GEE平台和CART方法的北京市土地解译

针对传统遥感影像解译效率较低、人力物力需求量大等问题,该文以谷歌地球引擎为依托平台,利用Landsat5TM影像,采用分类回归树算法对2010年北京市土地覆被/土地利用类型开展了解译研究,并从类型构成、类型混淆和空间一致性3个方面将解译所得LUC-2010产品与Globeland30-2010产品进行空间一致性分析。研究表明,谷歌地球引擎(GEE)平台通过编程运算,数据处理速度极快,大幅提高工作效率。解译产品与训练样本交叉验证的学习精度为94.2%。两套产品总体对比发现,林地、水体和耕地的空间一致性比率分别为84.28%、74.75%和73.56%;林地、水体和人工地表的地类纯净度分别为87.23%、77.04%和72.97%;总体分布空间一致性为74.0%。两套产品局部对比发现,LUC-2010产品分类结果更准确和精细,精度更高。

Land-cover and land-use classification in Beijing based on CART and GEE

Aiming at the problem that the traditional remote sensing image interpretation is slow,low efficiency,large manpower demand,the land-cover and land-use classification in 2010 of Beijing was interpreted in this paper,based on google earth engine(GEE)platform and classification and regression tree (CART)method,using Landsat 5 TM images and selecting normalized difference vegetation index(NDVI),normalized difference water index(NDWI)and digital elevation model(DEM)as test variables,seven types of cultivated land,woodland,grassland,wetland,water body,artificial surface and bare land were selected as classification system,after that,the spatial consistency analysis was carried out from type structure,confusion and spatial consistency between LUC-2010 products and GlobeLand30-2010 products.The results showed that google earth engine had outstanding advantages in remote sensing data analysis and processing at regional scales.The CART method had high accuracy of remote sensing classification,and the overall learning accuracy of the land cover products was above 94.2 ％.Overall comparison of two sets of products found that the spatial consistency ratios of woodland,water body and cultivated land were 84.28%,74.75% and 73.56% respectively;the purity of woodland,water body and artificial surface were 87.23％,77.04％ and 72.97％ respectively;the spatial consistency of the distribution was 74.0％.Local comparison of two sets of products found that theGlobeLand30-2010 product showed continuity was better;the LUC-2010 results had finer and higher accuracy.