摘要
了解非正式住区的位置规模、分布规律和未来土地需求对于减少贫困、优化土地利用和促进城市可持续发展具有重要意义。文章以撒哈拉以南非洲城市内罗毕和达累斯萨拉姆为例,基于Google Earth高分辨率影像、OpenStreetMap数据和建筑轮廓数据,利用机器学习分类等方法识别非正式住区,进而探究其空间分布规律和趋势。结果表明:1)在3种机器学习算法中,随机森林算法在非正式住区提取上的总体表现优于逻辑回归和支持向量机,Kappa系数达85%;建筑物特征对分类的总贡献度超过50%,起最关键作用。2)内罗毕和达累斯萨拉姆的非正式住区占城市总面积的比例分别为3.64%和21.69%,2个城市各识别出7个明显的集聚区,人口规模分别为3.47万~31.35万人、2.93万~129.41万人。非正式住区主要分布在缓坡、主要道路和水系沿线地区,且多位于城市中心区5 km以外,但在工业指向性上内罗毕显著高于达累斯萨拉姆。3)内罗毕非正式住区人均建筑面积仅为达累斯萨拉姆的一半,其人口拥挤和房屋紧凑程度更为严重。预计至2050年,2座城市将面临日益严峻的土地资源紧张困境与住区正规化挑战。因此,非正式住区的改造与升级亟待立足长远,采取前瞻性策略加以推进。
Informal settlements generally refer to low-quality residential areas built without government permission and characterized by poverty and insufficient basic services.Understanding their location,size,spatial distribution patterns,and future land demand is crucial for reducing poverty,optimizing land use,and promoting sustainable urban development.Using the sub-Saharan African cities of Nairobi and Dar es Salaam as case studies,based on multi-source open data,including Google Earth high-resolution imagery,OpenStreetMap data,and building footprint datasets,this study compared the performances of three machine learning classification methods(logistic regression,support vector machine,and random forest)in identifying informal settlements.After extracting informal settlements using the best-performing model,the spatial distribution patterns and trends of urban informal settlements in both cities were analyzed.Results showed the following:(1)The random forest model performed better than the support vector machine and logistic regression,exhibiting the highest overall accuracy and Kappa coefficient;hence,the random forest model had certain advantages in extracting informal settlements in different urban environments.The total contribution of four building features was greater than 50%,indicating that they played the most critical role in distinguishing informal settlements from other land types.(2)Informal settlements constitute 3.64%and 21.69%of the total areas of Nairobi and Dar es Salaam,respectively.Seven distinct agglomeration areas were identified in both Nairobi and Dar es Salaam,with population sizes of 34.7-313.5 and 29.3-1,294.1 thousand,respectively.In the two cities,informal settlements were mainly located in areas with gentle slopes,along rivers and main roads,and five kilometers away from the city center;however,informal settlements in Nairobi exhibited significantly higher industrial orientation than that in Dar es Salaam.(3)The population and building density of informal settlements in Nairobi were higher than those in Dar es Salaam,with the per capita building area in Nairobi being only half that of Dar es Salaam.This showed that the population congestion and housing compactness of informal settlements in Nairobi were more severe.With the increase of urban population,it is projected that by 2050,Nairobi and Dar es Salaam will add 30.26 and 169.96 km2 of informal built-up area,respectively.Under the prospect of informal settlement regularization,the future migrant population and corresponding land demand are expected to triple the current levels.In conclusion,we find a certain similarity in the spatial distribution patterns of informal settlements in Nairobi and Dar es Salaam;however,there are significant differences in their morphological patterns.In the future,both cities are expected to face increasing challenges,such as diminishing land resources and greater difficulties in regularizing informal settlements.Thus,informal settlements require targeted transformation and upgrading measures in combination with urban context differences and future development trends.
作者
朱静怡
陈爽
Zhu Jingyi;Chen Shuang(School of Geographical Sciences,Nanjing University of Information Science&Technology,Nanjing 210044,China;Research Centre of Urban Sustainable Development,Nanjing University of Information Science&Technology,Nanjing 210044,China)
出处
《热带地理》
北大核心
2025年第12期2265-2280,共16页
Tropical Geography
基金
国家自然科学基金项目(42161144003、41771140)。
关键词
城市非正式住区
机器学习
多源开放数据
土地资源
非洲
urban informal settlement
machine learning
multi-source open data
land resource
Africa
