Understanding human movement patterns at local, national and international scales is critical in a range of fields, including transportation, logistics and epidemiology. Data on human movement is increasingly available, and when combined with statistical models, enables predictions of movement patterns across broad regions. Movement characteristics, however, strongly depend on the scale and type of movement captured for a given study. The models that have so far been proposed for human movement are best suited to specific spatial scales and types of movement. Selecting both the scale of data collection, and the appropriate model for the data remains a key challenge in predicting human movements. We used two different data sources on human movement in Australia, at different spatial scales, to train a range of statistical movement models and evaluate their ability to predict movement patterns for each data type and scale. Whilst the five commonly-used movement models we evaluated varied markedly between datasets in their predictive ability, we show that an ensemble modelling approach that combines the predictions of these models consistently outperformed all individual models against hold-out data.

了解地方、国家和国际范围内的人类流动模式对于交通、物流和流行病学等一系列领域至关重要。有关人类运动的数据越来越多，当与统计模型相结合时，可以预测广泛区域的运动模式。然而，运动特征在很大程度上取决于给定研究捕获的运动的规模和类型。迄今为止提出的人类运动模型最适合特定的空间尺度和运动类型。选择数据收集的规模和适当的数据模型仍然是预测人类运动的关键挑战。我们在不同的空间尺度上使用了澳大利亚人类运动的两个不同数据源来训练一系列统计运动模型，并评估它们预测每种数据类型和规模的运动模式的能力。虽然我们评估的五种常用运动模型在数据集之间的预测能力存在显着差异，但我们表明，结合这些模型的预测的集成建模方法始终优于所有单独的模型（针对保留数据）。