Reproducibility package for the benchmark-relative WRF-Hydro–LSTM residual-correction framework in the Volta River Basin

Version 1.2 editor-ready reproducibility release for the benchmark-relative WRF-Hydro–LSTM residual-correction study in the Volta River Basin.

This release provides the reproducibility materials supporting the submitted manuscript, including dataset inventory, configuration files, model setup documentation, LSTM residual-learning architecture, evaluation settings, scripts, processed-data templates, locked result tables, figure-support materials, and crosswalk files linking manuscript tables, figures, datasets, and outputs.

The package supports reproduction and scrutiny of the AMJJASO benchmark-relative streamflow metrics, Q90/Q95 high-flow diagnostics, residual-baseline comparison, flood-area diagnostics, spatial-disagreement assessment, and satellite-supported spatial flood diagnostics.

Important interpretation note: G-RUN ENSEMBLE is used only as a runoff/reanalysis benchmark, not as observed discharge or independent validation. WRF-Hydro is retained as the process-based hydrological backbone. LSTM is used only as an additive residual-correction post-processor. Spatial flood diagnostics are benchmark-relative and satellite-supported, not full hydraulic inundation validation. No scientific results, model claims, figures, tables, or interpretation were changed in this editor-ready update.