Key finding
The FNRT (Fusion for Near-Real-Time) algorithm fuses three open-access satellite systems without requiring harmonised data cubes or training data. Tested across eight 100 x 100 km sites in the southwestern Amazon deforestation frontier, the algorithm detected 70% of disturbances within 30 days at 91% user accuracy and 85% within 60 days, with peak producer accuracy reaching 92%. The finding that optical sensors react faster than Sentinel-1 in the dry-season Amazon, but that SAR would be more valuable in persistently cloudy regions like Madagascar or coastal West Africa, provides practical guidance for system configuration by region.
What the research asks
Tropical forest monitoring systems face a fundamental trade-off: optical satellites like Landsat and Sentinel-2 provide detailed spectral information but are blinded by cloud cover, while radar satellites like Sentinel-1 see through clouds but provide less intuitive signals. Most operational systems use only one sensor type, limiting either temporal frequency or cloud-penetration capability. This study asked whether fusing data from all three open-access satellite systems could produce faster, more reliable near-real-time disturbance alerts than any single sensor alone, and how sensor contributions vary across seasons and regions.
What it finds
The FNRT (Fusion for Near-Real-Time) algorithm addresses one of the most persistent operational challenges in tropical forest monitoring: persistent cloud cover that blinds optical satellites for weeks at a time. By fusing data from three open-access satellite systems (optical Landsat and Sentinel-2 plus radar Sentinel-1) without requiring harmonised data cubes or training data, the algorithm is both flexible and scalable. Tested across eight 100 x 100 km sites in the southwestern Amazon, it detected 70% of tropical forest disturbances within 30 days at 91% user accuracy, and 85% within 60 days, with peak producer accuracy of 92%.
A key operational insight is that sensor performance varies by region and season: optical sensors react faster than Sentinel-1 in the dry-season Amazon where cloud cover is lower, but SAR would be the primary detection source in persistently cloudy regions like Madagascar or coastal West Africa. The full algorithm is publicly available on Google Earth Engine, making it immediately usable by national monitoring agencies and practitioners. With 65 citations, this is the most-cited near-real-time fusion paper in the tropical forest monitoring literature.
Why it matters
For Earth observation practitioners building operational forest monitoring systems, this paper is directly applicable. The algorithm runs on Google Earth Engine using only open-access data, so there are no licensing barriers to deployment. The paper provides clear performance benchmarks broken down by detection speed (30, 60, 90 days), sensor contribution (optical vs radar), and accuracy metric (user vs producer), giving system designers the evidence they need to choose sensor combinations for specific regions and monitoring objectives.
For carbon market MRV and REDD+ programs, faster detection means faster response. Reducing the detection delay from months to weeks enables enforcement agencies, project developers, and supply chain auditors to act before small clearings become large ones. The finding that Sentinel-1 radar provides independent detection capability in the cloudiest regions is especially valuable for the Congo Basin and insular Southeast Asia, where optical monitoring alone has consistently underperformed. The site currently has no paper covering near-real-time fusion methodology, and this fills a clear gap for practitioners building operational monitoring pipelines.