NYC Building Plume Detection from Skyline Imagery, 2013-2015

Ben Steers at New York University's Center for Urban Science and Progress created a dataset of over 1,100 annotated plumes from Manhattan building emissions. The dataset was generated by applying a trained deep convolutional neural network to archival skyline imagery captured at 0.1 Hz by the Urban Observatory. It contains detections of plumes, classified by color, from two periods: October 26 to December 31, 2013, and January 1 to March 13, 2015.

Use Cases

• Training object detection models for building plume emissions based on annotated plume imagery.
• Analyzing urban pollution sources and activity patterns based on plume ejection rates.
• Classifying plume types (e.g., carbon vs. water vapor) based on color features mentioned in the description.
• Monitoring environmental impacts of urban metabolism over time using the provided temporal coverage.

Strengths

• Contains over 1,100 actual annotated plumes for training.
• Includes sources of contamination (clouds, shadows, lights) to improve model robustness.
• Covers two distinct time periods spanning parts of 2013 and 2015.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Data may reflect geographic bias inherent to the Manhattan-only imagery.

Provenance

Source

New York University's Center for Urban Science and Progress (CUSP-UO)

Collection Method

Deep convolutional neural network applied to continuous skyline imagery from the Urban Observatory.

Time Range

October 26, 2013 to December 31, 2013; January 1, 2015 to March 13, 2015

Geography

Manhattan, New York City, USA