*Article* **E-Scooter Micro-Mobility Application for Postal Service: The Case of Turkey for Energy, Environment, and Economy Perspectives**

**Yunus Emre Ayözen 1, Hakan ˙ Inaç 2, Abdulkadir Atalan 3,\* and Cem Ça ˘grı Dönmez <sup>4</sup>**


**Abstract:** In this research, the advantages of the e-scooter tool used in the mail or package delivery process were discussed by considering the Turkish Post Office (PTT) data in the districts of Istanbul (Kadıköy, Üsküdar, Kartal, and Maltepe) in Turkey. The optimization Poisson regression model was utilized to deliver the maximum number of packages or mails with minimum cost and the shortest time in terms of energy consumption, cost, and environmental contribution. Statistical and optimization results of dependent and independent variables were calculated using numerical and categorical features of 100 e-scooter drivers. The Poisson regression analysis determined that the e-scooter driver's gender (*p*|0.05 < 0.199) and age (*p*|0.05 < 0.679) factors were not effective on the dependent variable. We analysed that the experience in the profession (tenure), the size of the area responsible, and environmental factors is effective in the e-scooter distribution activity. The number of packages delivered was 234 in a day, and the delivery cost per package was calculated as 0.51 TL (Turkish Lira) for the optimum values of the dependent variables. The findings show that the choice of e-scooter vehicle in the mail or package delivery process is beneficial in terms of time, cost, energy, and environmental contribution in districts with higher population density. As the most important result, the operation of e-scooter vehicles with electrical energy shows that it is environmentally friendly and has no CO2 emission. The fact that the distribution of packages or mail should now turn to micro-mobility is emerging with the advantages of e-scooter vehicles in the mail and package delivery. Finally, this analysis aims to provide a model for integrating e-scooters in package or mail delivery to local authorities, especially in densely populated areas.

**Keywords:** micro-mobility; e-scooter; postal service; Poisson regression; optimization; energy; cost; environment

#### **1. Introduction**

Micro-mobility seems to be a useful strategy for cities that want to reduce singleperson vehicle journeys and improve multimodal amenities [1]. Since the micro-mobility revolution is still in its infancy, it is an important topic of discussion in the literature, especially with the mobility sector changing rapidly and moving away from trend vehicle ownership, causing uncertainty about how this sector will develop to arise [2,3].

Entrepreneurs and authorized institutions are constantly searching for package and postal transportation vehicles. In particular, traffic density, one of the big cities' most significant problems, negatively affects package and mail delivery [4,5]. Significant congestion and dense urban structuring considerably impact distribution operations. To overcome such problems, although there are different opinions on the choice of distribution vehicles, enterprises and authorized institutions generally adopt scooter vehicles that are considered

**Citation:** Ayözen, Y.E.; ˙ Inaç, H.; Atalan, A.; Dönmez, C.Ç. E-Scooter Micro-Mobility Application for Postal Service: The Case of Turkey for Energy, Environment, and Economy Perspectives. *Energies* **2022**, *15*, 7587. https://doi.org/10.3390/en15207587

Academic Editor: Katarzyna Turo ´n

Received: 22 September 2022 Accepted: 11 October 2022 Published: 14 October 2022

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2022 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

within the scope of micro-mobility and work with electric energy [6]. The advantages and disadvantages of e-scooters in various aspects, such as social, environmental, economic, and energy effects, have been studied in the literature [7,8]. This study deals with the energy, economy, and ecological factors of e-scooter vehicle preference in package and mail delivery activities.

The latest strategy for city authorities that allows users to gain temporary access to modes of transport "as needed" is the use of delivery of a vehicle, bike, car, or another mode [9]. Electric micro-mobility systems such as e-scooters are used independently and as a shared service to provide sustainable mobility solutions for city logistics, particularly for certain classes of package delivery, user characteristics, and journey distances [10]. In particular, given the growth of e-commerce and the proliferation of new options for package delivery, it helps spread new alternatives such as e-scooters and speed delivery operations (such as bulk shipping) [10]. Conventional vehicles constitute 8–18% of urban traffic flows, significantly affecting the traditional vehicles (combivan, pick-up, etc.) used in package and mail distribution in traffic jams. However, it has been determined that the current road capacity has been reduced by 30% with micro-mobility applications [11,12].

Many advantages of using vehicles within the scope of micro-mobility in postal and package service have been discussed in the literature [13,14]. The most important benefit is that it provides fast and timely delivery of packages or mail with micro-mobility vehicles [15]. In one study, a project funded under the Smart Energy Europe program referred to micro-mobility tools for the fastest response in terms of time [16]. Another study mentioned the advantage of using e-scooter rides by calculating the average travel distance and running time of e-scooter journeys of 1.24 km and 7.55 min., respectively. However, another study determined that the time savings of e-scooter journeys in congested areas close to the city centre are limited, depending on the average cluster speed of e-scooter vehicles [17]. This study emphasizes that the time required to complete package and postal service operations is advantageous with the e-scooter [18].

Another advantage of using the e-scooter, which is among the micro-mobility tools, is that it keeps the energy consumption at a minimum level and reduces the distribution cost [19]. Most vehicles used in package or mail distribution activities need fuel energy [20]. Today, fuel type is more costly than electrical energy [21]. Gebhardt et al. argue that e-scooters used for package and mail delivery operations have lower land use consumption and significantly better energy efficiency than other motor vehicles [22]. In another study, it was determined that the e-scooter is more advantageous in terms of energy savings as a result of the test of e-scooter and other vehicles to estimate the amount of energy consumption in an area with a 0-degree slope [23]. In this study, the energy efficiency of the e-scooter delivery vehicle used in package and mail distribution is discussed in detail, especially in terms of cost.

Many studies explaining the effects of vehicles used for transportation or logistics purposes in many ways, especially in terms of environmental health and energy consumption, have been analysed with different methods. The common aspect of these studies draws attention to CO2 emission, one of the environmental factors. A study aimed to overcome the many uncertainties and complexities in the mix of economy-energy-environment systems, random CO2 emission, and water consumption control policies by integrating multi-objective programming, fuzzy linear programming, and multiple scenarios [24]. Another study focused on the environmental cost of CO2 emissions, aiming to break the barrier of the CGE (the computable general equilibrium) model and provide researchers with a CGE model with available code and data [25]. As there is an important link between environmental factors and energy consumption, Miao et al. emphasized that CO2 emissions, SO2 emissions, and atmospheric environmental inefficiency caused by primary energy use are the main causes [26]. Another study applied a multi-sector and multi-site dynamic recursive computable general equilibrium model to reduce coal consumption in order to reduce CO2 emissions to meet energy needs in the 2020–2030 period in China [27]. Using a multi-objective optimization model based on input–output analysis, Zhang et al. investigated China's energy, water consumption, and CO2 emissions values, including the high resolution of the country's electricity sector, in the period 2020–2030 [28]. Compared to studies dealing with the link between energy, cost, and environmental factors from different perspectives, this study also deals with the energy, cost, and environmental factors of e-scooter micro-mobility vehicle.

The environmental effects of e-scooter use are frequently discussed in the literature. At the beginning of environmental factors, the rate of CO2 emission has been tested by researchers with a wide range. One study highlights that approximately 5.8 kt of CO2 will be saved daily when e-scooter vehicles replace existing car trips [29]. Severengiz et al. investigated how using e-scooters for different purposes affect the crop greenhouse balance compared to alternative means of transportation by evaluating ecological factors [30]. Another study emphasizes that using e-scooters, among the smart mobility tools, will make urban life simpler, economical, and enjoyable with faster transportation, less congestion, and low CO2 emissions. The same study found that e-scooters emit almost 45% less CO2 than other vehicles, emphasizing that about 90% of people are exposed to air pollution [31]. According to the LCA results of personal e-scooter use, Moreau et al. calculated the environmental impact as approximately 67 g of CO2 emissions [32]. In another study, the authors highlight a net reduction in environmental effects when the e-scooter vehicle replaces the personal automobile vehicle, finding with Monte Carlo simulation models that 65% of the life-cycle greenhouse gas emissions associated with e-scooter use were higher than the modified set of transport models. In this study, the effects of e-scooters and other delivery vehicles on the environment were investigated by using e-scooter vehicles for package and mail distribution by government units [33].

With temporary delivery due to the increase in volume transported, operations involving a key logistics player often require electric-powered vehicles [34]. Scientific studies emphasized that tricycles or e-scooter vehicles provide significant advantages [35–37]. However, in using these vehicles, they must operate within the framework of some rules following the rules of the people and society [6]. In particular, in cities with crowded settlements, the authority departments need to exchange information with e-scooter companies to guide many driving rules and regulations, such as driving in the wrong direction, right of way, and speed [14]. In general, it can be said that adopting e-scooter vehicles in package and mail delivery significantly impacts the delivery system [36]. If generalization is made with the following important advantages in e-scooter preference, e-scooter vehicles:


and offer many advantages. Another advantage of e-scooter vehicle preference is customer and rider satisfaction [14,40]. According to research conducted in northern European and North American countries over the last decade, user satisfaction with electric scooters and the service delivery process is high overall [41–43]. A study conducted in the Netherlands observed user satisfaction in the service delivery process. It was discovered that user happiness was associated with the length of actual waiting times [44]. Chinese customers seem to prefer e-scooters over public transport because of customers' demand for more flexible, comfortable, and enjoyable (at a reasonable price) mobility [45]. This has led to an increase in the production of e-scooter vehicles that has contributed to customer satisfaction in recent years [46]. All these results mean that micro-mobility vehicles such as e-scooters will have more of a place in human life [47].

Scientific studies investigating the many advantages of using micro-mobility vehicles for different purposes in terms of environmental, cost, and energy have used different methods. Detailed information about the scientific studies in Table 1 is shared in order to reveal the difference between the Poisson optimization regression model method used in this study and other studies. The best statistical optimization model is the Poisson regression model method, especially for modelling situations that indicate the importance of the results of the objectives or output parameters in a subject [48]. In particular, we preferred methods for non-negative integer-valued variables that count information, such as specific counting data (such as the number of e-scooter vehicles and drivers) and the number of events occurring in a given time period (such as the number of packets delivered). Especially in a statistical or optimization model, if the values of the objective function or output parameter are positive and integer, the Poisson regression optimization model is preferred [49]. We contributed to obtaining numerical data of objective functions or output variables of micro-mobility vehicle applications (such as e-scooter, e-bike) with certain parameters with the statistical optimization model developed.

**Table 1.** Detailed information about the aims, methods, and factors of studies related to micromobility approaches and this study.


This study presents a case study of the advantages of using e-scooters vehicles in package and mail delivery services in Istanbul, Turkey. PTT provides package and mail distribution service with a total of 1915 vehicles throughout Istanbul megacity. Turkey postal service unit provides postal and package service, using approximately 1046 large vehicles (trucks, bus, van, combivans, minibus, etc.) and 655 motorbikes and other mobile vehicles. The use of e-scooters is used in four districts of Istanbul with a dense population. Around 161 e-scooters are allocated in these districts for postal and package services. The ratio of e-scooters to other vehicles is approximately 8.41% [59].

In this study, many independent parameters are considered to measure the effects of e-scooter vehicle choice in terms of environmental, economy (cost), and energy. These parameters are defined as the e-scooter driver's gender, age, professional experience, area of responsibility, and environmental factors. This study consists of five different sections. The literature review about the advantages and disadvantages of using e-scooter within the scope of micro-mobility is discussed in the introduction part of the study. The methodology developed for the study is discussed in the second section. The numerical results of the study are shared in the third section. Discussion of significant findings is debated in the fourth part of the study. The conclusion of the study is mentioned in the last part of the study.

#### **2. Materials and Methods**

This study aims to calculate the optimum values of the objective functions by developing optimization Poisson regression models using the data of the e-scooter application for package and mail distribution in four different regions of Istanbul with the highest population density. This part of the study consists of three main parts: data preparation, statistical analysis, and development of optimization models.

#### *2.1. Data*

PTT (International Logistics Services), which has been using the bicycle for years (one of the micro-mobility vehicles, which is an environmentally friendly and more efficient alternative in urban use), has started to use the e-scooter in urban distribution/delivery operations as of July 2021. Kadıköy, Üsküdar, Kartal, and Maltepe Postal Distribution Directorates affiliated with PTT Istanbul Regional Directorate were selected as pilots for the e-scooter, which is more suitable for the distribution of registered-unregistered mail and cargo-courier shipments under 2 kg/decis. The total population of these regions is approximately 2.02 million, constituting 12.73% of Istanbul's population as of 2021. The population density of these districts is 10,728 people per km squared. The pilot chosen areas for mail delivery with the e-scooter are illustrated in Figure 1.

One hundred e-scooters accompany one hundred drivers (each scooter belongs to only one driver, e-scooter vehicles are not shared between drivers) working in plot areas. The relationship between the age and occupational experience of 100 drivers employed in packages and mail delivery based on the delivery time by e-scooter is shown in Figure 2.

**Figure 1.** Plot zones selected for packages delivered with the e-scooter application.

**Figure 2.** The sort of the delivery drivers based on the delivery time by e-scooter.

The technical specifications of the preferred e-scooter ranges for the distribution process are shared in Figure 3.

**Figure 3.** Technical specifications of the e-scooter.

A total of 3000 data points were used, taking into account the 30-day working time of each driver. The average daily distribution numbers of 100 riders with the e-scooter are shown in Figure A1 in Appendix A Section. During the period in which the data were taken into account, a total of 351.180 distributions were made using e-scooters. The maximum distribution amount was computed as 8454, and the minimum distribution amount was calculated as 915. The monthly average distribution amount of these drivers was computed as 3511.8 packages.

For this study, the economic, energy, and environmental factors of both the e-scooter and other vehicles are utilized to compare the e-scooter distribution vehicle with other distribution vehicles. The monthly rental price of the vehicles, the energy (fuel and electric) used, the distribution flow, CO2 emission rates, and distance information are discussed in this study. The vehicles used in package distribution are given the combivan vehicle with a volume of 4 m3, and the motorcycle and e-scooter vehicles were used in the distribution at the PTT. Although many factors affect the performance of drivers in distribution planning with e-scooters, seven different parameters are considered in this research. Qualitative information about each parameter, such as variable type, units, status of variables, notation, and descriptive expressions, is shared in Table 2.


**Table 2.** Indicators of the packages delivered by the e-scooters.

\* Maintenance, repair, and insurance costs belong to the contractor company.

The driver's age, gender, and experience (tenure) factors, which are among the decision variables of the study and affect the number of packages distributed daily and monthly, are only for drivers using e-scooters. In addition, the amount of CO2 emission, an environmental factor that is thought to affect the amount of distribution, was also included in this study. The area for which each driver is responsible is taken into account in km2. Descriptive statistics of the data used for the decision variables are discussed in Table 3. Descriptive statistics data such as sample size, mean, standard deviation, first and third quartiles, variance, kurtosis, and skewness were analyzed for the data set of this study.


**Table 3.** Descriptive statistics of the variables of the e-scooter distribution.

Variable abbreviations: Delivery cost by package or postal vehicles: Cost\_e-S: e-Scooter cost, Cost\_m: motorcycle cost, Cost\_c: combivan cost. Statistical abbreviations: SE Mean, standard error mean; StDev, standard deviation; CoefVar, variance coefficient; Q1, the first quartile, Q3, the third quartile; IQR, interquartile range; N, number of samples.

Two methods, Poisson regression and response-optimization mathematical models, were used for the methodology of the study. Minitab-18 computer software, including statistical and optimization tools, was used to organize and analyze the raw data of the study [60]. There are theoretical explanations of the methods in the continuation of this section.

#### *2.2. Statistical Analysis*

The Poisson regression model, developed by Consul and Famoye (1992) and Famoye (1993), was used to model data for factors affected by a set of response variables [61]. The Poisson distribution regression model includes a series of statistical analyses for multipleaffected response variables and co-influencing variables in under- or over-dispersed count data. Generally, models are developed using maximum likelihood and moment methods in Poisson distribution regression analysis [62,63]. Poisson regression is one of the most preferred methods of analysis for modeling response variables with integer properties [64]. Poisson regression analysis was preferred because the data of the decision variables were integer in this study [65]. The vehicles (e-scooters) were used to make the distribution, and the number of packages and the drivers (human factor) who perform the distribution process represent the whole number. The Poisson regression model is formulated with the given by *f*(*μi*, *α*, *yi*) [61]:

$$f(\mu\_i, a, y\_i) = \left(\frac{\mu\_i}{1 + a \* \mu\_i}\right)^{y\_i} \* \frac{(1 + a \* y\_i)^{y\_i - 1}}{1 + a \* \mu\_i} \* \exp\left[\frac{-\mu\_i \* (1 + a \* y\_i)}{1 + a \* \mu\_i}\right] \tag{1}$$

where *yi* donates the response or dependent variable of the regression model with {*i* = 1, 2, . . . , *n*}. Independent or decision variables are defined as *xi* and the mean and variance of the Poisson distribution are the same as:

$$E(y\_i) = Var(y\_i) = \mu\_i \tag{2}$$

where the expected mean and variance value is defined as *E*(*yi*) and *Var*(*yi*), respectively.

$$
\mu\_i = \mu\_i(x\_i) \tag{3}
$$

then;

$$\mu\_{i}(x\_{i}) = \exp\left(\sum x\_{i\bar{j}}\beta\_{\bar{j}}\right), \ j = (1, \ 2, \ \dots, k) \tag{4}$$

where *β<sup>j</sup>* represents the coefficient of independent variables of the regression equation [66]. In order to calculate the maximum likelihood estimator in the Poisson regression model, the response variable *yi* must be in the form of non-negative integers (or count data). In this study, since the response variables are integers, the maximum likelihood function is given as follows [67]:

$$\psi(y\_i) = \exp\left[\frac{-\mu\_i \* (\mu\_i)^{y\_i}}{y\_i}\right] \tag{5}$$

s.t.

$$
\mu\_i > 0 \tag{6}
$$

where the likelihood function (*l βj* ) of the Poisson regression model is created as [67]:

$$\mathcal{U}(\beta\_{\dot{i}}) = \prod\_{i=1}^{n} \frac{\exp(-\mu\_{i} \* (\mu\_{i})^{\mathcal{Y}\_{i}})}{\mathcal{Y}\_{i}!} \tag{7}$$

then;

$$I(\beta\_{\bar{i}}) = \frac{\prod\_{i=1}^{n} (\mu\_i)^{y\_i} \exp(-\sum\_{i=1}^{n} \mu\_i)}{\prod\_{i=1}^{n} y\_i!} \tag{8}$$

Approximate tests are considered for testing the adequacy of a Poisson distribution regression model. We adopted e-scooter delivery data to evaluate and analyze the performance of the response variables and other decision variables (independent factors) proposed in this study. In addition, as a result of Poisson regression statistical analysis, the optimum values of the response functions and the decision variables were calculated using the restrictive data belonging to the decision variables.

#### *2.3. Optimization Models*

Optimization (mathematical) modeling is generally defined as expressing real-life problems with equations. Optimization models consist of four different steps: determination of decision variables, the definition of objective functions, creation of limits of decision variables, and regulation of sign directions of decision variables. Independent factors (or decision variables) are based on Poisson distribution regression analysis and optimization models. The objective function equation of an optimization model given the decision variable as *xij* is formed as follows [68]:

$$\text{objective}\_{\mathbf{o}} f = \sum\_{i=1}^{n} \sum\_{j=1}^{m} c\_{i}(\mathbf{x}\_{ij}) \tag{9}$$

where *ci* represents the coefficient of the decision variables with {*i* = 1, 2, . . . , *n*}. There are two versions of *x*, maximum and minimum. This version is preferred according to the purpose of the problem. Generally, the minimum preference is for the cost or time, while the maximum preference is for high-value purposes such as annual income or production amount [69]. Each optimization model has a limit of decision variables. These limits are defined as constraints in optimization models. In an optimization model, constraint equations are usually created as follows [70]:

$$\sum\_{i=1}^{n}\sum\_{j=1}^{m}a\_{i}(\mathbf{x}\_{ij})\begin{cases} \ge \\ = v\_{l\prime} \quad l = \{1, 2, \dots, L\} \\ \le \end{cases} \tag{10}$$

where *ai* signifies the coefficient of the decision variables in the equations of the constraints. The values of the constraints' limit are denoted by *vl*. The mixed-integer optimization model is created because some of the decision variables in the optimization models of this study are integers and others are natural numbers. The mixed-integer optimization model is constituted as [71]:

$$\text{objective}\_{\text{o}} \; f = \mathbf{x}^{t} \mathbf{Q} \mathbf{x} + q^{t} \mathbf{x} \tag{11}$$

s.t.

$$\begin{array}{c} A\boldsymbol{x} = \boldsymbol{v} \text{ (linear constraints)}\\ l \le \boldsymbol{x} \le \boldsymbol{u} \text{ (bound constraints)}\\ x^t Q\_i \boldsymbol{x} + q\_i^t \boldsymbol{x} \le b\_i \text{ (quadratic constraints)}\\ \text{Some (or all) of } \boldsymbol{x} \text{ values must be integer} \end{array} \tag{12}$$

The objective function of the optimization model of this study constitutes the equation obtained from the Poisson distribution regression model. Decision variables were defined as independent variables affecting the response factor. The mixed integer optimization model then turns out to be as follows [72]:

$$\{\text{objective}\_{\text{maxsize}}\{\text{Equation (11)}\}\}\tag{13}$$

s.t.

where, *xi* = *xage*, *xgender*, *xtenure*, *xarea*, *xCO*<sup>2</sup> . The MILP problem presented in this article is essentially an optimization problem, where the aim is to maximize the number of distributed packages by taking into account the effect of independent variables and which provides a set of feasible solutions belonging to the solution set within the limits of the decision variables of the system to be optimized. The desirability functions measure the degree of importance of the feasible values of the optimization model. For all feasible output values of the objective functions determined by the desirability equations, we ensure that the values suitable for the design factors are found simultaneously so that the optimization model reaches an optimal solution [73]. Desirability values (*di*) are calculated according to the following formulation [74]:

for maximization problems:

$$d\_i(y\_i(\mathbf{x})) = \begin{cases} 0 & \text{if } y\_i(\mathbf{x}) < l\_i \\ \left(\frac{y\_i(\mathbf{x}) - l\_i}{u\_i - l\_i}\right)^{r\_1} & \text{if } l\_i \le y\_i(\mathbf{x}) \le u\_i \\\ 1 & \text{if } y\_i(\mathbf{x}) \ge u\_i \end{cases} \tag{15}$$

for minimization problems:

$$d\_i(y\_i(\mathbf{x})) = \begin{cases} 1 & \text{if } y\_i(\mathbf{x}) < l\_i \\ \left(\frac{u\_i - y\_i(\mathbf{x})}{u\_i - l\_i}\right)^{r\_2} & \text{if } l\_i \le y\_i(\mathbf{x}) \le u\_i \\ 0 & \text{if } y\_i(\mathbf{x}) \ge u\_i \end{cases} \tag{16}$$

for target values of the objective functions:

$$d\_i(y\_i(\mathbf{x})) = \begin{cases} \text{if } y\_i(\mathbf{x}) < l\_i \\ 0 & - \\ \left(\frac{y\_i(\mathbf{x}) - l\_i}{u\_i - l\_i}\right)^{r\_1} & \text{if } l\_i \le y\_i(\mathbf{x}) \le T\_i \\ \left(\frac{u\_i - y\_i(\mathbf{x})}{u\_i - l\_i}\right)^{r\_2} & - \\ \text{if } T\_i \le y\_i(\mathbf{x}) \le u\_i \\ \text{if } y\_i(\mathbf{x}) \ge u\_i \end{cases} \tag{17}$$

where *li* and *li* are the upper and lower limit values of the desired response equation. The parameters of *r*<sup>1</sup> and *r*<sup>1</sup> express the importance of the response equations being close to the desired value [75]. We propose the limits of the independent variables for the plot regions, where the e-scooter application is planned with the Poisson distribution regression and optimization models we have developed.

#### **3. Results**

The numerical results of the study are discussed in this section. The statistical analyses and optimization results of the e-scooter data used for package and mail delivery were examined using the optimization Poisson regression distribution developed for the study. In addition, the numerical results were compared between the e-scooter vehicle and other distribution vehicles in terms of economic, environmental, and cost.

#### *3.1. Statistical Results of Poisson Distribution Regression Analysis*

Statistical data of the Poisson distribution regression analysis are given in Table 4. In the Poisson distribution regression analysis, data belonging to two dependent and five independent variables were used for statistical results. The regression analysis results for the dependent variable show that the independent factors have a significant effect with a *p*-value < 0.05. As a result of the statistical analysis, it has been determined that the amount of CO2 emission from the five independent variables, the driver's experience in the profession (tenure), and the size of the area that the driver is responsible for distribution affect the number of packages distributed. The significance levels of x*co*2, x*tenure*, and x*area* factors were calculated as 0.022, 0.001, and 0.001, respectively. The driver's experience in the profession (tenure) and the size of the area affect the cost of the packages distributed. The significance levels of x*tenure*, and x*area* factors were calculated as 0.0001, and 0.000, respectively. The effects of driver gender and age on dependent variables are limited based on the significance levels of the Poisson distribution.

Goodness-of-fit tests were used to determine whether the dataset used deviated in a way that the Poisson distribution did not predict. The model's data fit was assessed using the Pearson compatibility and Deviance tests. In these results, both goodness of fit tests had *p* values lower than the usual significance level of 0.05. Sufficient evidence has emerged to conclude that the number of predicted events does not deviate from the number of events observed. In addition, in terms of the accuracy and validity of the results of the Poisson distribution regression statistical analysis, the R-Squared (R-sq) and adjusted R-sq values for the packages delivered, which is a measure of goodness of fit for the model, were calculated as 90.24% and 90.21%, respectively. For the cost of the packages delivered, the value of the R-sq is computed as 90.24% and 90.19%, respectively.


**Table 4.** Poisson distribution regression analysis data of the e-scooter distribution.

SE Coef.: Coefficient of Standard Error, DF: Degree of Freedom, VIF: Variance Inflation Factor, R-Sq: R-squared, R-Sq (adj): Adjusted R-squared.

*3.2. Comparison of the Economy, Energy, and Environmental Dimensions of the E-Scooter Model with Other Vehicles Used in the Package Distribution*

In this section, we have analyzed the numerical results in terms of economic, energy, and environmental aspects so as to reveal the difference between the distribution provided by the e-scooter and the distribution operations performed with other vehicles. This section consists of three different subsections.

#### 3.2.1. Cost Analyses

In package or mail distribution, main expenses such as personnel, insurance, energy, car rental, and packaging are included in the general costs. In this study, it is understood that many advantages are obtained regarding cost with the e-scooter application in package distribution. As a result of distribution made by e-scooters and other vehicles, there are some differences between energy and package distribution costs (for example, distribution of more packages with the same personnel wage) apart from the common expenses. Table 5 includes the costs depending on the number of packages distributed with e-scooters and other vehicles. The cost information in this table represents the total cost of a package to the administration.

**Table 5.** The vehicle types in terms of the economic dimension.


\* Based on the Combivan, \*\* Based on the Motorcycle, \*\*\* TL: Turkish Lira.

We calculated that the delivery cost of a package with an e-scooter is 16 times more advantageous than a combivan and three times more than a motorcycle. Considering the daily distribution amounts, it was determined that the cost of distribution with the e-scooter decreased by 96.49% compared to the motorcycle and 99.51% compared to the combivan. Similarly, it was calculated that the cost of package distribution with a motorcycle decreased by 86.13% compared to a combivan. The hourly distribution amounts determined that the cost of distribution with the e-scooter decreased by 65.27% compared to the motorcycle and 95.18% compared to the combivan. Similarly, it was calculated that the cost of package distribution with a motorcycle decreased by 86.13% compared to a combivan. Depending on the package distribution quantity, the distribution costs of the package distribution vehicles are shown in Figure 4. The e-scooter's total distribution cost is minimal compared to other vehicles.

**Figure 4.** The total cost for a package delivered.

#### 3.2.2. Energy Analyses

The type of energy required for the e-scooter is electrical energy. The fuel type meets the energy supply of other vehicles. For daily use, e-scooter batteries are made ready before working hours. An extra full battery is allocated to drivers in case of an unexpected situation during working hours. The difference between the amount of energy consumption of the e-scooter and other distribution vehicles is shown in Table 6.


**Table 6.** The vehicle types in terms of energy dimension.

\* Based on the E-Scooter, \*\* Based on the Motorcycle, \*\*\* Based on the Combivan.

Regarding energy, the advantage of using e-scooters in mail or package delivery is very high compared to other vehicles. While the use of a motorcycle is 64.74% advantageous compared to the combivan vehicle, it has an 88.52% disadvantage compared to the escooter. Similarly, it has been calculated that the use of combivan has a 64.74% disadvantage compared to the motorcycle vehicle and 95.95% compared to the e-scooter. The amount of energy consumption required for the e-scooter is less than other vehicles. For the use of e-scoter, energy consumption is ten times less than the amount of energy required for a motorcycle and almost 30 times less than the amount of energy required for a combivan. According to the number of packages distributed in a month, the amount of energy required varies according to the vehicles. The energy change rates based on the number of packages delivered are shown in Figure 5.

**Figure 5.** (**a**) The cost of energy required for a package delivered, (**b**) The number of packages or mails distributed with the e-scooter micro-mobility.

#### 3.2.3. Environmental Analyses

The amount of CO2 emissions, which is one of the environmental factors that affects the amount of package distribution, varies considerably between distribution vehicles. The operation of e-scooter vehicles with electrical energy is environmentally friendly and the amount of CO2 emissions is very low. E-scooter use does not cause direct CO2 emission. However, this study did not consider CO2 emissions indirectly caused by e-scooter vehicles (e.g., battery charging, during the manufacturing process, transportation of e-scooter vehicles to users, etc.). The use of motorcycles and combivan vehicles in package distribution activities for many years has led to negative results in terms of environmental health.

The CO2 emission amounts of the motorcycle and combivan distribution vehicles are compared to the e-scooter distribution vehicle since the emission amount of the e-scooter distribution vehicle is low (Parameters that indirectly cause CO2 emissions from using the e-scooter vehicle were not considered, so the value of 0 was used in this study to reference the values of other vehicles). In the case of distribution by motorcycle, it has been calculated that the amount of CO2 emission has a 190% disadvantage compared to the e-scooter delivery vehicle and an advantage of 13.15% compared to the combivan vehicle. We have determined that Combivan preference for distribution activities is 200% and 15.15% disadvantageous compared to e-scooter and motorcycle distribution vehicles, respectively. Figure 6 includes the CO2 emission amounts of the motorcycle and combivan distribution vehicles, excluding the e-scooter distribution vehicle, according to the number of packages distributed.

**Figure 6.** CO2 emission values of motorcycle and combivan distribution vehicles depending on the package density distributed.

#### *3.3. Results of the Optimization Models*

Using mixed integer optimization models developed for this study, optimum values were obtained for objective functions and decision variables. Although these optimization models have the same constraints and decision variables, they have turned into a multiobjective optimization model type because they contain more than one objective function. Therefore, the optimum values obtained were also considered feasible values, as multiobjective optimization models also work like nonlinear optimization models. Generally, the results obtained in nonlinear optimization models are not optimum but feasible values.

Keeping the independent variables influencing the dependent variable at optimized values estimated by the desirability function approach allows one tfurther explore the effect of independent variables on individual output responses and overall desirability. In this study, Equations (14) and (15) were revised, and the following equation was used to obtain the desirability data obtained since two different objective functions were solved with the constraints consisting of five common independent variables:

$$2\*\text{Equation (14), 2\* Equation (15)}\tag{18}$$

With the e-scooter tool, the best 14 results (the feasible results after 14 iterations are the same) were obtained by running non-linear and mixed integer optimization models for the transportation of large numbers of packages in the package distribution service with minimum cost. These results are included in the solution set of the optimization model. The feasible results of the optimization models are depicted in Figure 7.

**Figure 7.** Optimum values of the number of packages distributed and distribution cost depending on the desirability values.

The optimum levels for a driver to deliver in a maximum time in a month, age, gender, experience in the profession, and area of responsibility were calculated as 22.63, F, 24.8 (maximum), and 0.113 km<sup>2</sup> (113,000 m2), respectively. Depending on the decision variables and objective functions, the optimum values of the age, gender, experience, and area size of a driver performing the distribution process in terms of the average values of the best 20 feasible results were calculated as 38.98, M/F, 14.91 years, and 0.953 km2, respectively. In the best 20 results of the optimization models, there were equal numbers of male and female drivers. The optimum values of the average number of packages distributed monthly and the monthly distribution cost, which are the objective functions, were calculated as 4685 and 2389 TL, respectively. The cost of distribution of a package to the administration was calculated as 0.509 TL based on the optimum results. We have determined that the effect of driver gender in the distribution process is low (unless physical characteristics are taken into account), and it is not essential in terms of distribution amount and cost.

#### **4. Discussion**

The Poisson distribution regression and optimization technique discussed in this paper is only applied to analyze the economic, energy, and environmental aspects of an e-scooter package delivery application in Turkey. The concept of micro-mobility has revealed that 40% of vehicle journeys worldwide are made at distances below 5 km, and only 5–10% of the fuel consumed in these vehicles is used to transport passengers. Micro-mobility has become an ideal system for journeys with vehicles traveling at a maximum speed of 20–25 km/h for distances up to 5–10 km. Considering that 70% of the world's population will live in cities and the use of individual vehicles will increase in 2050, the importance of micro-mobility will gradually increase in solving the problems caused by the number and density of vehicles.

The period in which the amount of distribution made by e-scooters is discussed and the amount of distribution made with other vehicles (including pedestrian distribution personnel) in the same period of the previous year are discussed. Data for both distribution types (e-scooter and other vehicles) are shown in Figure 8. In the same period, a total of 286,953 distributions were made with other vehicles and pedestrians. In comparison, a total of 351,180 distributions were made after using e-scooters, increasing the delivery performance by 22%. It should not be forgotten that with the pilot application started in Istanbul, the e-scooter not only increased the speed and distance traveled (~2x) during the day but also increased the comfort it provided to the pedestrian distribution personnel, who took thousands of steps.

**Figure 8.** The number of packages distributed with e-scooters and other vehicle delivery personnel.

Therefore, as PTT, we can say that we see e-scooter as an important alternative not only for pedestrian distribution but also for motorcycle and vehicle distribution. As a matter of fact, PTT plans to increase the number of e-scooters, which was 100 in July 2021, to 500 as of July 2022, upon high demand from the personnel. Among the results of this study, it is clearly seen that the e-scooter, which is an environmentally friendly vehicle with zero carbon emissions, is three times more efficient than motorcycle distribution and 16 times more efficient than vehicle distribution in terms of unit cost. In addition to the economic, energy, and environmental advantages of package and mail distribution, the e-scooter also provides the benefit of timely and fast distribution of packages or mail. We observed the advantage in the time factor by comparing two different years of a one-month period in which the data were taken into account. The average delivery time of a package or mail was established based on the following formulation:

$$p = 1 / \frac{\sum\_{i=1}^{n} n\_i / \sum\_{m=1}^{m} t\_m}{\sum\_{d=1}^{d} t\_d} \tag{19}$$

where the number of packages is symbolized by *p* and *p* = 1 is considered in this research. *ni* represents the number of packages delivered in a day. *tm* signifies the days of working in a month. *td* denotes the daily working hours of an employee performing the distribution process. Average delivery times of a packet or mail in terms of data for two periods are shown in Figure 9.

**Figure 9.** The delivery time of a package or mail on behalf of delivery vehicles.

Package or mail delivery times vary according to the distribution personnel due to factors such as gender, age, professional experience, and the extent of the responsible area. However, as a result of comparing the processing times of the same driver with different vehicles, it is understood that the e-scooter performs the distribution process in a faster time. While the average delivery time of a package or mail delivery is 4.462 min with the e-scooter of the driver who performs face distribution, the average delivery time is 5.364 min with the same driver with other vehicles. The delivery time of a package or mail is calculated to be shortened by approximately 0.902 min in terms of the means used for distribution. This period provides many benefits to the administration in terms of time, cost, and energy, as it is calculated for the distribution of numerous packages or mail. The use of other vehicles (especially combivan) in package or mail distribution, the physical structures of distribution locations such as road conditions, and parking problems are among the factors. However, considering that such problems are minimal with the e-scooter, the e-scooter makes a significant contribution to the delivery time of the package. Distance measures according to the type of vehicle used for a package are calculated as 0.61, 0.56, and 0.16 km for e-scooter, motorcycle, and combivan, respectively. We conclude that the most advantageous distribution vehicle is the e-scooter, as these vehicles differ in distance according to traffic, road, and building configurations.

This study has some limitations. While considering only the amount of CO2 emissions that cause air pollution, some factors such as temperature, humidity, pressure, and wind speed are not considered. Another gap in the research's scope is that the physical (negative consequences of factors such as weight and height in driving) and psychological factors of the drivers who carry out the distribution business were not taken into account. It is requested by the administration to carry out the distribution operations upon the instruction given to the drivers. The physical structures of the region, such as road conditions, building configuration types, and parking problems in the plot areas considered in the e-scooter application, are not included in the study.

#### **5. Conclusions**

The postal or package delivery process is seeking faster, safer, and less costly options. Traditional distribution tools are lacking in meeting the necessary needs in today's world. Today, in addition to conventional vehicles, the e-scooter preference, which is increasingly used in the postal service, provides significant advantages. Official institutions/organizations that prefer e-scooter vehicles for package and mail distribution support them, with the results of scientific studies showing that they have obtained many important benefits in terms of economy, environment, and energy. In particular, a significant reduction is achieved in CO2 emissions. In a project funded by Intelligent Energy Europe (IEE), researchers concluded that using electric micro-mobility vehicles in urban transport has a positive effect on reducing CO2 emissions and saving energy [36]. Ruesch et al. emphasized that using e-scooters is economically cheaper than other delivery vehicles to increase mail or mail delivery efficiency [76]. A report by the Swiss Federal Energy Office (SFOE) concluded that using e-scooters contributes to low energy consumption and CO2 emissions [77]. In general, many positive benefits are obtained by choosing e-scooters for package or mail delivery, and e-scooter vehicles are evaluated in many ways compared to traditional delivery vehicles with actual data in this study.

The e-scooter application has been started in four important districts of Istanbul, Turkey's most cosmopolitan city. This study offers the opportunity to compare and analyze the data of the results of the e-scooter application with the data of traditional transportation vehicles. This study calculated optimum results by examining the factors affecting e-scooter transportation using the optimization Poisson regression model. In particular, the values or types of factors that are effective for the e-scooter driver to deliver the maximum number of packages or mail in the shortest time have been determined. The delivery time with the e-scooter was calculated to be 16.81% faster than the delivery time with conventional vehicles. In addition, the number of packages delivered on time, as the number of deliveries made with e-scooters increased by 58.53% compared to traditional vehicles.

This study includes three primary parameters: energy, cost, and environmental effects of e-scooter use, provided that it is a short distance for logistics purposes. Using these parameters, we concluded that e-scooters are more advantageous than other delivery vehicles in terms of time and product (number of packages). This study's findings show that the average journey distance and travel time using the e-scooter is in the range of 0.113–1.98 km and 3.06 min [78]. An exemplary study of global findings showed that the average journey distance and travel time using an e-scooter was between 1.56 km and 10 min. However, in many studies, it has been noted that e-bikes, which are among the micro-mobility vehicles, cover a distance of 3.5 km in approximately 17.5 min [79]. The method we have developed shows that e-scooter vehicles, especially in courier services such as mail or package delivery, offer significant advantages to administrators compared to other studies. Therefore, the benefits of using micro-mobility vehicles such as e-scooters instead of traditional vehicles used in short-distance transportation contain great potential. This research presents the advantages of using e-scooters in urban package or mail delivery operations and offers models for future applications, making a significant contribution to the literature.

**Author Contributions:** Conceptualization, A.A. and C.Ç.D.; methodology, A.A.; software, A.A.; validation, A.A., C.Ç.D. and H.˙ I.; formal analysis, A.A.; investigation, A.A. and C.Ç.D.; resources, A.A., Y.E.A. and C.Ç.D.; data curation, Y.E.A., H.˙ I., A.A. and C.Ç.D.; writing—original draft preparation, A.A.; writing—review and editing, A.A.; visualization, A.A.; supervision, Y.E.A., H.˙ I. and C.Ç.D.; project administration, Y.E.A., H.˙ I. and C.Ç.D.; funding acquisition, Y.E.A. and H.˙ I. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** Not applicable.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **Appendix A**

Number of packages or mails distributed monthly for 100 e-scooter drivers employed at PTT are represented in Figure A1.

**Figure A1.** The number of packages delivered by the e-scooter.

#### **References**


### *Article* **Analysis and Evaluation of Methods Used in Measuring the Intensity of Bicycle Traffic**

**Piotr K ˛edziorek 1, Zbigniew Kasprzyk 2, Mariusz Rychlicki <sup>2</sup> and Adam Rosi ´nski 3,\***

<sup>1</sup> Heller Consult, Chałubi ´nskiego 8, 00-613 Warsaw, Poland


**Abstract:** The work presents the methods of collecting and processing data with the use of devices used in individual measurement methods. Based on the collected video materials, the number of vehicles was determined, which at both measuring points actually exceeded each of the tested cross-sections of the bicycle path. More precise determination of the means of transport was divided into three categories: bicycles, electric scooters, and PT (personal transporters). The data collected with the use of each of the devices was properly processed and aggregated into a form that allows for their mutual comparison (they can be used to manage the energy of electric vehicles). Their greatest advantages and disadvantages were indicated, and external factors that had an impact on the size of the measurement error were identified. The cost of carrying out the traffic volume survey was also assessed, broken down into the measurement methods used. The purpose of this paper is to analyse and evaluate the methods used to measure bicycle traffic volume. Four different measurement methods were used to perform the practical part, which included such devices as a video recorder, microwave radar, perpendicular radar, and a meter connected to an induction loop embedded in the asphalt. The results made it possible to select a rational method for measuring the volume of bicycle traffic. The measurements carried out allow optimization of bicycle routes, especially for electric bicycles. The results indicate the method of physical counting of vehicles from video footage, thanks to which it is possible to achieve a level of measurement accuracy equal to 100%.

**Keywords:** bicycle traffic measurement; vehicle counter; induction loop; video recording; perpendicular radar; microwave radar; energy management

#### **1. Introduction**

The dynamic development of urban infrastructure, combined with pro-ecological trends, has made bicycles become increasingly popular means of transport. This phenomenon became ever more evident during the SARS-CoV-2 pandemic [1–4]. The more frequent selection of a bicycle as means of transport increasingly more often translates to limited air pollution emissions [5–7] and improved quality and level of life [8–12]. Dense populations in cities and the growing number of vehicles face engineers with a number of complex challenges [13–15]. Information on bicycle traffic volume is one of the inherent elements of designing new bicycle lane network sections. Regular traffic measurements have to be conducted in order to obtain such data.

Traffic measurements have been one of the issues taken into account in planning and designing road networks for many years [16–20]. The process applies not only to motor vehicles, but also to bicycles and even pedestrians. Decisions on expanding a road network or implementing solutions aimed at improving traffic safety are made based on traffic volume information collected during bicycle traffic measurements [21–27]. One example of recording such data is the bicycle traffic study in the capital city of Warsaw, conducted

**Citation:** K ˛edziorek, P.; Kasprzyk, Z.; Rychlicki, M.; Rosi ´nski, A. Analysis and Evaluation of Methods Used in Measuring the Intensity of Bicycle Traffic. *Energies* **2023**, *16*, 752. https://doi.org/10.3390/en16020752

Academic Editor: Katarzyna Turo ´n

Received: 12 December 2022 Revised: 2 January 2023 Accepted: 5 January 2023 Published: 9 January 2023

**Copyright:** © 2023 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

in accordance with the guidelines of the Municipal Roads Authority in Warsaw. The first such measurement was conducted in 2007, and reports on bicycle traffic in the capital of Poland have been regularly developed based on acquired data since 2014 [28,29]. The annually collected traffic volume information enable conducting an effective analysis of forming trends and observed changes in the movement methods among the residents within the studied area [30–34]. This information can be used to manage the energy of electric vehicles [35–39].

#### **2. Study Area**

There are several methods applied in measuring bicycle traffic volume [40–44]. Devices automatically counting vehicles are gaining popularity with technical development [45–54]. Currently, the most frequently applied study methods include the observation and recording of the number of vehicles directly by an operator at the measurement point and counting objects in office conditions, based on acquired video footage from the measurement location. The second of these methods was used within the data collection method for the report "Pomiary Ruchu Rowerowego 2020" (Traffic volume measurements 2020), which involved a four-day measurement in 36 locations throughout Warsaw, in June and July of the given year. Vehicle volume was measured during morning and afternoon rush hours, i.e., 7–9 a.m. and 4–7 p.m., and the obtained data were aggregated into 15-minute intervals. The detailed categorization attributes involved traffic direction, sex, means of transport used, infrastructure type, wearing a helmet, and clothing type (sports/casual). Traffic volume information recorded using cameras [55,56] was expanded in the report with data from 35 automatic bicycle traffic measurement (ABTM) points, which counted vehicles passing through a given section via an induction loop embedded in the asphalt.

In order to discuss the whole issue of the analysis and evaluation of methods used in measuring bicycle traffic volume, the following organization of the article was adopted. First, an introduction is provided along with an analysis of the literature in the area under discussion. Then, the various methods used to record objects in selected locations are characterized. The next section presents the results of the measurements made, along with their analysis. The article concludes with conclusions, followed by a bibliography.

#### **3. Materials and Methods**

Determining the accuracy of measuring instruments required conducting a series of carefully planned activities [57]. Study implementation was divided into several stages. The first one involved determining bicycle lane sections, which were to be subject to the bicycle traffic volume study. Next, the radars and the camera were inspected in terms of damages and correct functioning. It was also checked whether batteries intended for powering the devices were charged, and the SIM card that was a component of one of the controllers used was topped up. The next stage involved setting up stations equipped with the measuring instruments at the selected locations. The task was completed a day prior to the planned start of the study. The correctness of data acquisition by the devices was checked several times in the course of the measurement. The next day, after the measurements were completed, both stations were disassembled, and the collected materials were copied to spare drives. The next stage involved converting the data collected by all the measuring instruments to a single, consistent format that ensured convenient implementation of subsequent analyses. This was followed by describing the methods used to record the objects at the selected locations.

#### *3.1. Video Recording and Vehicle Classification*

Bicycle traffic volume measurements resulted in a total of 96 h of video footage from two selected locations. The actual number of vehicles passing through the measurement section was manually determined based on this data. Furthermore, the authors adopted a categorization of the counted objects by means of transport, namely, a bicycle, electric scooter, and personal transporter. The measurement involved using a CCTV IP camera that records the image in 1080 P (1920 × 1080 pixels) at 30 fps, which is characterized with viewing angles of almost 100◦ in the horizontal plane and 30◦ in the vertical plane. The aggregation level for recorded video materials was limited to video length below 15 min. The video recorder capability range was additionally expanded through the application of external devices. These involved a multifunctional microcomputer containing a set of inputs, outputs, and modules. It was connected to the camera via an RJ-45 network cable. The microcomputer was connected to a laptop via a LAN cable and entering a specific IP number in the web browser enabled video preview in real time. In addition, a topped-up SIM card and an RTSP port enabled remote video preview in real time.

#### *3.2. Microwave Radar*

A microwave radar (Figure 1) classifies vehicles into appropriate classes based on technical parameters defined during traffic volume measurement. Its recording and classification options can accumulate data on daily traffic. The applied FSK technology enables measuring vehicle length and speed. The FSK technology also allows the selection of locations that do not enable operation of certain radar groups due to, for example, protective barriers or other reflective structures. A detailed 8 + 1 vehicle class classification pursuant to TLS 2012 (technical conditions of delivery for road stations) is based on classification features regarding vehicle length, distance from counter, and acoustic detection. Technical parameters are determined according to the class of vehicles. Vehicle classification was done according to the A1 accuracy according to TLS (specification issued by BASt, the German Federal Highway Administration), in particular, based on vehicle length, distance from the meter, and acoustic detection. The device can be additionally integrated with a microphone, which enables determining, apart from traffic data, sound intensity at a given measuring point. Access to data was possible via cellular transmission and a manual data transfer through a Bluetooth connection with a laptop. Receiving and analysing data was possible owing to an installed application that enables obtaining not only information on the activity status of individual systems, but also receive warning messages, e.g., in the event of low battery level. Various variants for exporting files with traffic data were available. This enabled a simple assessment and visualization of the data using third-party engineering software.

**Figure 1.** Microwave radar installed next to a bicycle lane (own study).

#### *3.3. Perpendicular Radar*

A perpendicular radar (Figure 2) used in the study recorded traffic volume information using a radio frequency of 24.125 GHz (K-band). It is also capable of measuring traffic volume, average speed, the individual speed of a vehicle, lane occupancy, and vehicle classification. The device sends two radio beams at different angles in the horizontal plane. As a result, each of the vehicles crosses both beams just once. The distance of a vehicle from the radar is determined based on the opening angle and the time between sending the beam, its reflection from the vehicle, and its return to the device. The distance, speed, and travel direction of recorded vehicles are determined based on the occupancy duration of each beam and the time between crossing both beams. The vertical range of the sent beam is an angle of 65◦. The radar is capable of detecting vehicles within its field of view, regardless of the weather and lighting conditions. This is because the radar electromagnetic wavelength is significantly greater than that of light; therefore, a radar beam can penetrate rain, snow, and even fog. Heavy rain or snowfall may slightly affect microwave radar test results.

**Figure 2.** Perpendicular radar and CCTV IP camera (own study).

#### *3.4. Induction Loop*

Induction loops are used in many transport-related fields, one of which is the issue of measuring bicycle traffic volume. Induction loops can detect vehicles both on a bicycle lane and bicycle traffic sections of a road. A properly designed and constructed loop will not be impacted even by larger vehicles passing nearby [58]. Magnetic sensors are among the instruments used to detect two-track vehicles. Unfortunately, they do not correctly record all passing vehicles made of aluminium. In order to solve this issue, loop operation was based on exciting eddy currents in conductive bicycle parts, which leads to a decrease in loop inductance. The current flowing through the coils of an inductive loop sensor creates a sine-alternating magnetic field (primary field) around it. When this field encounters a metal object nearby, it induces eddy currents therein, which then generate a sine-variable magnetic field that weakens the primary field. As a consequence, the resultant field also has a reduced modulus and is phase-shifted [59].

Owing to the application of this technology, such factors as the electrical conductivity of the material that the vehicle is made of, the distance from the loop, and its surface are of major significance. In the event of connecting two separate induction loops to a counter, bicycle lane traffic volume can be measured in both directions [60].

#### **4. Results**

Traffic volume data were collected by two equally equipped measuring stations. The chapter below presents measurement results broken down by measuring stations.

#### *4.1. Measurement of Traffic Volume at the Bicycle Lane at Stefana Banacha Street*

The stations used to measure bicycle traffic as part of the practical section of the diploma thesis were located within the capital city of Warsaw. The first one was located in the Ochota district, at the intersection of Stefana Banacha and Zwirki i Wigury streets ˙ (52.210646 N, 20.987321 E) (Figure 3). It was characterized by the immediate vicinity of a public park, Pole Mokotowskie, and was located approximately 3 km from the city center. Such locations are often characterized by the highest bicycle traffic volume in the direction of the city center from 7 a.m. to 9 a.m. (morning rush hours) and from 3 p.m. to 6 p.m. (afternoon rush hours) [61]. Increased traffic during these periods is caused by residents commuting to work and returning to their places of residence. The studied section was located on the northwestern part of the intersection and covered a bicycle lane fragment perpendicular to Stefana Banacha street. The measurement section was 30 m away from the intersection center with traffic lights, and there was a bus stop in its proximity. There was a wide pavement made of concrete slabs next to the studied bicycle lane section, which was not structurally separated from it. Minor snowfall was recorded on the day preceding the measurements. The weather was cloudy with clear spells and patchy rain during the night two days prior to the study period. The pavement of the studied section was wet for most of the study, and the average temperature during the day was 5 ◦C.

**Figure 3.** Location of the measuring station at Stefana Banacha street (own study, based on www. openstreetmap.org), accessed on 19 January 2022.

#### 4.1.1. Video Recording

The source data from the perpendicular and microwave radars and the coded video footage data were appropriately aggregated and compared. The collected video footage was used in order to obtain the number of vehicles actually passing the measurement section. In the case of the station at Stefana Banacha street, the studied measurement section was passed by 1701 vehicles over the 48 h of measurements. According to the adopted classification, this included 1587 bicycles, 104 scooters, and 10 personal transporters. Over the two days, 881 vehicles (including 815 bicycles) were recorded travelling in direction 1, i.e., towards Grójecka street, and 820 vehicles (including 772 bicycles) in direction 2, i.e., towards Zwirki i Wigury street (Figure ˙ 3). On 15 December, the section was crossed by 823 vehicles (including 759 bicycles) and on 16 December by 878 vehicles (including 828 bicycles). Tables 1 and 2 contain information on the structure type of recorded vehicles, broken down by directions and measurement days.

**Table 1.** Vehicle structure type at the bicycle lane along Stefana Banacha street, broken down by directions (own study).


**Table 2.** Vehicle structure type at the bicycle lane along Stefana Banacha street, broken down by measurement days (own study).


#### 4.1.2. Microwave Radar

During the 48 h of the measurement, the device recorded 1554 vehicles, which was 8.64% less than the number of vehicles that actually crossed the studied section. Based on the comparative analysis of data in the vehicle-by-vehicle structure with the video footage, it was concluded that the almost 9% difference resulted from the failure of the device to count a large number of scooters and personal transporters. Because the microwave radar does not classify such means of transport as a scooter or a personal transporter, all recorded vehicles were treated as bicycles in the course of further processing. The device also recorded the presence of three vehicles in the car category and nine in the motorcycle category. Due to the nature of the study, which involves only traffic along the bicycle lane, the aforementioned 12 records were classified as categorization errors and included in the comparative statement as bicycles. Under such an assumption, the device in question recorded 1554 bicycles, which compared to the actual number of bicycles of 1587 constituted a basis to adopt the estimated device measurement accuracy at almost 97.9%. It should also be emphasized that there were cases where scooters were recorded and classified by the microwave radar as bicycles, and where bicycles were not recorded at all. The analyses of the video recording indicate that such situations could have occurred for 30 to 40 vehicles, which amounts to approximately 2% of the entire research population. After taking these cases into account, the estimated measurement accuracy of the microwave radar in the bicycle category was determined at 96.0%. The measurement accuracies for direction 1 (Grójecka street) and direction 2 (Zwirki i Wigury street) compared to coded ˙ vehicles amounted to 99.0% and 93.4%, respectively. The device beam was directed not only on the bicycle lane, but also the pavement. Owing to the advancement of the used software, the device did not register pedestrians crossing the studied section in the output data. The graphs (Figures 4 and 5) below show the traffic volume measured at particular hours over the two measurement days. The first one shows a comparison of the number of

vehicles recorded by the microwave radar and the actual number of bicycles. The second one illustrates a comparison of the number of vehicles measured with the microwave radar and the number of vehicles crossing the measurement section.

**Figure 4.** Comparison of the number of vehicles recorded by the microwave radar with the actual number of bicycles (own study).

**Figure 5.** Comparison of the number of vehicles recorded by the microwave radar with the actual number of vehicles crossing the measurement section (own study).

Besides measuring the number of vehicles, the radar also measured their speed; in the case of the station at Stefana Banach street, the average value was 15.22 km/h. Factors that could have impacted this result include the vicinity of an intersection with traffic lights, which was located approximately 30 m from the measurement section. In addition, the studied bicycle lane section passed a pedestrian crossing. People moving with the use of vehicles should exercise particular caution due to the proximity of a bus stop and the resulting increased pedestrian traffic. The graph (Figure 6) below shows the distribution of vehicle speed recorded by the microwave radar.

**Figure 6.** Distribution of vehicle speed at the bicycle lane along Stefana Banacha street (own study).

#### 4.1.3. Perpendicular Radar

Over the two measurement days, the device registered 2187 records, which was 28.6% more than the actual number of vehicles and 37.8% more than the actual number of bicycles (Figure 7). After analysing video footage, it was concluded that the main reason behind the said differences was the counter recording pedestrians moving along the pavement adjacent to the bicycle lane. This happened despite setting out the detection area in the device calibration software. It should be noted that some people also moved directly on the bicycle lane. The device recorded the speed and length of objects crossing the measurement section, and thus, it was possible to discard records that did not fall within the expected value ranges. However, due to the nature of the conducted study, it was decided not to apply this step since the adopted analysis objective was to compare the data not modified in any way. The relative errors caused by the excessive number of recorded vehicles was 31.7% for direction 1 (Grójecka street) and 44.2% for direction 2 (Zwirki i Wigury street), ˙ relative to the actual number of bicycles.

#### 4.1.4. Induction Loop

The last vehicle recording method applied as part of the bicycle traffic volume study was an induction loop embedded in asphalt combined with a counter. A camera, which was one of the elements of the measuring station, was installed in such a way so that its recorded image displayed the total daily and annual numbers of cyclists on the pylon screen. Due to the contrast between the display (an integral part of the pylon) and the surroundings, the number of cyclists was visible on video footage only between 9.30 a.m. and 2.30 p.m. The cameras recorded the image through the entire 15 and 16 December and until 11 a.m. on 17 December. As a result, based on the available traffic volume data, it

was possible to calculate the number of vehicles recorded by the induction loop on both measurement days. The first conducted step was to determine the total number of recorded bicycles for each of the three days and for the entire year 2021. The calculations adopted the number of cyclists for 10 a.m. on 15–17 December. Next, in order to obtain the number of cyclists at midnight between the analysed days, the daily recorded volume was subtracted from the displayed annual volume. The difference between the annual number of cyclists at midnight between successive days enabled calculating the total number of cyclists recorded in one day, both for 15 and 16 December. It should be emphasized that the conducted measurement did not involve a division into travel directions. The counter connected to the induction loop recorded 775 vehicles on 15 December and 855 on 16 December.

**Figure 7.** Comparison of the number of vehicles recorded by the perpendicular radar with the actual number of bicycles (own study).

Therefore, over the two measurement days, the counter connected with the induction loop recorded 1630 vehicles. This amounted to 95.5% of the actual number of vehicles that crossed the measurement section. Compared to the actual 1587 bicycles, the device recorded 43 extra vehicles, which means a relative error of 2.7%. Similar to the microwave radar, there were situations where the counter connected to the induction loop failed to record passing bicycles and situations where it recorded passing scooters and personal transporters. Based on video analysis, it was concluded that the counter recorded 96% of the actual number of bicycles passing the measurement section, as well as 93.9% of the scooters and personal transporters.

#### *4.2. Measurement of Traffic Volume at the Bicycle Lane at Swi˛ ´ etokrzyska Street*

The second of the measurement locations was located in the Sr´ ódmie´scie district at 30 Swi ˛ ´ etokrzystka street, in the inner center of Warsaw (52.234959 N, 21.006358 E) (Figure 8). The studied section was located in the immediate vicinity of a pedestrian walkway, separated from the bicycle lane with a greenery belt in some places. An oblong residential building with several stores and eateries on the ground floor was located next to the section. Such locations are usually characterized by an above-average share of electric scooters in the vehicle type structure. Between the bicycle lane and the road, there was a paid car parking space dedicated for perpendicular parking. Due to the proximity of a parking meter and a minor park, one could expect increased pedestrian traffic in the vicinity of the studied area. The nearest intersections were located 120 m from the measurement section towards the east (with Marszałkowska street) and approx. 140 m to the west (with Raoul Wallenberg street). As with the measurements at Stefana Banacha street, the asphalt bicycle lane pavement was wet during the measurement, with no rainfall on either day. The average air temperature during the day, both on 15 and 16 December, was approx. 6 ◦C. Despite the measurement being conducted in the winter season, no snowfall was recorded during its course. The Figures below illustrate the location of the measurement station and photos showing its equipment.

**Figure 8.** Location of the measuring station at Swi ˛ ´ etokrzyska street (own study, based on www. openstreetmap.org accessed on 12 December 2022).

#### 4.2.1. Video Recording

According to the adopted assumption of conducting comparative analyses, the number of vehicles hand-coded from video footage fully represented the actual number of vehicles crossing the section located within the studied bicycle lane section. Over the two days of measurement covering the bicycle lane located by Swi ˛ ´ etokrzyska street, namely, the section between Marszałkowska street and Aleja Jana Pawła II, the authors recorded 1786 vehicles, including 1445 bicycles, 327 scooters, and 14 personal transporters. In direction 1, namely, Aleja Jana Pawła II, the studied section was crossed by 1158 vehicles (including 957 bicycles) and in direction 2 (Marszałkowska street), the equipment recorded 628 vehicles (including 488 bicycles). Tables 3 and 4 show the structure type of vehicles recorded in the bicycle lane at Swi ˛ ´ etokrzyska street, broken down by directions and measurement days.


**Table 3.** Vehicle structure type at the bicycle lane along Swi ˛ ´ etokrzyska street, broken down by directions (own study).

**Table 4.** Vehicle structure type at the bicycle lane along Swi ˛ ´ etokrzyska street, broken down by measurement days (own study).


#### 4.2.2. Microwave Radar

Over the 48 h of measurement (Figures 9 and 10), it recorded 1651 vehicles, including 1588 classified as bicycles, 57 at motorcycles, and 6 as cars on the dedicated bicycle path. Based on video footage analysis and according to the adopted procedure, 63 records involving motorcycles and cars were determined as categorization errors and included in further statements as bicycles. Under such an assumption, the counter recorded 1651 bicycles, which, compared to the 1445 bicycles that actually crossed the studied section, amounted to an absolute measurement error of the device of 14.3%. This was largely due to the fact that as many as 327 scooters as 14 personal transporters were recorded within the studied bicycle lane section, which translated to shares in the structure type of 18.3% and 0.8%, respectively. Similar to the measurement point at Stefana Banacha street, there were cases where electric scooters were classified by the counter as bicycles. The microwave radar placed on the bicycle lane at Swi ˛ ´ etokrzyska street recorded almost 79.2% of scooters and 95.6% of bicycles that crossed the measurement section.

Besides recording the very number of vehicles, the radar also measured their speed; in the case of the station at Swi ˛ ´ etokrzyska street, the average value was 20.14 km/h. This number was impacted by such factors as a good-quality asphalt pavement, distances from traffic flow intersections higher than 50 m in both directions, and a greenery belt separating the bicycle lane from the walkway. The graph (Figure 11) below illustrates the number of vehicles moving at particular speeds, rounded to whole numbers.

#### 4.2.3. Perpendicular Radar

Over the two days, the device recorded 1833 vehicles. However, the measurement was conducted without categorization. The device relative error caused by an excessive number of records amounted to 2.6% for all vehicles and 26.9% for bicycles. The studied bicycle lane section was separated from the walkway by a greenery belt. After analysing the video footage, it was concluded that due to the proximity of the car park, the studied section could have been crossed by pedestrians several dozen times. It was still a number several-fold lower than in the case of the measurement at Stefana Banacha street, where the walkway was not separated from the bicycle lane in any manner and the measurement station was located near a bus stop. In the graph (Figure 12) illustrating traffic volume over the two measurement days, the greatest differences in the number of vehicles recorded with the perpendicular radar and the actual number of vehicles can be seen between 2 and 3 a.m. on 15 December. After analysing the video footage, it was concluded that the measurement section was crossed by one bicycle, six electric scooters, and one pedestrian during that time. All other records were most likely recorded by the beam reflected from passing vehicles or other infrastructure elements located near the measuring device. The duration from 8 a.m. to 2 p.m. was where a significantly greater number of redundant objects recorded by the perpendicular radar was observed. Pedestrians constituted for the largest group during that time.

**Figure 9.** Comparison of the number of vehicles recorded by the microwave radar with the actual number of bicycles (the measuring station at Swi ˛ ´ etokrzyska street) (own study).

#### 4.2.4. Induction Loop

The traffic volume measured using an induction loop within the studied bicycle lane section at Swi ˛ ´ etokrzyska street was calculated using similar conversions, as in the case of the point at Stefana Banacha street. This action was forced by a large contrast between the numbers displayed on the pylon and the surroundings at night, which prevented reading the day traffic volume based on available video footage.

The counter connected with the induction loop recorded 1901 vehicles during the traffic volume measurements. This number is 31.6% higher than the actual number of bicycles, equal to 1445, and 6.4% higher than the actual number of vehicles crossing the section, which amounted to 1786. Additional numerical juxtapositions were found based on traffic data for 10 a.m. on both measurement days. On 15 December, the total number of vehicles recorded from midnight until 10 a.m. was 172 by the induction loop, whereas 166 crossed the measurement section. This means the relative error of the device was equal to 3.6%. On 16 December, this number was equal to 180 for the induction loop counter and 206 for vehicles hand-coded from video footage. Based on this part of the data, the vehicle measurement accuracy was 87.3%. Due to the observed irregularity, further steps were taken aimed at precise observation of the number of vehicles displayed on the pylon. In the course of the analyses, the authors noticed that there were situations at night where one bicycle was counted as several dozen bikes by the counter. Due to the high contrast of the display, it was impossible to accurately determine the number. The situation in question may result from a software error. In light of these circumstances, it was assumed that the

data collected by the induction loop in the bicycle lane section at Swi ˛ ´ etokrzyska street were not a reliable reference point to assess this measurement method.

**Figure 10.** Comparison of the number of vehicles recorded by the microwave radar with the actual number of vehicles (the measuring station at Swi ˛ ´ etokrzyska street) (own study).

**Figure 11.** Distribution of vehicle speed in the bicycle lane along Swi ˛ ´ etokrzyska street (own study).

**Figure 12.** Comparison of the number of vehicles recorded by the perpendicular radar with the actual number of bicycles (the measuring station at Swi ˛ ´ etokrzyska street) (own study).

#### *4.3. Summary*

The practical part of the described experiment involved collecting, processing, and comparing traffic volume data for the selected bicycle lane sections. The authors also described the nature of traffic in both locations. Each of the four used measurement methods differed in terms of data collection technology and data recording accuracy. A situation where, for various reasons, measurement data might not fully reflect the actual traffic volume was possible in the case of each method. Such cases include measuring equipment theft, physical device failure, interruption or absence of an active instrumentation power supply source, and the covering of the measurement section by a different object, e.g., a delivery truck. Traffic data covered by this study were collected under favorable weather conditions; however, due to the data collection method by the counters, these factors should not translate to lower or higher measurement accuracy. Tables 5 and 6 show processed and aggregated traffic volume data collected on 15 and 16 December by the measurement points located at Stefana Banacha and Swi ˛ ´ etokrzyska streets, broken down by measurement method.

**Table 5.** List of all vehicles recorded at the Stefana Banacha street point, broken down by applied measurement method (own study).



**Table 6.** List of all vehicles recorded at the Stefana Banacha street point, broken down by applied measurement method (own study).

#### **5. Conclusions**

The purpose of this study is to analyse and evaluate the methods used to measure bicycle traffic volume. Four different measurement methods were used in the study, which included devices such as a video recorder, microwave radar, perpendicular radar, and a meter connected to an induction loop embedded in the asphalt. Measurement through physical counting of vehicles based on video footage should be classified as a method that enables achieving a measurement accuracy of 100%. A properly trained operator is able to categorize passing cyclists according to numerous features, such as sex, bicycle type, helmet worn (or not), sports clothing, or movement direction. In addition, this measurement method enables applying significantly broader categorization options, e.g., vehicle type. This is of particular importance in the face of technological development and the increasing availability of such devices as electric scooters or other electric personal transporters. The obtained measurements can be used to manage the energy of electric vehicles. This measurement method enables studying traffic volume in many locations simultaneously.

The microwave radar was able to record over 95% of the bicycles crossing a measurement section. The applied technology does not lead to errors that involve identifying passing pedestrians as vehicles. The beam emission method enables determining the movement distance and direction of a passing vehicle, its speed, and its distance from the radar. This method is able to moderately record and identify electric scooters and personal transporters. A set of categories that can be assigned by the counter used in the study is limited only to bicycles in the case of a bicycle lane measurement. Furthermore, there may be situations where a person riding a scooter is classified as a cyclist. In light of the above, a microwave radar should achieve the greatest measurement accuracy in sections with the least number of vehicles other than bicycles. In the case of a more diverse type structure, vehicle identification accuracy of the device will decrease, and the error in determining the actual number of cyclists will increase.

The perpendicular radar device measured vehicle traffic volume with the lowest accuracy among the applied methods. The radar beam reflected off objects other than bicycles in many situations, which resulted in a large number of redundant records. Due to the design and data processing method based on beam parameters, the radar recorded almost all traffic, including pedestrians and cyclists, as well as cars passing at a distance of more than 4 m from the bicycle lane. Such situations happened despite setting out a measurement area in the radar-dedicated application. The device is not strictly intended for measuring bicycle traffic volume, which was reflected in practice. A relative measurement error between 25% and 40% is, in the case of many types of possible studies, a factor that disqualifies this measurement method in terms of recording bicycles.

The induction loop embedded in asphalt recorded over 95% of the bicycles crossing the measurement section. There were situations in the course of the study where bicycle presence was not counted, as well as cases where a bicycle was double-counted. Due to the work effort required to install a measurement station, the described measurement method is suitable for continuous, long-term measurement. Owing to a constant power supply and a design resistant to weather conditions, it satisfies fundamental requirements in a manner consistent with the assumptions. Apart from the bicycles themselves, the counter also recorded more than 90% of electric scooters and personal transporters. This means that, just like with a microwave radar, a higher share of non-bicycle vehicles may hinder determining the actual bicycle number. Given the vast measurement period while applying this method, studies aimed at controlling vehicle counting correctness should also be considered. A detailed observation of data displayed in one of the pylons participating in the study enables the conclusion that despite a relatively correct bicycle lane traffic recording, an incorrectly developed software may lead to displaying data encumbered with a significant error.

The measurements conducted made it possible to determine the advantages and disadvantages of the various methods of measuring bicycle traffic volumes. This can be used in the selection of a bicycle traffic volume measurement method for a specific bicycle route. The results of the bicycle traffic volume measurement allow the design and optimization of bicycle routes, including, in particular, for electric bicycles. This allows, on the one hand, for the rational placement of stations for the provision of electric bicycles, and on the other hand, for the optimization of the energy management system used for charging electric bicycles. The perpendicular radar measurement method showed the lowest accuracy in measuring the volume of bicycle vehicles among other methods in the same field. The measurement method of physically counting vehicles from video footage achieved a measurement accuracy level of 100%.

**Author Contributions:** Conceptualization, P.K. and Z.K.; methodology, P.K., Z.K. and M.R.; software, P.K. and Z.K.; validation, P.K., Z.K. and A.R.; formal analysis, P.K., Z.K. and M.R.; investigation, P.K., Z.K., M.R. and A.R.; resources, P.K. and Z.K.; data curation, P.K. and Z.K.; writing original draft preparation, P.K, Z.K. and M.R.; writing review and editing, P.K., Z.K., M.R. and A.R.; visualization, P.K. and Z.K.; supervision, P.K. and Z.K.; project administration, P.K., Z.K. and A.R; funding acquisition, A.R. All authors have read and agreed to the published version of the manuscript.

**Funding:** This work was co-financed by Military University of Technology under research project UGB 737.

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** The data presented in this study are available on request from the corresponding author.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


**Disclaimer/Publisher's Note:** The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.
