Potentials and limitations of large-scale mobile location GPS data for food access analysis

Introduction

Key codes used in the submitted paper Potentials and limitations of large-scale mobile location GPS data for food access analysis.

The data that support the findings of this study are available from Gravy Analytics. Restrictions apply to the availability of these data, which were used under license for this study. Aggregate-level data are available from the authors with the permission of Gravy Analytics per the data agreement with Gravy Analytics. We attached the corresponding data structure of the input and output of each script.

A workflow of the code is presented in the graph (Workflow_and_code.png).

Code files

1_stop_inference.py

Extract staypoints, positionfix linked with the staypoint and triplegs from GPS data.

• Input:

  • GPS data provided by Gravy Analytics (https://gravyanalytics.com/data-as-a-service/). Example:

      grid,latitude,longitude,time,user_id,week,geometry
      001eafd2-aa33-34c0-a022-26af9761ed22,30.35627,-81.54973,2022-09-03 04:24:38,2,0,POINT (-9078000.000000000 3549000.000000000)
      001eafd2-aa33-34c0-a022-26af9761ed22,30.35628,-81.54974,2022-09-03 04:24:48,2,0,POINT (-9078000.000000000 3549000.000000000)
    

• Output:

  • Trackintel framework components. For more information: https://trackintel.readthedocs.io/en/latest/index.html

  • staypoint:

      user_id,started_at,finished_at,geom id	
    

  • positionfix:

      user_id,flag,tracked_at,geom,staypoint_id,tripleg_id ID
    

  • tripleg:

      user_id,started_at,finished_at,geom,id	
    

2_home_inference.py

Based on the GPS data, infer device users' home locations.

Code based on algorithm by

*Zhao, X., Xu, Y., Lovreglio, R., Kuligowski, E., Nilsson, D., Cova, T.J., Wu, A., and Yan, X., 2022. Estimating wildfire evacuation decision and departure timing using large-scale GPS data. Transportation research part D: transport and environment, 107, 103277.*

Contains two functions for inferring home locations from GPS point:

1. Function1:

    input: each user's GPS, nighttime start and end hour.
    
    output: users home location inferred from nighttime stay
             location where user generated most GPS during nighttime .

2. Function2:

    input: each user's GPS.

    output: users home location inferred from weekend stay
             location where user generated most GPS during weekends

• Input:

  • GPS data (fields see above)

• Output:

  • Inferred home location for each device user

      user_id,home_lat,home_lon
    

3_food_stop_extraction.py

In this study, we used the North Florida food retailer location database by Erik Finlay at the University of Florida GeoPlan Center. For a quick look at the layers in the dataset: https://services.arcgis.com/LBbVDC0hKPAnLRpO/arcgis/rest/services/ACFS_Map/FeatureServer

Based on the extracted staypoints from 1_stop_inference.py and the food retailer location,

1. extract food related stops based on distance threshold
2. split the stops by users, later used as input for 4_food_trip_extraction.py
3. filter the stops, limiting 2-hour visit duration

• Input:

  • stay points (fields see above)

  • retail locations, reclassified from the input dataset and added a new TYPE field (Figure 4., Table 1. in submission):

      ObjectId,Store_Name,Longitude,Latitude,TYPE
    

  • threshold: numeric value, in meter. From all stay points, find out the stay points within the threshold of the food retails. We infer that the user has visited the location.

• Output:

  • food-related stops (Table 2. in submission)

      id,user_id,started_at,finished_at,lat,lon,retail_id,retail_lat,retail_lon
    

4_food_trip_extraction.py

Based on

1. the extracted triplegs from 1_stop_inference.py, and
2. the food stop extracted from 3_food_stop_extraction.py,

generate food trips. To do this, we used a sliding window method by setting the staying thresholds to a space of 100-meter radius, and a duration of at least 5 minutes and at most 720 minutes.

• Input:

  • triplegs (fields see above)
  • food-related stops (fields see above)
  • time window thresholds (explanation see above)

• Output:

  • food-related trips (Table 2. in submission)

      user_id,tripleg_ID,trip_started_at,trip_finished_at,trip,retail_id,retail_lat,retail_lon
    

5_metric_calculation.py

Based on

1. the user home location inferred with 2_home_inference.py
2. the retail location data
3. the extracted food-related stops from 3_food_stop_extraction.py
4. the extracted food-related trips from 4_food_trip_extraction.py

calculate the four food beahvior metrics at individual level.

• number of visit
• number of unique stores visited
• home-to-store distance (visited v.s. nearest)
• proportion of home-based visits

(Table 3., 4. and Figure 14. in sumission. See submission for detailed explanation of the metrics.)

6_visual_temporal.py

Based on the four food beahvior metrics calculated in 5_metric_calculation.py, generate line plots (curves) of

1. daily number of stops (Figure 11. in submission)
2. day of week pattern (Figure 10.3 in submission)
3. time of day pattern (Figure 10.1 and 10.2 in submission)

7_visualization_spatial.py

Based on the four food beahvior metrics calculated in 5_metric_calculation.py, calculate the tract-level aggregated average, and plot spatial distribution maps.

• Input:

  • calculated individual metrics (see 5_metric_calculation.py)
  • home location info (see 2_home_inference.py)
  • study area tracts shapefile (from US Census)

• Output:

  • spatial distribution maps of tract-level metrics (Figure 9 in submission)
  • other related figures (Figure 7., 8. and 13. in submission)

8_visualization_input.py

Plot the non-metrics-related datasets in maps.

• Input:

  • study area shapefile and ACS data
  • inferred user home location
  • food-related stops

• Output:

  • sociodemographic characteristics maps (Figure 3. in submission)
  • sampling rate map (Figure 5. in submission)