Establishing a baseline is crucial for examining changes in mobility patterns in response to external shocks and policy interventions, such as social-distancing policies and lockdown after the disease outbreak. It can be used for detecting displacement and returnees under and post disasters as well as measuring its impacts (Wilson et al., 2016; Li et al., 2019). As the human mobility is generally seasonal (Barrios, Bertinelli, & Strobl, 2011) (Wesolowski, Metcalf, et al., 2015), ideally, data of the same time window in previous years are useful for establishing a baseline, which enables to account for seasonality when compared (Statistics New Zealand, 2012). However, data for such a long time are often unavailable. In this case, impacts on the mobility patterns associated with the seasonality need to be carefully examined when interpreting detected changes under and post disasters.

To understand the usual mobility patterns, data of a pre-disaster period before the onset of disasters are often used for establishing a baseline. For example, it is recommended to use a period of at least four weeks before the event of interest, such as interventions to compute a baseline, against which the post-intervention conditions could be compared (Flowminder, n.d.). The COVID-19 community mobility reportby Google LLC uses the median day-value from the 5-week period in January 2020 for each day's baseline to examine changes in mobility patterns under the COVID-19 (Google, n.d.). The Facebook Movement Range Mapsof the Facebook Data for Good, uses the data of February as a baseline for computing changes under the COVID-19. For computing baselines, only data from the same time-of-day and day-of-the-week in the period preceding the crisis are used.

In countries where mobility is relatively high under normal conditions, it is important to account for the high baseline mobility. For example, in Nepal where population flows following the earthquake were normalized using pre-earthquake mobility estimate. These were computed as the changes in locations from a benchmark period just before the earthquake (Wilson et al., 2016).

  • No labels