Mapping Forest Disturbance And Recovery Histories Of Plantations In Southern China By Combining Remote Sensing Algorithms And Dense Landsat Time Series Observations

Forest ecosystem is the major component of the terrestrial biosphere, acting as one of thelargest carbon pools, and plays an important role in global carbon cycle. Forest disturbance andrecovery have been regarded as critical mechanisms for regulating carbon fluxes between theland surface and the atmosphere. Therefore, the balance of these processes is one control on netecosystem productivity. However，quantitative and spatio-temporally explicit information onforest disturbance and recovery is currently rare for most regions of the world. Remote sensingtechnology, as an important means of acquiring spatial and temporal change information has beenwidely used in monitoring forest changes, including distributions and structural parameters. Inthis paper, based on a Landsat time series stack consisting of annual scenes acquired from1986to2011, the LEDAPS algorithm was first used to convert the original scenes in the stack to thestandard surface reflectance images. Then the surface reflectance images were fed to LandTrendrand VCT models respectively to creat a suite of maps showing the locations and magnitudes offorest disturbances and recovery events in Fogang county, Guangdong province, followed byvalidating, forest disturbance occurrence year, magnitude and duration. Finally, the disturbanceidentificaiton precision of the two algorithms was evaluated. Relying on the history and status offorest disturbances in Fogang county, the driving forces contributing to plantations’ forestdisturbance were digged to develop corresponding forest management strategies andcountermeasures.Results suggested that the LEDAPS algorithms applied to standard and non-standardLandsat images were able to effectively retrieve the surface reflectances, and the extractedreflectance curves were similar to the actual spectral libaries.The results produced from LandTrendr algorithm showed forest disturbance in Fogangcounty was intense, the area of about1000ha per year. Anaysis of the area change trend of forestdisturbance and recovery revealed that in the late1980s to1990s, the disturbance and recoveryforest areas were less than those after2000, and the trend was gentler than that after2000. Since2000, the forest disturbance areas were gradually increasing and higher than those of forestrecovery, but the forest recovery areas were still going up. The results also showed that forestdisturbances in Fogang were mainly identified as abrupt events. Gradual forest disturbance andrecovery events existed but the overall areas were not large, highly different from the areas ofabrupt disturbance events. And before2000abrupt and gradual disturbance areas were almostequal, with a gentle change. Since2000, abrupt disturbance areas were greater than those ofgradual disturbances, but the trend fluctuated. Based on the visual interpretation of the originalsurface reflectance images, the disturbances mapped by the algorithm were in agreement with those visual interpretation results well.Forest change results mapped by VCT showed during the period1980s to1990s, forest areaincreased, while non-forest showed a trend of decline. Forest area of53488.62ha in2000dropped to42807.06ha in2011, especially from2005to2007, there had an abrupt decline inforest area, indicating that forest suffered a severe disturbance. After2009, forest started togradually recover. The highest level of forest disturbance rate occurred in2008because there wasa large-scael severe disaster of low temperature, snow and freeze sweeping sourthern China anddestroying a large number of forests, including timber and economic forests, bamboo forests andnewly established plantations with different degrees of damage. In the hybrid validation approach,using pre-and post-disturbance Landsat images could easily identify disturbances, and thedisturbance year map determined the locations and the years of the first disturbance eventoccurring. At the per-pixel level, the overall accuracy was about71.51%and the Kappacoefficient was0.6738, indicating a high precision of disturbance information extraction.In summary, it was quite necessary to use dense time series remote sensing images toautomatically map forest disturbance and recovery events in wide plantations dominated regions.These automatic algorithms used in this paper can give qualitative, locational and quantitativeforest change results to the land use decision-makers and conservation communities, enabling thestrategic development of sustainable forest management, providing effective data support toevaluate forest productivity and forest ecosystem health.