Authors: Xuchao Yang*, , Qian Chen, Ocean College, Zhejiang University, Tingting Ye, Ocean College, Zhejiang University, Naizhuo Zhao, Center for Geospatial Technology, Texas Tech University, Wenze Yue, Department of Land Management, Zhejiang University
Topics: Applied Geography, Remote Sensing, China
Keywords: GDP, random forests, Points of interest, remote sensing, China
Session Type: Paper
Start / End Time: 9:55 AM / 11:35 AM
Room: Senate Room, Omni, West
Presentation File: No File Uploaded
Nighttime light (NTL) imagery is the most commonly used data for mapping of GDP over large areas. However, the spatial and textural features derived from medium-spatial-resolution NTL are not good proxies for disaggregating agricultural GDP and are inadequate to differentiate the speciﬁc distribution of secondary and tertiary sectors. Land use and vegetation cover data have a natural advantage in mapping production of primary sector because forestry, farming, livestock, and fishing highly relate to the natural resources. Points of interest (POI), a typical geospatial big data with location information and textual description of category, can distinguish business, industrial, and commercial areas, and therefore benefit the precise mapping of GDP from secondary and tertiary sectors. In this study, a random forest (RF) algorithm was used to disaggregating the 2010 Chinese census GDP data by county to 1 km × 1 km grid. The RF method is characterized by a flexible framework that allows varying data types (i.e. multi-source remote sensing data and geospatial data) to interact with each other. Individual RF model for each economic sector was constructed to explore the non-linear relationships between varied covariates and economic production from different sectors. By fusing POIs of varying categories, spatial distribution of economic activities from the secondary and tertiary sectors can be distinguished effectively. Comparing to previous studies, the strategy of developing different RF model for different sectors generated more rational distribution of GDP. The POIs and methodology from this study can be easily adopted for mapping other socioeconomic parameters in other countries.