Prototype software for a forest fire workstation has been developed for automatic detection and monitoring of forest fires. Data from the AVHRR (Advanced Very High Resolution Radiometer) sensor of the NOAA satellites are utilized as the primary input.
The NOAA images are transferred from a receiving station to the workstation via a data transfer network. The images are checked for missing or erroneous lines due to reception errors. Image data are geo-coded using orbital data of the satellite. The geo-coded image data are searched for fire pixels. Detection of fire pixels is based on thresholding AVHRR channel 3 (middle infrared, central wavelength 3.5 µm). Contiguous areas of fire pixels are grouped into fire patches. For each fire patch, an alert message is generated and transmitted to a recipient via electronic mail. In the box below an example of such a message is given:
From: VTINSX::palokuva “Metsapalokuva” 26-OCT-1993 11:02:15.69 To: rauste CC: Subj: FIRE_Alert
A possible forest fire has been detected in data set: n119308121446 (Acquired on 1993-08-12 at 14:46) Channel-3 minimum: 98, 8 pixels Co-ordinates: Northing: 4377.8, Easting: 119.0 (line: 865.9, column: 242.1)
Best regards: Forest-fire workstation (at 93-10-26 11:03)
The automatic forest-fire monitoring system was tested during the summer season 1993. In Finland, summer 1993 was more rainy than e.g. summer 1992. The test area covered southern Finland, Estonia, Latvia, Lithuania, and parts of Russia. NOAA image data from 1 June to 20 August (with some breaks of a few days) were obtained from the Finnish Meteorological Institute. The system reported 66 NOAA images as containing a possible fire. These 66 images were inspected individually. In four cases, the reported fire was considered to be real fire. Two of the fires could not be verified with fire authorities because the fires were located outside the Finnish territory. One of the fires, which was verified with fire authorities, covered an area of 30 ha. The system has also been tested over the territory of Greece.
Most of the cases where fire was reported by the automatic monitoring system were (specular) reflections from clouds or water. In future development, when the imaging geometry is taken into account in search of fires, the number of false alarms can be reduced significantly.
In forest fire monitoring for fire-fighting activities, it is essential that data in the 3.5 µm band (e.g. NOAA AVHRR channel 3) are available. This spectral band enables the day-time detection of relatively small forest fires. Most forest fires tend to start in the day time. If forest fires are only monitored using visible-wavelength data acquired during the hours of darkness, the fires have a long time to spread before they get detected. This makes the extinguishing of the fires more difficult. In high-latitude boreal forests, the acquisition of night-time images is difficult around mid-summer due to short nights.
The development project of the forest fire workstation was ordered by the Rescue Department of the Finnish Ministry of the Interior.
Häme, T. and Y. Rauste 1993. Multitemporal satellite data in forest mapping and fire monitoring, a paper presented at the International Workshop on “Satellite technology and GIS for Mediterranean forest mapping and fire management”, Thessaloniki, Greece, 4-6 November 1993, 13 p. (see report on page 29).
Rauste, Y. 1993. Remote sensing workstation for forest fire fighting – fire detection/monitoring system. Internal report, 17p.
From: Yrjö Rauste Address: VTT Technical Research Centre of Finland Instrument Laboratory P.O. Box 107 SF – 02151 Espoo