Near Real Time Fire Detection Service via the PROBA-V Mission Exploitation Platform

In the framework of the PROBA -V Mission Exploitation Platform (PROBA-V MEP), the “Detection of fires and burned areas” research activity is part of the PROBA-V MEP Third Party Services aimed at better facilitating the exploitation of PROBA-V data across the EO open science community. Progressive Systems carried out this research activity dedicated to support the Centre de Suivi Ecologique (CSE, Senegal) participation in the Monitoring for Environment and Security in Africa (MESA) project through the development of a fire detection and burned areas characterization service over the “Economic Community of Western Africa States” (ECOWAS). The Fire Detection algorithm is based on a modified implementation of a temporal Kalman filter which is capable to detect hotspots in near real time from Meteosat Second Generation (MSG) Spinning Enhanced Visible and Infrared Imager (SEVIRI) geostationary multispectral data. The first component of the algorithm consists of a clear-air anomaly detection making use of multispectral Kalman features. After that, the second component is capable to classify identified anomalies between clouds and hotspots. The service takes in input SEVIRI multispectral data provisioned in near real time by EumetCast every 15 minutes. A direct access to PROBA-V archive of vegetation index and burned area products is then provisioned via the PROBA-V MEP infrastructure. In order to initialize the algorithm over the entire domain, a background model has been retrieved for each pixel and considered channel to depict the daily radiance trend in time of nominal clear air observations. Average values have been calculated, for each channel used by the Kalman filter, by exploiting the EUMETSAT’s Cloud Mask products to filter out anomalies from SEVIRI measurements and consider only clear-sky conditions. Main outputs of the algorithm are fire detections given in tabular and vector formats containing information such as fire ID, geolocation and confidence level together with PROBA-V derived NDVI and NDWI index estimations. Moreover the system is capable to compute a “Fire Occurrence” product over a defined composite period that complements the available PROBA-V Burnt Area product. The main code has been developed in Python while wrapper scripts have been written in BASH. The service prototype has been deployed within a Virtual Machine equipped with 4vCPUs and 8GB RAM within the PROBA-V MEP. Such resources are sufficient to guarantee the near real-time processing over the Western Africa area according to the input product delivery every 15 minutes. First investigations on clear sky classification of MSG scenes over ECOWAS region have shown a strong correlation with respect to EUMETSAT’s Cloud Mask products. Furthermore preliminary Fire Detection comparisons with respect to EUMETSAT’s FRP products has shown a fairly good agreement within hotspots having similar confidence level. Fine tuning of clear-sky and anomaly thresholds is required together with a validation of fire detections with respect to other products (e.g. MODIS FIRMS). Finally further activities, such as field validation campaign and CSE staff training, are planned to validate results and gather feedback from local stakeholders.