Predicting and Generating Urban Human Mobility Flows with Random Forests and Land Cover Data

Models of human mobility are crucial to understanding transit needs, demographic trends, and disease outbreaks, among much else. Since the start of the COVID-19 pandemic, new datasets have arrived from public and private sources, and the literature has advanced with new modelling techniques using deep learning and machine learning. Inspired by these breakthroughs, this paper seeks to develop a mobility model of mobility that competes with the traditional gravity model in both accuracy and ease of use. To do so, we employ Copernicus land cover data and random forest models, tested using recent data collected from over 400 million mobile phones in Spain. Our model improves on the gravity model in both prediction of future flows and generation of flows for unknown locations.