Following the severe 2018 fire season, a fully automated satellite based system for early wildfire detection was implemented in Sweden. Currently, it uses the VIIRS instrument onboard the three Suomi-NPP/JPSS spacecraft. The Swedish wildfire suppression strategy relies heavily on fast response, requiring accurate geo-location, low latency, and a small number of false alarms. Here, we describe the system, measures taken to reduce false and multiple alarms, and assess the outcome from the first years of operation. We also describe validations of (1) detections against large experimental wood crib fires with well-defined heat release rates and temperature distributions and (2) the 41 brightest wildfires detected in 2023 and their relation to forest characteristics. Average end-to-end latency from overpass to alert is 17 min. The satellites satisfactorily reproduce location and pixel temperatures, although an increasing forest timber density reduces the pixel temperature. Verified false alarms accounted for approximately 10% of the detections since 2022, mainly attributed to reflections on large arrays of solar panels, large greenhouses, or large metal farmhouse rooftops. Although satellites still fail to detect many fires due to unfavorable timing, cloud cover or sensitivity, 29% of the wildfire detections submitted an alarm before any other means of detection. High fire localization accuracy was also confirmed, further reducing suppression time, even when the satellite system is not the first to trigger an alarm. Regional coverage is further expected to increase substantially with the addition of the first satellite in the European Metop-SG programme, Metop-D, launched in August 2025. • We describe an automated system for wildfire detection from satellite observation to alarm. • The detections from the first three years of operation of the system are assessed and validated. • The VIIRS active fire algorithm is validated against well-defined wood crib fires. • There is potential both in expanding the system and performing local optimization of the active fire algorithm.
Dybbroe et al. (Sun,) studied this question.