The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Download
Publications Copernicus
Download
Citation
Articles | Volume XL-7/W3
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XL-7/W3, 765–770, 2015
https://doi.org/10.5194/isprsarchives-XL-7-W3-765-2015
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XL-7/W3, 765–770, 2015
https://doi.org/10.5194/isprsarchives-XL-7-W3-765-2015

  29 Apr 2015

29 Apr 2015

Real-time stream processing for active fire monitoring on Landsat 8 direct reception data

C. Böhme, P. Bouwer, and M. J. Prinsloo C. Böhme et al.
  • Pinkmatter Solutions, Pretoria, South Africa

Keywords: Real-time Processing, Near-real-time processing, NRT, Fire Detection, Fire Notification, Landsat 8

Abstract. Some remote sensing applications are relatively time insensitive, for others, near-real-time processing (results 30-180 minutes after data reception) offer a viable solution. There are, however, a few applications, such as active wildfire monitoring or ship and airplane detection, where real-time processing and image interpretation offers a distinct advantage. The objective of real-time processing is to provide notifications before the complete satellite pass has been received. This paper presents an automated system for real-time, stream–based processing of data acquired from direct broadcast push-broom sensors for applications that require a high degree of timeliness. Based on this system, a processing chain for active fire monitoring using Landsat 8 live data streams was implemented and evaluated. The real-time processing system, called the FarEarth Observer, is connected to a ground station’s demodulator and uses its live data stream as input. Processing is done on variable size image segments assembled from detector lines of the push broom sensor as they are streamed from the satellite, enabling detection of active fires and sending of notifications within seconds of the satellite passing over the affected area, long before the actual acquisition completes. This approach requires performance optimized techniques for radiometric and geometric correction of the sensor data. Throughput of the processing system is kept well above the 400Mbit/s downlink speed of Landsat 8. A latency of below 10 seconds from sensor line acquisition to anomaly detection and notification is achieved. Analyses of geometric and radiometric accuracy and comparisons in latency to traditional near-real-time systems are also presented.