Origin–destination (OD) matrices are one of the key elements in travel behaviour analysis. For decades, transportation researchers have mostly used data obtained by active solicitation such as surveys to construct these matrices but new data sources like automatic fare collection (AFC) are now available and can be used to measure OD flows. As a result, a more heterogeneous corpus of data sources is now available to estimate travel demand. However, little research examines how comparable the estimated demands may be. In this paper, three data sources namely a household travel survey, a large scale origin–destination survey and entry only automated fare collection are processed to derive typical weekday public transit OD trip matrices. Various elements of the resulting matrices are then compared. While all the matrices share some common characteristics, there are also substantial differences that must be addressed. AFC data is not error-free and needs to be supplemented with data from other sources to construct a representative OD trip matrix. This is because not all destinations can be inferred, the smart card penetration rate is far less than 100% and fare evasion cannot be ignored. Our empirical results suggest that scaling an AFC matrix with automated passenger counts may be a viable solution. The results also indicate that the household travel survey significantly underestimates the volume of public transit trips compared to the other sources. The findings of this research contribute to a better understanding of the available data sources for public transit demand estimation. They can help practitioners to improve the quality and accuracy of OD matrices.

出发地目的地（OD）矩阵是旅行行为分析中的关键要素之一。几十年来，交通运输研究人员大多使用主动征集获得的数据（例如调查数据）来构建这些矩阵，但是现在已经有了诸如自动票价收集（AFC）之类的新数据源，可用于测量OD流量。结果，现在可以使用更加异构的数据源来估计旅行需求。但是，很少有研究检查估计需求的可比性。在本文中，处理了三个数据源，即家庭旅行调查，大规模原点-目的地调查和仅入口自动票价收集，以得出典型的平日公交OD出行矩阵。然后比较所得矩阵的各个元素。尽管所有矩阵都有一些共同的特征，但是也存在必须解决的实质性差异。AFC数据并非没有错误，需要用其他来源的数据进行补充，以构建代表性的OD跳闸矩阵。这是因为并非可以推断出所有目的地，智能卡普及率远低于100％，并且不能忽略逃票。我们的经验结果表明，使用自动乘客计数来缩放AFC矩阵可能是可行的解决方案。结果还表明，与其他来源相比，家庭旅行调查显着低估了公共交通旅行的数量。这项研究的结果有助于更好地理解可用于公共交通需求估算的数据源。