Narrowing the gap for city building height predictions.

Area of Science:

• Geospatial Science
• Urban Planning
• Remote Sensing

Background:

• Accurate 3D urban data is vital for sustainable development and hazard resilience.
• High-resolution satellite imagery is often inaccessible in developing countries, leading to data gaps.
• Existing methods for urban height estimation face limitations in accessibility and accuracy.

Purpose of the Study:

• To assess the feasibility of using high-resolution satellite-derived Digital Elevation Models (DEMs) for measuring urban vertical structures.
• To evaluate the accuracy of deep learning models in predicting building heights from satellite imagery.
• To address data scarcity and analytical biases in urban development studies for Global South cities.

Main Methods:

• Utilized 1.5m resolution DEMs derived from satellite imagery to measure building heights in Nairobi, Kathmandu, and Quito.
• Employed a deep learning model trained on high-resolution satellite imagery for height prediction.
• Compared results with published modelled heights and Google's Open Buildings 2.5D Temporal Dataset.

Main Results:

• Building heights were determined with a Mean Absolute Error (MAE) of less than 1m using DEMs.
• The deep learning model achieved an MAE of 2.2-7.0m for building height prediction.
• Google's Open Buildings dataset improved predictions but tended to overestimate heights.

Conclusions:

• Satellite-derived DEMs and deep learning offer a viable solution for high-resolution 3D urban analysis in data-scarce regions.
• Accurate building height data enables better quantification of population, hazard exposure, and material consumption.
• Local evaluation of deep learning models is crucial to mitigate potential biases and ensure reliable data.