VASA: Visual Analytics for Spatial Association

Exploring Spatio-Temporal Patterns

of Human Mobility

Evgeny Noi, Alexander Rudolph, Somayeh Dodge

Movement perspectives

Lagrangian perspective: Observing movement from the perspective of the moving entity; following the entity along its track over time

​

GPS Trajectories of an albatross (90 min resolution)

Eulerian perspective: Observing movement at fixed locations; recording presence of moving entities at certain locations

​

Aggregate Movement Indices

March 1, 2020

Department of Geography

Background

• Analysis of aggregated movement became crucial during the COVID-19 pandemic to investigate viral transmission and non-pharmaceutical interventions limiting human mobility.
• Due to privacy concerns (Bluetooth and GPS) researchers had to rely on aggregated movement data
• Opportunity to explore relationship between COVID-19 and mobility

Department of Geography

Regional COVID-19 mobility response

Department of Geography

Problem

• Measuring Mobility
• Mobility cannot be measured directly (cost, sampling, privacy)
• There is no unit of mobility (distance, exposure, dwell time, % sheltered)
• Mobility is inherently complex, heterogeneous and uncertain
• Assessing and Visualizing Mobility
• Conventional GIS tools
• Analysis across scales poses challenges
• Spatio-temporal (auto)correlation is hard to disentangle and grasp

Department of Geography

Data

• 26 indicators from across 10 data sources (Apple, SafeGraph, Cuebiq, Google, etc.)
• Missing data, gaps, measurement errors
• Spatial and temporal coverage
• Bias(edness)

Noi, E., Rudolph, A., & Dodge, S. (2022). Assessing COVID-induced changes in spatiotemporal structure of mobility in the United States in 2020: a multi-source analytical framework. International Journal of Geographical Information Science, 36(3), 585-616.

Department of Geography

Spatial Coverage (% of complete records per county)

✔️

✔️

✔️

✔️

Department of Geography

Temporal Coverage (% of complete observations per day)

✔️

✔️

✔️

✔️

Department of Geography

Bias(edness)

Only had access to the userbase across four data source.

​

Large metropolitan areas pose challenges to representation

Limited userbase:

• Cook, IL (Chicago)
• Los Angeles, CA (Los Angeles)
• Harris, TX (Houston)
• Maricopa, AZ (Phoenix)

Elevated userbase:

• Tarrant, TX (Fort Worth)

Census POP

Userbase

Over-represented in sample

Under-represented in sample

Department of Geography

VASA - Visual Analytics for Spatial Association

• Free and open source in Python
• Ideally suited for exploratory spatio-temporal data analysis of areal data
• Capture spatial and temporal structure of mobility
• Streamlines comparative visual analysis of spatio temporal data.
• Two visual displays to assess spatio-temporal patterns of data
• Recency and consistency map (RECO)
• Line-path scatter plots (LIPA)

Noi, E., Rudolph, A., & Dodge, S. (2023). VASA: an exploratory visualization tool for mapping spatio-temporal structure of mobility–a COVID-19 case study. Cartography and Geographic Information Science, 1-22.

Department of Geography

Global and Local Indicators of Spatial Association

The product of the value at location i with its spatial lag, the weighted sum of the values at neighboring locations.

Department of Geography

Spatial Weights

• Contiguity-based spatial weights (Queen, Rook, Bishop)
• Distance-based spatial weights (Euclidean, Manhattan, GC, kNN)
• We consider local neighborhood only spatial units located within a certain user-defined threshold distance (d) of spatial unit i.
• We consider local neighborhood only k-closest spatial units located in the vicinity spatial unit i.

Department of Geography

Typical Workflow for VASA

• Specify temporal and spatial unit of analysis and aggregate your data
• Specify spatial weights specification for your data (Queens, Rooks, kNN, etc.)
• Calculate Local Moran’s I for each spatial unit (i) at time (t), classify the units into hotspots/coldspots
• Calculate summary statistics for VASA: consistency and recency for each hotspots and coldspots
• Visualize calculated summary statistics using visual displays

Department of Geography

RECO - Recency and Consistency Map

(a) Centroids are generated and mapped as circle markers; (b) LISA is run to generate hotspots/coldspots, which are mapped to red/blue colors; (c) Recency is mapped to color value; (d) Consistency (number of times a county (or other object-level unit) is a cluster) is mapped to a marker size; (e) Recency and consistency are finalized in a RECO map.

Department of Geography

LIPA - Line-Path Scatter Plots

(1) significant coldspot, �(2) neither a hotspot/coldspot, �(3/4) - various behavior

Red for hotspots, blue for coldspots

Annotation of events (COVID-19 and state of emergency in CA)

Department of Geography

Department of Geography

Spatio-temporal patterns across Georgia

(2020, Cuebiq and SafeGraph, % sheltered)

​

Atlanta

National Average

State Average

Department of Geography

Spatio-temporal patterns across California

(2020-22, Apple and Google, visits*)

• Coldspots in the Bay Area: in Alameda, San Francisco, Contra Costa, San Mateo, and Marin county (IT jobs)

​

• Apple misses recent hotspots in NorCal: in Shasta, Humboldt, Lassen, Modoc, and Siskiyou counties (wineries and camp sites in numerous national forests).

​

Apple

Google

Department of Geography

Spatio-temporal patterns across California

(2020-22, Apple and Google, visits*)

Apple mobility plateaus around May 2021 (fails to capture Delta, Omicron and BA.1/2)

Apple

Google

Hotspots are better captured in Google data.

Department of Geography

Concluding Remarks

• Mobility data is heterogenous: spatial and temporal variations depend on the metric used
• Spatial and temporal coverage of the data varies and may affect research outcomes
• VASA provides a convenient way to assess spatial and temporal patterns in data across data sources and across scale

Department of Geography

Department of Geography

Hotspot/coldspot Recency and Consistency map

Department of Geography

Line-path of Hotspots/coldspots over time

Department of Geography