RST-FIRES, an exportable algorithm for early-fire detection and monitoring: description, implementation, and field validation in the case of the MSG-SEVIRI sensor

Wildfires are a worldwide phenomenon with local and global effects. Many satellite-based methods for fire detection and monitoring have been developed to exploit data acquired by sensors onboard polar orbiting platforms. Their relatively low temporal resolution (hours) is, however, decidedly not adequate for detecting short-living events or fires characterised by a marked diurnal cycle and rapid evolution times. Geostationary satellites have very high temporal resolution of 30 to 2.5 min and could, in principle, be more suitable for providing timely alerts and facilitating possible mitigation actions. However, such short revisit times are coupled with spatial resolutions of 3–5 km, which are coarser than those of the polar orbiting sensors generally having 1 km capability in the infrared region. This could represent a significant limitation for small fire detection and precise localization. However, unlike polar orbiting systems, the geostationary attitude assures very stable observational conditions at the pixel level, including fixed view angles, same time of day, and same ground resolution cell dimensions. This can be crucial for fire-detection methods based on multi-temporal analyses. This work describes in detail the Robust Satellite Techniques for FIRES detection and monitoring (RST-FIRES), a multi-temporal change-detection technique, and its application to the data of Spinning Enhanced Visible and InfraRed Imager (SEVIRI) on board the Meteosat Second Generation (MSG) platform. Its performance in terms of reliability and sensitivity was verified by > 20,000 SEVIRI images collected throughout the day during a four-year-collaboration with the regional Civil Protection Departments and other Local Authorities of two Italian regions which provided about 950 near real-time ground and aerial checks of the RST-FIRES detections. The results indicate a mean commission error rate of 18%, ranging from 3% to 30% according to the specific experimental campaign, with an average omission error of 44%, ranging from 3% to 66% depending on the campaign and the validation source used. This study fully demonstrates the added value of the RST-FIRES technique for the early warning of fire events. During the considered validation campaigns, RST-FIRES provided the sole fire alert in 348 cases; in further 227 cases, its warnings were given > 1 h before any other source of information was provided. Finally, this paper presents and discusses a comparison among RST-FIRES and other SEVIRI-based fire detection products. On the whole, RST-FIRES is shown to be 3 to 70 times more sensitive than all of the other considered SEVIRI-based products. This satisfactory result was not completely unexpected, considering the quite high false positive rate exhibited by RST-FIRES and the fact that other algorithms were designed to work in a wider area, from continental to the SEVIRI full disk.