Hydrogeomorphic flood priors: data, code, and validation outputs across six satellite-mapped events
Description
This dataset supports the study of the geographic transferability of static hydrogeomorphic flood priors across independent satellite-mapped flood events. The analysis combines hydrogeomorphic predictors derived from JRC Global Surface Water and Copernicus DEM with independent flood delineations from six Copernicus Emergency Management Service (CEMS) Rapid Mapping activations in Spain, Italy, and Mozambique. The audited base workflow was developed in northern Colombia and evaluated using a frozen-validation design: model specification, preprocessing, scaling parameters, coefficients, event definitions, and quality-control rules were fixed before external event evaluation. Version 2 preserves that primary experiment and adds a diagnostic extension designed to explain the observed geographic-transfer behaviour without reselecting or recalibrating the frozen primary model. The Version 2 extension adds operational effort-recovery curves, area-to-50%-recall and area-to-80%-recall measures, event-level predictor-ablation deltas, 500-replicate paired spatial-block bootstrap intervals, decomposition of the frozen linear predictor, and development-to-transfer performance-gap summaries. These outputs quantify both the operational value that remains after geographic transport and the changing contribution of hydrographic and terrain predictors across external events. The repository contains the audited MATLAB workflow, reproducibility instructions, model-selection and external-validation outputs, spatial-bootstrap results, environmental-support and observation-footprint diagnostics, product provenance, checksums, figure- and table-ready outputs, and the Version 2 geographic-transfer diagnostic extension. Raw third-party satellite and emergency-mapping products are not redistributed and should be obtained directly from their original providers. Version 2 adds a reproducible geographic-transfer diagnostic extension while preserving the original frozen primary model and six-event evaluation. New materials include operational effort-recovery curves, area-to-50%-recall and area-to-80%-recall, event-level ablation deltas, 500-replicate paired spatial-block bootstrap intervals, frozen-score contribution decomposition, and development-to-transfer gap summaries. No external event label was used to reselect, recalibrate, or replace the primary model. No study threshold, event selection, target-support rule, or original QA decision was changed.
Files
Steps to reproduce
1. Download and extract the audited base MATLAB package retained from Version 1 into a folder with read/write permission. 2. Use MATLAB R2026a Update 4 and the toolboxes documented in the Version 1 repository. Read the base reproducibility documentation before execution. 3. Open the extracted project as the MATLAB Current Folder. All configuration paths must remain relative to the project; do not add personal absolute paths to the configuration files. 4. Reproduce the base frozen workflow by running: RSE_EJECUTAR_TODO_AUTOMATICO 5. Complete the frozen QA-review step exactly as documented in Version 1, rerun the base workflow, and confirm: CODE PREFLIGHT: PASS SYNTHETIC TEST PASSED OUTPUT VALIDATION: PASS AUTOMATIC PIPELINE: PASS 6. Confirm that the completed runtime contains: results/models/Analysis_Index.mat results/models/Frozen_Development_Model.mat and that the pipeline identifier is: RSE_MULTIREGION_2.2.7_CEMS_RASTER_QUERY_AUDITED 7. Copy EJECUTAR_A3_EJRS_NOVELTY_EXTENSION_FINAL.m from the Version 2 extension package into the root of the complete audited runtime. 8. Run: EJECUTAR_A3_EJRS_NOVELTY_EXTENSION_FINAL 9. The extension first replays the six frozen external-event metrics and the previously audited development-only ablations. Execution stops if these invariants are not reproduced. 10. After the replay checks pass, the extension calculates operational effort-recovery curves, area-to-50%-recall and area-to-80%-recall, event-level ablation deltas, 500 paired spatial-block bootstrap replicates per event and diagnostic comparison, frozen linear-predictor contribution decomposition, and development-to-transfer summaries. 11. Compare the regenerated outputs with EJRS_NOVELTY_V2_SUMMARY.txt and the CSV and figure files included in the Version 2 extension package. 12. Raw JRC Global Surface Water, Copernicus DEM and CEMS source products are not redistributed. Obtain them from the original providers through the related links supplied with the dataset.
Institutions
- Universidad Nacional de ColombiaBogota D.C., Bogotá