Determining the Impacts of COVID-19 on Urban Deliveries in the Metropolitan Region of Belo Horizonte Using Spatial Analysis

Abstract

The COVID-19 pandemic has changed lifestyles, with consequent impacts on urban freight movements. This paper analyzes the impacts of COVID-19 on urban deliveries in the Belo Horizonte Metropolitan Region, Brazil. The Lee index and the Local Indicator of Spatial Association were calculated using data on urban deliveries (retail and home deliveries) and COVID-19 cases. The results confirmed the negative impacts on retail deliveries and the positive impacts on home deliveries. The spatial analysis demonstrated that the most interconnected cities presented more similar patterns. At the beginning of the pandemic, consumers were considerably concerned about the virus spread, and the changes in consumption behavior were slow. The findings suggest the importance of alternative strategies to traditional retail. In addition, the local infrastructure should adapt to the increased demand for home deliveries during pandemics.

The COVID-19 pandemic has changed people’s lifestyles ( 1 ) and also affected urban freight movements. In Latin America, the first COVID-19 case was confirmed on February 26, 2020, in São Paulo, Brazil. Brazilian cities then followed other countries’ examples to reduce the virus spread, for example, closing non-essential services and advising city residents to stay home. The lockdown measures shifted consumer behavior by affecting non-essential traditional retail, the supply chain, and urban freight transport. The initial motivation of this paper was to understand the impacts of closing non-essential shops on urban deliveries during the first epidemiological weeks (from March 15 until April 30, 2020) of the COVID-19 pandemic in Brazil.

Epidemics disrupt the urban freight distribution system ( 2 ). Furthermore, it is challenging to keep the system working during epidemics because of the magnitude and the complexity of the related problems ( 3 ). For example, a pandemic could create food shortages in the U.S. food system as it may reduce labor availability by 25% ( 3 ). The demand for freight transportation also decreased as a result of the COVID-19 outbreak in Australia. The outbreak reduced the revenues of the transport industry, including the air, freight, and public transport sectors ( 4 ). Moreover, epidemics can disrupt consumer behavior and affect the number of urban deliveries ( 5 ). However, most of the literature related to urban freight transport (UFT) disruption focuses on the impacts of natural disasters, such as urban flooding ( 6 ). Several authors reported that e-commerce ( 7 – 9 ) and freight operations ( 10 ) increased because of the COVID-19 pandemic. However, the UFT literature on pandemics is still limited ( 4 ) and has two main goals: (i) measuring the impacts of pandemics on UFT and (ii) proposing UFT solutions during pandemics.

Different studies evaluated the impacts of pandemics on the UFT system. For example, the Middle East Respiratory Syndrome (MERS) outbreak changed consumer expenditures in South Korea. A study showed that psychological factors and the fear of contracting MERS altered consumers’ behavior ( 5 ). In the U.S., households changed their spending patterns as the media released more news about the impacts of COVID-19 in their living areas ( 11 ). As a result, households spent more on home goods and decreased their spending on restaurants, retail, air travel, and public transport ( 11 ). Heterogeneity was found in spending patterns depending on demographic characteristics but there were similarities in relation to household income ( 11 ). In Spain, in the Madrid Central Area, daily e-commerce parcels doubled during the COVID-19 pandemic, which increased the size of the light commercial fleet needed to deliver them ( 9 ). In Brazil, the COVID-19 pandemic negatively affected the transportation sector. It reduced delivery operations in urban areas because of local travel restrictions and lockdown measures ( 12 ). In France, the COVID-19 pandemic affected the urban distribution by changing in-store shopping to e-commerce. This occurred mainly for non-food orders, but other logistic distribution issues were also observed ( 13 ). The pharmaceutical and the food sectors were assessed in relation to the inequality of online home deliveries for residents on the west coast of Sweden during the COVID-19 pandemic ( 14 ). Results showed that geographic, demographic, and health-related reasons contributed to increasing the marginalized share of the population as a result of the pandemic ( 14 ). During the COVID-19 pandemic in India, the type of goods, routing efficiency, and fulfillment timeline were the major factors that affected last-mile deliveries from e-commerce, the food sector, and the retail sector ( 15 ).

Other studies proposed UFT solutions. For instance, mobile warehouses were viewed as a last-mile logistics solution during COVID-19 ( 16 ). Furthermore, with the UFT disruptions and challenges in public transportation, integrating freight in transit networks within a “Physical Internet” could be a solution during pandemics ( 17 ). The COVID-19 pandemic could also stimulate circular economy strategies for reverse logistics ( 18 ). The City of Montreal, Canada, implemented an on-demand delivery service initiative using cargo bikes. The system delivered goods from closed businesses to nearby consumers during the first wave of the COVID-19 pandemic ( 19 ). The study found that population density, the number of web business transactions, and COVID-19 cases contributed to the success of the initiative ( 19 ). In addition, in a medium-sized Brazilian city, consumers are willing to adopt new delivery services in a post-pandemic world ( 20 ).

Despite these studies, some critical COVID-19 freight transportation issues reported by the Transportation Research Board Executive Committee ( 21 ) still need to be addressed. Some of these topics include the freight activity indirectly, concerning the impacts of technology and automation on freight activities ( 21 ). This report has an entire section related to goods movements, which includes (i) the decline in cargo movements, (ii) the high dependency of the U.S. inventory on imports from Asian countries, (iii) the distribution of vaccines and medical supplies to the most remote places of the country, and (iv) the increase in online shopping and home deliveries that directly affect the local infrastructure ( 21 ). These concerns are also relevant in the Brazilian scenario.

Based on these critical issues and motivated by the changes in consumer behavior, the increase in home deliveries, and the shrinkage of retail deliveries, this paper analyzed the impacts of the COVID-19 pandemic on urban deliveries. The analysis was conducted for the Belo Horizonte Metropolitan Region (BHMR), Brazil. The following research questions were proposed: Is the increase in home deliveries related to the COVID-19 pandemic? Is the increase in retail deliveries related to the reduction in home deliveries? Is the increase in home deliveries related to lockdown measures? To answer these questions, retail delivery and home delivery data from a delivery company were used. The analysis period consisted of data from January to April 2020, and the data were analyzed using spatial analysis.

The results indicated a reduction in retail deliveries because of COVID-19 cases and lockdown measures. In contrast, home deliveries, which represent e-commerce for non-essential goods, increased in the first epidemiological weeks. Furthermore, a negative correlation was observed between retail and home deliveries in the seventh epidemiological week. These results are presented in detail in the following sections. The findings are consistent with Jung et al. ( 5 ), who found that retail deliveries decreased during the MERS outbreak, whereas home deliveries increased. This finding is relevant and shows the importance of new strategies alternative to traditional retail during pandemics. Implementing green logistics strategies could reduce the number of home deliveries and, consequently, last-mile distribution costs. In this new scenario, contactless delivery lockers could be a potential delivery solution for larger Brazilian cities.

Study Area and Data

The BHMR has 4.58 million inhabitants. It is the third-largest metropolitan area in Brazil and the seventh-largest in Latin America. The BHMR has 34 municipalities, which are distributed across 9,460 km². Within the region, 47% of the population live in Belo Horizonte, which is the capital of the State of Minas Gerais and the most important city in the BHMR. The 34 BHMR municipalities are illustrated in Figure 1. The municipalities are considerably different from each other since the BHMR was created mainly under political interests, and limited technical metrics or physical integration considerations were taken into account in the process. The core BHMR municipalities (i.e., Belo Horizonte, Betim, Contagem, Nova Lima, Ribeirão das Neves, and Santa Luzia) are better integrated and have considerably higher populational densities, average incomes, and gross domestic products (GDP). The city of Belo Horizonte was conceived to be the state’s administrative center and it currently represents the central service hub of Minas Gerais. Betim and Contagem have been consolidated as the industrial hub of this metropolitan region. Nova Lima is a historical city, and it has been associated with mining exploitation since the colonial era. Ribeirão das Neves and Santa Luzia are commuter towns and have industrial hubs with a relevant size. These municipalities combined concentrate most of the BHMR population and income, and their transportation network is integrated considering both road and railroad infrastructures. Conversely, the municipalities located on the border (Baldim, Capim Branco, Florestal, Itaguara, Itatiaiuçu, Jaboticatubas, Nova União, Raposos, Rio Acima, and Rio Manso) were added to the BHMR to receive financial incentives. Their integration with the other municipalities is poor even in road and railroad infrastructure. These cities have a predominantly agricultural economy, which makes them different from the others in the BHMR. The lack of transportation integration is shown in Figure 2, which depicts the regional transportation within the BHMR ( 22 ). These different characteristics (e.g., the differences in residents’ lifestyles, the economic activities, and the integration between the cities) directly affected how the COVID-19 virus spread in the first epidemiological weeks and how each city reacted to slow down the virus.

Figure 1.

Belo Horizonte Metropolitan Region (BHMR).

Figure 2.

Transportation infrastructure in the Belo Horizonte Metropolitan Region (BHMR).

The first COVID-19 case in the BHMR was reported on March 16, 2020. This day was considered to determine the COVID-19 epidemiological weeks: the first epidemiological week (EW) started on March 15, 2020. Figure 3 shows the number of new cases in the first seven EWs in the BHMR (i.e., from March 15 until April 30, 2020). The first cases were reported in Belo Horizonte, and they spread to the neighboring cities over time. Until the end of April, the official weekly statistics for COVID-19 were reliable. However, with changes in the Brazilian Ministry of Health, discrepancies were observed in the weekly data.

Figure 3.

COVID-19 cases in the first seven epidemiological weeks in Belo Horizonte Metropolitan Region (BHMR).

For this reason, this study is limited to the first seven EWs (Table 1). The study focuses on COVID-19 cases since the first COVID-19-related death was reported only on March 29, 2020, and 25 deaths were registered in the BHMR until the end of April. Therefore, COVID-19-related deaths were not considered in the analysis. No more than 110 new cases were registered in a single EW, and many cities did not even present new cases during the first seven EWs. The data demonstrated the efficiency of containing the COVID-19 spread in this specific region during the analysis period, as only six out of the 34 municipalities presented new cases of COVID-19. The BHMR is not well connected in relation to transportation infrastructure. For example, Ribeirão das Neves can be considered a suburb of Belo Horizonte, but no cases were reported in this city during these weeks. The virus spread mainly within the most connected municipalities, which present higher transportation flows between them.

Table 1.

Calendar Dates of Epidemiological Weeks (EW) (2020)

Epidemiological week	From	To
EW1	March 16	March 22
EW2	March 23	March 29
EW3	March 30	April 05
EW4	April 06	April 12
EW5	April 13	April 19
EW6	April 20	April 26
EW7	April 27	April 30

With the increase in COVID-19 cases and the first reported deaths in Brazil, many cities implemented lockdown measures to contain the spread of the virus. The lockdown measures had the most significant impact on non-essential services, which were closed. Commercial establishments that sold food, beverages, and medicines remained open. Figure 4 shows the duration of lockdown in the BHMR. Most municipalities introduced lockdowns during EW2 and EW3. Since the cases and deaths did not follow the increasing trend that was observed internationally (at that time, some countries were experiencing their first peak in cases and deaths), the duration of the lockdown measures was relatively short, except for Belo Horizonte and Nova Lima, where the lockdown was longer. Moreover, these municipalities, especially Belo Horizonte, adopted a more conservative approach to dealing with COVID-19. The local government implemented the lockdown when the first cases were reported and decided to reopen only if health metrics allowed. In the case of Belo Horizonte, this first lockdown lasted until the end of May 2020 ( 23 ). The municipalities with lower integration to Belo Horizonte and in the outer limits of the BHMR (e.g., Baldim, Itaguara, Itatiuiçu, and Rio Manso) did not even implement lockdowns during these weeks because they did not report any cases in this period. In addition, the lockdowns lasted for just a couple of weeks, which is consistent with the rest of the country: small lockdowns were implemented, and the population was resistant to restrictive measures.

Figure 4.

Duration of lockdown measures in the Belo Horizonte Metropolitan Region (BHMR).

Figure 5 shows the evolution of urban deliveries in the BHMR. Data from one delivery company that operates in the region were used. The biggest cities concentrate most of the urban deliveries. The number of urban deliveries was similar in January and February, where varying patterns were observed among the cities. February, March, and April presented similar patterns among the cities; however, urban deliveries (i.e., retail deliveries and home deliveries) increased by 7%. Retail deliveries did not include non-essential goods. Home deliveries represent most of the urban deliveries: 75% in January, 76% in February, 77% in March, and 86% in April.

Figure 5.

Evolution of total deliveries by one delivery company in the Belo Horizonte Metropolitan Region (BHMR).

Figures 6 and 7 present the evolution of retail deliveries and home deliveries, respectively, in the first seven EWs. Figures 8 and 9 present the evolution in relative values, that is, positive values indicate growth and negative values indicate decrease. It should be noted that the company does not deliver to some cities on the edge of the BHMR, which are represented by the lighter color. EW1 had the highest number of retail deliveries. Retail deliveries between EW1 and EW2 decreased by 41%, and by 38% between EW1 and EW7. EW6 had the lowest number of retail deliveries (a 50% reduction in comparison with EW1). For home deliveries, EW3 and EW6 presented a peak in Belo Horizonte, and a similar pattern was observed in the other cities. Home deliveries increased by 13% between EW1 and EW3 and increased by 7% between EW1 and EW7. EW3 presented the peak of home deliveries, as the data showed 50% more home deliveries in comparison with EW1.

Figure 6.

Retail deliveries by one delivery company for the first seven epidemiological weeks in the Belo Horizonte Metropolitan Region (BHMR).

Figure 7.

Home deliveries by one delivery company for the first seven epidemiological weeks in the Belo Horizonte Metropolitan Region (BHMR).

Figure 8.

Evolution of the relative retail deliveries for the first seven epidemiological weeks in the Belo Horizonte Metropolitan Region (BHMR).

Figure 9.

Evolution of relative home deliveries for the first seven epidemiological weeks in the Belo Horizonte Metropolitan Region (BHMR).

From this brief overview of the data, the following hypotheses were proposed: (i) there is a correlation between COVID-19 cases and urban deliveries, (ii) there is a correlation between COVID-19 cases and lockdown measures, (iii) both retail deliveries and home deliveries had a similar pattern of reduction in the first weeks of the COVID-19 pandemic, and (iv) the first weeks of the COVID-19 pandemic were followed by a reduction in urban delivery activities. These hypotheses were addressed with the research method hereafter.

Research Method

This paper investigates the association between COVID-19 and urban deliveries in the BHMR using local spatial statistics. The data contain spatial information and attributes to produce a local statistic and local parameters, which allows estimation of their variations and patterns across space. The use of local statistics eliminates false assumptions from global statistics and makes it possible to understand the differences between heterogeneous cities ( 11 ). The effects of COVID-19 on retail deliveries and home deliveries were investigated using two spatial methods: the Lee index (L) and the Local Indicator of Spatial Association (LISA). The Lee index measured the correlation between COVID-19 cases and urban deliveries. LISA, which is based on the bivariate Moran index ( $I_i^B$ ), identified clusters between home and retail deliveries. Both spatial techniques were applied to the data for each EW.

The Lee Index (L)

The Lee index (L) is a local spatial Pearson index that captures the degree of the numerical correspondence between two values at a location. This index multiplies two z-scores by their means and standard deviations. The local Lee index is an association between the Moran’s (I) statistic and the Pearson’s correlation coefficient (r). The Lee index is obtained using Equation 1:

$$L_{X,Y}=\frac{\sum _i({\tilde{x}}_i-\overline{x})({\tilde{y}}_i-\overline{y})}{\sqrt{\sum _i{(x_i-\overline{x})}^2}{}^{*}\sqrt{\sum _i{(y_i-\overline{y})}^2}}$$ (1)

where

${\tilde{x}}_i$ and ${\tilde{y}}_i$ = the elements for location $i$ in X and Y in the spatially lagged vectors,

$\overline{x}$ and $\overline{y}$ = the mean values of the attribute,

$x_i$ and $y_i$ = the attribute values at location $i$ ( 12 ).

The Lee index was calculated using the spdep package ( 13 ) in the R environment. This index is complementary to the Pearson’s correlation coefficient r, and it has the power to capture the spatial dependency of the dataset ( 12 ). The interpretation of the Lee index is similar to the Pearson’s correlation coefficient: negative values indicate negative correlation, and positive values indicate positive correlation.

LISA based on the Bivariate Moran Index

The local Moran index is an inferential spatial statistic to calculate the local spatial correlation. The main applications are cluster identification and tracing spatial outliers ( 14 ). The bivariate local Moran index is a local statistic that measures the degree to which one variable x is correlated with a spatially lagged (i.e., average value at a nearby location) different variable y ( 15 ). Equation 2 shows the general form of the bivariate local Moran index ( $I_i^B)$ .

$$I_i^B=c{x}_i\sum _j{w}_{\mathrm{ij}}{y}_j$$ (2)

where

$w_{\mathrm{ij}}$ = the spatial weights matrix,

$x_i$ = the variable at location $i$ ,

$\sum _j{w}_{\mathrm{ij}}{y}_j$ = the spatial lag of the other variable $y_j$ ,

$c$ = a constant estimate of the variance.

LISA was developed by Srinivas and Marathe ( 16 ) based on this indicator. It is used to identify contiguous location sets in which the LISA has statistical significance. The patterns are identified by four types of clusters: High-High, High-Low, Low-High, or Low-Low. The clusters follow the rules presented in Table 2. Clusters with no significance level are not mapped. The mathematical formulation of this method, along with a detailed example, can be found in the paper by Srinivas and Marathe ( 16 ).

Table 2.

Local Indicator of Spatial Association (LISA) Cluster Description

Cluster type	Region characteristic	Nearby region characteristic
High-High	High value	High value
High-Low	High value	Low value
Low-High	Low value	High value
Low-Low	Low value	Low value

The Moran values were calculated using the spdep package ( 13 ) in the R environment. The cluster quartiles were built with a 95% significance interval, and the weight matrix was built using the cities as the spatial units and the queen contiguity (i.e., sharing a border or a point). Only the clusters with p-values lower than 0.05 were presented in the results.

Two types of LISA clusters based on the bivariate Moran index were applied. The first was a bivariate analysis, where the variables “home deliveries” (HD) and “retail deliveries” (RD) were used. For the LISA between HD and lagged RD, the High-High cluster indicates the neighbors where the HD and the overall RD are high, so the adjacent neighbors are high. The Low-Low cluster has the inverse interpretation. The relationship between HD and overall RD is low, so the adjacent neighbors are low. The High-Low cluster indicates that the relationship between HD and overall RD is high, so the adjacent neighbors are low. Finally, the Low-High cluster shows that the relationship between HD and overall RD is low, so the adjacent neighbors are high. The same logic can be applied for the LISA between RD and lagged HD. This step is related to the third hypothesis of this study.

The second bivariate LISA cluster was a time analysis. Two situations were considered: HD at time t₁ and lagged HD at time t₀, and RD at time t₁ and lagged RD at time t₀. This type of cluster enables us to observe the time dependency of any increase in HD and RD separately, and it shows a spatial dependency. The High-High cluster indicates an increase over time in the neighbor and the adjacent neighbors. The Low-Low cluster indicates a reduction over time in the neighbor and the adjacent neighbors. The High-Low cluster indicates an increase over time in the neighbor and a decrease in the adjacent neighbors. The Low-High cluster indicates a reduction over time in the neighbor and an increase in the adjacent neighbors. This step is related to the fourth hypothesis of this study.

Results

Figure 10 presents the spatial correlation between COVID-19 cases and retail deliveries with the Lee index. A negative correlation is observed in some cities, such as Belo Horizonte, and a strong positive correlation is observed in its neighbors, such as Betim and Ribeirão das Neves. Similar patterns were observed in all EWs. Thus, COVID-19 cases influenced retail deliveries during the timeline considered in the analysis.

Figure 10.

Spatial correlation between COVID-19 cases and retail deliveries.

Figure 11 shows the relationship between lockdown measures and retail deliveries, and no patterns were observed among the EWs. However, EW1 and EW7 had a stronger negative correlation between lockdown measures and retail deliveries when compared with the other epidemiological weeks. Furthermore, especially in EW7, a stronger negative correlation was observed when most cities had no lockdown measures. Thus, the lockdown measures could have increased stock for non-essential shops, while the demand for goods decreased some weeks after the lockdown measures. Therefore, COVID-19 cases and lockdown measures impacted retail deliveries in the BHMR.

Figure 11.

Spatial correlation between lockdown measures and retail deliveries.

Figure 12 shows the spatial correlation between home deliveries and COVID-19 cases. A negative correlation was observed in Belo Horizonte, which is explained by low COVID-19 case counts in the first EWs. However, similar patterns were observed in all EWs. This result shows that COVID-19 cases did not influence home deliveries during the first EWs in the BHMR.

Figure 12.

Spatial correlation between COVID-19 cases and home deliveries.

Similarly, there were no clear patterns for the correlation between lockdown measures and home deliveries over the first EWs, as shown in Figure 13. A negative correlation was observed in almost all municipalities in EW7, that is, when non-essential shops reopened and home deliveries increased. Thus, despite the correlation, the lockdown measures did not influence home deliveries.

Figure 13.

Spatial correlation between lockdown measures and home deliveries.

Figure 14 shows the Lee index between home deliveries and retail deliveries. A positive correlation was observed in most BHMR cities between EW2 and EW5, and the correlation was negative in EW1 and EW7. In EW7, the negative correlation likely resulted from the increase in retail deliveries and from the decrease in home deliveries. That week, the lockdown measures were lifted in most cities. The Lee index between home deliveries and retail deliveries in the first EWs shows the importance of e-commerce for non-essential shops at this period. Therefore, COVID-19 cases influenced retail deliveries and home deliveries of non-essential goods in the BHMR.

Figure 14.

Spatial correlation between retail deliveries and home deliveries.

Figure 15 shows the LISA cluster results for the analysis between home deliveries in neighboring cities and retail deliveries in the city of interest. The results showed that retail deliveries in Belo Horizonte, Contagem, and Ribeirão das Neves had a Low-Low cluster. The delivery pattern in these three cities is influenced by their surroundings, as they all present similar delivery patterns. Ibirité presented a High-Low cluster, which indicates that the relationship between home deliveries and lagged retail deliveries in this city is high, but low in its surroundings. Ibirité has a high pattern for home deliveries while the surrounding municipalities have low influence. Therefore, the delivery pattern in this city is not influenced by its surroundings.

Figure 15.

Local Indicator of Spatial Association (LISA) cluster between home deliveries and lagged retail deliveries.

Figure 16 shows the LISA clusters considering retail deliveries in the cities and lagged home deliveries in the neighboring cities. Low-Low clusters are observed in Belo Horizonte and Contagem from EW1 to EW4 and in Belo Horizonte, Contagem, and Ribeirão das Neves from EW5 to EW7. The Low-Low clusters indicate that the relationship between retail deliveries in each city and lagged home deliveries is low. The same relationship was observed for the neighboring municipalities. Therefore, the proximity between these three cities influences the delivery patterns.

Figure 16.

Local Indicator of Spatial Association (LISA) cluster between retail deliveries and lagged home deliveries.

Additionally, High-Low clusters were identified: Ibirité, Ribeirão das Neves, and Vespasiano from EW1 to EW4 and Ibirité and Vespasiano from EW5 to EW7. The High-Low cluster indicates that these cities’ delivery patterns are not influenced by their surroundings. Therefore, the relationship between their retail deliveries and lagged home deliveries is high, but the surroundings have a low influence. In both LISA analyses, the Low-Low clusters were related to the three cities with higher flows of people, especially employees. Therefore, similar consumption and delivery patterns were expected in these three cities given that the flow of people and goods between these three cities is very intense. The LISA cluster results show a strong connection between these cities. In addition, they indicate that their populations behave similarly even during a pandemic, where lockdown measures were implemented to different degrees by the different municipalities. This cluster of cities shows that city-specific lockdown measures to contain the spread of the virus may not be effective. Thus, at a metropolitan level, lockdown measures should be coordinated between cities, especially those with higher levels of integrated transportation flows. With this cluster analysis, similar reduction patterns were observed across the different weeks, which corroborates the third hypothesis of this study.

Figure 17 shows the time dependency results for RD, which were conducted using LISA based on the bivariate Moran index. The results demonstrated that, while COVID-19 was spreading, the urban retail deliveries reduced, especially for the biggest centers (e.g., Belo Horizonte and Betim). Furthermore, High-Low clusters were identified in Ibirité and Ribeirão das Neves. This indicates that these municipalities increased retail deliveries in the subsequent weeks, and, in contrast, the neighboring cities had a reduction.

Figure 17.

Time-dependent Local Indicator of Spatial Association (LISA) cluster for retail deliveries.

Finally, Figure 18 depicts the time dependency of home deliveries. It shows the tendency of reduction in Belo Horizonte, Betim, Ribeirão das Neves, Ibirité, and Vespasiano during the study period. These findings indicate that these cities continued their main economic activities because the lockdown measures only targeted the service sector.

Figure 18.

Time-dependent Local Indicator of Spatial Association (LISA) cluster for home deliveries.

These time dependency clusters confirmed the fourth hypothesis, that there was an overall reduction in both retail deliveries and home deliveries at the beginning of the COVID-19 pandemic. This result may be related to (i) conservative behavior caused by the uncertainty of not knowing what could happen in the next days or (ii) the prioritization of essential goods that could be bought in supermarkets.

Discussion

Lockdown measures affected non-essential shops, and the data showed that the COVID-19 pandemic indirectly affected e-commerce services related to non-essential goods, which are represented by home deliveries in this study. The results also showed that retail deliveries were affected by COVID-19 cases, which is consistent with previous research ( 10 ). Jung et al. ( 5 ) indicated the potential of shifting consumer demand from traditional stores to online stores to reduce possible exposure during a pandemic. This study found that the COVID-19 pandemic indirectly affected retail deliveries, which is similar to previous research ( 5 ). Therefore, it is essential to identify strategies that can help to guarantee the survival of the traditional retail trade in a pandemic.

Online marketplaces could be one option for small and medium retailers. Most of these retailers consist of independent companies. Online marketplaces are available in digital applications or on websites that offer goods from different sellers. Online marketplaces could be an alternative to implement online sales and explore new markets, which go beyond traditional retail. Examples of online marketplaces in Brazil are Mercado Livre, Amazon, OLX, and Magalu. Thus, combining online and traditional retail sales may be an option for the survival of traditional retail. During pandemics, local shopping spaces could be converted into warehouses. Additionally, omnichannel strategies could improve customer experience with online/offline sales. A similar strategy was suggested by Jung et al. ( 5 ).

However, e-commerce strategies have logistical implications and increase home deliveries. Since it is crucial to enhance sustainable UFT, green city solutions need to be supported. In addition, as contactless deliveries have become part of urban deliveries during the COVID-19 pandemic ( 7 ), lockers could be one solution to reduce the number of home deliveries and, consequently, last-mile costs. Security is a major challenge in implementing this solution in Brazil. However, omnichannel (i.e., enable the companies to meet customer demand concerning the purchase and delivery experience) and marketplace lockers can be connected during and post-COVID-19 pandemic as part of the new normal. Thus, this connection may replace some home deliveries after the COVID-19 pandemic.

It is important to note that click-and-collect delivery systems were not considered as an alternative when the retail sector was closed because of the pandemic. Lockers could be a more suitable solution to replace home deliveries, especially in the largest cities, as long as they are situated in locations of easy access at any time of day or night, in any condition. However, shippers need to experience a cultural shift since they intend to offer the best customer service. In the case of Brazilian e-commerce, this is related to home deliveries.

The COVID-19 pandemic provided a lesson for UFT: innovation and flexibility must be the new normal for organizations ( 7 ). UFT will be needed in any situation, including catastrophes, pandemics, natural disasters, and others. At any moment, including these unprecedented situations, people need goods for survival. The COVID-19 pandemic highlighted the current lack of UFT planning by public and private companies.

UFT contributes to economic development. Therefore, identifying strategies to support consumer behavior in future pandemics could reduce economic disruption at local and national levels in these situations ( 5 ).

Conclusion

The COVID-19 pandemic has changed people’s lives as it directly affected access to goods, especially food. Considering these changes, this paper investigated the impacts of the COVID-19 pandemic on urban deliveries during the first seven EWs in the BHMR. Urban deliveries were composed of retail deliveries and home deliveries, which represent e-commerce for non-essential goods. This study used the Lee index, which calculates the correlation between two variables, and the LISA based on the bivariate Moran index, which identifies spatial patterns among the EWs.

The results showed that COVID-19 cases and lockdown measures had negative impacts on retail deliveries. In contrast, home deliveries increased over the weeks. However, the time dependency analysis showed patterns of reduction for the association between space and home deliveries. A negative correlation was identified between retail deliveries and home deliveries in EW7. The analysis period considered in this paper is a limitation of the study. However, the results clearly show the changes in consumer behavior when traditional retail outlets were closed by lockdown measures.

Increases in home deliveries affect the local infrastructure ( 10 ). The results indicate that policies to reduce the impact of this kind of delivery are essential. In Brazil, initiatives such as click-and-collect and parcel lockers are restricted to a small number of cities, and they need to be expanded for more efficient last-mile deliveries. The increase in home deliveries also demonstrates a change in consumer patterns, which could consequently change land use. Therefore, the public administration needs to be aware of this issue to avoid depreciation in some city areas.

Since this study was conducted using data from the beginning of the pandemic, future research should investigate these relationships throughout the course of the COVID-19 pandemic. After more than a year since the beginning of the pandemic, significant changes are noticeable, including changes in people’s lifestyles and better knowledge among the population in relation to COVID-19 risks. The virus spread is also considerably different at the time of writing, when all cities currently have a high number of daily cases and lockdown measures have been implemented to varying degrees. In addition, most of the traditional retail has closed to some extent, and e-commerce is increasing daily. What are the effects of COVID-19 on urban deliveries after one year of the pandemic? This is a proposed research question for future studies.