Waste Management 33 (2013) 793–806

Contents lists available at SciVerse ScienceDirect

Waste Management

journal homepage: www.elsevier .com/locate /wasman

Operation costs and pollutant emissions reduction by definition of newcollection scheduling and optimization of MSW collection routes using GIS. Thecase study of Barreiro, Portugal

Zdena Zsigraiova a, Viriato Semiao b,⇑, Filipa Beijoco b

a Department of Furnaces and Thermal Technology, Technical University of Košice, Slovakiab TU Lisbon, Instituto Superior Tecnico, IDMEC, Dept. Mech. Engn., Av. Rovisco Pais, P1049001 Lisbon, Portugal

a r t i c l e i n f o a b s t r a c t

Article history:Received 7 February 2012Accepted 25 November 2012Available online 23 December 2012

Keywords:Glass-wasteGIS route optimizationCosts reductionEmissions abatementDynamic load

0956-053X/$ - see front matter � 2012 Elsevier Ltd. Ahttp://dx.doi.org/10.1016/j.wasman.2012.11.015

⇑ Corresponding author. Tel.: +351 218417726; faxE-mail address: ViriatoSemiao@ist.utl.pt (V. Semia

This work proposes an innovative methodology for the reduction of the operation costs and pollutantemissions involved in the waste collection and transportation. Its innovative feature lies in combiningvehicle route optimization with that of waste collection scheduling. The latter uses historical data ofthe filling rate of each container individually to establish the daily circuits of collection points to be vis-ited, which is more realistic than the usual assumption of a single average fill-up rate common to all thesystem containers. Moreover, this allows for the ahead planning of the collection scheduling, which per-mits a better system management. The optimization process of the routes to be travelled makes recourseto Geographical Information Systems (GISs) and uses interchangeably two optimization criteria: totalspent time and travelled distance. Furthermore, rather than using average values, the relevant parame-ters influencing fuel consumption and pollutant emissions, such as vehicle speed in different roadsand loading weight, are taken into consideration. The established methodology is applied to the glass-waste collection and transportation system of Amarsul S.A., in Barreiro. Moreover, to isolate the influenceof the dynamic load on fuel consumption and pollutant emissions a sensitivity analysis of the vehicleloading process is performed. For that, two hypothetical scenarios are tested: one with the collected vol-ume increasing exponentially along the collection path; the other assuming that the collected volumedecreases exponentially along the same path. The results evidence unquestionable beneficial impactsof the optimization on both the operation costs (labor and vehicles maintenance and fuel consumption)and pollutant emissions, regardless the optimization criterion used. Nonetheless, such impact is partic-ularly relevant when optimizing for time yielding substantial improvements to the existing system:potential reductions of 62% for the total spent time, 43% for the fuel consumption and 40% for the emittedpollutants. This results in total cost savings of 57%, labor being the greatest contributor, representing over€11,000 per year for the two vehicles collecting glass-waste. Moreover, it is shown herein that thedynamic loading process of the collection vehicle impacts on both the fuel consumption and on pollutantemissions.

� 2012 Elsevier Ltd. All rights reserved.

1. Introduction gaining particular relevance, and waste management has emerged

For sustainability reasons modern societies have to mitigate theenvironmental impact of their daily activities, which are potentialgenerators of ecological imbalances. Nowadays, it is commonly ac-cepted that no efforts should be spared to minimize the negative im-pact on the environment of the human activities and recourse mustbe made to all possible tools (technology, economy, energy, socialsciences, and healthcare and scientific research) to attain suchobjective.

Amongst the vast range of daily human activities that may beharmful to the environment, waste generation and disposal are

ll rights reserved.

: +351 218475545.o).

as a crucial activity to mitigate such hazardous effects.Because waste management is an intensive energy-consuming

activity, aiming at preserving sustainability of life on earth and cre-ating better habitats, efficiency improvement of waste manage-ment systems and related processes is a priority. In fact, highlyefficient systems allow maintaining both adequate favorable con-ditions to human life and availability of resources, and simulta-neously preserving climate.

From the economic point of view, municipalities are commonlystakeholders (and, not seldom, stockholders) in waste manage-ment activities. Hence, and although profit may not be a primarygoal, all the costs must be taken into account and reduced when-ever possible, so that benefits can be achieved in both financialperformance and environmental trustworthiness.

794 Z. Zsigraiova et al. / Waste Management 33 (2013) 793–806

Municipal solid waste (MSW) collection and transportation is awaste management activity that contributes substantially to haz-ardous environmental impact due to fuel consumption and conse-quent pollutant emissions, and can absorb a considerable share ofthe total system budget (Dogan and Suleyman, 2003; Ghose et al.,2006; Nguyen and Wilson, 2010). Although fuel consumption islargely determined by the vehicle’s travelled distance, other factorssuch as speed, acceleration/deceleration, road inclination and vehi-cle load can affect it considerably (Tavares et al., 2008; Kuo, 2010;Kuo and Wang, 2011; Xiao et al., 2012).

In vehicle fleet systems, as MSW collection and transportationcan be considered, the environmental impact of operation is, tosome extent, a controllable function of vehicle routing and sched-uling decisions. In spite of that, only a few works have addressed sofar the environmental impact in the optimization of vehicle fleetrouting and scheduling (Dessouky et al., 2002; Armstrong andKhan, 2004).

Vehicles assigned to collect MSW, e.g. with a self-compactor orwith a crane arm, represent considerable operating expenses.While loading and unloading containers, trucks have to keep theirengines running, producing constant exhaust emissions. The non-transportation time, including time spent for load/unload opera-tions and other idling times can reach as much as 50% of the totaltime of waste collection in cities with high population density andhigh traffic congestion (Faccio et al., 2011). Therefore, it is crucialfor the reduction of operating costs and vehicle emissions to designefficient collection strategies in terms of the optimized vehiclerouting, reducing both the transportation time and the numberof load/unload stops (McLeod and Cherrett, 2008).

Unloading waste collection containers that are only partiallyfull seems an unnecessary waste of resources. One way to reducecosts in MSW collection and transportation activities is to optimizethe number and location of the collection containers and their col-lection frequency (Kulcar, 1996; Chang and Wei, 1999; Badran andEl-Haggar, 2006; Johansson, 2006; Lin et al., 2010; Faccio et al.,2011).

The waste collection problem is stochastic by nature as theamount of MSW is highly variable and the accumulation of wastedepends on several factors such as the number of inhabitants percontainer, GDP per capita, lifestyle, and season (Nuortio et al.,2006). In spite of that, there is a generalized lack of reliable infor-mation on the mass and volume of waste in individual containers.Most of the works reported in literature used constant average val-ues for the model parameters within selected areas or even in thewhole area under study (e.g. Teixeira et al., 2004). To overcomethis, Nuortio et al. (2006) used the average accumulation rate ofwaste in each container type estimated separately based on thehistorical weight and route. Johansson (2006), in turn, introduceda dynamic scheduling and routing for waste collection based onthe real-time information provided by a level sensor placed insidethe containers. Faccio et al. (2011) introduced an innovative frame-work to implement the modern traceability devices in waste col-lection (e.g. volumetric sensors, RFID, GPRS and GPS) andpresented a multi-objective routing model for waste collectionbased on the integration of real-time traceability data inputs,including real-time bin level status and real-time vehicle position.With these approaches the obtained results for routing optimiza-tion are more reliable than those obtained with static methods.Although the implementation requires a considerable investmentin equipments, it can be compensated by the gained technicaland economic benefits (Faccio et al., 2011).

Optimization of routes for waste collection and transportationis another way to reduce operation costs that can be complemen-tary to the previous one. Such optimization process can be per-formed either alone while keeping unchanged the existingnetwork of containers (Komilis, 2008; Simonetto and Borenstein,

2007), or by combining both strategies of route optimization andcontainers network adjustments by replacing and reallocatingthem more adequately (McLeod and Cherrett, 2008; Zamoranoet al., 2009).

Vehicle routing for waste collection can be formulated either asan Arc Routing Problem (ARP) (Dror, 2000), where vehicles have totraverse a set of streets, or as a Node Routing Problem (NRP), wherevehicles have to visit a number of points (Kulcar, 1996).

The graph theory defines a transportation network (a graph) asa system of interconnected elements, edges and nodes, the edges(or arcs) representing street segments and the nodes standing forjunctions. Thus, the ARP is a natural way to model waste collectionproblems in cases where most or all containers along a given streetsegment must be collected at the same time, and most of the streetsegments must be traversed by the collection vehicle as in verydensely populated areas. If there is a vehicle capacity constraint,such problems may be modelled as a Capacitated ARP (CARP)(Amponsah and Salhi 2004; Wøhlk, 2008).

When applying NRP specific weights of waste are identified ascollections at a number of specified points on the network. Noderouting seems to be more appropriate for collection of recyclablesallowing for considering each container separately (with its spe-cific fill-up rate data when available) and exact stop points of thevehicles to collect each container, hence, providing a more detailedmodelling (Nuortio et al., 2006).

The node routing counterpart of the CARP is the Vehicle RoutingProblem (VRP). The basic VRP is one of the most widely studiedproblems in combinatorial optimization, where the required ser-vice is performed by a fleet of vehicles (Eksioglu et al., 2009). Itsapplication to waste collection and transportation systems com-prises a certain cost in making a vehicle to travel between the de-pot and the collection points. Such cost is minimized for a givenroute (sequence of all visited collection points by a certain vehicle,starting and ending at a depot). Hence, the VRP objective is to serveall collection points, optimizing the total cost of the routes of allvehicles in the fleet (one route per vehicle). When routing is con-strained by the vehicle’s capacity, working hours or other factors,the problem falls into the class of Capacitated VRP (Longo et al.,2006).

As the routing problem is computationally very demanding(belonging to the set of NP-Hard problems), and cannot be solvedby optimal (exact) methods for large practical systems, heuristics(e.g. Nearest Neighbour Insertion, Giant Tour based algorithms orClarke & Wright’s Savings algorithm) and meta-heuristics (e.g.Ant Colony algorithms, Tabu Search, Simulated Annealing orGenetic algorithms) are used for this purpose yielding anas-optimal-as-possible solution (Laporte et al., 2000). For example,Kuo and Wang (2011) proposed a method for solving the VRP min-imizing the fuel consumption applying Tabu Search to optimize therouting plan.

Whatever are the approaches used to solve the routing prob-lems and the collection frequencies, MSW management systemsare strongly of spatial nature (treatment and disposal facilitieslocations, containers distribution and vehicle routing). Therefore,it is possible, and advisable, to take advantage of new technologieslike GIS that proved to be a powerful tool with ability to providethe detailed spatial information and its effective handling (Keenan,1998). Several works applied GIS to general vehicle traffic studies(e.g. Taylor et al., 2000; Ericsson et al., 2006) or to waste collectionbins distribution and vehicle routing (e.g. Tarantilis et al., 2004;Ghose et al., 2006; Tavares et al., 2009; Zamorano et al., 2009 orZsigraiova et al., 2009). It is now widely accepted in the wastemanagement community that effective decision making requiresimplementation of vehicle routing techniques where GIS can alsobe easily and advantageously integrated (e.g. Apaydin and Gonullu,2008 or Tavares et al., 2009).

Z. Zsigraiova et al. / Waste Management 33 (2013) 793–806 795

The present work is focused on the collection and transporta-tion of a glass-waste system in a Portuguese selected area. In Por-tugal the collection and transportation systems of MSW consist ofnetworks of containers for both undifferentiated collection andselective collection of recyclables. The containers for recycledmaterials like glass, plastics and paper/cardboard are organizedin the so-called ecopoints. Each ecopoint normally comprises onecontainer for each of the previously mentioned fractions; never-theless, some ecopoints can have multiple containers if necessary,i.e. more than one container for a given fraction, or no containersfor a particular waste type. In general, fleets of heavy-duty vehiclesto collect different types of waste (undifferentiated or recyclable)fulfill predefined itineraries with a predefined collection frequency.While undifferentiated solid waste collection is under the respon-sibility of municipalities, there are multi-municipal or inter-municipal systems for recyclable waste, managed separately byprivate companies.

Within this scope, this paper presents an innovative methodol-ogy for the reduction of the total operation cost (including partialcosts for vehicles maintenance, labor and fuel) involved in thewaste collection and transportation. Its innovative feature liesmainly in combining collection scheduling based on historical dataof the individual container fill-up rates with vehicle route optimi-zation in GIS environment. The proposed collection scheduling ismore realistic than the usual assumption of a single average fill-up rate value applied to all the system containers and allows formore adequate and ahead collection planning. As far as GIS is con-cerned, it significantly simplifies the vehicle routing optimizationprocess, representing and visualizing efficiently and convenientlythe obtained results. In addition, for calculating the fuel consump-tion and intrinsic amount of released pollutant emissions origi-nated from vehicles operation during waste collection andtransportation, the main influencing factors and their variationare taken into consideration, rather than using their average val-ues. Such factors are the different vehicle speeds correspondingto different types of roads, and the changing loading weight as aconsequence of picking-up waste at each visited container. Tothe best of the present authors’ knowledge, the methodology pro-posed herein, combining route optimization in GIS environmentwith prior collection scheduling, has not previously been reported.The established methodology was applied to study the system ofthe glass-waste collection and transportation of the waste collec-tion of Amarsul S.A., in Barreiro. As direct effects of the above com-bined optimization, expected fuel consumption decrease and,consequently, a decrease of the pollutant emissions (CO, CO2,NOx and PM) of the heavy-duty vehicles in the fleet assigned forthe task are analyzed.

The particular objectives of the present work are:

(i) To optimize a system for glass-waste collection and trans-portation by the definition of both the best collection fre-quencies (according to the individual fill-up rate of eachcontainer) and the shortest vehicle collection routes (accord-ing to the optimization criterion applied), taking intoaccount the available vehicles in the fleet and the main fac-tors influencing fuel consumption and related emissions(different speeds in different roads and changing loadingweight).

(ii) To quantify the impact of the system optimization on thereduction of operation costs (maintenance, labor and fuelconsumption) and emissions.

(iii) To study the influence of vehicles dynamic load on fuel con-sumption and pollutant emissions, namely CO, CO2, NOx andPM (particulate mater), by comparing their values alongequivalent routes with constant and dynamic loads.

2. Methodology

An optimization process aims at ensuring the effective use ofthe available resources: material, financial and human. In this par-ticular case, the applied methodology brings together a GIS tool forroute optimization, taking into account the available vehicles inthe company fleet and a set of ecopoints to be visited, and the def-inition of the best collection frequencies for each glass-waste con-tainer individually, according to its statistical fill-up rate historicaldata (SIGRSAMARSUL, 2010), keeping unchanged the existing net-work of containers.

2.1. Scheduling the waste collection circuits of the studied area

The municipality of Barreiro, with an area of 31.6 km2 and morethan 80,000 inhabitants, was selected as the case study for thepresent research. Integrated in Setubal district, the municipalitybelongs to the metropolitan area of Lisbon, located on the southbank of the river Tagus, as illustrated in Fig. 1. It lies about40 km from the capital, Lisbon, and about 35 km of Setubal, thecapital of the district.

In Barreiro municipality the collection and treatment of recycla-ble waste is provided by the private company Amarsul S.A. The col-lected waste is transported to a sorting station (see Fig. 1), where itis prepared for further treatment or reuse.

The waste is collected from returnable containers with capaci-ties of 2.5 or 3 m3 that are distributed along 240 ecopoints spreadout through the entire municipality, as depicted in Fig. 1. The con-tainers are emptied by compacting vehicles in the case of paperand plastics, or by open trucks for glass.

There are presently five circuits defined for collection of recyc-lables covering the whole area of the Barreiro municipality. Eachcircuit corresponds to a different set of ecopoints, grouped accord-ing to the pre-determined collection frequencies of the containers.As far as the collection frequency is concerned, each circuit has aspecific frequency for paper, plastics and glass, coming out fromlong-standing observation experience. For glass-waste, the collec-tion frequencies by circuit are presented in Table 1.

It can be noticed from this table that, depending on the circuit,the glass-waste is collected every 3 (circuits A–C) or 4 weeks (cir-cuits D and E), and the cycle of collection frequencies lasts12 weeks.

The available collection vehicle fleet includes two heavy-dutytrucks with open cases, vehicles A and B, capable of loading ca. 5and 7 tonnes of glass-waste, respectively. The trucks characteris-tics can be found in Table 2.

As far as the vehicle routing is concerned, since the actual col-lection routes (i.e. the sequence of visited ecopoints) of each circuitare not fixed, the drivers have liberty in choosing the sequence ofthe ecopoints to be visited.

In spite of the long-standing observation experience, the cir-cuits and collection frequencies practiced by Amarsul do not com-ply with the actual fill-up rates of each individual glass-wastecontainer. Because of that, and in the frame of a joint research pro-ject SIGRSAMARSUL, the actual fill-up evolution of the containersdeployed by Amarsul in Barreiro was characterized applying a sta-tistical analysis based on the historical data covering several years(SIGRSAMARSUL, 2010). The obtained results, validated in thescope of the project, were provided for the present work in theform of mean daily fill-up increments of each glass container inthat area, a (m3/day). It represented the starting point for the pres-ent work and allowed defining new collection circuits with attrib-uted glass-waste containers that serve as basis for the optimizationof the collection routes of the vehicles.

Fig. 1. Location map of the studied region of Barreiro and of the sorting station Amarsul with the distribution of the ecopoints.

Table 2Vehicle characteristics.

Vehicle feature Units Vehicle A Vehicle B

Fuel consumed – DieselCargo box – Fixed container without

compactionTotal weight Tonnes 15 19Curb weight Tonnes 9.8 12.1Maximum load Tonnes 5.2 6.9Total volume m3 15 20Maximum volume of glass wastea m3 11.5 15.2

a The maximum volume of glass waste that each truck can transport is deter-mined by its maximum load and the average density of the cullet. The maximumcullet load filling the entire cargo box is much higher than the maximum allowedvehicle load.

Table 1Original circuits schedule for glass-waste collection by Amarsul in Barreiro.

Circuit Week of collection

1 2 3 4 5 6 7 8 9 10 11 12 13a

A � � � � �B � � � �C � � � � �D � � �E � � � �

a The schedule of circuits for week 13 starts a new cycle and is equal to theschedule for week 1.

796 Z. Zsigraiova et al. / Waste Management 33 (2013) 793–806

As just mentioned, the glass-waste collection circuits have fre-quencies of 3 or 4 weeks (see Table 2) in the current system. How-ever, the historical data on the containers replenishment evidencesthat only 27 containers out of the 230 reach the set-out rate of atleast 70% that makes them ready to be emptied every 4 weeks(±1 day). The collection frequencies of the remaining containersvary from 5 to 28 weeks. These results revealed that neither the

actual collection frequencies nor distribution of the glass-wastecontainers is adequate.

To define new more adequate glass-waste collection circuits thefollowing modelling assumptions were defined by Amarsul: asscheduling thumb-rule containers to be emptied are those with aset-out rate from 70% to 100% of their total volume. If D (days) isthe elapse period from the last visit to a specific container, its ex-pected fill-up volume status d (m3) is given by d = a D.

2.2. Vehicle routing optimization in GIS environment

Since one of the main objectives of this study was to evaluatethe potential effectiveness of different waste collection strategiesto be applied to an actual waste management system, rather thanto contribute to vehicle routing theory, commercially-availablerouting software, the Network Analyst (NA) extension of ESRI’s Arc-GIS� ArcMap 9.3, was used. The features of the NA that make it par-ticularly suitable are the use of street-level mapping, in the form ofnetwork datasets, allowing for detailed routes to be calculated andto replicate certain characteristics, e.g. multiple depot locations ordifferent vehicle types, while obeying the restrictions specified bythe user. It uses the Dijkstra algorithm to determine the shortestpath source–destination matrix of the network, and Tabu Searchmeta-heuristic as the VRP solving method. The commercial charac-ter of the software ESRI’s ArcGIS� ArcMap 9.3 and its extensions dic-tates the protection of the detailed algorithms used for proprietyreasons.

A realistic representation of the situation studied in this work isperformed based on available digitized two-dimensional maps andgeo-referenced information (see Fig. 1) provided in a set ofshapefiles.

As already mentioned, a detailed validation of the existing col-lection system in terms of the routes actually taken by the driverswas not possible due to the fluctuations in the collection se-quences, since the routes taken around the specified collection areaare left to the discretion of the company drivers. This means that

Table 3Values of the coefficients appearing in Eq. (4) for heavy-duty vehicles with grossweight in the range 7.5–16 tonnes (Hickman et al., 1999).

Pollutant Parameter

CO k1 = 3.08; a = �0.0135; b = 0; c = 0; d = �37.7; e = 1560; f = �5736k2 = 1.03; r = 9.77 � 10�4; s = 0; t = 0; u = 0

CO2 k1 = 871; a = �16; b = 0.143; c = 0; d = 0; e = 32031; f = 0k2 = 1.26; r = 0; s = 0; t = �2.03 � 10�7; u = �1.14

NOx k1 = 2.59; a = 0; b = �6.65 � 10�4; c = 8.56 � 10�6; d = 140; e = 0;f = 0k2 = 1.19; r = 0; s = 0; t = 0; u = �0.977

PM k1 = 0.0541; a = 1.51 � 10�3; b = 0; c = 0; d = 17.1; e = 0; f = 0k2 = 1.02; r = 2.34 � 10�3; s = 0; t = 0; u = 0

Table 4Values of the coefficients appearing in Eq. (4) for heavy-duty vehicles with grossweight in the range 16–32 tonnes (Hickman et al., 1999).

Pollutant Parameter

CO k1 = 1.53; a = 0; b = 0; c = 0; d = 60.6; e = 117; f = 0k2 = 1.17; r = 0; s = 0; t = 0; u = �0.755

CO2 k1 = 765; a = �7.04; b = 0; c = 6.32x10�4; d = 8334; e = 0; f = 0k2 = 1.27; r = 0; s = 0; t = 0; u = �0.483

NOx k1 = 9.45; a = �0.107; b = 0; c = 7.55x10�6; d = 132; e = 0; f = 0k2 = 1.28; r = 0; s = 0; t = 0; u = �0.874

PM k1 = 0.184; a = 0; b = 0; c = 1.72x10�7; d = 15.2; e = 0; f = 0k2 = 1.24; r = 0; s = 0; t = 0; u = �1.06

Z. Zsigraiova et al. / Waste Management 33 (2013) 793–806 797

there are no fixed routes. In a stage of the research project thatsupported this paper, there was an attempt to obtain the actualroutes travelled by the drivers by using an on-board GPS registra-tion system. However, the obtained results were sometimesambiguous impeding to unequivocally state the exact routes con-stituting the vehicles’ path. The reason was that the position regis-tration time interval (1 min) of the GPS equipment was too largefor urban area roads, particularly for short ones. As a result, some-times it was not possible to indentify exactly which street (of sev-eral short possible streets) the vehicle passed.

Because of that, and for comparison purposes, before analyzingthe different waste collection strategies under investigation, theimprovements that could be made by simply optimizing the exist-ing collection system were quantified. This was undertaken byoptimizing the routes taken on each of the existing circuits apply-ing the shortest distance criterion, which is named hereafter as thebaseline case.

Then, new waste collection strategies are proposed and routestaken on the newly designed circuits are compared to the baselinecase. The idea is to ensure that the obtained benefits are over andabove those available from optimizing the routes in the existingcircuits. In addition, this approach ensures a fair comparison ofthe collection strategies.

The routes for the new circuits are optimized for the shortesttravelled distance and, alternatively and not simultaneously, forthe smallest spent time and the results are compared in terms oftotal cost and pollutant emissions to assess the most attractivealternative.

As far as the working time is concerned, and in order to avoidovertime due to the heavy burden of manpower costs, the com-pany Amarsul established 5 h (with a tolerance of 10 min) as themaximum duration of each collection round trip, where a tripcan be executed either in the morning or in the afternoon shift.The total trip time also includes the idling time, i.e. the time spentat each ecopoint to unload waste containers (a constant containercollection time of 4.5 min per container), at the sorting station tounload the truck (a constant empty time of 15 min per visit), andthe planned resting period for the vehicle crew (45 min per trip).

2.3. Calculation of the pollutant emissions and fuel consumption ofcollection vehicles

Both travelled distance and actual operation conditions for a gi-ven vehicle influence fuel consumption during waste collectionand transportation and, therefore, the amount of emittedpollutants.

These effects are determined herein making recourse to MEETmethodology provided by Hickman et al. (1999).

The trucks used for waste collection and transportation in Bar-reiro municipality are classified by MEET in the category of dieselheavy-duty vehicles, and the EURO IV legislation class is utilized.

The values of selected pollutants – NOx, CO2, CO and PM – cal-culated for each road segment are stored in the spatial databaseto be used later during the route solver procedure. Once an optimalroute is determined according to the selected optimization crite-rion, the corresponding amount of emissions are evaluated basedon it.

The fuel consumption retrieved from MEET is given by:

Cfuel ¼Mfuel

½CO�MCOþ ½CO2 �

MCO2þ ½HC�

MHCþ ½PM�

MPM

� �qfuel

ð1Þ

In the previous equation, Cfuel represents the fuel consumption (l),[i] is the emitted mass of pollutant i (g), HC stands for unburnedhydrocarbons and PM for the particulate matter, Mi is the molarmass of pollutant i (g/mol), and qfuel is fuel density (g/l).

According to MEET, pollutant emissions are calculated by Eqs.(2)–(5), where Ei, Ei,hot and Ei,cold are the total emissions of pollutanti (g), the hot emissions of pollutant i (g), and the cold emissions ofpollutant i (g), respectively; ei,c is the hot emission factor for pollu-tant i corrected for load (g/km), and dtr is the travelled distance(km); v is the mean velocity (km/h), k1, a, b, c, d, e, f and k2, r, s, t,u are coefficients depending on the vehicle total weight, and z isthe fraction of transported load; ei,cold is the cold emission factorfor pollutant i (g/cold start), and N is number of cold starts.

Ei ¼X

vehicle route

ðEi;hot þ Ei;coldÞ ð2Þ

Ei;hot ¼ ei;cdtr ð3Þ

ei;c ¼ k1 þ av þ bv2 þ cv3 þ dv þ

ev2 þ

fv3

� �

� k2 þ rv þ sv2 þ tv3 þ uv � 1

� �zþ 1

h ið4Þ

Ei;cold ¼ ei;coldN ð5Þ

The values of the coefficients of Eq. (4) are displayed in Table 3for heavy-duty vehicles with gross weight in the range 7.5–16 ton-nes, and in Table 4 for heavy-duty vehicles with gross weight in therange 16–32 tonnes (Hickman et al., 1999).

The values of the coefficients of Eq. (5) are displayed in Table 5for both types of vehicles (Hickman et al., 1999).

It should be noted that, although vehicle acceleration/decelera-tion may have impact on fuel consumption and vehicle emissions(Nguyen and Wilson, 2010) this was not considered herein. How-ever, the effect of different velocities on fuel consumption and pol-lutant emissions was taken into account since velocities of vehiclestravelling on urban roads, outside-urban roads and highways areactually different. As a result, a specific mean velocity value wasassigned to each segment of the road network: the mean vehiclevelocity of 30 km/h for circulation in urban roads, 50 km/h for cir-culation outside-urban roads and 70 km/h on the highways.

798 Z. Zsigraiova et al. / Waste Management 33 (2013) 793–806

2.4. Calculation of the cost of the service

In order to identify the possible factors influencing the collec-tion and transportation costs of recyclable glass-waste, an analysisconsidering the established goal and the defined boundaries of thesystem under study, and the discussions with Amarsul, yielded thecost function expressed by Eq. (6). The cost function comprises thelabor cost and vehicle operation cost, including maintenance. Oneshould stress that the capital cost is of importance but it was nottaken into consideration since the goal established for the studiedcase was to optimize an already existing system (with the invest-ments already done in the past) rather than to design a new one.

TC ¼X

j

Xi

fLCjtij þ CFfcidle;ijtidle;ij þ ðMCj þ CFfcijÞ dijg ð6Þ

In Eq. (6), TC is the total cost (€), LCj is the specific labor cost (€/h) ofcrew in vehicle j, tij is the total time (h) spent by vehicle j in route i,CF is the specific fuel cost (€/l), fcidle,ij is the specific fuel consump-tion (l/h) in idling mode for vehicle j in route i, tidle,ij is the total time(h) spent in idling mode by vehicle j in route i, MCj is the specificmaintenance cost (€/km) for vehicle j, fcij is the specific fuel con-sumption (€/h) in travelling mode for vehicle j route i, and dij isthe distance travelled (km) by vehicle j in route i.

Furthermore, the total cost components expressed by Eq. (6)can be distinguished as dynamic and static. The dynamic cost is re-lated to the vehicle travelling between ecopoints or between thelatter and the sorting station and is calculated as the sum of thetransportation cost based on the travelled distance and the trip-duration cost. In turn, the static cost is related to the vehicle wasteloading at the ecopoints, in idling mode, and to the waste droppingoff at the sorting station. In this case, the cost is the sum of thevehicle operation cost and those of the duration of uploading andthe duration of dropping off. Both previous costs, dynamic and sta-tic, are governed by two parameters: the vehicle operation cost(that includes specific fuel consumption, costs of fuel and vehiclemaintenance) and the manpower cost.

The values used herein for the specific costs of labor (LCj in €/h),fuel (CF in €/l) and maintenance (MCj in €/km) for both vehicles Aand B are displayed in Table 6. For proprietary reasons, the valuespresented in Table 6 are not exactly the actual ones, but they are aclose approximation.

The distances required for evaluating the total cost are obtainedfrom the available road network geo-database in GIS that stores

Table 6Actual approximate specific cost values of Amarsul.

Labor cost(€/h)

Fuel cost(€/l)

Maintenance cost(€/km)

Vehicle A 20 0.8 0.16Vehicle B 20 0.8 0.17

Table 5Values of the coefficients appearing in Eq. (5) for heavy-dutyvehicles (Hickman et al., 1999).

Pollutant Vehicle type(tonnes)

Parameter ei,cold

(g/cold start)

CO 7.5–16 616–32 6

CO2 7.5–16 30016–32 500

NOx 7.5–16 �216–32 �5

PM 7.5–16 0.616–32 0.6

the length of each road element. In turn, the time spent duringeach travel is calculated from the mean vehicle velocity values.

2.5. Influence of the dynamic load

In this work, and contrasting with some previous works in thisarea mentioned in Section 1, the studied system is given one moredegree of freedom by allowing the vehicle load to vary along thewaste collection trip as such variation can contribute with up to20% of released hot emissions (Hickman et al., 1999). The vehicleload is considered to change each time the glass-waste from a differ-ent container is uploaded, which influences the fuel consumptionand pollutant emissions on the subsequent road section travelled.

For the sake of the results intelligibility, the study of the dy-namic load influence on fuel consumption and pollutant emissionsis performed by virtually locating the sorting station closer to thecollection area than it is in reality. This was done in order to min-imize the potential impact of the considerably long distance be-tween the sorting station and the first container visited (andbetween the last container visited and the sorting station in theway back – see Fig. 1), once the actual dynamic load effects couldbe concealed by the dominating effect of global fuel consumptionat this part of the collection round trip.

This influence study is performed for one of the new establishedcircuits consisting of 50 ecopoints. First, the vehicles routes are opti-mized for the shortest time and the fuel consumption is calculatedassuming that the glass-waste volume uploaded at each visited eco-point is constant and equal to 1 m3 (i.e. the actual fill-up rate of thecontainers was not considered in this case) – the baseline case.

Then, two contrasting scenarios were designed to perform thesimulations that yield the dynamic load effects on the fuel con-sumption and pollutant emissions, considering an exponential dis-tribution of the collected waste amount along the collection route.The first scenario assumed that the collected volume of waste in-creases exponentially along the collection route, whereas the sec-ond scenario assumed the exponential decrease for the sameroute. This means that the volume of glass-waste added to thevehicle case at each ecopoint is exponentially more (or less forthe second scenario) than that of the preceding ecopoint, assumingthat all the material in the container is always unloaded and trans-ferred to the collection vehicle. Both scenarios keep invariable thetotal waste volume collected per route and the sequence of visitedecopoints, i.e. they are the same as those obtained in the case of theconstant load (the baseline case).

These exponential increase or decrease was considered in orderto clearly evidence the dynamic load effect. The reason is that itscontribution is much less evident than that of the routeoptimization.

3. Results and discussion

3.1. Scheduling the waste collection circuits

The method applied herein benefits from the specific cumula-tive data about each container without the need for investmentsin new equipments as suggested by Johansson (2006) or Faccioet al. (2011). This innovative aspect lies in the use of a distinctivefill-up rate for each glass-waste container, which contrasts withthe works that used a single average fill-up rate value for all thecontainers in the system (e.g. Teixeira et al., 2004; Simonetto andBorenstein, 2007; Komilis, 2008; Apaydin and Gonullu, 2008 orTavares et al., 2009).

The new proposed glass-waste collection schedule representsan enhanced pattern of frequencies repeating only after 288 weeksand with a total of 58 new collection circuits, which contrasts with

Table 7Examples of optimized circuits scheduling for glass collection according to the fill-uprates of containers.

Circuit No. of ecopoints Collection weeks

C1 50 1; 145C6 38 7; 79; 115; 151; 223; 259C8 44 10C10 22 13(1); 85(1); 121(1); 157(1); 229(1); 265(1)C11 37 13(2); 85(2); 157(2); 229(2)C13 19 17; 113; 161; 257C36 32 67; 103; 139; 211; 247; 283C54 31 121(2); 256(2)C58 48 140; 284

Z. Zsigraiova et al. / Waste Management 33 (2013) 793–806 799

the 12-week cycle and five circuits presently used by Amarsul. Inthis case study, the planning cycle was not set a priori as it waspossible for example in the works of Teixeira et al. (2004) or Nuor-tio et al. (2006), since the schedule intervals of the glass-wastecontainers vary significantly, from 5 to 28 weeks. The new 58 cir-cuits were generated according to the specific fill-up rate of eachcontainer: at the date a glass-waste container becomes filled atleast 70% of its volume, as described in Section 2.1., it is includedin that day circuit. As it can be seen in Fig. 1, the glass-waste con-tainers are located both in densely populated regions (where thecontainers concentration is quite large) and rural areas (very spar-sely located). This means that the fill-up rates have a quite dis-persed distribution and, therefore, the number of circuits is quitelarge and their repetition only occurs when the required fulfillingcriterion is attained.

The new collection scheduling methodology presented con-trasts, for example, with that of Apaydin and Gonullu (2008)where, for an actual MSW collection/hauling system, route optimi-zation was performed solely for the already existing collection cir-cuits without considering the individual fill-up rates of containers.

These results suggest a future analysis of the glass-waste con-tainers distribution and an eventual redistribution.

Generally, each week has a 5-h collection period, althoughweeks without collection (e.g. weeks 24, 30, 48 and 54) as wellas those with two collection periods (e.g. weeks 13, 21 and 85), oc-cur during the cycle. Table 7 presents a few examples of the 58optimized circuits to evidence the previous explanations.

3.2. Vehicle routing optimization

After obtaining the previously described new glass-waste col-lection scheduling, vehicles routings in the new circuits were opti-mized alternatively, but not simultaneously, for the shortesttravelled distance and spent time with the application of the ESRI’sArcGIS� ArcMap 9.3 extension Network Analyst.

As far as the vehicle path is concerned, each vehicle alwaysstarts the waste collection trip at the Amarsul sorting station site,emptying then the glass-waste from the containers sited in definedlocations of a given circuit, being finally driven back towards thesorting station making a closed loop – see Fig. 2, showing routesfor both vehicles in a given circuit.

Whenever a vehicle reaches its allowed collection capacity, it isdriven back to the sorting station where it is emptied. In this situ-ation, and whenever the total duration of both trips will not exceedthe limit time, the same vehicle returns to collect waste from therest of the ecopoints of the circuit. In contrast, whenever the totaltime for both trips exceeds the maximum time allowed, a differentvehicle completes the collection circuit.

From the above reasoning it is clear that complete waste collec-tion in a circuit can be, sometimes, accomplished in more than oneroute for each vehicle or, alternatively, by more than one vehiclewhen duration limit is exceeded. This is illustrated in Fig. 2.

In order to compare the performance of the new proposed col-lection and transportation system for glass-waste with that exist-ing presently in Barreiro municipality, weekly average values oftime spent in travelling – travel time, total time (including the timespent at ecopoints, at the sorting station and for the staff rest),travelled distance, fuel consumption, pollutant emissions amountsand costs were computed and the results are displayed in Table 8.It should be noted that the three last parameters were determinedapplying the methodology presented in the previous section forboth the existing and new proposed systems.

In general, and as revealed by Table 8, substantial reductionsare attainable with the new proposed circuits either optimizingroutes for distance or for time.

In fact, there are reductions of 49% and 62% for the total timespent when optimizing with shortest distance and shortest time,respectively; and 45% and 57% for the total cost with the sameoptimizations.

In turn, for the pollutant emissions, reductions vary between27% and 30% when optimizing with shortest distance and between43% and 50% when optimizing with shortest time. These reductionsare a direct consequence of the fuel consumption reductions (28%for distance optimization and 43% for time optimization).

These results are in agreement with those presented in the workof Apaydin and Gonullu (2008), where the authors applied the short-est distance criterion to optimize an actual MSW collection/haulingsystem in Trabzon City (Turkey) to minimize emissions. For theirstudied system, if the optimized routes were used instead of theones being actually used, distance and time could be decreased onaverage by some 25% and 44%, respectively, accompanied by a con-siderable decrease in pollutant emissions originated from vehicleexhaust. However, rather than considering variable vehicle velocityand load as in the present work, their calculations were based onfixed average values for the specific fuel consumption and emissionsreleased as a function of the travelled distance.

Although, in the present work, both optimization results showgreat improvements, the most significant potential benefits comefrom the optimization for time as will be seen below from the anal-ysis of Figs. 3–5. This is not surprising since the route optimizationfor time takes into account different contributions: travelling time,idling time and different vehicle velocities according to the roadcategory (Badran and El-Haggar, 2006 or Komilis, 2008). In con-trast, route optimization for the shortest distance does not takeall those factors into account – see, for example, the work of Apay-din and Gonullu (2008), who estimated the emissions exclusivelyfor the travelling phase of the waste collection procedure. Notethat MSW collection vehicles spend a considerable part of theoperation time working in the idling mode during containers emp-tying and waste discharge (Fig. 3). Thus, this activity is responsiblefor significant amounts of consumed fuel and released emissions.

Fig. 3 shows the potential gains for the travelled time, totaltime, total travelled distance and fuel consumption with the opti-mization of the system under study. As it can be seen, the timespent by vehicles in travelling to collect glass-waste could be re-duced by around 32% and 59% optimizing for distance and for time,respectively. In turn, the total travelled distance may decreasearound 25% for the both optimization scenarios. This reinforcesthe more effective results when optimizing routes for time.

Fuel consumption, contributing also to the total cost, could bereduced by almost 43% with the optimization for time, reachinga decrease of around 28% when the optimization for distance isconcerned. Summing up, routes optimization for time produceslarger reduction values of the operating variables (time, distancesand fuel consumption).

As far as the emissions are concerned, the pollutants that showgreater relative reductions are PM (50% when the routes are opti-mized for time), as can be seen in Fig. 4 that shows the potential

Fig. 2. Illustration of collection routes for a selected circuit attributed to vehicle A (highlighted in blue) and to vehicle B (highlighted in green). (For interpretation of thereferences to colour in this figure legend, the reader is referred to the web version of this article.)

800 Z. Zsigraiova et al. / Waste Management 33 (2013) 793–806

gains for the pollutant emissions. Nonetheless, the emittedamounts of all the pollutants decreased more than 26% and morethan 40% when the routes are optimized for distance and for time,respectively, as it can be seen in Fig. 4. In conclusion, routes

optimization for time produces larger reduction values of the stud-ied pollutant emissions.

Moreover, the implementation of the routes optimized for timecould result in considerable savings that can go up to 57% of the

Table 8Weekly average results for the new system optimized for distance, for time and the corresponding values for the Amarsul system.

New system optimized for DISTANCE New system optimized for TIME Existing system in Amarsul(optimized for distance – BASELINE)

Vehicle A Vehicle B Total Vehicle A Vehicle B Total Vehicle A Vehicle B Total

Travelled time (on roads) (min/week) 130 120 250 72 78 150 213 152 365Total time (min/week) 197 198 395 137 160 297 405 375 780Travelled distance (km/week) 55.8 53.2 109.0 53.3 56.1 109.4 81.8 63.2 145.0Fuel consumption (l/week) 14.5 20.0 34.5 10.4 16.9 27.3 23.2 24.4 47.6

Pollutant emissions (g/week)CO 182 235 417 134 184 318 285 302 587CO2 37,926 52,422 90,348 27,103 44,327 71,430 60,463 64,098 124,561NOx 445 662 1107 295 524 819 707 805 1512PM 45 44 89 29 34 63 72 55 127

Cost (€/week)Fuel consumption 14.1 19.5 33.6 10.1 16.4 26.5 22.6 23.8 46.4Maintenance 10.6 11.7 22.3 10.1 12.3 22.4 15.5 13.9 29.4Labor 80.2 80.6 160.8 55.6 64.7 120.3 164.9 152.6 317.5Total 105.0 111.8 216.8 75.8 93.5 169.3 202.9 190.3 393.2

Fig. 3. Comparison of the differences between the travelled times, total times, travelled distances and fuel consumptions between both new systems optimized for distanceand time and the actual system. (A) Absolute differences. (B) Relative differences.

Z. Zsigraiova et al. / Waste Management 33 (2013) 793–806 801

Fig. 4. Comparison of the pollutant emissions (CO, CO2, NOx, and PM) amounts between both new systems optimized for distance and time and the actual system. (A)Absolute differences. (B) Relative differences.

802 Z. Zsigraiova et al. / Waste Management 33 (2013) 793–806

total cost as shown in Fig. 5 that shows the potential gains for thetotal cost and the different contributing parcels. This figure alsoevidences that the labor cost is the parcel that contributes the mostfor those savings by exhibiting a reduction of 62%. In conclusion,routes optimization for time produces larger reduction values ofthe operating costs.

Altogether, the previous results represent over €11,000 of costsavings per year for the two vehicles (A and B) collecting glass-waste when circuits are optimized by time.

The presented results clearly evidence that the optimization ofthe studied system demonstrates a positive influence on the oper-ation costs (maintenance, labor and fuel consumption) and trans-port related emissions, regardless the optimization criterion used(shortest distance or shortest time), being particularly relevantfor optimization for time and yielding substantial decreases oftheir values compared to those of the existing system in operation.

Besides travelled distance and time, fuel consumption have alsobeen utilized as an optimization criterion. Tavares et al. (2009) pro-

posed the use of a GIS 3D route modelling software for waste col-lection and transportation allowing driving routes to be optimizedfor minimum fuel consumption while taking into account the ef-fects of road inclination and vehicle weight. Kuo and Wang(2011) presented a method for solving the VRP minimizing the fuelconsumption for transportation applications. Alike in the presentwork, the three factors that greatly affect fuel consumption oftransportation, transportation distance, speed and loading weight,were taken into consideration. In both mentioned works, the re-sults showed that the optimized routing plans provided substantialimprovements over those based on minimizing transportationdistances.

In the present application, it was established with the engineersfrom the waste management company (Amarsul) that time shouldbe used as the optimization criterion due to its importance andlimiting character in the waste collection and transportation logis-tics, since manpower and idling time of vehicles have a relevantcontribution to global operating costs.

Fig. 5. Comparison of the fuel, maintenance, labor and total costs between both new systems optimized for distance and time and the actual system. (A) Absolute differences.(B) Relative differences.

Z. Zsigraiova et al. / Waste Management 33 (2013) 793–806 803

3.3. Influence of the dynamic loading process

To investigate and understand the effects of the dynamicchange of the vehicle loading process on pollutant emissions andfuel consumption a sensitivity analysis was carried out for one ofthe new established circuits consisting of 50 ecopoints.

After obtaining the vehicles routes optimized for the shortesttime under the assumption of an average constant load incrementof 1 m3 at every visited ecopoint, two simulation scenarios weretested: the first was designed assuming that the collected volumeincreases exponentially along the visited containers; the other wasdesigned assuming that the collected volume along the visited con-tainers decreases exponentially. However, for comparison pur-poses, the total collected waste volume per route and thesequence of visited ecopoints were kept unchanged in both scenar-ios, i.e. the same as those obtained in the case of the constant loadincrement.

The path travelled by one of the glass-waste collection vehicles(vehicle B) is depicted in Fig. 6.

Table 9 shows the numerical results obtained. The values com-prise the transportation related pollutants emissions and fuelconsumption.

In the case of constant load increment at each ecopoint theloading process is evenly distributed along the route. Contrastingwith this, in the case of the exponentially increasing load the vehi-cle takes the major portion of glass-waste close to the end of theroute. This means that the vehicle travels with a significant loadonly on a short distance (compared to the entire route length) be-fore unloading it at the virtual sorting station.

On the other hand, when the load decreases exponentially, thevehicle collects the major load fraction from the few first eco-points, right at the beginning of the route. As a consequence, thevehicle travels with the major portion of the glass-waste along asubstantial part of the route.

The varying load distribution along the route evidencedclearly the effect of the dynamic load on costs (mainly forthe fuel consumption) and on pollutant emissions as can be seenin Fig. 7.

Fig. 6. Collection route attributed to vehicle B in studying the dynamical load effects.

804 Z. Zsigraiova et al. / Waste Management 33 (2013) 793–806

Comparing the case of the exponentially increasing load withthat of constant load increments, one can observe a decrease forboth the emitted pollutants and fuel consumption for the studiedcircuit: from 2.3% to 4% for the different pollutants and a value of3.8% for the fuel consumption.

As expected, an opposite tendency is observed for the exponen-tially decreasing load, where the values of the all observed param-eters increase relatively to the constant load distribution case:from 1.9% to 3.7% for the different pollutants and a value of 3.6%for the fuel consumption.

Fig. 7. Effects of the dynamical load on pollutant emissions and fuel consumption (FC) in two designed scenarios: exponential increase and exponential decrease of theloading process.

Table 9Comparisons of the pollutants emissions and fuel consumption.

Volume to collect constant: 1 m3 per collection point Volume to collect increasing exponentially Volume to collect decreasing exponentiallyEmissions (g) Emissions (g) Emissions (g)

PollutantCO 151.7 148.2 154.6CO2 200958.9 193199.7 208149.8NOx 2148.0 2062.2 2228.1PM 82.9 80.2 85.5

Consumption (kg) Consumption (kg) Consumption (kg)

Fuel 64.1 61.7 66.4

Z. Zsigraiova et al. / Waste Management 33 (2013) 793–806 805

Although the differences may not be particularly sharp, theinfluence of the dynamic load of the vehicle on the fuel consump-tion and pollutant emissions is unquestionable as also reported inthe works of Tavares et al. (2008), Kuo (2010), Kuo and Wang(2011) or Xiao et al. (2012). Thus, in terms of lower fuel consump-tion and less associated amount of pollutants emitted, the load dis-tribution along the collection route plays an important role.Although the inclusion of this parameter into the optimization pro-cess turns the problem quite more complex (e.g. multi-objectiveoptimization), it appears to be a more promising solution thanthe common minimization of travelled distance or travelled time.

4. Conclusions

The present work presented an innovative tool to improvewaste management in its most budget-weighty activity – wastecollection and transportation. Its innovative feature is representedby the combination of vehicle route optimization in GIS environ-ment with that of waste collection scheduling based on historicaldata of the individual fill-up rate of each container, which hasnot been previously reported. The use of this approach ensuresan effective use of the available material (truck fleet and wastecontainers) and the financial and human resources involved inwaste collection and transportation activities. Furthermore, andin contrast with the works using average values of relevant param-eters, the influence of different transportation speeds and changingloading weight on the fuel consumption and amount of released

pollutant emissions were taken into account. This is anotherimportant aspect of the present work approaching the obtained re-sults to real-life situations. With such an optimization tool, for thestudied case of the glass-waste collection and transportation of theAmarsul S.A. system in Barreiro, Portugal, it was possible to attain aconsiderable reduction in the operation costs (maintenance, laborand fuel) due to the reduction of the total collection time and fuelconsumption and, consequently, reductions of the pollutant emis-sions (CO, CO2, NOx and PM).

This achievement constitutes a considerable improvement ofthe previous practice employed in the waste management com-pany, which now possesses a supporting tool to analyze and quan-tify modifications to the collection system as well as to re-plan thevehicle routing.

The use of a distinctive fill-up rate for each glass-waste con-tainer brought a system representation more realistic than thoseusing a single average fill-up rate value for all the containers.

In addition, routing optimization in GIS environment evidencedthe benefits of its use: it significantly simplifies the process, evenfor large sets of data for transportation network information andlocation of the ecopoints like the ones used herein; it computeseasily the source–destination cost matrix; it finds the solution toVRP smoothly using the heuristic Tabu Search technique; it effi-ciently and conveniently represents and visualizes the obtainedresults.

Finally, the effect of the dynamic load of the collection vehicleson fuel consumption and pollutant emissions was analyzedthrough a sensitive analysis using two hypothetical scenarios.

806 Z. Zsigraiova et al. / Waste Management 33 (2013) 793–806

The results confirmed the important influence of vehicle load onfuel consumption and subsequent emissions release. Therefore,this factor should be taken into account when dealing with VRPso that fuel consumption is estimate more accurately.

Although applied to the glass-waste case, the tool presentedherein is easily extendable to collection of any other type ofMSW providing that the individual fill-up rates of the containersare available.

Acknowledgements

The authors would like to acknowledge the financial supportfrom Amarsul S.A. and the National Strategic Reference Frameworkprogramme (QREN – Quadro de Referencia Estrategico Nacional)for the present research through the national Project No. 1527,‘‘SIGRSAMARSUL – Information system for management of recycla-ble waste collection’’.

References

Amponsah, S.K., Salhi, S., 2004. The investigation of a class of capacitated arc routingproblems: the collection of garbage in developing countries. WasteManagement 24, 711–721.

Apaydin, O., Gonullu, M.T., 2008. Emission control with route optimization in solidwaste collection process: a case study. Sadhana – Academy Proceedings inEngineering Sciences 33 (2), 71–82.

Armstrong, J.M., Khan, A.M., 2004. Modelling urban transportation emissions: roleof GIS. Computers, Environment and Urban Systems 28, 421–433.

Badran, M.F., El-Haggar, S.M., 2006. Optimization of municipal solid wastemanagement in Port Said – Egypt. Waste Management 26, 534–545.

Chang, N.B., Wei, Y.L., 1999. Strategic planning of recycling drop-off stations andcollection network by multiobjective programming. EnvironmentalManagement 24, 247–263.

Dessouky, M., Rahimi, M., Weidner, M., 2002. Green Transit Scheduler: amethodology for jointly optimizing cost, service, and environmentalperformance in demand-responsive transit scheduling. Final Report toMetrans Transportation Center, University of Southern California, Los Angeles,and California State University, Long Beach. <http://www.metrans.org>.

Dogan, K., Suleyman, S., 2003. Report: cost and financing of municipal solid wastecollection services in Istanbul. Waste Management & Research 21, 480–485.

Dror, M. (Ed.), 2000. Arc Routing – Theory, Solutions and Applications. KluwerAcademic Publishers.

Eksioglu, B., Vural, A.V., Reisman, A., 2009. The vehicle routing problem: ataxonomic review. Computers & Industrial Engineering 57, 1472–1483.

Ericsson, E., Larsson, H., Brundell-Freij, K., 2006. Optimizing route choice for lowestfuel consumption – potential effects of a new driver support tool.Transportation Research Part C 14, 369–383.

Faccio, M., Persona, A., Zanin, G., 2011. Waste collection multi objective model withreal time traceability data. Waste Management 31, 2391–2405.

Ghose, M.K., Dikshit, A.K., Sharma, S.K., 2006. A GIS based transportation model forsolid waste disposal – a case study on Asansol municipality. WasteManagement 26, 1287–1293.

Hickman, J., Hassel, D., Joumard, R., Samaras, Z., Sorenson, S., 1999. Methodology forcalculating transport emissions and energy consumption. Report No. SE/491/98.Transport Research Laboratory, Crowthorne, UK.

Johansson, O.M., 2006. The effect of dynamic scheduling and routing in a solidwaste management system. Waste Management 26, 875–885.

Keenan, P.B., 1998. Spatial decision support systems for vehicle routing. DecisionSupport Systems 22, 65–71.

Komilis, D.P., 2008. Conceptual modeling to optimize the haul and transfer ofmunicipal solid waste. Waste Management 28, 2355–2365.

Kulcar, T., 1996. Optimizing solid waste collection in Brussels. European Journal ofOperational Research 90, 71–77.

Kuo, Y., 2010. Using simulated annealing to minimize fuel consumption for thetime-dependent vehicle routing problem. Computers & Industrial Engineering59, 157–165.

Kuo, Y., Wang, C.-C., 2011. Optimizing the VRP by minimizing fuel consumption.Management of Environmental Quality: An International Journal 22 (4), 440–450.

Laporte, G., Gendreau, M., Potvin, J.Y., Semet, F., 2000. Classical and modernheuristics for the vehicle routing problem. International Transaction inOperational Research 7, 285–300.

Lin, H.Y., Chen, G.H., Lee, P.H., Lin, C.H., 2010. An interactive optimization system forthe location of supplementary recycling depots. Resources, Conservation &Recycling 54, 615–622.

Longo, H., de Aragao, M.P., Uchoa, E., 2006. Solving capacitated arc routing problemusing a transformation to the CVRP. Computers & Operations Research 33 (6),1823–1837.

McLeod, F., Cherrett, T., 2008. Quantifying the transport impacts of domestic wastecollection strategies. Waste Management 28, 2271–2278.

Nguyen, T.T., Wilson, B.G., 2010. Fuel consumption estimation for kerbsidemunicipal solid waste (MSW) collection activities. Waste ManagementResearch 28, 289–297.

Nuortio, T., Kytojoki, J., Niska, H., Braysy, O., 2006. Improved route planning andscheduling of waste collection and transport. Expert Systems with Applications30, 223–232.

SIGRSAMARSUL, 2010. Information system for management of recyclable wastecollection. Project No. 1527 (2008–2010), Funded by National StrategicReference Framework Programme QREN, Portugal.

Simonetto, E.O., Borenstein, D., 2007. A decision support system for the operationalplanning of solid waste collection. Waste Management 27, 1286–1297.

Tarantilis, C.D., Diakoulaki, D., Kiranoudis, C.T., 2004. Combination of geographicalinformation system and efficient routing algorithms for real life distributionoperations. European Journal of Operational Research 152, 437–453.

Tavares, G., Zsigraiova, Z., Semiao, V., Carvalho, M.G., 2008. A case study of fuelsavings through optimization of MSW transportation routes. Management ofEnvironmental Quality: An International Journal 19 (4), 444–454.

Tavares, G., Zsigraiova, Z., Semiao, V., Carvalho, M.G., 2009. Optimisation of MSWcollection routes for minimum fuel consumption with 3D GIS modelling. WasteManagement 29, 1176–1185.

Taylor, M.A.P., Woolley, J.E., Zito, R., 2000. Integration of the global positioningsystem and geographical information systems for traffic congestion studies.Transportation Research Part C 8, 257–285.

Teixeira, J., Antunes, A.P., Sousa, J.P., 2004. Recyclable waste collection planning – acase study. European Journal of Operational Research 158, 543–554.

Wøhlk, S., 2008. An approximation algorithm for the capacitated arc routingproblem. The Open Operational Research Journal 2, 8–12.

Xiao, Y., Zhao, Q., Kaku, I., Xu, Y., 2012. Development of a fuel consumptionoptimization model for the capacitated vehicle routing problem. Computers &Operations Research 39, 1419–1431.

Zamorano, M., Molero, E., Grindlay, A., Rodriguez, M.L., Hurtado, A., Calvo, F.J., 2009.A planning scenario for the application of geographical information systems inmunicipal waste collection: a case of Churriana de la Vega (Granada, Spain).Resources, Conservation & Recycling 54, 123–133.

Zsigraiova, Z., Tavares, G., Semiao, V., Carvalho, M.G., 2009. Integrated waste-to-energy conversion and waste transportation within island communities. Energy34, 623–635.