Reproducibility materials for benchmarking a transparent flood-susceptibility overlay against HAND in two Papua New Guinea basins

# Reproducibility materials for flood-susceptibility benchmarking in Papua New Guinea Version: 1.8 Release date: 2026-08-11 Creators: Yang You, Haobo Liu Affiliation: Shanghai Meteorological Service, Shanghai, China ## Summary This record supports the manuscript *Benchmarking a transparent flood-susceptibility overlay against HAND: preregistered evaluations in two Papua New Guinea basins*. It contains the registered configurations, analysis code, processed geospatial layers, validation samples, spatial-bootstrap outputs, figure-source data, final figures and machine-readable tables used in the Laloki and Sepik evaluations. The record is designed for two levels of reuse: 1. direct checking of reported values from the deposited processed data; and 2. reconstruction from the public upstream datasets using the deposited scripts and configurations. Raw copies of third-party Earth-observation and geospatial products are not redistributed. Their product identifiers and access routes are listed in `THIRD_PARTY_DATA.md`. ## Directory structure | Directory | Contents | |---|---| | `config/registration_1` | Laloki preregistration, event inventory, analysis domain, parameter logs and checksums | | `config/registration_2` | Sepik preregistration, domain files, event inventory, frozen coefficients and checksums | | `code/laloki_gee` | Google Earth Engine and local static-layer preparation | | `code/laloki_water_reference` | Event-water processing, patch review and multi-product reference construction | | `code/laloki_statistics` | Validation sampling, ROC-AUC, spatial bootstrap, sensitivity and machine-learning analyses | | `code/laloki_robustness` | Scale-dependence, negative-control and robustness analyses | | `code/sepik` | Sepik factor preparation, HAND construction, scoring and sensitivity analyses | | `code/figures` | Figure and exposure-processing scripts | | `data/laloki` | Laloki processed rasters, event references, validation samples and statistical outputs | | `data/sepik` | Sepik 250 m comparison layers, GFD references, domains and final statistics | | `tables` | Tables 1–11 exported as UTF-8 CSV, including subpanels | | `figures` | Final main and supplementary figures in PNG format | | `environment` | Software and dependency information | `DATA_DICTIONARY.md` describes the principal files and variables. `FIGURE_SOURCE_MAP.csv` maps every figure to its deposited source data and rendering script. `TABLE_SOURCE_MAP.csv` provides the corresponding table mapping. ## Direct verification ### Laloki primary endpoint The principal processed inputs are: - `data/laloki/validation/11_validation_samples.parquet`; - `data/laloki/validation/11_validation_samples_schema.json`; - the final event references in `data/laloki/event_reference/tripartite`; and - `data/laloki/validation/12_primary_endpoint_FINAL.json`. The primary and secondary statistics are implemented in `code/laloki_statistics`. Spatial-bootstrap replicates and complete result tables are deposited beside the primary endpoint. ROC-AUC calculations use the standard half-credit convention for tied scores. ### Sepik comparison The 250 m predictor-complete comparison layers are in `data/sepik/aggregated_250m`; the MODIS-derived flood-reference layers are in `data/sepik/flood_reference`. The registered result is stored in `data/sepik/results/registered_statistics.json`. Completed sensitivities are stored in `WO-L_all_results_v2_1.csv` and its JSON counterpart. ### Tables and figures All numerical tables are available as CSV in `tables`. Final figures are in `figures`, and the source mapping is recorded in `FIGURE_SOURCE_MAP.csv`. The lightweight check `python3 code/verify_reported_values.py` verifies the principal deposited estimates and the expected figure and table counts. ## Full reconstruction sequence 1. Set up the Python environments described in `environment` and install WhiteboxTools 2.4.0. 2. Recreate the Laloki static layers with `code/laloki_gee` using the frozen configuration in `config/registration_1/00_config.json`. 3. Recreate event-water products and the consensus proxy references with `code/laloki_water_reference`. 4. Export the validation sample and run the scripts in `code/laloki_statistics`. 5. Run the deposited robustness scripts when reproducing the exploratory and diagnostic analyses. 6. Recreate the Sepik layers and statistics with `code/sepik` and the second registration files. 7. Render the figures using `code/figures`. The scripts are preserved as the executed project snapshot and grouped by analysis stage for clarity. Some retain `/home/PapuaNewGuinea` and the original project output locations. To rerun those scripts elsewhere, update the root constant and map the original stage directories to the corresponding folders in this record. External services require the user's own Earth Engine or provider authentication; no credentials are included. ## Integrity `FILE_INVENTORY.csv` lists the deposited payload files, byte sizes, media extensions and individual SHA-256 digests. `MANIFEST_SHA256.txt` records the SHA-256 digest of every deposited file except the manifest itself. Use the following command from the record root: ```bash sha256sum -c MANIFEST_SHA256.txt ``` ## Related registrations and code - Laloki preregistration: https://doi.org/10.17605/OSF.IO/M7Z5B - Sepik preregistration: https://doi.org/10.17605/OSF.IO/5DMWT - Public code history: https://github.com/whelveleon/PNG-flood ## Citation You Y, Liu H. 2026. Reproducibility materials for benchmarking a transparent flood-susceptibility overlay against HAND in two Papua New Guinea basins, version 1.8. Zenodo. The DOI will be assigned when this record is published. ## Licences Author-created data, figures and documentation are released under Creative Commons Attribution 4.0 International. Author-created source code is released under the MIT License. These licences do not replace the terms of the upstream third-party datasets described in `THIRD_PARTY_DATA.md`.