面向人群迁移的事件-地理过程模型研究

Human movement including migration and mobility is a typically complicated geographical process. A deep understanding of features and underlying mechanism of human migration and mobility is still crucial and challengeable, and increasingly important for a range of issues regarding of management, planning and decision-makings. The primary objective of the present project is to develop an approach of Even-Geographical Process modelling for human migration and mobility in order to overcome the limited capabilities in quantitative expression and analysis of traditional methods derived from observed data of human movement straightforwardly. Even-Geographical Process model is based upon the combination of measured events of displacement and corresponding geographical process involving reparametrized covariance of driving forces. A series of analysis method and mining approach derived from the Even-Geographical Process model will be proposed as well for quantitatively characterizing features and patterns of human mobility. Using the Even-Geographical Process model and observed datasets for three distinct processes of human migration and mobility (i.e. census based floating population, seasonal migration during Lunar New Year from Baidu Big Data and geo-located Weibo derived human mobility), we will comparatively investigate the features and patterns of human movement with different temporal scales and diverse impact factors across China to promote understanding of complicated human migration and mobility. This project will provide methodological insights into the modelling and potential for improving the understanding of complicated human migration and mobility in present-day China.

人群的迁移（包括迁徙和移动）是典型的多因素共同作用下的复杂地理过程，深入分析迁移的基本特征和理解迁移的基本规律不仅是相关研究领域所关心的热点问题和面临的挑战，也对管理、规划和相关决策制定有着重要的意义。本研究将突破现有的人群迁移模型在表达能力和分析能力方面的限制，改变以拟合观测数据为目标的直接建模方式，在迁移事件的观测集合上通过结合数据本身的统计特征和迁移地理过程的影响因子来建立参数化的分布表达，从而建立用于定量表达人群迁移的“事件-地理过程”模型，并在事件-地理过程模型的基础上发展相应的分析方法。利用事件-地理过程模型对中国三个不同的人群迁移过程（包括基于普查的人口流动、基于百度大数据的春节迁徙和基于微博的人群移动）进行实证和应用研究，促进对中国复杂的人群迁移动特征和格局的深入理解，从而实现对迁移建模分析方法研究以及迁移规律发现研究的创新。

本项目针对人群迁徙和移动的定量估算与建模，通过结合事件-地理过程框架和Proportional Odds假设来建立迁移概率，实现了对多尺度迁移过程和驱动因素的时空耦合模拟建模。利用基于社交媒体的位置大数据并结合春节返乡这一传统现象，对整个中国范围的农村人口减少在空间分布和相对强度上获取了有效的定量估算以及分布的区域特征。同时，利用基于社交媒体的位置大数据，首次获取了月尺度夜间辐射信号与人群动态在不同空间尺度上的定量关系，并发现了正的响应特征的尺度离散性以及与城市规模和特征相关的区域变异性，进一步揭示了夜间辐射信号在精细的时空尺度上对于城市人口减少的定量响应关系，并系统地分析了基于信号变化检测不同尺度人群减少的有效性和限制性。在综合人类活动引起夜间辐射强度、不透水面比例以及植被指数等信息的基础上，提出了新的人类居住合成指数HSCI，使得可以对各种类型的人类居住区的空间分布和对应的社会经济活动强度进行空间清晰的综合定量刻画，并且可以有效地克服单一指数的饱和、扩散以及强度缺失等问题。同时提出基于局部亮度极值流域的分割算法，并结合二阶指数衰减模型，实现了对夜间辐射影像的面向城市空间结构的提取方法。在此基础上，进一步为基于轨迹流数据的人群移动和分布动态研究提供了新的时空耦合分析和制图方法。