Univariate time series typically represent the values of a single quantitative variable over time. However, in certain applications, such as hydrology, there may be multiple observations for each point in the index set of the time series. Common approaches to dealing with this issue involve summarizing the observations using functions such as averages, maximums, and minimums. However, such approaches can lead to loss of valuable information. In this paper, we propose using boxplots to represent multiple observations for each time point, and constructing a time series of boxplots. This enables the use of multivariate time series analysis techniques to extract information from the boxplot time series. To demonstrate the effectiveness of this approach, we apply singular spectrum analysis to model a real time series of water table depths in Iran. Our results suggest that the use of boxplots can lead to improved modeling and forecasting of time series in cases where there are multiple observations for each time point.
