A Multi-Modal, Multi-Temporal, Multi-Resolution Benchmark Dataset for Building Height Estimation.

Area of Science:

• Remote Sensing
• Geospatial Analysis
• Urban Planning

Background:

• Accurate building height data is vital for urban planning and monitoring.
• Traditional methods using aerial imagery and LiDAR are expensive and slow for large-scale, frequent assessments.
• Satellite data combined with Deep Learning (DL) offers a scalable solution, but lacks comprehensive open-source datasets.

Purpose of the Study:

• Introduce M4Heights, a novel multi-modal, multi-resolution, and multi-temporal dataset for building height estimation.
• Facilitate the advancement of DL models for accurate and scalable building height estimation.
• Support diverse urban planning and monitoring applications.

Main Methods:

• Compiled a dataset of approximately 1 million images from Sentinel-1 SAR and Sentinel-2 MSI satellite data.
• Integrated high-resolution aerial orthophotos and ground truth building height maps.
• Included the largest associated multi-image super-resolution dataset for enhanced height estimation.

Main Results:

• M4Heights covers diverse architectural styles, urban densities, and terrain across Estonia, Netherlands, and Switzerland.
• The dataset supports various modeling approaches and enables extensibility to new regions.
• It provides a foundation for improving DL-based building height estimation.

Conclusions:

• M4Heights addresses the critical need for large-scale, open-source training data in building height estimation.
• The dataset will accelerate the development and deployment of advanced DL models for urban monitoring.
• It offers significant potential for sustainable urban planning and management.