Expanding food bank's distribution program during Covid-19 pandemic
Network and suitability analysis to find 10 new sites for food distribution using ArcGIS
Capstone Project - GIS UC Davis-Coursera Specialization
As a consequence of the COVID-19 pandemic food insecurity has increased worldwide. High unemployment rates have impacted many families and individuals’ access to food. Food banks have been essential to fulfill the needs of a growing hungry population during this crisis providing food to millions of Americans.
This project aims firstly to determine how much of the food insecure population of Santa Clara and San Mateo counties (California, USA) is currently served by the local food bank; then it seeks to identify areas in which 10 additional distribution sites could be created in order to expand the spatial reach of the food distribution program.
Results
In the San Mateo and Santa Clara counties, 194 census tracts qualified as potentially at risk of food insecurity. These included low-income tracts in which the median yearly income was below $82,216; and tracts with high rates of unemployment. Out of the estimated total population of almost 2.7 million people, about 1 million residents (37%) were located in census tracts at-risk of food insecurity. The Food Bank serving the population in need had already established 77 food distribution sites throughout the two counties. Virtually the entire population at risk (99%) resided within a 3-mile radius of an existing distribution site, with only two tracts in San Mateo County outside that coverage.
After the COVID-19 pandemic began, the country had registered an increase in unemployment rate reaching 15% at its peak in April 2020. According to the Congressional Research Report published in January 2021, the population that experienced relatively higher rates of unemployment were part-time workers, workers with low educational attainment, and racial and ethnic minorities. After screening the population using these criteria, 46 new tracts were identified.
The selection of suitable sites was based on their accessibility, namely vicinity to a bus stop (≤0.5 miles) and walkability score of the area. 673 sites qualified as suitable locations based on these criteria.
Once the suitable new distribution sites were identified, Network Analysis was used to determine the optimal locations for 10 additional sites. An impedance distance cut-off of 9 miles was necessary to reach 88% of population miles.
Interactive map in ArcGIS Online
Workflow
Created a feature class with existing food distribution sites by geocoding physical addresses using TEXAS A&M Geoservices
Identified the census tract of low-income community at risk of food insecurity: income ≤80% of the metro area median income (U.S. Code Section 45D(e)), OR ≥10% of unemployment rate
To determine how much of the population is currently served by the food bank, the existing distribution sites were buffered at 3miles and 6miles.
The total population for each category at risk was adjusted for the share of working population for each tract as follow: (Total population working age / Total population) * Total population risk category
Then the tracts with predominantly at risk categories were selected. The cut-off for each category was established as follow: for Race, the majority-minority census tracts criteria was used; the national averages for part-time workers and education attainment were used as cut-off
The tracts that qualified as existing_food_insecure were erased from the selection, and newly_food_insecure_tracts layer was generated.
Accessibility was evaluated based on the vicinity to a bus stop and the walkability index of the area
Bus stops were clipped to the area of interest, then buffered at 0.5 miles and clipped again
The more walkable areas (index ≥5.76) were clipped from the suitable_step1 layer (suitable_step2)
Then suitable_step2 was clipped from school_candidate_sites to identify the suitable_candidate_sites
Network Analysis, Location-Allocation was used to identify the optimal location for 10 new food sites
To convert the OSM street data to network format and for topology correction, the Feature to Edge tool (GISF2E) was used (Karduni et al., 2016). The output edge and node files and the original road shapefile were imported in a new feature dataset, from which a new network dataset was generated
Location/Allocation solver: the existing_sites were used as required facilities, school_candidate_sites as candidate facilities. Other settings were: Problem Type was Maximize Attendance, Travel from Demand to Facility; 'Working population' used as weight; Lenght (meters) was used as impedance, tested cut-offs of 3, 5, and 9 miles
Out of the 240 demand points, six resulted allocated but not connected. Eventually, those non connected points were repositioned using Select/Move Network Location Tool. The points were moved less than 300 meters from the centroids
After the repositioning, the Solver successfully connected all the demand points to 65 required and 10 candidate facilities.
Limitations
National unemployment statistics were used here, but it has not been confirmed that the national trends are representative of the area of interest. In addition, it is unknown if site accessibility by individuals without private vehicles is typically considered a significant criterion for the selection of all distribution sites. The use of this criteria may have been too conservative, leading to the underestimation of suitable candidate sites.
In at least three instances, the tract centroids were located far from inhabited areas. This may have impacted the allocation of the population to the nearest site, overestimating the distance between them and potentially forcing the cut-off to expand behind real needs. In addition, a 9-mile distance between the tracts and the distribution sites may not actually be acceptable. It might be preferable to instead add more sites to cover the population at risk.
In addition, during the Location-Allocation analysis, the set number of candidate sites was arbitrarily set at 10 under the assumption that larger number of sites would have meant a significant increase in operation costs and management. However, this assumption may not hold in real life.
Finally, during the analysis it appeared that the network connectivity had to be reestablished for some of the distribution sites and it is reasonable to think that some issues remained undiscovered. Therefore it is plausible to expect that using a different network data set may yield different results.
References
Impact of COVID-19 on food security and nutrition. Committee on World Food Security, HLPE. FAO, September 2020 http://www.fao.org/3/cb1000en/cb1000en.pdf
Hunger spikes, demand rises for US food banks. Silvia Martinelli, 14 December 2020, BBC News. https://www.bbc.com/news/world-us-canada-55307722
About us: Bringing healthy food to Silicon Valley. Access date: April 13 2021. https://www.shfb.org/about-us/
Unemployment Rates During the COVID-19 Pandemic: in Brief. Congressional Research Service. January 12 2021. https://fas.org/sgp/crs/misc/R46554.pdf
Feeding America. Access date: April 13 2021. https://www.feedingamerica.org/take-action/coronavirus
Second Harvest Food Bank of Silicon Valley website
TEXAS A&M Geoservices
Median household income for the San Jose- San Francisco – Oakland CA Census Reporter
U.S. Internal Revenue Code Section 45D(e)
majority-minority definition wikipedia
For the missing values, the OpenStreetMap wiki page was used to understand the type of road and eventually determine the maxspeed for each type
Karduni, A., Kermanshah, A. & Derrible, S. A protocol to convert spatial polyline data to network formats and applications to world urban road networks. Sci Data 3, 160046 (2016)
Disclaimer
This project was designed and developed as capstone project for the GIS Specialization (UC Davis, Coursera). The only purpose of the study is to showcase skills acquired. It must not be considered for any other use. The analysis is based on data freely available online. The project’s author serves as volunteer at the Second Harvest food bank. However, she has no access to any confidential data or information.