Building height and roof material dataset - Dominica, Pichelin and Roseau, 2018, 2024

Science Case Name

 Multi-hazards in the Caribbean SIDS

Dataset Title

Building height and roof material dataset - Dominica, Pichelin and Roseau, 2018, 2024

Dataset Description

Building height predictions of a monocular depth estimation network trained on UAV imageries of Pichelin (2024) and Roseau (2018) trained on Google’s Open Building 2.5D Temporal dataset ( https://sites.research.google/gr/open-buildings/temporal/) and roof material classification using the model provided in GFDRR’s Caribbean Rooftop Classification (https://github.com/GFDRR/caribbean-rooftop-classification):

·        pichelin_network_height_rooftype_predictions.gpkg – Geopackage containing building height estimation and roof material classification of Pichelin, Dominica (2024).

·        roseau_network_predicitons.gpkg – Geopackage containing building height estimation and roof material classification of Roseau, Dominica (2018).

Key Methodologies

A monocular depth estimation network called MiDaS (https://pytorch.org/hub/intelisl_midas_v2/) and the roof type and material classification network of GFDRR based on ResNet-50.

Temporal Domain

2018 and 2024

Spatial Domain

Roseau and Pichelin, Dominica

Key Indicators

Building height and roof material

Data Format

A vector as a Geopackage (.gpkg)

Source Data

Openaerialmap UAV image of Roseau (2018), UAV images of Pichelin (2024) captured using a DJI drone, and Google’s Open Building 2.5D Temporal dataset

Accessibility

10.5281/zenodo.17100331

Disclaimer – This dataset is provided solely for research purposes. It is based on available data and has not been verified or approved by any local authority. The dataset represents the outcome of the applied methodology and research judgments, and does not reflect the official policies or positions of the authorities of Dominica.

Stakeholder Relevance

The building height and roof material dataset can be used for hazard mapping and risk management. For example, they can be integrated into the digital twin environment of buildings-at-risk.

Limitations/Assumptions

Temporal alignment of images used and building height dataset the network was trained on.

Additional Information

Contact Information

Bergado, John Ray (UT-ITC)