基于多源数据融合的地表覆盖数据重建研究进展综述

地表覆盖数据对于全球环境变化、生物多样性和发展政策制定有着重要意义。遥感已成为获取地表覆盖数据的重要手段。而目前的地表覆盖数据产品,如Globe Land30、FROM-GLC、MODIS Collection5、MODIS Cropland、Glob Cover、GLC2000等,存在数据精度不高、数据间一致性较差、与统计数据差异较大等问题。因此,基于多源数据融合的数据重建方法成为目前研究中的热点问题。文章检索了近10年关于多源数据融合在地表覆盖数据重建中的应用的相关文献,概括了多源数据在数据重建中的应用现状,并对基于多源数据融合的地表覆盖数据重建方法进行了归纳总结,重点评述了不同方法的特点及应用情况,阐明了各种方法的优势与不足,同时对存在的问题进行探讨并展望了未来基于多源数据融合的地表覆盖数据重建研究的发展方向。基于多源数据融合的数据重建方法包括基于多源遥感数据融合法以及基于多源遥感和非遥感数据融合法。该文在对基于多源遥感数据融合的数据重建方法进行论述时,主要讨论了其中应用最广泛的两种融合方法:基于数据一致性的融合法和基于回归分析的融合法。对于其他基于多源遥感数据融合的数据重建方法,如基于D-S证据理论融合法、基于数据集成融合法、基于统计模型融合法,也列举了最具代表性的相关文献进行论述。在对基于遥感数据和非遥感数据融合的数据重建方法进行论述时,主要讨论了其3种空间分配方法:完全依赖法、部分依赖法、动态依赖法。在对目前研究进行探讨的过程中,该文对其研究区域、数据源、地表覆盖类型、空间分辨率、融合方法和文献来源进行总结分析,并重点就融合方法展开讨论。围绕各种融合方法在数据重建中的运用,该文归纳出目前研究中存在的主要问题:研究对象和区域上的不足,研究区多为全球及欧美,其他区域的研究过少,研究对象多为所有地表覆盖类型和森林,对耕地和草地的研究过少;融合算法上的不足、重建结果精度上的不足。最后,指出基于多源数据融合的数据重建方法未来的发展方向,即综合运用两类方法,得到具有详细完整空间信息的长时间序列的地表覆盖数据集。

PROGRESSES IN LAND COVER DATA RECONSTRUCTION METHOD BASED ON MULTI-SOURCE DATA FUSION

Land cover data is of great significance for global environmental change, biodiversity and policy-making. Remote sensing has been demonstrated an important method to obtain land cover data. Nowadays, there are varieties of remote sensing products, such as GlobeLand30, FROM-GLC, MODIS Collection5, MODIS Cropland, GlobCover, GLC2000. However, these previous remote sensing products exist some limitations, such as poor products accuracies, low consistency between different products and big discrepancy compared with statistic data. In this paper, the related literatures on the land cover data reconstruction methods based on the multi-source data fusion in the latest decade were sorted out, application of the multi-source data reconstruction methods were summarized, characteristics and development of different fusion methods in the latest researches were discussed intensively, and the advantages and disadvantages of various methods were concluded. This paper also discussed the limitations of current data reconstruction researches and proposed some important directions for future studies of the multi-source data fusion. In terms of the land cover reconstruction method based on multi-source remote sensing data fusion, this paper mainly discussed two widely-used methods: data consistency and regression analysis. In addition to the two common methods, other methods on multi-source remote sensing data fusion were also summarized, such as Dempster-Shafer evidential reasoning, data integration, statistical models, and the synergistic combination theory. In terms of the reconstruction method of fusing multi-source remote sensing and non-remote sensing data, three allocation methods were respectively discussed: the complete dependence method, partial dependence method and dynamic dependence method. The study region, used data source, targeted land cover types, spatial resolution, fusion methods and literature resources were respectively summarized and analyzed, among which the fusion methods were intensively discussed. At the end of this paper, the limitations and problems of the latest research were concluded: the study objectives were limited to the forest and all land cover types while the studies on the grassland and cultivated land were scarce. Most studies focused on European and American regions. Besides, the fusion methods were limited, and the accuracies of the reconstruction methods were not satisfactory. The future study directions and priorities of the reconstruction method were also proposed, such as integrating two methods to product a hybrid land cover map with a long time series and detailed spatial information.