| Myanmar is one of the countries with the highest forest coverage in Southeast Asia.However,with Myanmar’s rapid political and economic development in recent years,more and more virgin forests have been cut down to meet the expansion of rubber forest land,and large areas of natural forests have disappeared.The local environment is under great pressure,and problems such as soil degradation and reduction of biodiversity have gradually emerged.A timely grasp of the distribution range and dynamic changes of rubber forests will provide basic data support for the reasonable planning of rubber forest planting and environmental protection;The development of remote sensing technology and the application of multi-source remote sensing images provide convenient,fast and reliable technical measure for rubber forest dynamic monitoring.Taking Mon State of Myanmar as the study area,this paper uses the Sentinel-1/2 image with 10 m resolution to extract the planting range of rubber forest with high precision,then uses the Landsat image with30 m resolution from 2000 to 2019 to construct time series to detect and classify the continuous change of land use,and finally detects the spatio-temporal expansion change of rubber forest land in the past 20 years.The main research results are as follows:(1)The land cover classification method based on the time series statistical characteristics of Sentinel-1 images can effectively reduce the impact of terrain changes,suppress noise,make land classification patches more complete,and overcome the limited number of high-quality optical image acquisitions in cloudy and rainy areas.It can be used for high-precision extraction of forest land in tropical areas.Experiments show that the overall accuracy of forest/non-forest land classification based on time-series statistical characteristics increases by 26.66 percentage points compared with single-phase classification,the Kappa coefficient increases by 0.55,and the accuracy of forest producers and users increases by 29.76 and 25.89 percentage points,respectively.(2)Combining Sentinel-1 and Sentinel-2 image data,mining rich spectral information,spatial texture information features,can realize the complementation of information between different types of sensors,and improve the recognition accuracy of rubber forests.The producer accuracy and user accuracy of rubber forest information extraction based on the optimized combination of spectrum,texture and Sentinel-1 image backscattering features are 86.77% and 89.13%,respectively,and they are increased by 3.7 and 6.93 percentage points respectively compared with the combination of spectral and texture features;And they are increased by 5.29% and 8.92%,respectively compared with only based on spectral characteristics.(3)The continuous change detection of rubber forest based on the CCDC algorithm greatly simplifies the data preprocessing process.It can make full use of the available images of the Landsat series to construct a dense time series,realize the high-precision extraction of rubber forest land,and automatically identify land use changes.We can get the land cover map of the specified time,greatly enhance the consistency of the rubber forest extraction time and space,and finally analyze the time and space distribution changes of the rubber forest in the study area. |
