Measuring the Accuracy of Forest Change Detection at Multi-temporal Resolutions using the Breaks for Additive Season and Trend Algorithm, a Case Study of Forests in Northern California.
Forest disturbances are closely connected to the global climate, key ecological processes, and can directly impact human livelihoods. Reliable and systematically acquired information about the timing of forest change, is essential to making promptly decisions that help to prevent further forest loss. A wealth of landcover change-detecting techniques exist, including near-real-time monitoring systems of forest disturbance. However, the assessment of their performance is often inconsistent and rarely comprises a precise measurement of the time between the event and its detection. This study implements the Breaks for Additive Season and Trend (BFAST) algorithm to temporally segment optical (Sentinel-2) and SAR (Sentinel-1) satellite-data time-series and performs a multi-temporal resolution accuracy assessment of the results. Relatively high accuracies of change-detection (88% User’s accuracy) are found using optical data and less successful values are returned from the SAR derived product (38% User’s accuracy). While the accuracies decrease with assessments at higher-temporal resolutions for data from both sensors, the product of Sentinel-1 time-series demonstrates a more consistent accuracy across all of the temporal scales. With an increase of SAR data availability in the nearest future, such assessments are necessary to better understand SAR data capabilities and support its potential implementation as a globally-forest-monitoring system. Furthermore, the combination of BFAST and Sentinel-2 data has demonstrated potential for precise mapping of forest change at higher-temporal resolutions, complementing the existing key global datasets (i.e. Global Forest Watch (GFC)) that are released on an annual basis.