Estimating commuting matrix and error mitigation – A complementary use of aggregate travel survey, location-based big data and discrete choice models

Abstract The prevalence of location-based big data has opened a new research frontier for estimating origin–destination commuting matrices for cities where granular flow data are not yet available from official sources. However, investigations into estimation errors and potential correction methods have been rare in the literature. To address the research gap, this paper first compares the performance of two estimated commuting matrices for Shanghai, derived by two distinct matrix estimation methods, namely a big-data approach using mobile phone signalling data and a discrete choice model for simulating the residential location of commuters. The empirical results indicate an outstanding analytical complementarity of the two approaches. A novel method is then proposed for mitigating the errors associated with the big-data approach. The proposed method features a selective blending of the big-data based flow estimation and the model-based estimation. By comparing the blended flow estimation with benchmark travel statistics, we find that the proposed method would significantly reduce the estimation errors and hence improve the robustness of the estimated matrix. It is expected that the proposed method will set a new standard for correcting potential errors in big-data based flow estimation.