Forecasting Municipal Solid Waste Generation in MajorEuropean Cities

P. Beigl, G. Wassermann, F. Schneider and S. Salhofer

Institute of Waste Management, BOKU – University of Natural Resources and Applied Life Sciences,Vienna, Austria

Abstract: An understanding of the relationships between the quantity and quality of environmentally relevantoutputs from human processes and regional characteristics is a prerequisite for planning and implementingecologically sustainable strategies. Apart from process-related parameters, continuous and discontinuoussocio-economic long-term trends often play a key role in the assessment of environmental impacts. This paperdescribes the development of a prognosis model for municipal solid waste (MSW) generation in Europeanregions. The objective is to assess future municipal waste streams in major European cities. We thereforefocussed on cities, which face significant social and economic changes, e.g. in central and east European(CEE) countries. The investigations covered waste-related data and a broad set of potential influencingparameters that contained commonly used social, economic and demographic indicators as well as previouslyproved waste generation factors. An extensive database was created with an annual time series up to 32 yearsfrom 55 European cities and 32 countries. The evaluation of this historic time series and the cross-sectionaldata by means of multivariate statistical methods has unveiled significant relationships between the status ofregional development and municipal solid waste generation. We identified a core set of significant indicators,which can describe a long-term development path that predetermines the level of waste generation. Thesefindings concerning this analogy have been integrated in an econometric model for European cities.

Keywords: Waste management; Municipal solid waste; Waste generation; Modelling; Forecasting

1. INTRODUCTION

The development of waste management modelsover the last decades can be characterised by anincreasing level of integration of related processeswith consideration of environmental, economic andsocial aspects. The genesis of these decisionsupport tools [Björklund, 2000] is reflected byextending system boundaries, which are shown inFigure 1. In early models, attention was paid to theproblems in subsystems, e.g. routing of vehiclesand location of treatment and disposal facilities,with a focus on only a few criteria (e.g. costs).Recently, waste management models have startedto evaluate entire waste management systems,considering broad sets of quantitative andqualitative criteria.

Up to now, most of these decision support tools forwaste management planning use the amount ofwaste generation as the given input parameter[Björklund, 2000; White et al., 1999]. Thus theimpacts of demographic, social and economicdynamics as well as other factors (e.g. consumptionpatterns or waste prevention) are not taken into

consideration for the accurate assessment of futurewaste generation.

Waste management system

Products

Energy

Costs

Products

Energy

Emissions

Municipal solid waste generation

Socio-economicchanges

Demographicdynamics

Consumptionpatterns

Wasteprevention

Incr

easi

ngm

odel

inte

grat

ion

Sub-

syst

em1

Sub-

syst

em2

Sub-

syst

emx

. . . .

Figure 1. System boundaries in waste managementmodels with different level of integration.

This paper aims to identify parameters which helpto explain the present situation and to assess thefuture amount of municipal solid waste (MSW)generated per capita in different European cities.

This study is part of the European Commissionproject “The Use of Life Cycle Assessment Toolsfor the Development of Integrated Waste

Management Strategies for Cities and Regions withRapidly Growing Economies.” The goal of thisproject is to develop a highly integrated decisionsupport tool for cities in southern, central and eastEuropean countries which helps to evaluate wastemanagement options with regard to environmental,economic and social criteria. In this context, MSWforecasts should arouse the consciousness of themunicipal decision makers to implementecologically sustainable measures (e.g. increasingrecycling quotas).

2. METHODOLOGICALCONSIDERATIONS

Due to the high heterogeneity of municipal solidwaste streams and the diversity of their waysthrough the economy, the identification ofparameters is a highly complex problem. Anoverview of studies in this scientific field by Beiglet al. [2003] describes previous approaches, whichcan be classified by the type of model:

• Input-output models: Here the input of thewaste generator is assessed by usingproduction, trade and consumption data aboutproducts related to the specific waste streams.

• Factor models: These models focus onanalyses of the factors, which describe theprocesses of waste generation. Examples ofproved parameters are e.g. the income ofhouseholds, dwelling types or the type ofheating.

Based on this comparative study, only a fewmethodological procedures came intoconsideration for application within the aimedforecasting model for cities. This was due to thefollowing reasons:

• Level of aggregation: The identification ofparameters has to be based on a database,which describes regional peculiarities. Theexclusive use of national aggregates in input-output models [Patel et al., 1998] is notappropriate for explaining regional dynamics.Therefore preference was given to factormodels that focus on socio-economic anddemographic indicators available at a regionallevel [Bach et al., 2003].

• Predictability of parameters: The selectionof model parameters has to prioritiseparameters at the city level, which can beforecasted with a relatively high accuracy anda long forecasting horizon. Examples of suchparameters with high inertia are thepopulation age structure, household size orinfant mortality rate [Lindh, 2003].

• Applicability refers to the user-friendlinessof the aimed forecasting tool. Thereforemethods that provide easily available,standardised secondary data have to befavoured over elaborate and time-consumingqualitative approaches such as the Delphimethod [Karavezyris, 2001].

Based on these considerations, the amount andcomposition of municipal solid waste werehypothesised in this approach dependent uponeasily available socio-economic and demographicparameters. Under the assumption of an analogousdevelopment of regional characteristics and MSWgeneration, this could explain regional differencesbetween cities as well as long-term changesconcerning a city by means of an ex post analysis.

3. DATA

Due to their relevance for waste managementplanning, the waste potentials of the main materials-- such as organic material, paper and cardboard,plastics and compounds, glass or metals -- weredefined as variables to be explained. As the wastepotential of a certain waste stream (apart fromillegally dumped waste) contains the separatelycollected material and a partial stream of theresidual waste, data about the collected quantitiesof residual waste and the separately collectedmaterials as well as about the composition ofresidual waste (derived from sorting analyses) wereboth defined in order to be collected as data.

The investigation covered the collection andinspection of the mentioned waste-related data aswell as the data in terms of economic, demographicand social indicators in all major European citieswith more than 500,000 inhabitants. Together withsix regional partners we co-operated with local cityrepresentatives who provided us with waste-relatedand socio-economic data at the city level.Additionally, national data were obtained frominternational organisations, such as the UnitedNations or OECD.

To enable the analysis of regional dynamics, thecollection of data covered the years from 1970 to2001. This formed the basis for the hypothesis ofthe existence of a long-term development path,which is based on the analogous time-shiftedchanges of different regions.

Finally, it was possible to collect data about themunicipal solid waste quantities (including generaldata at the city and country level) in 55 major cities(out of a total of 65) in the EU-15 and 10 CEEcountries with an average time-series length of tenyears. In terms of the waste potential of the main

materials, only 45 data sets from 31 cities wereavailable.

4. STATISTICAL EVALUATION

There are remarkable differences in the MSWgeneration rates as well as in the growth rates inEuropean cities. As an example, a comparison ofeconomic areas in the year 2000 shows that majorEU-15 cities were characterised by far higherMSW generation rates (510 kg/cap/yr) than theCEE cities (354 kg/cap/yr), while from 1995 to2001 annual growth in CEE cities is more thantwice as high (4.3%) as in cities of EU-15countries (1.8%).

Therefore several bivariate and multivariatestatistical analyses were carried out to identifyindicators with a significant impact on MSW gen-eration and composition. Table 1 shows the con-sidered socio-economic indicators, which wereavailable as a time-series at both the city andnational level.

Table 1. Available socio-economic indicators.

Available indicators at city and national level

• Population • Population density

• Population age structure(0 to 14 years / 15 to 59years / 60 and more years)

• Sectoral employment(Agriculture / Industry /Services)

• Gross domestic product • Infant mortality rate

• Overnight stays • Life expectancy at birth

• Average household size • Unemployment rate

Firstly single regression analyses (using Kolmogo-rov-Smirnov tests) proved that the single para-meters at the city level explain only anunsatisfactory part of the intertemporal andinterregional variance. The infant mortality rate(R²=0.37, n=86) as the most significant indicatorperformed even better than the commonly usedgross domestic product (R²=0.22, n=86) as well asall other mentioned urban indicators. This ismainly due to the fact that the (usually available)mean urban gross domestic product is a lessmeaningful indicator of the social standard than theinfant mortality rate as it does not reflect the socialand economic inequality, which is especially highwithin CEE cities [Förster et al., 2002].

Secondly the hypothesis of a general long-termdevelopment path concerning urban wastegeneration was tested by means of multivariateanalyses. A hierarchical cluster analysis wastherefore carried out to categorise each case(representing a city in a certain year) within

homogeneous groups with a similar social andeconomic standard, here defined as the ‘prosperitylevel.’ Four national development indicators (seeTable 2) were selected as cluster criteria in order toexplain this latent ‘prosperity level’ variable.

Table 2 shows the assumed prosperity indicatorsand the MSW generation rates, which prove thehypothesised relationship. Low MSW generationrates coincide with low gross domestic products,high infant mortality rates and agriculturally domi-nated economy and vice versa. Similar results wereobtained by a principal component analysis. Ananalysis of variance using One-way-ANOVA con-firmed the rejection of the null hypothesis, whichstates the identity of the group means, with an F-value of 61.7 and an F-significance of below 0.1%.

Table 2. Municipal solid waste generated anddevelopment indicators.

National Prosperitydevelopment levelindicators andMSW generation

Low Medium

High Veryhigh

Gross domesticproduct per capita1

5841 11400 19418 21317

Infant mortality rate2 15.0 8.7 7.6 5.5

Labour force inagriculture (%)

24.0 18.7 4.8 3.2

Labour force inservices (%)

44.4 52.2 59.4 66.2

Municipal solidwaste (kg/cap/yr)

287 367 415 495

1 USD Purchasing power parities at 1995 prices2 Per 1,000 births

Based on this prosperity-related classification, theanalysis of the waste potential data unveiled long-term trends in the municipal solid waste streams(Figure 2). While the generation of paper andcardboard, glass, plastics and compounds in

0

100

200

300

400

500

low medium high very highProsperity level

MSW

[kg/

cap/

yr]

Otherfractions

Metals

Plastics and compounds

Glass

Paper andcardboard

Organicwaste

Figure 2. Municipal solid waste streams atdifferent prosperity levels.

prosperous cities is significantly higher both inabsolute (per capita) and relative (masspercentage) figures, the amount of organic waste

generated is very similar in these four city groups.Additionally the impact of the remaining MSWindicators (see Table 1) was tested and is closelyrelated to the model development described below.

5. MODEL

5.1 Approach

The developed model was designed for repeateduse by municipal representatives to appropriatelyassess the future municipal waste streams of majorEuropean cities. In order to support decisionsconcerning waste management strategies, theplanning horizon was defined by 15 years.

Due to the background of these addressed users(municipal representatives), a model concept wascreated in order to enable a suitable compromisebetween ease of use and forecasting accuracy.Usability primarily refers to the preference formodel input parameters that are adequatelypredictable at a city level, such as demographic orsocial indicators.

The selection of the applied approach was basedon recent forecasting methodology [Armstrong,2001]. The size and type of database as well as theexisting knowledge of relationships were thecriteria for the method selection.

Thus different methodological approaches wereselected for the following two modules: an ‘MSWgeneration module’ forecasting the total MSWgeneration rate and an ‘MSW composition module’assessing the future mass percentage of main wastestreams within the municipal solid waste. Theextensive database with time series and cross-sectional data from the total MSW quantitiesallowed the implementation of an econometricmodel. The small database concerning wastepotentials enabled the application of only thecomparably simple extrapolation method.

Figure 3 shows the elements of the overall model.The first calculation step is similar for bothmodules and is based on the findings for the long-term analogies between prosperity and MSWgeneration. A given city in a defined future yearwill be assigned to one of four prosperity groupsusing forecasts or assumptions for development-related indicators, such as the gross domesticproduct or infant mortality rate.

The MSW generation module is an econometricmodel, which consists of a system of multiplelinear equations for each city group. Due to theshort length of the available time series, theclassical econometric approach, rather than the

vector autoregression (VAR) approach, wasimplemented.

The MSW composition module is based on theassumption that prosperity-related changes in thewaste composition are similar. Due to the limitedwaste potential data, the objective lies on a roughtrend estimation represented as default values.

Development indicators

MSW-related parameters

Mod

el in

put

Stage of prosperity?

Equation - group 1

Equation - group 2

Equation - group x

:

Econ

omet

ric m

odel

Long

-term

an

alog

ies

MSW composition by main fractions

MSW growth rate per capita

Mod

el o

utpu

tExtrapolation

Figure 3. MSW generation model.

5.2 Implementation issues

The main issues concerning the implementationrefer to the estimation of the econometric modelusing socio-economic data, which are representedas cross-sectional time series. The followingpotential problems were considered to avoidmisspecifications of the model [Armstrong, 2001]:

• Collinearity of parameters: Social andeconomic indicators often are highlycorrelated. The inclusion of too manyvariables within multiple regression modelstypically causes collinearity problems leadingto ill-conditioned models. A suitable measurefor the identification of collinear variables isthe variance inflation factor that was thereforeused.

• Autocorrelations between residuals ofneighbouring cases often occur duringanalyses of time series or structured data.This can depend on wrongly assumedfunctional relationships or on measurementerrors during the data collection. Thusresidual analyses including the use of Durbin-Watson coefficients were applied.

• Outliers: The full consideration of outliers,which can occur due to measurement errors (a

well-known problem in waste management),deteriorates the accuracy of regression modelestimations. Hence the median absolutepercentage error (MdAPE) [Armstrong,2001] was used to avoid theoverrepresentation of unrealistic values.

5.3 Procedures

The attribution of the cases according theprosperity level was based on the classificationmentioned in Chapter 4. The initial specificationincluded the indicators listed in Table 1. The finalmodel was selected by backward regression usingthe ordinary least squares method. To avoidautocorrelation and collinearity problems, Durbin-Watson statistics, collinearity tests and residualanalyses were carried out.

6. RESULTS

The estimated equations of the final MSWgeneration model for cities are represented by:

tt GDPMSW ⋅+= 014.05.359 (1)( )t

urbINFlog1.197 ⋅−

for cities with very high prosperity,

tt GDPMSW ⋅+= 016.05.276 (2)( )t

urbINFlog5.126 ⋅−

for cities with high prosperity, and

( )tnat

t INFMSW log6.3757.360 ⋅−−= (3)tt HHSIZEPOP ⋅−⋅+ − 9.12393.8 5915

tLIFEEXP⋅+ 7.11

for cities with low or medium prosperity,

where MSWt is the municipal solid wastegenerated per capita and year, GDPt is the nationalgross domestic product per capita at 1995purchasing power parities, INF is the infantmortality rate per 1,000 births in the city (INFurb)or in the country (INFnat), POP15-59

t is thepercentage of the population aged 15 to 59 years,HHSIZEt is the average household size andLIFEEXPt is the life expectancy at birth and t is theyear.

Concerning the definition of prosperity levels,Table 3 shows the approximate threshold valuesfor three national indicators among the city groups.For exact calculations, a set of canonicaldiscriminant functions was determined to allocate acity with a given (present or future) characteristic.

All parameters are significant at the 5% error level.Only the ‘infant mortality rate’ parameter is log

transformed because of its obviously exponentialnature. The model explains 65% of the variation ofthe MSW generation rate per capita between citiesand in time. The model was validated with a hold-out sample which included 59% of all cases. Theout-of-sample error represents a median absolutepercentage error of 8.0%, providing a useful modelfor waste management planning.

Table 3. Approximate threshold values betweencity groups (national indicators).

Prosperity Gross domestic Infant Labour forcelevel product1 mortality rate2 in agriculture (%)

Low7,100 12.0 21.4

Medium13,800 8.1 10.5

High20,200 6.3 4.0

Very high1 USD Purchasing power parities per capita at 1995 prices2 Per 1,000 births

In the following, the factors are described whichwere found to have a significant impact on theamount of municipal solid waste:

• Gross domestic product: This commonlyused indicator proved to be a significantfactor in cities with a high prosperity, but notfor cities with a lower economic output. Thisis due to the fact that the high regionalincome inequality in CEE countries [Försteret al., 2002] causes a big gap between theusually available mean values compared withthe clearly lower median values, which is amore meaningful, but rarely observedindicator for the social well-being and livingstandard.

• Social indicators: Previous studies neverused the infant mortality rate and lifeexpectancy parameters to indicate MSWgeneration, but they showed a remarkableability to serve as an additional or alternativevariable for the gross domestic product. Theadvantages of their use are the highexplanatory power for regional welfare, thegood availability of data, the high quality ofdata (due to easy compilation withoutcomplicated definitions) and the relativelygood predictability on the part of citymunicipalities.

• Age structure: The positive relationshipbetween the percentage of the medium agegroup and MSW generation confirmed theprevious studies [Sircar et al., 2003; Lindh,2003].

• Household size: As in the study of Dennisonet al. [1996], the significantly negative

relationship between the average householdsize and MSW generation was analysed on aregional scale.

7. CONCLUSION AND OUTLOOK

At present environmental models, such as in wastemanagement, are characterised by a high level ofsophistication in terms of the technical andenvironmental optimisation within the drawnsystem boundaries, but also by an almost completelack of consideration of social, demographic andeconomic border conditions. In waste managementthe usual municipal and regional waste forecasts(as the key starting point for the development ofwaste management plans) regard the totalpopulation as the only input parameter[Karavezyris, 2001]. As a consequence thesestrongly simplified assumptions can favour falseestimations concerning future waste treatmentcapacities and dimensioning of restructuringactivities.

This approach aimed to integrate the impacts ofregional socio-demographic and economicdynamics on the municipal solid waste generation.A qualitative, analogy-related approach wascombined with the econometric methodology inorder to assess these relationships for 55 veryheterogeneous European cities in a long-termperspective. The results showed that this model canbe implemented as a helpful decision support toolfor municipal waste planners.

In future work, this model will be statisticallyrefined (especially the composition module) andrealised in aJava environment. The beta versionwill be tested in six European cities from regionswith rapidly growing economies to verify itspracticability.

6. ACKNOWLEDGEMENTS

This study is part of a project initiated within theEuropean Commission's Fifth FrameworkProgramme (EVK4-CT-2002-00087).

The authors wish to thank the following people fortheir dedicated work during the course of theinvestigation (in alphabetical order): HeinrichRiegler (BOKU – University of Natural Resourcesand Applied Life Sciences, Austria); Emilia denBoer, Jan den Boer (Darmstadt University ofTechnology, Germany); Panagiotis Panagiota-kopoulos, Nantia Tsilemou (Democritus Universityof Thrace, Greece); Pouwel Inberg, Marjakke vander Sluis, Wim van Veen (De Straat Milieu-adviseurs, the Netherlands); Lothar Schanne(novaTec, Luxembourg); Francesc Castells,

Montse Meneses, Julio Rodrigo (UniversitatRovira i Virgili, Spain); Iwona Mackow, PawelMrowinski, Marta Sebastian and Ryszard Szpadt(Wroclaw University of Technology, Poland).

8. REFERENCES

Armstrong, J.S., Principles of forecasting: ahandbook for researchers and practitioners,Kluwer Academic Publishers, Boston, 2001.

Bach, H., A. Mild, M. Natter, and A. Weber,Combining socio-demographic and logisticfactors to explain the generation andcollection of waste paper, Resources,Conservation and Recycling, In Press, 2003.

Beigl, P., G. Wassermann, F. Schneider, and S.Salhofer, Municipal solid waste generationtrends in European countries and cities, paperpresented at the 9th International WasteManagement and Landfill Symposium, CISA,S. Margherita di Pula (Cagliari), Sardinia,Italy, October 6-10, 2003.

Björklund, A., Environmental systems analysis ofwaste management – Experiences fromapplications of the ORWARE model, Ph.D.thesis, Royal Institute of Technology,Stockholm, 2000.

Dennison, G.J., V.A. Dodd, and B. Whelan, Asocio-economic based survey of householdwaste characteristics in the city of Dublin,Ireland, II. Waste quantities. Resources,Conservation and Recycling, 17, 245-257,1996.

Förster, M., D. Jesuit and T. Smeeding, Regionalpoverty and income inequality in central andeastern Europe: Evidence from theLuxembourg income study, SyracuseUniversity, New York, 2002.

Karavezyris, V., Prognose von Siedlungsabfällen:Untersuchungen zu determinierendenFaktoren und methodischen Ansätzen, TKVerlag, Neuruppin, 2001.

Lindh, T., Demography as a forecasting tool,Futures, 35, 37-48, 2003.

Patel, M.K., E. Jochem, P. Radgem and E.Worrell, Plastic streams in Germany – ananalysis of production, consumption andwaste generation, Resources, Conservationand Recycling, 24, 191-215, 1998.

Sircar, R., F. Ewert, and U. Bohn, GanzheitlichePrognose von Siedlungsabfällen, Müll undAbfall, 1, 7-11, 2003.

White, P., M. Franke and P. Hindle, Integratedwaste management: A lifecycle inventory,Aspen Publishers, Maryland, USA, 1999.