PADR-Net reproducibility archive: Physics-Informed Reservoir Learning for Shallow-Water Flood Modelling

Reproducibility archive for “Physics-Informed Reservoir Learning for Shallow-Water Flood Modelling” This archive provides the reproducibility package for the manuscript “Physics-Informed Reservoir Learning for Shallow-Water Flood Modelling.” It supports the full computational workflow of PADR-Net — the Physics-Aware Deep Reservoir Network — a physics-informed learning framework that combines an Echo State Network reservoir with a shallow-water-equation residual penalty for flood-depth reconstruction, hydrodynamic consistency assessment, and flood-severity prediction. The study evaluates PADR-Net on controlled synthetic flood regimes and on a harmonized archive of 243 African flood events spanning 2000–2024 across three macro-regions: West Africa (Niger–Benue), East Africa (Nile headwaters), and Southern Africa (Limpopo–Zambezi). The archive is designed to make the manuscript’s numerical experiments transparent, repeatable, and auditable by providing the scripts, processed metadata, result tables, environment files, and publication figures needed to reproduce the reported analyses. This reproducibility package contains: Python scripts for data preparation, model training, ablation analysis, transfer validation, uncertainty analysis, and figure generation. Pre-computed result tables for the M0–M8 ablation experiments, nested predictor comparisons, physics-weight sensitivity, leave-one-region-out validation, leave-one-year-out validation, and bootstrap confidence intervals. Publication figures in multiple formats, including PNG, SVG, and EPS. Environment files, including environment.yml and requirements.txt, to reproduce the computational setup. Metadata for the 243-event African flood inventory, including regional assignment, event timing, severity class, and validation split information. Download instructions and source metadata for public external datasets, including ERA5 reanalysis data from the Copernicus Climate Data Store. Scripts to regenerate the main tables and publication figures from the pre-computed outputs without requiring users to re-download all raw external data. The main publication outputs can be reproduced from the pre-computed tables using: conda env create -f environment/environment.yml conda activate padrnet python code/scripts/06_make_figures.py The archive is organized so that reviewers and readers can reproduce the manuscript tables and figures directly from the released result files, while also retaining the scripts required to rerun the model pipeline when the external datasets are available locally. Key reported results include the full PADR-Net configuration using rainfall, hydroclimatic memory, exposure, and hydrodynamic descriptors (R+M+E+H). In the archived experiment set, the selected physics weight is lambda* = 0.1, with Spearman rank correlation rho = 0.671, PR-AUC = 0.703, and NSE_depth = 0.643. The archive also includes the ablation outputs used to evaluate the incremental role of hydrodynamic descriptors and the physics penalty. In addition, the accompanying theoretical analysis shows that the severity head is invariant to the physics weight, allowing the physics regularization to be tuned for depth reconstruction and hydrodynamic consistency without directly changing the severity-ranking head. Code repository: github.com release: github.com: base-attentive.readthedocs.… data access: cds.climate.copernicus.eugl… Flood Database: global-flood-database.cloud… This archive is intended to address reproducibility, code transparency, and validation traceability for the PADR-Net flood-modelling experiments.