Near-real-time monitoring of land disturbance with harmonized Landsats 7–8 and Sentinel-2 data
Rong Shang, Zhe Zhu, Junxue Zhang, Shi Qiu, Zhiqiang Yang, Tian Li, Xiucheng Yang
Abstract
Land disturbance can increase carbon emissions, cause detrimental environmental impacts, and threaten human life and property. Monitoring land disturbance in near-real-time is essential to mitigate their negative effects and prevent future losses. However, rapid and timely monitoring of land disturbance at a high spatial resolution is in its infancy. Here, we developed an algorithm for Near-Real-Time MOnitoring of laNd dIsturbance based on Time-series of harmOnized Reflectance (NRT-MONITOR) from Landsats 7–8 and Sentinel-2 data at 30-m spatial resolution. It incorporates an online recursive algorithm called Forgetting Factor to improve efficiency in the determination of land disturbance to get fast detection based on the harmonized data. This algorithm is developed and validated by using 1200 samples created from the harmonized Landsats 7–8 and Sentinel-2 time series from 2015 to 2019 within the conterminous United States (CONUS). An overall accuracy of 70% has been achieved for monitoring a variety of land disturbance types. NRT-MONITOR improves the processing efficiency (11.5 times faster) compared to the COLD algorithm (Zhu et al., 2020). The mean time lag of NRT-MONITOR, defined as the delta days of confirming a land disturbance after its occurrence, is only 35 days, which is achieved by using the harmonized Landsats 7–8 and Sentinel-2 observations and reduced number of clear observations (from six to four) needed to confirm a land disturbance. Finally, NRT-MONITOR can be integrated into an alerting system to provide potential land disturbance probability maps that are updated every three days.