Uncovering the Nature of Urban Land Use Composition Using Multi-Source Open Big Data with Ensemble Learning
Detailed information on urban land uses has been an essential requirement for urban land management and policymaking. Recent advances in remote sensing and machine learning technologies have contributed to the mapping and monitoring of multi-scale urban land uses, yet there lacks a holistic mapping...
Guardado en:
| Autores principales: | , , , , , , |
|---|---|
| Formato: | article |
| Lenguaje: | EN |
| Publicado: |
MDPI AG
2021
|
| Materias: | |
| Acceso en línea: | https://doaj.org/article/3c6a17472d6c4aadbbd865e0b2743c23 |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| Sumario: | Detailed information on urban land uses has been an essential requirement for urban land management and policymaking. Recent advances in remote sensing and machine learning technologies have contributed to the mapping and monitoring of multi-scale urban land uses, yet there lacks a holistic mapping framework that is compatible with different end users’ demands. Moreover, land use mix has evolved to be a key component in modern urban settings, but few have explicitly measured the spatial complexity of land use or quantitively uncovered its driving forces. Addressing these challenges, here we developed a novel two-stage bottom-up scheme for mapping essential urban land use categories. In the first stage, we conducted object-based land use classification using crowdsourcing features derived from multi-source open big data and an automated ensemble learning approach. In the second stage, we identified parcel-based land use attributes, including the dominant type and mixture mode, by spatially correlating land parcels with the object-based results. Furthermore, we investigated the potential influencing factors of land use mix using principal components analysis and multiple linear regression. Experimental results in Ningbo, a coastal city in China, showed that the proposed framework could accurately depict the distribution and composition of urban land uses. At the object scale, the highest classification accuracy was as high as 86% and 78% for the major (Level I) and minor (Level II) categories, respectively. At the parcel scale, the generated land use maps were spatially consistent with the object-based maps. We found larger parcels were more likely to be mixed in land use, and industrial lands were characterized as the most complicated category. We also identified multiple factors that had a collective impact on land use mix, including geography, socioeconomy, accessibility, and landscape metrics. Altogether, our proposed framework offered an alternative to investigating urban land use composition, which could be applied in a broad range of implications in future urban studies. |
|---|