - 7 cm) | instantaneous - str | surface net thermal radiation | J/m^2 | surface | accumulated - strd | surface thermal radiation | J/m^2 | surface | accumulated | downwards - swvl1 | volumetric soil water | m^3/m^3 | soil layer (0 - 7 cm) | instantaneous - swvl2 | volumetric soil water | m^3/m^3 | soil layer (7 - 28 cm) | instantaneous - swvl3 | volumetric soil water | m^3/m^3 | soil layer (28 - 100 cm) | instantaneous - swvl4 | volumetric soil water | m^3/m^3 | soil layer (100 - 289 cm) | instantaneous - t2m | 2m temperature | K | 2m | instantaneous - tp | total precipitation | m | surface | accumulated - u10 | 10 metre U wind component | m/s | 10m | instantaneous - v10 | 10 metre V wind component | m/s | 10m | instantaneous 2. Missing data codes/symbols: NaN. E DATA-SPECIFIC INFORMATION FOR ECMWF-HRES Forecasts: 1. Variable list including full names and definitions of column headings for tabular data: - e | total evaporation | m of water equivalent | surface | accumulated - sf | snowfall | m of water equivalent | surface | accumulated - sp | surface pressure | Pa | surface | instantaneous - ssr | surface net solar radiation | J/m^2 | surface | accumulated - str | surface net thermal radiation | J/m^2 | surface | accumulated - t2m | 2m temperature | K | 2m | instantaneous - tp | total precipitation | m | surface | accumulated 3. Missing data codes/symbols: NaN. E DATA-SPECIFIC INFORMATION FOR CPC: 1. Variable list including full names and definitions of column headings for tabular data: - precip | total precipitation | mm/day | surface | accumulated 3. Missing data codes/symbols: NaN. E DATA-SPECIFIC INFORMATION FOR LISFLOOD static data: 1. Variable list including full names and definitions of column headings for tabular data: see the documentation in https://data.jrc.ec.europa.eu/dataset/68050d73-9c06-499c-a441-dc5053cb0c86 3. Missing data codes/symbols: NaN. E DATA-SPECIFIC INFORMATION FOR HydroRIVERS static data: 1. Variable list including full names and definitions of column headings for tabular data: see the list in ./Licenses/HydroATLAS_HydroRIVERS/Overview_NeuralFAS_HydroRIVERS_static.ods 3. Missing data codes/symbols: NaN. E DATA-SPECIFIC INFORMATION FOR RiverMamba Reforecasts: 1. Variable list including full names and definitions of column headings for tabular data: - dis24 | river discharge forecasts | m^3/s | surface | averaged over 24 hours 3. Missing data codes/symbols: NaN.