By Johannes Eberenz (Germany)
Monitoring forest cover change in near real-time is crucial for timely detection of deforestation. Integration of remote-sensed medium-resolution optical and SAR data can lead to denser time-series in tropical regions with high cloud cover, and thus potentially improve detection speed and accuracy. I developed methods for near real-time deforestation monitoring with integrated multi-temporal Landsat NDVI and ALOS PALSAR L-band HVHH backscatter ratio. Change detection was based on BFAST monitor and data-driven thresholds; fusion was performed at data- and decision level. The methods were validated with 3-monthly reference from an evergreen plantation site on Fiji (2859 ha). Impact of increased cloud cover was studied by testing the methods also with an artificially increased level of missing NDVI data. NDVI – HVHH data fusion at decision-level with threshold-based change detection was found to improve the detection accuracy and speed slightly, when using the original NDVI data (ca. 53% missing data): The overall accuracy with fused data reached 94.4%, compared to 93.8% when using the only SAR data. The detection time-lag could be reduced by 0.2 month at equal accuracy. For 90% missing optical data, the advantages of data-fusion were even smaller. Motivated by short-comings of the common accuracy measures for multi-temporal near real-time change detection, I also propose a statio-temporal change detection accuracy assessment method.
Keywords: Multi-sensor data fusion; ALOS PALSAR; Landsat; Time series; Change detection; Near real-time; Deforestation; Sensor interoperability; BFAST monitor; MulTiFuse; Multi-temporal change validation.