GlobeLand 30和自发地理信息的对比分析研究
作者简介:马京振(1993-),男,博士生,研究方向为多源数据融合与处理。E-mail: zb50mjz@163.com
收稿日期: 2018-01-26
要求修回日期: 2018-06-04
网络出版日期: 2018-09-25
基金资助
国家自然科学基金项目(41571399)
Comparison Analysis of GlobeLand 30 and Volunteered Geographic Information
Received date: 2018-01-26
Request revised date: 2018-06-04
Online published: 2018-09-25
Supported by
National Natural Science Foundation of China, No.41571399.
Copyright
地表覆盖数据是关于土地利用信息的重要来源,在地理国情监测、生态环境保护等方面发挥着重要的作用,目前遥感影像解译、实地测量是该数据生产的主要手段,但是仍然存在一定的局限性。随着Web2.0、互联网技术以及各种GPS设备的快速发展传播,普通大众也可以参与公众制图,志愿者用户的参与能够有效判定地表类型的空间分布和属性特征,提高地表覆盖制图的分类精度。本文以自发地理信息中最成功的项目OpenStreetMap为例,与中国新研制的全球最高30m分辨率地表覆盖数据产品GlobeLand 30进行对比分析,首先对数据进行相应的预处理和拓扑检查,然后建立两种数据的要素对应关系,最后生成误差矩阵并分析两种数据的一致性。实验结果表明:① OpenStreetMap数据缺失的部分主要是耕地类型,其草地和水体要素比GlobeLand 30更加丰富;② 2种数据的一致性较好为75%左右,其中林地和人造地表的精度较高,耕地和水体次之,草地较差;③ 重点对不一致区域的地表类型进行判断验证,能够发现GlobeLand 30数据中的错误分类,为进一步修改和优化提供依据。本文研究表明,自发地理信息中包含丰富的地表覆盖信息,能够给地表覆盖制图及评价验证带来巨大的发展潜力。
关键词: 地表覆盖; 自发地理信息; GlobeLand 30; OpenStreetMap; 一致性
马京振 , 孙群 , 徐立 , 温伯威 , 李元復 . GlobeLand 30和自发地理信息的对比分析研究[J]. 地球信息科学学报, 2018 , 20(9) : 1225 -1234 . DOI: 10.12082/dqxxkx.2018.180077
Land cover data, which plays a significant role in national geographical condition monitoring, ecological environmental protection and some other areas, is an important resource of the information on land use. At present, land cover data is produced mainly through the interpretation of remote sensing imagery and field measurement, and some limitations still exist to a certain extent. With the rapid development and wide spread of Web2.0, internet technology and various kinds of GPS equipment, the general public have the opportunity to participate in crowd sourced mapping. Volunteer users can identify the spatial distribution and attributive characters of the land cover effectively. Therefore, the classification accuracy of land cover map can be improved in the meantime. In this paper, OpenStreetMap, the most successful item of volunteered geographic information, was taken as an example on the comparison analysis with GlobeLand 30, the newly developed land cover data produced in China with 30m resolution. Firstly, the data was preprocessed and topologically checked, and then the feature relationship was established. Finally, a confusion matrix was built to analyze the consistency between the two kinds of data. The experimental results show that the missing part of the OpenStreetMap is mainly cultivated land, and the grassland and water elements are more abundant than those of GlobeLand 30. The consistency of OpenStreetMap and GlobeLand 30 is high with a value of 75%. Forest and artificial surface have the highest accuracy, and cultivated land and water body take the second place, while grassland possesses the worst consistency. The key point is to verify and determine the land cover type within the inconsistent areas, and try to find classification errors of GlobeLand 30 so as to provide basis for further modification and optimization. Volunteered geographic information contains abundant land cover information, so it can provide great potential for the development and evaluation of land cover maps. The research methods and conclusions of this paper can provide basis for exploring the application of OpenStreetMap data for land cover mapping, and provide support for assessment and improvement of the classification accuracy of GlobeLand 30.
Fig. 1 The global land cover GlobeLand 30 map图1 全球地表覆盖数据GlobeLand 30分布示意图 |
Fig. 2 The land use map图2 土地利用分布图 |
Fig. 3 The flow chart of experimental data processing图3 实验数据处理流程图 |
Fig. 4 Sliver polygons and overlapping polygons of OSM areas图4 OSM面状要素的破碎和重叠多边形实例图 |
Tab. 1 Features corresponding relation of GlobeLand 30 and OSM表1 GlobeLand 30和OSM的要素对应关系表 |
OSM | GlobeLand 30 | 重分类 | |
---|---|---|---|
Key(要素类) | Key Values(属性值) | ||
amenity | arts_center, bank, bar, café, car_rental, cinema, clinic, college, community_centre, courthouse, crematorium, crypt, dentist, embassy, fast_food, ferry_terminal, fire_station, fuel, gym, hospital, internet_cafe | 80人造地表 | 5人造地表 |
building | apartments, house, garage, residential, cathedral, chapel, church, civic, commercial, hangar, hospital, hotel, industrial, kiosk, mosque, office, public, retail, school, shrine, stadium, synagogue, temple, train_station, transportation, warehouse | 80人造地表 | 5人造地表 |
landuse | allotments, farm, farmland, orchard, vineyard | 10耕地 | 1耕地 |
forest, scrub | 20森林,40灌木地 | 2林地 | |
grass, meadow, flowers, greenfield, plants | 30草地 | 3草地 | |
reservoir, pond | 50湿地,60水体 | 4水体 | |
residential, cemetery, commercial, industrial, military, retail, brownfield, construction, depot, quarry | 80人造地表 | 5人造地表 | |
natural | forest, scrub, wood | 20森林,40灌木地 | 2林地 |
grassland | 30草地 | 3草地 | |
mud, wetland, bay, water, riverbank | 50湿地,60水体 | 4水体 | |
heath, beach, sand, shingle, bare_rock, scree, glacier | 70苔原,90裸地,100冰川和永久积雪 | 6其他 | |
pofw | buddhist, christian, christian_catholic, jewish, muslim | 80人造地表 | 5人造地表 |
traffic | dam, fuel, marina, parking, service, weir | 80人造地表 | 5人造地表 |
water | dock, reservoir, river, water, wetland | 50湿地,60水体 | 4水体 |
Tab. 2 Pixel numbers and error coefficients of features表2 要素像元数量及误差系数统计表 |
类型 | 维也纳 | 巴塞罗那 | |||||
---|---|---|---|---|---|---|---|
OSM | GlobeLand 30 | 误差系数/% | OSM | GlobeLand 30 | 误差系数/% | ||
1耕地 | 2 357 067 | 6 588 479 | 64.22 | 63 678 | 662 979 | 90.40 | |
2林地 | 2 729 166 | 2 237 169 | 18.03 | 1 093 207 | 1 204 142 | 9.21 | |
3草地 | 328 699 | 39 632 | 87.94 | 14 886 | 39 100 | 61.93 | |
4水体 | 358 031 | 277 560 | 22.48 | 18 241 | 2651 | 85.47 | |
5人造地表 | 1 292 383 | 1 076 367 | 16.71 | 616 343 | 790 409 | 22.02 | |
6其他 | 5919 | 66 | 98.88 | 3369 | 8128 | 58.55 | |
总数 | 7 071 265 | 10 219 273 | 30.80 | 1 809 724 | 2 707 409 | 33.16 |
Fig. 5 The land cover map of GlobeLand 30 and OSM图5 GlobeLand 30和OSM地表覆盖分布图 |
Tab. 3 The confusion matrix of Vienna表3 维也纳地区误差矩阵 |
OpenStreetMap | ||||||||
---|---|---|---|---|---|---|---|---|
1耕地 | 2林地 | 3草地 | 4水体 | 5人造地表 | 6其他 | 使用者精度% | ||
GlobeLand 30 | 1耕地 | 2 245 656 | 790 002 | 194 314 | 51 202 | 359 563 | 4630 | 61.60 |
2林地 | 35 055 | 1 845 935 | 98 264 | 41 782 | 99 355 | 483 | 87.04 | |
3草地 | 302 | 2036 | 1895 | 4864 | 27 126 | 46 | 5.22 | |
4水体 | 1720 | 6038 | 2708 | 235 316 | 7704 | 299 | 92.72 | |
5人造地表 | 67 090 | 55 563 | 27 962 | 18 930 | 795 999 | 451 | 82.40 | |
6其他 | 17 | 18 | 0 | 31 | 0 | 0 | 0 | |
生产者精度% | 95.57 | 68.38 | 0.58 | 66.83 | 61.72 | 0 |
Tab. 4 The confusion matrix of Barcelona表4 巴塞罗那地区误差矩阵 |
OpenStreetMap | ||||||||
---|---|---|---|---|---|---|---|---|
1耕地 | 2林地 | 3草地 | 4水体 | 5人造地表 | 6其他 | 使用者精度/% | ||
GlobeLand 30 | 1耕地 | 52 694 | 154 054 | 6123 | 3901 | 58 607 | 754 | 19.08 |
2林地 | 3022 | 876 764 | 1259 | 3799 | 37 814 | 1755 | 94.85 | |
3草地 | 982 | 11 976 | 697 | 3088 | 6365 | 167 | 2.99 | |
4水体 | 42 | 617 | 9 | 1378 | 147 | 196 | 57.68 | |
5人造地表 | 6933 | 47 673 | 6783 | 4935 | 510 108 | 486 | 88.42 | |
6其他 | 5 | 2121 | 12 | 281 | 2434 | 11 | 0.23 | |
生产者精度/% | 82.75 | 80.20 | 4.68 | 7.93 | 82.88 | 0.33 |
Tab. 5 Comparison of assessment indicators between Vienna and Barcelona表5 维也纳和巴塞罗那的评价指标对比 |
城市 | OA/% | AD/% | QD/% | Kappa系数 |
---|---|---|---|---|
维也纳 | 72.98 | 8.51 | 18.32 | 0.6043 |
巴塞罗那 | 79.74 | 7.96 | 12.29 | 0.6487 |
Fig. 6 The bar chart of producer and user accuracy图6 生产者精度和使用者精度图 |
Fig. 7 Spatial distribution of agreement and disagreement图7 GlobeLand 30和OSM数据一致/不一致空间分布 |
Fig. 8 The inconsistent water between OSM and GlobeLand 30图8 OSM面状水体与GlobeLand 30不一致区域 |
The authors have declared that no competing interests exist.
[1] |
|
[2] |
[
|
[3] |
[
|
[4] |
[
|
[5] |
[
|
[6] |
[
|
[7] |
|
[8] |
[
|
[9] |
|
[10] |
|
[11] |
|
[12] |
|
[13] |
|
[14] |
|
[15] |
[
|
[16] |
|
[17] |
|
[18] |
|
[19] |
OpenStreetMap Wiki Map Features. Available online: https://www.openstreetmap.org/wiki.
|
[20] |
[
|
[21] |
|
/
〈 | 〉 |