Key Technology Of Time Series Regional Land Cover Mapping

Land cover products are fundamental materials for land resource management,ecological environment monitoring and global change studies.Remote sensing technology has been the effective method of large-scale land cover mapping.Currently,the global land cover maps produced by lower spatial resolution remote sensing images have been played an important role in various applications.Owing to the the increasement of the acquisition approaches and the reduction of the cost of the satellite images with higher spatial resolution,it is realizable to produce the land cover maps with high resolution.Compared with the global land cover maps,a regionally produced time-series land cover maps would better reflect the regional dynamic change and make contributions to the regional planning and management.Most of the land cover products have low consistency.The quality of the remote sensing images,classification algorithms and the discrepancies of the class definition all lead to the issue.The traditional classification strategies rarely take into account the temporal correlation and spatial continuity.Great discrepancies can be discovered between different land cover products,and a number of illogical transitions exist between different time points from the same land cover products.In addition,the accuracy and reliability of time series land cover products can hardly be evaluated objectively and exactly,which influences the usage of the time series land cover products to some extent.Aiming at the special issues and common issues of time series land cover mapping,the paper focused on the time series land cover mapping and the assessment strategy.The main work includes:1.Introduce the process of several statistical learning methods in time series land cover mapping.Compare the properties of Hiden Markov Model,Conditional Random Fields and Maximum a Posteriori-Markov Random Fields in the time series labelling.Analyze the boundedness of undirected graph model in the post-classification of time series land cover maps and the advantages of di-graph model in time series land cover labelling.2.Focusing on the problems of the spatial-temporal continuity of time series land cover products,the paper took into account the spatial-temporal factors in time series land cover mapping.The study improved the hierarchical seasonal land cover products by employing the logical rules of the land cover change.In the interannual land cover mapping,the paper exploited the hidden Markov model that took into account both the temporal relationships and spatial continuities.In the proposed algorithm,the transition matrix of the land cover types were decided by the local environment,logical rules and the previous land cover change studies.The ovservation probabilities were weighted by the spatial consistency,instead of the way of introducing the weighting parameters into the model or algorithm.As a result,the proposed model simultaneously took into account the temporal relationships and spatial continuities without complicating the model.3.Based on the propobility calculation problem of hidden Markov model,the paper proposed a novel strategy in the assessment of time series land cover maps.The assessment strategy was independent of samples,instead,the temporal relationships,spatial continuities and the classification performance in each time point were simultaneously token into account.The quality of the pixels in the products were evaluated by the joint probabilities.To test the proposed strategies,the study produced the seasonal and interannual land cover products of Poyang Lake area based on the proposed mapping strategy.Further,the paper assessed the products by the proposed assessment strategy.The experiments demonstrated the effectiveness of the proposed strategies.