Revisiting Food Deserts in North Carolina, USA, Using a Cloud-Based

Real-Time Quality Assurance/Quality Control (QA/QC) Tool

Timothy Mulrooney

and Isabel Gutierrez

Department of Environmental, Earth and Geospatial Sciences, North Carolina Central University,

Durham, North Carolina, U.S.A.

Keywords: Food Environment, Geographic Information Systems, Quality Assurance, Quality Control, Spatial Data

Quality.

Abstract: In the study of the food environment, little research has explored the spatial data quality of store locations

which impacts the spatial representation of the food environment. In this paper, we created a cloud-based

tool that can inspect, edit and create new supermarkets in real-time which changes the complexion of the food

environment. Comparisons were made between data supplied between a CAB (Commercially Available

Business) Database and those corrected after field verification. Results showed differences between the food

environment using the data provided and the actual food environment after QA/QC, with a general

underestimation of those who are truly food needy due to errors of temporal accuracy, misattribution and

geocoding in the original data provided.

1 INTRODUCTION

An underlying theme of underrepresented and

marginalized communities across the United States is

differential access to community amenities. In

particular, healthy food is one of these amenities to

which these communities have poorer access.

Organizations such as the United States Department

of Agriculture (USDA) has utilized the term food

desert to highlight regions within low-income

communities located far from fresh and healthy

sources of food in the form of supermarkets and

supercenters. These Low Income/Low Access

(LILA) regions can visualized through the USDA

Food Access Atlas (https://www.ers.usda.gov/data-

products/food-access-research-atlas/go-to-the-atlas/)

at the census tract level. Furthermore, data which

compose these maps include more than 140 attributes

across 72,000 census tracts that can be downloaded,

analyzed and mapped within the confines of a GIS

(Geographic Information System).

The USDA helps determine access by its

physical proximity to supermarkets using geographic

measurements. The data on which this proximity is

measured changes on a regular basis due to the

https://orcid.org/0000-0001-9333-9641

https://orcid.org/0000-0001-6413-5463

closing and opening of new stores, and is further

exacerbated by the fidelity of those data on which

measurements are based. An understudied tenet of

food environment research is an overall assessment

and evaluation of the spatial data quality, in this case

the supermarkets store data used to measure this food

access. This assessment has been easier with custom

phone applications that can access data stored in the

cloud to inspect, verify, edit and re-attribute the

spatial data used to represent supermarkets and the

larger food environment in general. These errors of

omission and commission can have a distinct impact

on these regions highlighted as Low Access by the

USDA Food Access Atlas and those regions that are

truly low access using the most current data. In this

study, supermarkets for a 5-county region in North

Carolina, United States, are brought into a custom

field application that can explore various accuracies

(horizontal, temporal, attribute) of existing data to

answer the question of to what extent do real-time

QA/QC techniques impact the spatial and

statistical representation of the food environment.

After a comprehensive QA/QC is run on the data

using this phone application, newly-analyzed Low

Access (LA) and then LILA regions using these

Mulrooney, T. and Gutierrez, I.

Revisiting Food Deserts in North Carolina, USA, Using a Cloud-Based Real-Time Quality Assurance/Quality Control (QA/QC) Tool.

DOI: 10.5220/0011713500003473

In Proceedings of the 9th International Conference on Geographical Information Systems Theory, Applications and Management (GISTAM 2023), pages 115-122

ISBN: 978-989-758-649-1; ISSN: 2184-500X

 2023 by SCITEPRESS – Science and Technology Publications, Lda. Under CC license (CC BY-NC-ND 4.0)

115

corrected data are compared to LA and LILA data

utilized using the USDA Food Access Atlas. Using

statistical and geostatistical tools, the level of

agreement and disagreement between USDA Food

Access maps and maps using corrected data will be

measured to explore if, where and how these

differences exist across the study area.

2 LITERATURE REVIEW

Spatial data quality is the result of frameworks

designed to ensure newly created data are correct

(Quality Assurance) while identifying existing data

that are incorrect (Quality Control). Although the

QA/QC of spatial data within a GIS is required as per

Federal Geographic Data Committee (FGDC)

standards and various organizations have processes in

place to ensure the various accuracies are adhered to

that best fit their needs, resources and limitations, it

has not been at the forefront of GIS research when

compared to other facets of Geographic Information

Science. GIS data, subsequent analysis and products

of this analysis such as decisions and maps are only

as good as the data on which it is based. Newcomer

and Szajgin (1984) and later Heuvelink (1998)

showed inaccuracies in original GIS data exacerbated

data quality throughout the life of a GIS project,

culminating in unreliable analysis and maps.

QA/QC procedures have been applied to digital

data related to the food environment. Liese et al.

(2010) and Auchincloss et al. (2012) explored the

quality of purchased retail location data, referred to as

Commercially Available Business (CAB) data.

These CAB data serve as a baseline for data

QA/QCed in the field in this project. Other studies by

Mendez et al. (2016), Rummo et al. (2015), Han et al.

(2012) and Hosler and Dharssi (2010) were

performed for Pittsburgh Durham, Chicago, Albany

and respectively. All cited some degree of difference

between CAB data versus field-based and automated

methods. Sharkey and Horel (2009) verified the

addresses of food sources provided from independent

sources such as Internet telephone directories,

telephone directories and the Texas Department of

Agriculture. They found 18.9% of food sources

provided via this public data could not be verified.

Furthermore, they found 35.7% of food sources

within their study area were only identified through

ground-truthing, representing errors of omission. In

another study by Lake et al. (2012), field verification

was performed on 21 different food source categories

(Restaurant, Pub/Bar, etc.) across different

permutations of socio-economic status (SES) and

population density (urban, rural, mixed). For the rural

low SES, more than one third (36%) of food sources

provided could not be found in the field (i.e., error of

commission). Not only is access and availability

compromised in marginalized areas, but the quality of

data as well. In North Carolina, Vilme et al. (2020)

complemented CAB data developed by

ReferenceUSA (the predecessor to DataAxle) with in

situ verification through Google or the facility’s web

site. They further utilized the Jackson Heath Study

Retail Store classification to derive favorable,

unfavorable and unknown categories from 15

different classifications. These categories will be

important in this study as census tracts will be

denoted as LA vs. not LA or LILA vs. not LILA based

on proximity measures provided by the USDA and

then recreated using QA/QCed data.

3 STUDY AREA

As part of a larger research project into large-scale

data quality issues across North Carolina’s food

environment, a 5-county study area in central North

Carolina was created across the counties of

Alamance, Chatham, Orange, Person and Yancey

Counties. This study area was selected due to its 1)

proximity to the author’s host institution so field

QA/QC could be performed 2) an area that has a

manageable number of supermarkets that could be

handled within the scope of this project and 3) the

combination of rural to suburban and urban regions in

Table 1: Summary of Study Area Using USDA Food Access Atlas Data.

Urban Non-Urban Study Area

# Census Tracts 46 44 90

Total Po

ulation 204,064 207,556 411,620

% Minorit

(

Non-White

)

31.7 21.5 26.7

Median Famil

Income $79,003 $79,905 $79,449

Poverty Rate 19.4 11.5 15.5

% Kids (Under age 17) 21.6 22.1 21.9

% Seniors (Over age 65) 13.1 14.7 13.9

% Grou

Quarters 6.3 1.53 3.96

GISTAM 2023 - 9th International Conference on Geographical Information Systems Theory, Applications and Management

116

the area. This includes the cities of Burlington (2020

pop. 57,303) and Chapel Hill (61,960). Utilizing 2010

census data via the USDA Food Access Atlas, the

study area’s 90 census tracts contain a 2010

population of 411,620. Tracts range in size of .26 sq.

miles (.67 sq. km) in Chapel Hill to 160.82 square

miles (416.51 sq. km) in rural Chatham County.

Populations range from 1,450 to 8,760 per census

tract. Within these data provided via the Food Access

Atlas is a flag (1 = yes, 0 = no) to denote if a census

tract is urban, as well as well as information about

income, food availability, and related socio-economic

factors such as age, race, incomes and ethnicity in a

spreadsheet format across more than 140 attributes.

Table 1 highlights the composition for the study area.

4 DATA AND METHODS

Data from the USDA Food Access Atlas were

downloaded, brought into a GIS and mapped for the

study area. Also included in the aforementioned

socio-economic-demographic variables (Table 1) are

metrics related to those census tracts that are Low

Access (LA) and Low Income/Low Access (LIIA).

According to the USDA (https://www.ers.

usda.gov/data-products/food-access-research-

atlas/documentation/), LA is defined as “a tract with

at least 500 people, or 33 percent of the population,

living more than 1 mile (urban areas) or 10 miles

(rural areas) from the nearest supermarket.” LILA are

defined to be census tracts that satisfy both Low

Access (LA) and Low Income (LI), which represent

tracts where the “annual family income at or below

200 percent of the Federal poverty threshold for

family size.”

Table 2: Information about study area using USDA Food

Access Data.

Urban Rural Total

# LA Tracts 29 6 35

# LILA Tracts 14 4 18

ulation LA 130,870 26,642 157,512

ulation LILA 66,262 18,372 84,634

4.1 Development of QA/QC Tool

Data related to supermarkets were utilized by point

data provided by DataAxle. These data were queried

using their NAICS (North American Industry

Classification Standard) code which classifies

business establishments by their primary economic

activity. According to the database, there are 104

stores classified as supermarkets within the study

area. These data were exported to the cloud that could

be accessed using desktop applications such as ArGIS

Pro, online applications such as ArcGIS Online as

well as online and smartphone applications such as

Esri Field Maps. These field maps have advantages

over applications such as Survey123 which create

data from scratch in that data can be added to the

existing database or edited from data brought in by

the data creator. Furthermore, additional fields can

be added to data where Survey123 does not allow for

those on-the-fly changes after features have been

created. This application has simple drop-down

menus to answer questions related to temporal,

attribute and positional accuracy of the data in

question. It also allows images of the site to be

captured and attached to data records.

Figure 1: Esri Field Maps Application.

A Standard Operating Procedure (SOP)

document was developed to maintain consistency in

data collection. Data QA/QC took place over the

Spring of 2022 using a combination of actual field

visits complemented with virtual field visits using

GoogleMaps and NCOneMap data where updated

imagery were available using the latest imagery

available through the North Carolina Department of

Transportation imagery service (https://

services.nconemap.gov/secure/rest/services).

4.2 Creation of Low Access Tracts

After QA/QC, 84 supermarkets were identified within

in the study area. From these 84 supermarkets, GIS

calculations were performed on the data using the

same methodology as the USDA Food Access

calculations. The methodology used was 1) the study

area is divided into ½ kilometer square grids and then

2) the distance to the nearest supermarket is measured

from the center of the grid to the center of the grid

with the nearest supermarket. The distances were

then grouped at the census tract level which contains

estimates on population. This was done using the

Revisiting Food Deserts in North Carolina, USA, Using a Cloud-Based Real-Time Quality Assurance/Quality Control (QA/QC) Tool

117

Create Fishnet function to create 37,816 grids within

the study area. The Near function was used to

calculate the distance between grid centroids and the

center of the grid within the nearest supermarket.

Lastly, the Spatial Join function was used to group

grid centroids with the calculated distance within

each census tract. Urban tracts whose average

distance was more than 1 mile was calculated as LA

while rural tracts whose distance was more than 10

miles was denoted as LA. Those tracts that are now

denoted as LA were compared to the existing LI

Tracts from the USDA tract-level data to delineate

new LILA tracts.

4.3 Comparison of USDA Map and

Newly Created Low Access Map

LA census tracts according to the USDA (Figure 2)

and then using the new calculations after QA/QC

(Figure 3) were created. Maps of LILA tracts

according to the USDA Food Access Atlas and their

QA/QCed counterparts were also created.

When compared visually, they have tremendous

aesthetic value, but little computational value. In

response, the Jaccard Index or Jaccard Similarity

Index is a statistic for gauging the similarity and

diversity of sample sets. The Jaccard Index has been

traditionally used in object detection in digital images

and even raster GIS data. In this research, this metric

is useful since LA and LILA are Boolean values (1 or

0) instead of continuous numeric values where

regression or other statistical measures could be used.

It measures the intersection (values that are common

between two different methods) when compared to

the union (all values between different methods) for

all 90 census tracts within the study area. The Jaccard

Index ranges between 0 (complete dissimilarity) to 1

(complete similarity). Another test for similarity that

can be computed within the confines of a GIS is

McNemar’s test, which creates a χ

statistic and

accompanying p-value for statistical significance on

paired nominal data, in this case true (1) and false (0)

values created for Low Access and LILA between the

before and after QA/QC datasets. It expands upon the

Jaccard Index by breaking down the indiviudal

complements (tracts that do not intersect) from the

Jaccard Index calculation and uses a contingency

table to determine where two attributes/maps for the

same group of enumeration units disagree with each

other with statistical significance

igure 2: Low Access

Tracts as per USDA Food

ccess Atlas.

Figure 3: Low Acces Tract

after QA/QC.

Practically applied, the visualization of these

changes can be articulated through a drive-time map

created using data from before QA/QC and after

QA/QC. While it is difficult to determine which

points were used in the determination of the USDA’s

Food Access database, the before and after

supermarket stores taken from DataAxle data were

utilized using the Network Analyst tool’s function of

Service Area to create a polygon representing a 10-

minute drive-time from supermarkets and then

compared based on census block group and block

level data taken from the census. Not only can these

drive-time maps be visualized, but the impacted

populations calculated while summarizing the types

of errors taken.

5 RESULTS

Table 3 highlights a summary of both the number of

census tracts and population considered to be LA and

LILA from before QA/QC (Using USDA Food

Access Atlas) and after QA/QC using data checked in

the field. While the number of census tracts impacted

remain almost the same, the reconfiguration of these

census tracts highlights a 2.71% decrease in LA

populations between USDA Food Access values and

those calculated after QA/QC. Furthermore, the

population denoted as LILA according to the USDA

Food Access Atlas is 11.84% more than LILA

counterparts after QA/QC. Overall, given the

decreased number of supermarkets found in the field

after QA/QC (84) versus the original number of

supermarkets (104), there is a general overestimation

of food needy (both Low Access and LILA) regions

(except for LILA urban) using the USDA Food

Access when compared to data after QA/QC at this

scale.

GISTAM 2023 - 9th International Conference on Geographical Information Systems Theory, Applications and Management

118

Table 3: Summary of Results from Before and After QA/QC.

Before QA/QC After QA/QC

Urban

ural Total Urban

ural Total

# Low Access Tracts 29 6 35 29 6 35

# LILA Tracts 14 4 18 15 1 16

ulation of Low Access Tracts 107,336 37,360 144,696 132,136 24,920 157,056

ulation of LILA Tracts 67,331 18,581 85,912 70,484 5,547 76,031

5.1 Jaccard Index

As applied to the USDA LA tracts against the newly-

created LA tracts using the methods described above

results in a Jaccard Index of .867. In the 12 cases of

disagreement between the two sets, six were the result

of previous LA regions that were no longer Low

Access after QA/QC. The other six were denoted as

Low Access after QA/QC after not being identified as

Low Access in the original USDA data. Calculating

the Jaccard Index for LILA results in a value of .933.

In cases of disagreement, two census tracts not

identified as LILA in USDA data were denoted as

LILA after QA/QC while four census tracts lost their

status of LILA after QA/QC.

5.2 McNemar’s Test

McNemar’s test highlighted 12 disagreements from

before QA/QC. Two separate McNemar’ tests were

run on the Low Access and LILA variables. Tests of

statistical significance calculate a χ

statistics as the

probability of the each outcome occuring independent

of each other through its discordants. With a χ

statistic value of .083 and p-value of .772, there is not

enough evidence to support a difference in marginal

probabilities for LA between the original data and

QA/QCed data. For LILA, the χ

statistic value is

.167 resulting in a p-value of .683. As a result, there

is not enough evidence to show significant

differences in the number and probability of LA and

LILA regions within the study area before and after

QA/QC.

5.3 Drive Time Map

Drive-time maps visualize the practical challenges of

accessing healthy food and providing an overall

complexion of the food environment understandable

to all level of users. Using Esri’s Network Analyst

tool, a 10-minute drive time was calculated around

the 104 stores (Figure 4) that existed in the original

database and then the 84 stores resulting after QA/QC

as shown in Figure 5.

To increase granularity, block group level data

igure 4: 10-minute drive

ime to supermarkets of data

efore QA/QC.

Figure 5: 10-minute driv

time to supermarkets of dat

after QA/QC.

were agglomerated from the 267 block groups and

11,138 blocks composing the 90 census tracts within

the study area. Using the Select by Location and

Statistics tools, information is highlighted about the

populations within 1 mile, 5 miles and the 10-minute

drivetime from supermarkets before and after

QA/QC. As highlighted in the results in Table 4, the

population calculated to be within a 5-mile distance

of supermarkets using non-QA/QCed data is

approximately 7.0% more than its QA/QCed

counterparts. The difference for a 10-minute drive

time map (Figure 5) represents a 7.22%

Table 4: Summary of various buffers and drive-times before

and after QA/QC of supermarkets.

Before

QA/QC

After

QA/QC

# of su

ermarkets 104 84

Population within 1 mile of

supermarket*

219,944 195,380

Population within 5 miles of

supermarket*

372,051 347,713

Population outside 5 miles of

supermarket*

39,697 64,035

Population within 10-minute

drive of su

ermarket*

384,373 358,481

Population outside 10-minute

drive of supermarket*

27,375 53,267

* Based on block level data

Revisiting Food Deserts in North Carolina, USA, Using a Cloud-Based Real-Time Quality Assurance/Quality Control (QA/QC) Tool

119

overestimation of non-QA/QCed data versus its

QA/QCed counterparts. As a result, more than 26,000

people within the study are who are estimated to be

living within a 10-minute drive of a supermarket

using one set of data who do not live within this

threshold using field-checked data. CAB data grossly

overestimates food-secure populations and

underestimates the number of people living in food-

needy regions by almost half (27,375 vs. 53,267)

based on supermarket data that exists in the field.

6 DISCUSSION

While this study is meant to estimate the food

environment and simulate those methods from the

USDA Food Access Atlas to create comparative

statistics through the lens of supermarkets,

supermarkets do not represent the entire food

environment. While food can be found in such

disparate places such as restaurants, laundromats and

home improvement stores, stores such as Dollar

General, not represented in supermarkets stores, are

gaining a foothold in areas overlooked by major

supermarkets and grocery stores. Many of these

Dollar Generals stores provide staples such as

vegetables, fruits, milks and eggs that are indicative

of supermarkets and grocery stores and a healthy food

environment.

Between 2009 and 2021, just the number of

Dollar General stores have more than doubled (17 to

37) in the study area and 12 out of the 35 census tracts

denoted as LILA within study area contain a Dollar

General. Future food environment studies should

include stores such as Dollar General which provide

alternatives to supermarkets and smaller grocery

stores that are also affordable.

igure 6: Healthy and fresh food offerings in Dolla

General store within study area.

Besides the McNemar’s Test, this research

highlighted differences in the represented and real

food environments using maps and descriptive

statistics. More robust statistics with statistical

significance such as those using a two-tail t-test

exploring differences in socio-economics across

LILA regions using CAB data versus ground-truthed

data (for example, exploring median household

income in LA regions from the USDA Food Access

Atlas via Figure 2 versus the median household

income in Low Incomes from data extracted from this

research via Figure 3) may better reinforce the need

for ground-truthed data. Other research (Real and

Vargas 1996) has explored the conversion of the

Jaccard Index to p-values. However, these topics are

a subject for future research.

Data were analyzed at the census tract level

because data provided by the USDA Food Access

Atlas is provided at that scale. While these tract-level

LI and LILA designators can be applied to the census

block groups that lie within them, LI and LILA be

calculated using the USDA methodology from

QA/QCed data, accumulating statistics or making

comparisons using socio-economic information at the

block group level can be problematic because of the

reliability of data. Socio-economic data are collected

through the American Community Survey (ACS).

Within ACS data, three classes of reliability exist:

High, Medium and Low. In general, reliability of data

collected at the census tract level is much better than

counterparts at the block group level.

Included in these data is a flag (1 = yes, 0 = no)

to denote if a census tract is urban. This flag can be

problematic because census tracts that are not urban

should not be automatically considered rural although

they are applied this way in this research. There is a

continuum of urban to rural and many agree that there

is no single definition of rural that best encapsulates

the concept of rural across various applications, needs

and scales (Nelson et al., 2021; Coburn et al., 2007).

There are up to nine different definitions of the term

rural used by the U.S. federal agencies. With the

variety of quantitative definitions, the important

questions arise on the consistency of the major

operational definitions of rural and the practical

implications of the differences in identifying rural

populations based on alternative, commonly used

quantitative criteria for rurality highlighted in this

research. One recent study by the research team

(Mulrooney et al. 2023) showed the application of the

term rural utilizing the USDA RUCA (Rural-Urban

Commuting Area) best aligns with other definitions

of rural, and future applications of these data should

somehow incorporate this application with existing

Food Desert Atlas data.

7 CONCLUSIONS

Deterministic data models can model the food

environment given well-understood rules, parameters

and data. In this study, low access (LA), low income

GISTAM 2023 - 9th International Conference on Geographical Information Systems Theory, Applications and Management

120

(LI), low income and low access (LILA) can be

extracted from existing data via the USDA Food

Access Atlas based on access to supermarkets as part

of a larger study on food deserts. However, little

work has been studied to understand the accuracy of

supermarket data on which this low access is based

and how this accuracy is manifested in changed or

compromised food environments based on input data

assumed to be correct and data which have been field

checked. The assessment of these data has been

easier with custom phone applications that can access

data stored in the cloud to inspect, verify, edit and re-

attribute the spatial data used to represent

supermarkets and the larger food environment in

general. In this study, we utilized real-time QA/QC

procedures merging hand-held phone applications

and cloud data to 1) explore errors of omission and

commission for Commercially Available Business

(CAB) Databases and their counterparts QA/QCed in

the real world 2) measure the differences in the CAB

database and data after QA/QC and 3) explore the

spatial differences in the food environment as a result

of the differences in these two sets of data.

In this study, supermarket data extracted from

DataAxle were checked in the field to explore errors

of omission and commission. Based on the QA/QCed

data, new Low Access (Figure 6) and LILA maps

were created based on the methodology to create

these data at the census tract level and compared to

the original USDA Food Access Atlas (Table 3). At

the census tract scale, results highlight a general

overestimation of food needy populations when

compared to data calculated using supermarkets

currently in the field, but even greater

overestimations of rural food needy populations

(18,581 estimated using USDA Food Access Atlas vs.

5,547 using QA/QCed data). Jaccard Indices for both

Low Income (.867) and LILA (.933) also indicate

general agreement between the two sets of data, as

well as the McNemar’s Test which highlight there is

not enough evidence to show significant differences

in the number and probability of Low Access and

LILA regions within the study area before and after

QA/QC.

Probably most accentuated were drive-time maps

and accompanying tables comparing the CAB data

versus QA/QCed counterparts through the mapped

food environment. Most obvious in these maps are

differences in southern Alamance and Caswell

Counties, as well as southeastern Chatham County,

which indicated compromised food environments

after QA/QC. DataAxle data had indicated these rural

regions did in fact contain supermarkets and grocery

stores while QA/QC unearthed the contrary.

In summary, this research has higlighted the

following:

• Phone applications such as Esri Field Maps or

Survey123 are relatively easy to create and allow

for real-time attribution/reattribution and

creation of cloud-based data that can be analyzed

in the field and can easily be integrated into

applications such as utility mapping and

inspections.

• QA/QC procedures found 20 less supermarkets

in the study area after QA/QC (84) compared to

the data provided in the CAB (104). Reasons for

these differences included 1) the business was

not a supermarket 2) the point in the CAB was

actually a residential address 3) the food source

in the CAB was permanently closed and 4) the

point did not exist in the CAB database,

highlighting an error or omission.

• The one error of omission occurred in Chatham

County in the town of Pittsboro. However, it was

located close to other grocery stores and did not

impact the overall food environment.

• While the food environments before and QA/QC

generally agreed with each other statistically,

there appeared to be an overestimation of food

accessible populations (i.e., an underestimation

of food needy populations) using CAB data

compared to its QA/QCed counterparts.

• Major differences in the food environment were

found in rural areas in southern Alamance and

Caswell Counties, as well as southeastern

Chatham County due to supermarkets that were

found not to exist after QA/QC.

With the interoperability and relative ease of

powerful desktop applications and cloud-based data

that can be updated in real-time, on-the-fly food

environment maps can be created using the latest and

most updated data from the field. These maps can

guide policy and facilitate decisions regarding those

who are represented as food needy through

applications such as the USDA Food Access Atlas

versus those who truly food needy based on real-time

data extracted through the applications and analysis

as part of this research.

ACKNOWLEDGEMENTS

The project was supported by the Agricultural and

Food Research Initiative Competitive Program of the

USDA National Institute of Food and Agriculture

(NIFA), grant number 2021-67021-34152. This

material is also supported by the National Science

Foundation under Grants No. 1824949 and No.

Revisiting Food Deserts in North Carolina, USA, Using a Cloud-Based Real-Time Quality Assurance/Quality Control (QA/QC) Tool

121

2226312. Any opinions, findings, and conclusions or

recommendations expressed in this material are those

of the author(s) and do not necessarily reflect the

views of the NSF and USDA.

REFERENCES

Auchincloss, A., Moore, K., Moore, L. and Diez-Roux, A.

(2012). Improving retrospective characterization of the

food environment for a larger region in the United State

during a historic time period. Health and Place, 18(6),

1341 – 1347.

Coburn, A. F., MacKinney, A. C., McBride, T. D., Mueller,

K. J., Slifkin, R. T., Wakefield, M. K. (2007). Choosing

rural definitions: Implications for health policy. Rural

Policy Research Institute Health Panel (Issue Brief #2).

Felgueiras, Carlos Alberto et al. (2017). Modeling and

visualization of uncertainties of categorical spatial data

using geostatistics, 3D planar projections and color

fusion techniques. Proceedings of GEOINFO 2017,

152 – 162.

Han, E., Powell, L., Zenk, S., Rimkus, L, Ohri-Vachaspati,

P. and Chaloupka, F. (2012) Classification bias in

commercial business lists for retail food stores in the

U.S. International Journal of Behavioral Nutrition and

Physical Activity, 46, https://doi.org/10.1186/1479-

5868-9-46.

Hosler, A. and Dharssi, A. (2010). Identifying retail food

stores to evaluate the food environment. American

Journal of Preventive Medicine, 39(1), 41 – 44.

Heuvelink, G.B.M. (1998). Error propagation in

environmental modeling with GIS. New York, NY:

Taylor and Francis.

Lake A.A., Burgoine T., Stamp E. and Grieve, R. (2012).

The foodscape: classification and field validation of

secondary data sources across urban/rural and socio-

economic classifications in England. International

Journal of Behavioral Nutrition and Physical Activity,

9(1), 37.

Liese, A., Clabianchi, N., Lamichhane, A, Barnes, T.

Hibbert, J., Porter, D., Nichols, M. and Lawson, A.

(2010). Validation of 3 food outlet databases:

completeness and geospatial accuracy in rural and

urban food environments. American Journal of

Epidemiology, 172(11), 1324 – 1333.

Mendez, D., Kim, K., Hardaway, C. and Fabio, A. (2016).

Neighborhood racial and socioeconomic disparities in

the food and alcohol environment: Are there difference

by commercial data sources? Journal of Racial and

Ethnic Health Disparities, 3, 108 – 116.

Mulrooney, T., Liang, K., Kurkalova, L., McGinn, C. and

Okoli, C. (2023). Quantitatively Defining and

Mapping Rural: A Case Study of North Carolina.

Journal of Rural Studies, 97, 47 – 56.

Nelson, K.S., Nguyen, T.D., Brownstein, N.A., Garcia, D.,

Walker, H.C., Watson, J.T. and Xin, A. (2021).

Definitions, measures, and uses of rurality: A

systematic review of the empirical and quantitative

literature. Journal of Rural Studies, 82, 351-365.

Newcomer, J.A. and Szajgin, J. (1984). Accumulation of

thematic map errors in digital overlay analysis. The

American Cartographer, 11(1), 58 – 62.

Real, R. and Vargas, J.M. (1996). The Probabilistic Basis

of Jaccard’s Index of Similarity. Systematic Biology,

45, 380-385.

Rummo, P., Albrecht, S., & Gordon-Larsen, P. (2015).

Field validation of food outlet databases: the Latino

food environment in North Carolina, USA. Public

Health Nutrition, 18(6), 977 – 982.

Sharkey, J.R., Horel, S., Han, D. and Huber, D. (2009)

Association between neighborhood need and spatial

access to food stores and fast food restaurants in

neighborhoods of Colonias. International Journal of

Health Geography, 8(9), 1 - 17.

Vilme, H., Paul, C.J., Duke, N.N., Campbell, S.D., Sauls,

D., Muirur, C., Skinn, A.C., Boswort, H., Dokurugu,

Y.M, and Fay, J.P.(2020). Using geographic

information systems to characterize food environments

around historically black colleges and universities:

Implications for nutrition interventions. Journal of

American College Health, 70(3), 818-823.

GISTAM 2023 - 9th International Conference on Geographical Information Systems Theory, Applications and Management

122