High-Resolution (30 m) Flood Conditioning Raster Layers for Flood Susceptibility Modelling in Pakistan

This dataset provides national-scale, high-resolution geospatial raster layers developed to support flood susceptibility modelling and related geospatial analyses in Pakistan. The collection accompanies the peer-reviewed study:

Waleed, M., & Sajjad, M. (2…

The dataset contains 13 raster layers at a nominal spatial resolution of 30 metres. Ten layers represent the final flood conditioning features used to train the Light Gradient Boosting Machine (LightGBM) and eXtreme Gradient Boosting (XGBoost) models reported in the associated publication. Three additional layers are included to support methodological transparency, sensitivity analysis, and further research applications.

Flood conditioning features used in the published models

1. Aspect — terrain aspect in degrees
2. Curvature — surface curvature
3. Distance to drainage — Euclidean distance to drainage lines in metres
4. Distance to rivers — Euclidean distance to major rivers in metres
5. Distance to roads — Euclidean distance to roads in metres
6. Elevation — terrain elevation in metres
7. NDVI — Normalized Difference Vegetation Index
8. Rainfall frequency above 10 mm — number of rainfall events exceeding the 10 mm threshold during 2010–2022
9. Slope — terrain slope in degrees
10. Topographic Wetness Index — terrain-based indicator of potential water accumulation

These ten variables correspond to the final flood conditioning features documented in the associated article.

Supplementary raster layers

The dataset also includes:

• Rainfall frequency above 50 mm, provided as an alternative rainfall-threshold layer for sensitivity analysis;
• Maximum rainfall intensity, which was initially evaluated as a candidate predictor but excluded from the final models because of its high correlation with rainfall frequency;
• OpenStreetMap-derived water presence, provided as an auxiliary categorical water layer.

The supplementary layers were not all included as predictors in the final published machine-learning models and should therefore be distinguished from the ten principal flood conditioning features when reproducing the study.

Technical specifications

• Geographic coverage: Pakistan
• Nominal spatial resolution: 30 metres
• Coordinate Reference System: EPSG:3395 — WGS 84 / World Mercator
• File format: Cloud Optimized GeoTIFF
• Stored data type: Unsigned 16-bit integer (uint16)
• NoData value: 65535

To improve storage efficiency and web-based access, continuous raster values were linearly scaled to the range 0–65534, with 65535 reserved for NoData. The original values can be reconstructed using:

Original value = SCALE_SRC_MIN +
                 (Stored value / 65534) ×
                 (SCALE_SRC_MAX − SCALE_SRC_MIN)

The values of SCALE_SRC_MIN and SCALE_SRC_MAX are stored as metadata tags within each GeoTIFF. Categorical rasters, including the water-presence layer, retain their original categorical values and were not linearly scaled.

A Python notebook, reconstruct_original.ipynb, is included to assist users in converting the scaled rasters back to their original value ranges.

Dataset contents

• aspect.tif
• curvature.tif
• distdrainage.tif
• distriver.tif
• distroads.tif
• elevation.tif
• ndvi.tif
• rfreq_10_201022.tif
• rfreq_50_201022.tif
• rmax.tif
• slope.tif
• twi.tif
• waterosm.tif
• README.md
• reconstruct_original.ipynb

Intended uses

The dataset may support:

• Flood susceptibility and hazard modelling;
• Comparison and benchmarking of machine-learning models;
• Feature importance and sensitivity analysis;
• Hydrological and terrain-based spatial analysis;
• Disaster-risk planning and preparedness;
• Reproducibility and extension of the associated research;
• Development of regional or local flood susceptibility models;
• Education and training in geospatial artificial intelligence.

Users seeking to reproduce the published results should use the ten principal conditioning features and follow the preprocessing, sampling, normalization, and machine-learning procedures described in the associated article and project repository.

Related resources

Associated article:
Waleed, M., & Sajjad, M. (2025). High-resolution flood susceptibility mapping and exposure assessment in Pakistan: An integrated artificial intelligence, machine learning and geospatial framework. International Journal of Disaster Risk Reduction, 121, 105442. DOI: 10.1016/j.ijdrr.2025.105442

Published flood susceptibility maps:
Waleed, M., & Sajjad, M. (2025). High-Resolution (30 m) Flood Susceptibility Mapping Data of Pakistan. Zenodo. DOI: 10.5281/zenodo.18513602

Source code, notebooks, and supporting materials:
GitHub repository: waleedgeo/fsm-pk