以下文章来源于地球空间信息科学学报GSIS ，作者北海道大学
引言
土地利用与土地覆盖（LULC）地图是展现城市环境动态变化的重要基础，可以用于解决城市扩张及其生态影响等社会问题。
然而，由于城市环境空间结构复杂、建筑材料种类繁多以及卫星图像覆盖范围有限等，混合像元问题在撒哈拉沙漠以南的非洲城市遥感图像中显著存在，使得LULC分类在基于遥感的城市研究中是一项具有挑战性的任务。数据源、分类特征和分类器的选择都影响着LULC分类与应用的效果。
日本北海道大学Armstrong Manuvakola Ezequias Ngolo和Teiji Watanabe教授综合运用GIS、遥感和机器学习技术，针对混合像元问题提出了一种基于光谱指数的LULC分类方法，并利用PQk-means（一种无监督学习算法）评估了其有效性，随后采用逻辑回归模型监测安哥拉罗安达的城市扩张情况，探究了潜在的驱动因素。文章研究结果有利于环境学家和城市规划者理解与监测城市变化情况。
相关研究成果凝练于学术论文Integrating geographical information systems, remote sensing, and machine learning techniques to monitor urban expansion: an application to Luanda, Angola，发表于国际SCI期刊Geo-Spatial Information Science(地球空间信息科学学报，GSIS)。
欢迎扫码查看原文
+
引用本文
Armstrong Manuvakola Ezequias Ngolo & Teiji Watanabe (2022) Integrating geographical information systems, remote sensing, and machine learning techniques to monitor urban expansion: an application to Luanda, Angola, Geo-spatial Information Science, DOI: 10.1080/10095020.2022.2066574.
研究区域与方法
研究区域
研究区域为罗安达北部5274km2的地区。罗安达位于安哥拉西部大西洋沿岸，面积18826km2（2014），人口 6,945,386（2016），现由七个自治市组成。
近年来，安哥拉政府正积极制定相关发展计划，旨在规定新的城区边界，控制城市的无序扩张。
研究数据
使用的数据包括：美国宇航局（NASA）Landsat 系列卫星数据，欧空局（ESA）哨兵一号SAR数据，日本宇航局（JAXA）的ALOS DSM数据。
研究方法
采用的研究方法主要包括：
① 基于Landsat 数据，计算五种光谱指数（NDBI, MEI, VIGS, DBI, QzCal）并作为无监督分类的输入，同时加入SAR数据以改善分类效果。
② 利用无监督分类器PQk-means生成LULC地图，并将其与当前最优的K-means无监督分类器、SVM等监督分类器的结果进行比较，以评估有效性。
③ 基于逻辑回归模型，探究城市LULC地图变化情况（2000-2008、2008-2018年LULC的累积变化）与潜在驱动因素（人口密度、海拔、坡度、到主要道路的距离、到住宅区的距离）之间的关系。
前沿观点
遥感技术能从上层空间采集数据，机器学习算法可以从中提取知识，然后在GIS环境叠加其他数据集（如道路、河流与样本点）进行空间分析操作，可以生成具备较高可解译性的可视化产品。
Remote sensing allows the collection of data from upper space and ML algorithms can help to extract knowledge from its products that can later be manipulated by overlaying with a cartographic dataset (e.g. road, river, sample points) derived from a GIS environment to produce a merged product that can be visualized for further interpretation.
比较PQk-means与其他无监督、监督方法的分类结果：
① 对于2018年LULC地图，PQk - means和K - means无监督算法分别取得了93%和95%的总体精度。PQk - means相较于K - means具有更高的计算效率。
② 在监督分类中，SVM、RF和MLC算法分别取得了89%、87%和78%的总体精度。
Compare the classification effect of PQk-means with other unsupervised and supervised methods :
① For the LULC map 2018, 93% and 95% of overall accuracy were achieved corresponding to PQk-means and K-means unsupervised algorithms, respectively. PQk-means has higher calculation efficiency, compared with K-means.
② From the supervised classification, 89%, 87%, and 78% of overall accuracy were achieved using SVM, RF, and MLC algorithms, respectively.
对于特定区域，如果选择了合适的光谱指数和分类器，基于光谱指数的无监督分类方法能够有效实现LULC分类。
对于无监督的LULC分类任务，首选K-means算法。但是考虑到存在大量的输入（光谱指数），PQk-means是一个可行的替代方案，它能够提供类似于K-means的良好分类效果，同时具有更高的内存和计算效率。
Unsupervised classification utilizing spectral indexes can effectively classify LULC if appropriate spectral indices and classifiers are chosen for a specific area.
For unsupervised LULC classification, the K-means algorithm is preferred. However, considering a large number of inputs (spectral indexes), PQk-means is an alternative selection because it is more efficient in memory and computing, and gives good accuracy similar to K-means.
基于LULC分类结果估算罗安达三个年份（2000，2008，2018）的建成区面积：
① 针对2000年，估算的建成区面积约为94 km2，而城市扩张图集(AUE)估算的为171.75 km2 。造成差异的原因可能是AUE的分类标准略显武断。
② 针对2008年，估算的建成区面积约为197 km2 ，而基于Envisat MERIS高分辨率300m影像（2009年）估算的为207 km2。
③ 针对2018年，估算的建成区面积为468 km2，而ESA (2017)基于非洲高分辨率原型地图上估算的2015/2016年建成区面积为407 km2，两者非常接近。
Estimate built-up area of Luanda for three years (2000, 2008, 2018) based on LULC classification results:
① For 2000, the estimated built-up area is around 94 km2, while the Atlas of Urban Expansion (AUE) estimated the built-up area to be 171.75 km2. The reason for the difference may be that AEU's standard of classification is somewhat arbitrary.
② For 2008, the estimated built-up area is around 197 km2 while the map based on Envisat MERIS fine resolution 300 m image of 2009 estimates 207 km2.
③ For 2018, the estimated built-up area is 468 km2 that is comparable with the 407 km2 of the 2015/2016 built-up area estimated by ESA (2017) on its high-resolution prototype map of Africa.
探究罗安达城市扩张的驱动因素：
① 罗安达2000年至2018年期间的城市扩张主要由已建成住宅区和主要道路驱动。
② 从2000年到2018年，海拔对城市扩张的影响由正面变成负面，影响力有所降低，而坡度的影响由负面变成正面，但影响力仍保持在较低水平。
③ 生成的概率图显示，在政府确定了住房计划的地区，城市发展的概率很高。
④ 政府实施的政策决定了城市建成区的面貌，因此在未来的城市建设中，交通网络的规划预计将成为城市发展的主要动力，并为新型高密度和高质量发展预留空间。
Explore the driving factors of urban expansion in Luanda :
① The urban expansion in Luanda of during the period 2000–2018 was mainly driven by the proximity to the already established residential areas and to the main roads.
② The variable elevation changes from a positive and high influence on urban expansion to a negative and low influence, while the variable slope changes from a negative and low influence to a positive and still low influence during the period 2000–2018.
③The generated probability maps show high probability of urban growth in the areas where government had defined housing programs
④ Since government-imposed rules determines the way the built-up area looks like today, the transport network is expected to become the main driver of the urban construction and to be reserved for new dense and high-quality development.
作者简介
Armstrong Manuvakola Ezequias Ngolo
在安哥拉罗安达的阿戈斯蒂尼奥内托大学理学院获得地质学学士学位，在日本北海道大学环境科学学院获得环境科学硕士学位。研究兴趣包括面向地球科学的GIS/遥感技术、应用型机器学习和网络制图。
Armstrong Manuvakola Ezequias Ngolo received his undergraduate degree in Geology from the Faculty of Science, Agostinho Neto University, at Luanda, Angola and his Master degree in Environmental Sciences from Hokkaido University, Graduate School of Environmental Science, at Hokkaido, Japan. His research interests are GIS/Remote sensing applied to Geosciences, Applied Machine Learning, Web Cartography.
Teiji Watanabe
在加州大学戴维斯分校获得博士学位，现为北海道大学环境地球科学学院环境地理学教授、全球土地计划（GLP）日本办公室主任。
Teiji Watanabe received his PhD degree from the University of California at Davis. He is a professor of Environmental Geography at Hokkaido University, Faculty of Environmental Earth Science, and the director of Global Land Programme (GLP) Japan Nodal Office.
转自：“测绘学术资讯”微信公众号
如有侵权，请联系本站删除！
