Abstract:

Following the emergence of large numbers of spatiotemporal datasets, the literatures related to spatiotemporal data analysis increase rapidly in recent years. This paper reviews the literatures and practices in spatiotemporal data analysis, and classifies the methods available for spatiotemporal data analysis into seven categories: including geovisualization of spatiotemporal data, time series analysis of spatial statistical indicators, coupling spatial and temporal change indicators, detection of spatiotemporal pattern and abnormality, spatiotemporal interpolation, spatiotemporal regression, spatiotemporal process modelling, and spatiotemporal evolution tree. We summarized the principles, input and output, assumptions and computer software of the methods that would be helpful for users to make a choice from the toolbox in spatiotemporal data analysis. When we handle spatiotemporal big data, spatial sampling appears to be one of the core methods, because (1) information in a big data is often too big to be mastered by human physical brain, so has to be summarized by statistics understandable; (2) the users of Weibo, Twitter, internet, mobile phone, mobile vehicles are neither the total population nor a random sample of the total population, therefore, the big data sample is usually biased from the population, and the bias has to be remedied to make a correct inference; (3) the data quality is usually inconsistent within a big data, so there should be a balance between the variances of inferences made by using data with various quality and by using small but high quality data.

Key words: spatiotemporal data, spatiotemporal pattern, spatiotemporal process, spatiotemporal mechanism, sample, target population, big data