A GIS-based zoning of illegal dumping potential for efficient surveillance.pdf

8/19/2019 A GIS-based zoning of illegal dumping potential for efficient surveillance.pdf

1/12

A GIS-based zoning of illegal dumping potential forefficient surveillance

Tomohiro Tasaki a,*, Takatsune Kawahata a, Masahiro Osako a, Yasuhiro Matsui b,Susumu Takagishi c, Akihiro Morita c, Shigeki Akishima c

a Research Center for Material Cycles and Waste Management, National Institute for Environmental Studies,

16-2 Onogawa, Tsukuba, Ibaraki 305-8506, Japanb Okayama University, 3-1-1, Tsushimanaka, Okayama 700-8530, Japan

c PASCO Co., Higashiyama Building, 1-1-2, Higashiyama, Meguro-ku, Tokyo 153-0043, Japan

Accepted 4 January 2006Available online 13 March 2006

Abstract

To assist in the efficient surveillance against illegal dumping, this study examined and evaluated two methods to illustrate the illegaldumping potential of sites using GIS (Geographic Information System) data. One approach focused on the occurrence of illegal dumpingsites; the other on the size of the illegal dumping. Both approaches to zoning were implemented for the Kanto region of Japan, utilizingseven or eight major geographical attributes most closely related to illegal dumping. The zoning results revealed the areas requiringpatrols against illegal dumping. Evaluation of the zoning results using the ROC (Receiver Operating Characteristic) curve showedthe number of illegal dumping sites detectable under certain surveillance conditions and that the size-based zoning was superior, but thissuperiority was insignificant for revealing sites with higher potential for large illegal dumping, for which it would be sufficient to use the

occurrence-based zoning. The evaluation also showed the contribution of each geographical attribute. Finally, application of the ROCcurve to the surveillance planning process was examined, which enables the total social cost of pollution by illegal dumping, rehabilita-tion of dumping sites, and illegal dumping surveillance to be minimized. 2006 Elsevier Ltd. All rights reserved.

1. Introduction

Illegal dumping causes severe environmental damage.According to an investigation conducted by the Ministryof Environment of Japan (1999–2001), approximately400,000 tonnes of illegally-dumped industrial wastes are

discovered annually in Japan, with four out of every 10such sites yet to be rehabilitated. Precautionary measuresare no doubt important and several previous researchefforts (Choe and Fraser, 1999; Fullerton and Kinnaman,1995; Sigman, 1998; Xepapadeas, 1991) showed that botha decrease in waste disposal fees and an increase in finesfor illegal dumping were key factors in preventing illegaldumping. However, neither one appears to be a break-

through in preventing illegal dumping. One reason is thatwaste disposal fees have been increasing due to stricter reg-ulations concerning waste treatment in the interest of betterenvironmental conservation. Another reason is that theJapanese government has already raised the maximum ille-gal dumping fine for a company several times, by more

than a 100-fold, to 100 million yen (approximately US$1million).An alternative measure against illegal dumping is sur-

veillance. Choe and Fraser (1999) pointed out its impor-tance based on the results of their theoreticalexamination on garbage pricing. As an example of a localgovernment that opted for intensive surveillance, Nara pre-fecture (Japan Industrial Newspaper, 2002) designated 750sites for ordinary surveillance. However, progressive localgovernments such as Nara prefecture selected patrol sitesbased only on past experience rather than objective site

0956-053X/$ - see front matter 2006 Elsevier Ltd. All rights reserved.

doi:10.1016/j.wasman.2006.01.018

* Corresponding author. Tel.: +81 29 850 2988; fax: +81 29 850 2830.E-mail address: [email protected] (T. Tasaki).

www.elsevier.com/locate/wasman

Waste Management 27 (2007) 256–267

mailto:[email protected]:[email protected]


2/12

data. Moreover, surveillance by local governmental officersalone involves budgetary and time constraints. Consideringthe fact that many local governments in Japan have createdand utilized patrolling and reporting systems hiring ex-police officers and collaborating citizens (Japan IndustrialWaste Technology Center, 2002), it would be useful to

show people areas where waste is frequently illegallydumped. Actually, one community organization, NewSpirit (East St. Louis, IL, USA) uses hand-drawn commu-nity maps to identify illegal dumping area (US EPA Region5, 1998). However, these maps merely showed discoveredillegal dumping sites, and did not illustrate undiscoveredillegal dumping sites or the illegal dumping potential of every site within a particular area.

This study therefore developed a method of zoning toillustrate the illegal dumping potential of sites using theGeographic Information System (GIS), in order to findundiscovered illegal dumping sites, improve the efficiencyof surveillance, and inform and warn governmental officers

of the sites with high illegal dumping potential. Withregard to the use of information technology in wastemanagement, for example, Chang et al. (2001) created anInternet Web-based information system to exchange infor-mation about scrap vehicles in Taiwan, including spatialdata on the location of recyclers; Kontos et al. (2005) eval-uated the suitability of selecting an optimal site for landfillusing GIS; and the US EPA Region 5 (2000) developed theIllegal Dumping Economic Assessment (IDEA) model toconduct a cost analysis of cleaning up illegal dumping sites.Unlike these previous works, this study focused on surveil-lance against illegal dumping and tried to provide another

technique and complementary information for measuresagainst illegal dumping.

This paper is organized as follows. Firstly, we discussthe methodology of zoning. Secondly, we create zoningmaps employing two approaches and using seven or eightmajor geographical attributes most closely related to illegaldumping. The first approach focuses on the number of ille-gal dumping sites, while the other focuses on the size of theillegal dumping. The target area is the Kanto region(around Tokyo). Thirdly, we evaluate the two zoning meth-ods using the ROC (Receiver Operating Characteristic)curve to determine the superiority of the zoning methodsand the contribution of each geographical attribute (detailsof the ROC curve will be explained in Section 2.5). Finally,we examine an application of the ROC curve for planningillegal dumping surveillance based on the costs of surveil-lance, rehabilitation, and environmental pollution.

2. Methods

2.1. Zoning methods for surveillance against illegal dumping

This study examined two zoning methods, based oneither the occurrence or the size of illegal dumping depend-ing on the geographical attributes of individual sites, out of

four possible zoning methods that were considered. The

degree of necessity for illegal dumping surveillance can beexpressed, similarly to the concept of risk, as:

ðDegree of necessity for surveillanceÞ

¼ ðProbability of occurrence of illegal dumping at a

certain siteÞ ðExpected size of illegal dumping

when it occursÞ ð1Þ

The occurrence and the size of illegal dumping are there-fore two elements of zoning; the other two are geographicalattributes of the site and the conditions of waste transpor-tation around the site. Geographical attributes, whichdetermine visibility of the site, etc., affect both the occur-rence and the size of illegal dumping. The conditions of waste transportation also affect both the occurrence andthe size: the more waste that is transported near a site,the higher the probability and larger the size of illegaldumping expected there. Hence, the following four combi-nations are the basis of the four zoning methods consid-

ered: (1) geographical attributes and occurrence, (2)geographical attributes and size, (3) waste transportationand occurrence, and (4) waste transportation and size.However, there are two disadvantages associated with zon-ing based on waste transportation. One is the difficulty of obtaining data concerning waste transportation. The otheris that the utility value of the zoning result would quicklydecrease because the conditions of waste transportationtend to change rapidly compared to the geographical attri-butes of sites. Thus, we did not examine the two zoningmethods based on waste transportation in this study.

It is not necessarily better to take both the occurrence

and the size into account simultaneously for the zoningbased on the geographical attributes, since some govern-ment officers have expressed the opinions: ‘‘the mere detec-tion of illegal dumping is significant, regardless of the size’’and ‘‘existing waste attracts more waste, and an illegaldumping site tends to expand if countermeasures fail atan early stage,’’ as well as ‘‘since illegal dumping is an ille-gal act, surveillance should be extensive regardless of thesize and consecutive effects.’’ In view of these opinions,we examined two methods of zoning based on occurrenceand size individually.

2.2. Zoning method based on occurrence of illegal dumping

The illegal dumping occurrence potential, which wastaken as the probability of occurrence of illegal dumping,was defined as:

Z ¼Y

a

P ai ð2Þ

where P ai is the relative probability of occurrence of illegaldumping, defined as:

P ai ¼ ð Dai= N aiÞ=ð DT= N Þ ð3Þ

where Dai is the number of illegal dumping sites under Cat-egory i of Geographical attribute a within the target area,

DT is the total number of illegal dumping sites within the

T. Tasaki et al. / Waste Management 27 (2007) 256–267 257


3/12

target area, N ai is the number of sites under Category i of Geographical attribute a within the target area, and N T isthe total number of sites within the target area. The unitof site in this study was municipality for the regional-levelanalysis and 25-m mesh for the site level. (Japan hasapproximately 3300 municipalities, each with an average

land area of about 100 km2

.) A relative probability exceed-ing 1 indicates that the site is more frequently used for ille-gal dumping than average. The reason for using relativeprobabilities is to correct the difference in the number of sites under each individual category within a geographicalattribute, thus ensuring that the probabilities of a categorydo not appear higher simply due to the greater number of sites under the category in question.

The primary reasons to select this approach stemmedfrom the characteristics of illegal dumping data, eventhough other zoning approaches may be applicable. Theavailable data for illegal dumping sites can be classifiedinto two types: (a) data concerning sites where illegal

dumping has been discovered and reported, and (b) dataconcerning other sites. Typical approaches in this caseinvolve the use of discriminant analysis or a quantificationmethod of the second type, using the data in both (a) and(b) as external criteria for the occurrence or otherwise of illegal dumping. However, these methods would be inap-propriate because the (b) data do not indicate that illegaldumping will not occur at the site in the future, and thedata also include illegal dumping sites that are undiscov-ered or not reported.

Therefore, this study examined a zoning approach thatuses the (a) data of illegal dumping sites only. This

approach assumes that a site with similar geographicalattributes to those of discovered illegal dumping sites hasa higher illegal dumping potential. In other words, siteswith a higher illegal dumping potential include both ‘‘thosenot yet used for illegal dumping but where there is futureillegal dumping potential’’ and ‘‘those actually used for ille-gal dumping but as yet undiscovered or unreported.’’ Thevalidity of the assumption will be examined in Section3.3.1 using illegal dumping data of different years; one setof data is for known illegal dumping and the other forunknown or future illegal dumping. This study used multi-plication in Eq. (2) for integrating different geographicalattributes because it was more appropriate than additionas discussed in Appendix 1.

2.3. Zoning method based on size of illegal dumping

A similar methodology was applied for the zoning of illegal dumping based on size owing to the aforementionedcharacteristics concerning available data on illegal dump-ing sites, but only the data of large illegal dumping wereused. The relative probabilities of occurrence of large ille-gal dumping sites were calculated by using Eq. (3), andthen the illegal dumping potential in terms of size wasobtained from Eq. (2). Here, large illegal dumping was

defined as 300 tonnes or more of illegal dumping, based

on the histogram of illegal dumping data shown inFig. 1. The illegal dumping data was that from FY 2000and 2001 in three local prefectures: Chiba, Ibaraki, andTochigi (the top three in terms of the amount of illegally-dumped waste for 1999–2002 in the Kanto region, whichis the region around Tokyo that generates the most waste

in Japan). Fig. 1 shows that 300 tonnes or more of illegaldumping accounted for 15% in terms of the number of cases of occurrence but no less than 75% by weight, andthat 1000 tonnes or more of illegal dumping accountedfor only 3% of the overall total number of the cases. Itwould thus be inappropriate to use data from this lattercategory given the scarcity of data available.

2.4. Data concerning illegal dumping sites for zoning

Data were collected on 608 illegal dumping sites at theregional level in units of municipality, as well as data on421 sites at the site level in mesh units of 25 m. The former

data were collected by the Ministry of Environment of Japan (2000–2002) for FY 2000. The latter data were col-lected by the authors in collaboration with three local pre-fectures: Chiba, Ibaraki, and Tochigi, for FY 2001.Prefectural officers were asked to mark all illegal dumpingsites on a map of scale 1:5000 or 1:50,000, which were thenused to create GIS data (using ArcInfo 8.1 (ESRI) as theGIS software package). Although site-level data are espe-cially important when deciding the patrolling routes, regio-nal-level data were also used for compensatory purposes,given the absence of site-level data on social and economicaspects relating to illegal dumping.

Data were also collected concerning 31 geographicalattributes such as population density, livelihood, industry,crime, land use, accessibility (distance from a road),

0

20

40

60

80

100

120

140

160

180

≤

0 1

≤

0 3

≤ 1 0 0

≤

0 0 3

≤ 1 , 0 0 0

v o

e r

0 , 1

0 0

Size of illegal dumping (tonnes)

F r

q e

c n e u

y

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

C u m u l a t i v e % i n

t e m r s o f w e i g h t

Fig. 1. Histogram of illegal dumping in three prefectures (Chiba, Ibaraki,

and Tochigi).

258 T. Tasaki et al. / Waste Management 27 (2007) 256–267

http://-/?-http://-/?-


4/12

topographical features, distance from police stations andadministrative boundaries. These data were attached toeach site datum on illegal dumping. Results from a previ-ous study (Tasaki et al., 2004) concluded that seven majorattributes (population density, unemployment rate, incomeper capita, distances from main roads (national and main

local roads) and roads (over 3 m in width), distance fromforest edge, and topographical features) were the most clo-sely related to illegal dumping among the 31 relevant geo-graphical attributes studied (the relative probabilities forthese data are shown in Section 3.2). The remaining 24attributes either did not adequately explain the phenomenaof illegal dumping reported by local government officers(for explanation of illegal dumping, see US EPA Region5 (1998, 1997) and (National Institute for EnvironmentalStudies, 2002), or were highly correlated (based on theresults of correlation and factor analysis) with the sevenaforementioned major attributes. Therefore, these sevengeographical attributes were used for the occurrence-based

zoning. For the size-based zoning, inclinations (in gradientsof 50 m) were also used, showing the difference between therelative probabilities overall and those for large illegaldumping, as shown in Section 3.2.

Incidentally, regional-level data concerning illegaldumping within certain municipalities included two ormore illegal dumping sites. No distinction was made of the actual number of illegal dumping sites within a munic-ipality during the occurrence-based zoning process, i.e.,this study considered the ‘‘occurrence’’ only for a coupleof reasons. The first was related to the characteristics of the data: the number of illegal dumping sites did not neces-

sarily correlate with the degree of occurrence of illegaldumping. A difference in such numbers between municipal-ities may represent differences in the intensities of surveil-lance between the municipalities. The other reason wasrelated to the coherency between analyses at the regionallevel and the site level. If the number of dumping incidentsin a municipality is taken into account for regional-levelanalysis, the number of illegal dumping acts at a certain sitemust be taken into account for site-level analysis. This isdifficult because there are no data to distinguish each indi-vidual dumping act and to determine the number of indi-vidual illegal dumping acts.

There is a similar issue with regional-level data concern-ing the size of illegal dumping. Regional-level size data rep-resent the total amount of illegally-dumped wastes in amunicipality. The size data of municipalities that have twoor more illegal dumping sites may be the sum of small illegaldumping sites. In order to completely exclude such data of small illegal dumping sites from the analysis, the averageamount of illegally-dumped waste was used instead for cal-culating the relative probabilities for the size-based zoning.

2.5. Evaluation of zoning results

For evaluating zoning results, the concept of the ROC

curve was applied, which is used in clinical decision-making

for comparing and determining the superiority of diagnos-tic tests. Other techniques for the evaluation, such as Pear-son’s correlation coefficient, could have been used.However, the ROC curve was chosen for the following rea-sons: it can provide much information regarding the rela-tionship between discovery of illegal dumping and patrol

conditions; it can be used for simple patrol planning as dis-cussed in Section 3.4; and it is an established useful methodin clinical science. An ROC curve is formulated by plotting‘‘sensitivity’’ (representing the percentage of real detectionamong ‘‘positive’’ results of a diagnostic test.) on the y-axisand ‘‘false positive rate’’ (the percentage of misjudgmentamong the ‘‘negative’’ results of a diagnostic test) on thex-axis for different diagnostic conditions. Let us considerthe case that patrolling is conducted for sites whose illegaldumping potentials exceeds a certain threshold value andthat a matrix of a zoning result and patrol condition iswritten in the form of Table 1 (here, a b, c d , anda + b + c + d is the total number of sites within a target

area: approximately 28 million for this study). Then, thesensitivity y is expressed by Eq. (4) and represents the per-centage of illegal dumping sites to be discovered becausea + c is the total number of illegal dumping sites withinthe target area. The false positive rate x is expressed byEq. (5) and represents the percentage of sites under surveil-lance because a + b is the total number of patrolled siteswithin the target area.

y ¼ a=ða þ cÞ ð4Þ

x ¼ b=ðb þ d Þ ða þ bÞ=ða þ b þ c þ d Þ ð5Þ

As the threshold value of patrolling for illegal dumping po-

tential decreases from its maximum value, x and y both in-crease, whereas y/x decreases because patrolling is startedfrom the site with highest illegal dumping potential to thatwith lower potential. The locus of (x, y) forms the ROCcurve, which is convex to the upper left. A more convexcurve to the upper left indicates a superior result, the sameas in the evaluation of an ROC curve – higher sensitivityand a lower false positive rate. If the ROC curve forms astraight line of 45, it means that patrolling either the high-er potential sites or patrolling the lower potential sites dis-covers the same number of illegal dumping sites, and hencethe zoning result is useless.

We prepared two sets of site-level data for evaluation.One set was data for 421 sites in FY 2001, while the otherwas data for 193 sites for FY 2000. Both data were col-

Table 1Number of sites to be patrolled and illegal dumping sites to be discovered

Illegal dumping No il legal dumping

Patrolled a bNot patrolled c d

a, b, c, and d represent the number of sites.a + b + c + d : total number of sites within the target area.

a + c: total number of illegal dumping sites.a + b: total number of patrolled sites.

a: the number of discovered illegal dumping sites.


http://-/?-http://-/?-


5/12

lected from the three prefectures as mentioned in Section2.4.

3. Results

3.1. Relative probabilities of occurrence of illegal dumping

The relative probabilities of occurrence of all illegaldumping and large illegal dumping are shown in Fig. 2,with major differences between all and large illegal dump-ing shown in histograms of population density, income,depth from forest edge, topographical features, and incli-nation. With regard to population density, urban areaswith over 1000 people/km2 naturally meant a lower proba-bility of illegal dumping, while conversely, rural areas withdensities of 100–1000 people/km2 had higher probabilities.In sparsely populated areas, however, with 100 people/km2

or less, the probabilities decreased once again. Since suchareas were probably remote, with lower accessibility, this

decreased the probabilities of illegal dumping. The distri-bution of large illegal dumping seems to be shifted to theleft and to be narrower in width, compared to that of allillegal dumping, meaning that while large illegal dumpingrequires more space to dump (the left-shift in the histo-gram), dumping in remote areas requires more moneyand time, which decreases the probability thereof.

With regard to income, poorer areas showed a propen-sity toward having a higher probability in general. Munic-ipalities with an income under 1 million yen/capita,however, showed a higher probability of overall illegaldumping but lower probabilities for large illegal dumping.

Since municipalities with an income under 0.5 million yen/capita accounted only for 0.5% of all municipalities inJapan, this is interpreted as indicating that large illegaldumping had simply not yet occurred in these municipali-ties. However, it is difficult to explain the phenomenon of lower probabilities for municipalities with an income inthe range of 0.5–1 million yen/capita.

With regard to depth from the forest edge, which wasselected as a parameter concerning visibility, areas awayfrom the forest (over 100 m) showed a lower probability.Areas near a forest edge (within 100 m) had higher proba-bilities due to reduced visibility, but those actually insidethe forest had rather lower probabilities, despite the lackof visibility. This phenomenon can be explained as follows.Dumping inside a forest would require both deforestationand a road traversing the forest to access the dumping site.The preparation costs required probably reduce the proba-bility of illegal dumping, especially for large illegaldumping.

With regard to topographical features, hills showed thehighest probability. When probabilities between all andlarge illegal dumping categories were compared, the prob-ability of large illegal dumping on a hill was even higher,while that for a flat/marsh area was lower. To dump a largeamount of waste, it is more convenient to utilize natural

geography, and dumping a considerable amount of waste

in a flat/marsh area would require digging holes and/orfurther piling up of waste. A similar explanation appliesto slopes. Areas with an inclination of 3–10 had a higherprobability of large illegal dumping compared to all illegaldumping, probably due to the propensity to utilize the nat-ural geology for dumping purposes. In areas with an incli-

nation exceeding 10, however, it is physically moredifficult to dump and the relative inaccessibility of suchareas probably reduces the probability of illegal dumping.

The probabilities of the other geographical attributesdiffered little between all and large illegal dumping. Theprobability of unemployment showed that poor areas havea higher probability of illegal dumping. Although areaswith an unemployment rate of over 6% have different ten-dencies, only three prefectures of all 47 in Japan met thiscriterion. Thus, the conclusion was that unemploymentdata for the area were not representative of areas as awhole. Distance from main roads and roads also affectsillegal dumping, since these factors affect accessibility to

the potential dumping site. Most illegal dumping occursin areas within 100 m from these roads.

3.2. Results of zoning

The results of occurrence-based zoning in the Kantoregion are shown in Fig. 3, with darker areas representingsites with a higher illegal dumping potential. (The zoningresults were divided into 10 ranks using ArcInfo’s equal-division function. Because there were a number of areaswith the same zoning value Z , the area of each rank wasnot the same.) The overall characteristics and tendencies

of zoning results were the same for occurrence-based zon-ing and size-based zoning. The black and white areasshown in Fig. 3 were located in specific regions, rather thanbeing randomly distributed. Most of the black areas werein the three prefectures of the eastern and northern partsof the Kanto region: Chiba, Ibaraki, and Tochigi, whichcorrelate to the current state of illegal dumping in Japan.These areas are outside Tokyo and Yokohama (a city inthe eastern part of Kanagawa prefecture), the most urban-ized areas in the Kanto region, where zoning potentialswere much lower. The whitest area (Rank 1) in theupper-left area of Fig. 3, also outside Tokyo and Yoko-hama, was an uninhabitable mountainous area.

3.3. Evaluation of zoning results

3.3.1. Occurrence-based zoning

The result of the evaluation of occurrence-based zoningis shown in Fig. 4. Here, data for all illegal dumping in FY2001 and FY 2000 were used for the evaluation. The ROCcurves of the FY 2000 data showed superiority, but this isinsignificant. The important point is that the difference inROC curves indicates the robust accuracy of the zoningresult for different years and supports the assumption men-tioned in Section 2.2. By patrolling approximately 30% of

the total area, 60–75% of illegal dumping sites would be



6/12

discovered, and patrolling approximately 20% of the totalarea would result in the discovery of 45–60% of illegaldumping sites.

In order to determine the contribution of each geo-graphical attribute to the zoning result in terms of occur-

rence, the ROC curves for each geographical attribute

were calculated in the same way, and then the area betweenthe ROC curve and 45 line was calculated. (The 45 linedoes not represent a relationship between a geographicalattribute and the occurrence of illegal dumping.) Theresults are shown in Table 2. Population density, distance

from the forest edge, topographical features, and distance

Fig. 2. Relative probabilities of occurrence of illegal dumping for major geographical attributes of all and large illegal dumping (R: regional level, S : sitelevel).



7/12

from roads were the primary factors affecting occurrence-based zoning results and can be individually interpretedas factors of the intensity of surveillance, visibility, naturalgeography and accessibility, respectively. On the otherhand, economic factors (i.e., unemployment rate andincome per capita) had a relatively insignificant effect.(The non-positive figure for income per capita indicatesthat it includes errors of data and, more importantly, thatthis geographical attribute does not affect the occurrence of illegal dumping at least in the Kanto region. This does notcontradict the result of analysis for all municipalities in

Japan (Fig. 2).) The distance from main roads also did

not contribute any more than that from roads, implyingthat local rather than regional accessibility is more relatedto illegal dumping. Incidentally, the furthest left plot in theROC curve concerning distance from roads is at approxi-mately 65% of the x-axis and represents areas within100 m of roads. This means that areas within 100 maccount for most of those in the Kanto region and suchinformation involving patrols of the area within 100 m isless helpful for efficient surveillance.

3.3.2. Size-based zoning

The evaluation results of the size-based zoning areshown in Fig. 5. Here, data concerning large illegal dump-ing (300 tonnes or more) in FY 2001 were used for evalu-ation. Comparing the ROC curve of occurrence-basedzoning with size-based zoning, the latter was shown to besuperior because it was positioned slightly further to theupper-left. However, with regard to higher zoning ranks,i.e., to the left of the 5th plot from the left, the differencewas slight. This means that size-based zoning would be use-less for identifying a higher potential for large illegal dump-

ing, considering the fact that zoning maps take time and

Fig. 3. Zoning results of illegal dumping potential based on occurrence.

20% % o f i l l e g a l d u m p i n g s i t e s

Data o f FY200 1

Data o f FY200 0

45-degr ee line

0%

40%

60%

80%

100%

0% 20% 40% 60% 80% 100%

t o b e d i s c o v e r e d

% of sites under surveillance

Data of FY2001

Data of FY2000

45-degree line

Fig. 4. Comparison of ROC curves of occurrence-based zoning fordifferent years.

Table 2Area of ROC curve of each geographical attribute for occurrence-basedzoning

Distance from forest edge 0.166Population density 0.155Topographic features 0.141Distance from roads 0.138Inclination 0.088Distance from main roads 0.079

Unemployment rate 0.042Income per capita 0.017



8/12

money to create and also the opinions of the governmentofficers (see Section 2.1), e.g., that ‘‘the mere detection of illegal dumping is significant, regardless of the size.’’ Itwould thus be sufficient to use occurrence-based zoning.

ROC curves for each geographical attribute were calcu-lated, and only those differing between occurrence-basedand size-based zoning are shown in Fig. 6. The differencescame in the middle of the y-axis for population density, inthe upper range of the y-axis for the distance from the for-est edge, and in the lower range of the y-axis for inclina-tion. Contrary to the authors’ expectations, the ROCcurves of topological features did not differ. Due to the dif-

ference in the ROC curves of Fig. 5, located in the upperrange of the y-axis, the distance from the forest edge prob-ably affected and contributed to the superiority of the size-based zoning result.

3.4. Application for the planning of illegal dumping

surveillance

Finally, the ROC curve was applied to the planning of illegal dumping surveillance. Let us consider a target areaA (km2) and the number of illegal dumping sites D. Thetotal social cost C is expressed as the sum of the costs of rehabilitation, environmental pollution by illegal dumping(external costs), and surveillance as:

yD sr þ ð1 y Þ D sd þ xA P ¼ C ð6Þ

where r is the unit cost of rehabilitation of illegally-dumpedwaste per amount ($/ton), d is the unit cost of pollution byillegal dumping when it is not rehabilitated ($/ton), P is theunit cost of surveillance per area ($/km2), and s is the aver-age size of the illegal dumping.

However, not all illegal dumping sites subject to surveil-lance will be found and not all discovered illegal dumpingsites will be rehabilitated. Thus, Eq. (6) is rewritten usingthe probability of successful discovery z and the percentage

of rehabilitation of discovered dumping sites m as:

mzyD sr þ ð1 mzy Þ D sd þ xA P ¼ C ð7Þ

Transforming the formula,

y ¼ AP

Dsmz ðd r Þ x þ

Dsd C

Dsmz ðd r Þ

y ¼ p

mz Dr x þ

d t

mz Dr

where p ¼ P = M ¼ P =ð Ds= AÞ

t ¼ C = Ds ¼ mzyr þ ð1 mzy Þd þ xp

Dr ¼ d r

ð8Þ

where p is the unit cost of surveillance per amount of ille-gally-dumped waste in the target area ($/ton), M is theamount of illegally-dumped waste per area in the targetarea (ton/km2), t is the total cost per amount of illegally-dumped waste in the target area ($/ton), and Dr is the so-cial gain from the rehabilitation of illegal dumping peramount of waste removed ($/ton).

The total cost t should be minimized for optimization.Given a certain positive value of the slope of the line of Eq. (8), the contact point between the ROC curve andthe line of Eq. (8) indicates an optimum condition withthe largest y-intercept of the line of Eq. (8) and the lowesttotal cost (Dr is positive under normal conditions because if it is negative, illegal dumping sites should not be rehabili-tated; besides the other parameters, p, m, z, are all posi-tive). Let us estimate the surveillance condition using thedata shown in Table 3. The target area was the three pre-fectures, Chiba, Ibaraki, and Tochigi, and these data werecollected from the three prefectures. For the data notobtained, other prefectural data or all-Japan data wereused. The ROC curve of occurrence-based zoning withFY 2000 data was used. For the case d = 1.3r (Dr = 1.3),the slope is 0.94. Two lines of Eq. (8) with this slope areshown in Fig. 7 with the ROC curve. The upper-left brokenline is that when the total cost t is US$123/ton, and the

contact point with the ROC curve represents the optimum

0%

20%

40%

60%

80%

100%

0% 20% 40% 60% 80% 100%

Based on occu r rence

Based on siz e

45-degr ee line

0%

20%

40%

60%

80%

100%

0% 20% 40% 60% 80% 100%


% o f i l l e g a l d u m p i n g

s i t e s t o b e

d i s c o v e r e d

Based on occurrence

Based on size

45-degree line

Fig. 5. Comparison of ROC curve of results of occurrence-based and size-based zoning.



9/12

condition. The lower-right broken line is that when thetotal cost is $126/ton, and the two intersection points, Aand B , are the feasible solutions. As shown in Table 4, if the surveillance area is smaller than the optimum surveil-lance condition of 31%, despite the reduced respective costs

of surveillance and rehabilitation from $4.6/ton to $1.6/tonand from $37/ton to $18/ton, the increased pollution raisesthe total cost by $3/ton. If the surveillance area is largerthan the optimum surveillance conditions, despite the factthe pollution cost by illegal dumping is reduced from$82/ton to $63/ton, the increase in other costs raises theoverall cost by $3/ton. This difference may be regarded asthe margin of error and is insignificant due to the errorof cost data. If so, the patrol condition should be chosenfrom the range of the line AOB.

Incidentally, t can be rewritten as:

t ¼ d mzy Dr þ xp ð9Þ

Eq. (9) indicates that a certain percentage increase in eitherone of m, z, or y decreases the total cost to the same degree.Besides, z is further affected by the unit cost of surveillance:how carefully and often surveillance is conducted.

Next, the optimum surveillance condition was calcu-

lated for several different cases, since the data concerningthe unit cost of pollution by illegal dumping d , the unit costof surveillance p, and the probability of successful discov-ery of illegal dumping z, were all uncertain. The resultsof calculating the slope of the line of Eq. (8) and the costand patrol area under optimum conditions are shown inTable 5. Firstly, as the unit cost of pollution by illegaldumping d increased from 1.3r (Case 1) to 10r (Case 4),the slope decreased from 0.94 to 0.03 and the optimumpatrol area expanded from 31% to 100%, resulting in bothincreases in the discovery of illegal dumping and surveil-lance cost. The increase in total cost (4.4-fold), however,

was less than that in the unit cost of pollution (7.7-fold).

All illegal dumping sites

Large illegal dumping sites

45-degree line

Inclination

0%

20%

40%

60%

80%

100 %

0% 20% 40% 60% 80% 100 %


% o f i l l e g a l d u m p i n g s i t e s t o

b e d i s c o v e r e d

% o f i l l e g a l d u m p i n

g s i t e s t o

b e d i s c o v e r e d

Distance from forest edge

20%

40%

60%

80%

100 %

0% 20% 40% 60% 80% 100 %


Population density

0%

20%

40%

60%

80%

100 %

0% 20% 40% 60% 80% 100 %


0%

% o f i l l e g a l d u m p i n g s i t e s t o

b e d i s c o v

e r e d

Fig. 6. ROC curves of each geographical attribute for both occurrence-based and size-based zoning.



10/12

Secondly, as the unit cost of surveillance p increased 4-foldfrom 0.5 p (Case 1 0 and 4 0) to 2 p (Case 100 and 400), the totalcost increased 1.1-fold for the cases of d = 1.3r, as well as

1.04-fold for the cases of d = 10. Thirdly, as the probability

Table 3Data for estimating optimal surveillance conditions

Parameter Remarks or references

Target area A (km2) 17,500 Chiba, Ibaraki, Tochigiprefectures

No. of illegal dumpingincidents D ()

289 Discovered in the threeprefectures

Average size of illegaldumping s (ton)

595 Discovered in the threeprefectures

Unit cost (US$/ton)Rehabilitation of

illegally-dumpedwaste r

100 Ministry of Environment (AllJapan)

Surveillance perweight p

14.7 Calculated from the otherparameters

Surveillance per areaP ($/km2)

145 Proxy data (Nara prefecture)

Probability of successfuldiscovery z

75%

% of rehabilitation of discovered dumping

sites m

75%

Table 4Calculated cost of three points in Fig. 7

Optimized unit cost (US$/ton)

Total Surveillance Rehabilitationof dumping sites

Pollution byillegal dumping

O 123 4.6 37 82A 126 1.6 18 106B 126 11 52 63

T

a b l e 5

R

e s u l t s o f s l o p e a n d i n t e r c e p t o f E q .

( 8 ) a n d

o p t i m a l s o l u t i o n

C

a s e

P a r a m e t e r

S l o p e o f

E q .

( 8 )

%

o f a r e a t o

b e p a t r o l l e d

%

o f i l l e g a l

d u m p i n g

d i s c o v e r e d

O

p t i m i z e d u n i t c o s t [ $ / t o n ]

U n i t c o s t o f

p o l l u t i o n b y i l l e g a l

d u m p i n g ( $ / t o n )

U n i t c o s t

o f

s u r v e i l l a n c e

p e r w e i g h t ( $ / t o n )

%

o f d i s c o v e r y o f

i l l e g a l d u m p i n g

T

o t a l

S u r v e i l l a n c e

R e h a b i l i t a t i o n o

f

d u m p i n g s i t e s

P o l l u t i o n b y

i l l e g a l d u m p i n g

1

1 . 3 r

p

7 5

0 . 9

4

3 1

5 0

1

2 3

4 . 6

3 7

8 2

2

2 r

p

7 5

0 . 2

8

6 7

6 8

1

5 9

9 . 9

5 1

9 8

3

5 r

p

7 5

0 . 0

7

7 2

6 9

3

0 5

1 1

5 1

2 4 3

4

1 0 r

p

7 5

0 . 0

3

1 0 0

7 0

5

4 2

1 5

5 3

4 7 5

1

0

1 . 3 r

0 . 5 p

7 5

0 . 4

7

6 7

6 8

1

2 0

4 . 9

5 1

6 4

1 0 0

1 . 3 r

2 p

7 5

1 . 8

7

2 2

4 1

1

2 7

6 . 5

3 0

9 0

4

0

1 0 r

0 . 5 p

7 5

0 . 0

2

1 0 0

7 0

5

3 5

7 . 4

5 3

4 7 5

4 0 0

1 0 r

2 p

7 5

0 . 0

6

1 0 0

7 0

5

5 7

3 0

5 3

4 7 5

A

1 . 3 r

p

1 0 0

0 . 7

0

4 8

8 5

1

1 9

7 . 1

6 0

5 3

A

0

1 . 3 r

p

5 0

1 . 4

0

2 2

4 1

1

2 7

3 . 2

2 0

1 0 4

D

1 0 r

p

1 0 0

0 . 0

2

1 0 0

7 0

3

8 5

1 5

7 0

3 0 0

D

0

1 0 r

p

5 0

0 . 0

5

1 0 0

7 0

7

0 0

1 5

3 5

6 5 0

C

o s t s a r e i n U S d o l l a r s .

0%

20%

40%

60%

80%

100%

0% 20% 40% 60% 80% 100%


%

o f

e l l i

g a

d l

u

p m

n i g

t i s

e

o t s

e b

t i n o m

o

d e r

Da ta o f FY 2000 in F ig. 4.

Eq . (8)

Data of FY 2000 in Fig. 4.

Eq. (8)

O (31%, 71%)

A

(11%, 35%)

B

(78%, 98%)

Fig. 7. Application of an ROC curve to the planning of surveillanceagainst illegal dumping.



11/12

of successful discovery of illegal dumping z decreased from100% (Case A and D) to 50% (Case A 0 and D 0), the totalcost increased 1.1-fold for the cases of d = 1.3r, Cases Ato A 0, as well as 1.8-fold for the cases of d = 10, Cases Dto D 0.

All of these results show that the degree of increase of

total cost was smaller than that of each cost parameter,namely that this method is useful for reducing increasedtotal costs related to illegal dumping. This application isbased on a simplification, however, that does not take intoaccount the detailed distribution of each level of illegaldumping potential (exact patrol routes) and differences inthe rehabilitation and social costs of illegal dumping sitesin different sites in the target area. Nevertheless, it enablesthe calculation of a rough estimate in an improved surveil-lance planning, as shown above.

4. Conclusion

To assist in efficient surveillance against illegal dumping,two methods of zoning used to illustrate the illegal dump-ing potential of sites using GIS data were examined andevaluated. The main results were as follows:

1. Zoning results showed areas that should be patrolledagainst illegal dumping.

2. The size-based zoning was superior for surveillanceagainst large illegal dumping, but this superiority wasinsignificant for revealing sites with higher potentialfor large illegal dumping. It would thus be sufficient touse occurrence-based zoning.

3. An ROC curve of a zoning result can clarify the numberof detectable illegal dumping sites based on certain sur-veillance conditions.

4. The ROC curve can be used for surveillance planning tominimize the total social cost concerning illegal dumpingand appropriate countermeasures.

Data of the geographical attributes used in the paperwere limited only to available GIS data, and did notinclude important data thought to be related to illegaldumping according to governmental officers, such as therelation of dumpers to the landlord of a site. If these GISdata are obtained and used in the zoning, the ROC curvewould show superior results. The multiplication approachfor integrating individual geographical attributes has roomfor improvement. Clarifying the detailed mechanism of ille-gal dumping and deriving a more exact function of the ille-gal dumping potential with geographical attributes wouldimprove the accuracy of the methodology and the zoningresults.

Acknowledgements

This research was conducted in a 3-year project, Devel-opment of illegal dumping surveillance system using artificial

satellites, under the Ministry of Environment of Japan. We

are grateful to the project members and members of theboard including experts and local governmental officers,as well as those individuals in charge of patrolling for ille-gal dumping from the local governments visited during thissurvey.

Appendix 1

Three zoning approaches can be used to integrate sev-eral factors of concern and obtain a certain kind of score,especially when the exact function of a target phenomenonis unknown. The first approach involves summing the fac-tors (in this study, the relative probabilities of geographicalattributes) as shown in Eq. (A1). The second approachinvolves multiplying the factors as shown in Eq. (A2).

Z S ¼X

a

P ai ðA1Þ

Z M ¼Ya

P ai ðA2Þ

The third approach, called the ‘‘matrix approach’’ by theauthors, involves cross-tabulation in two or more dimen-sions for all factors and all of their categories. Determiningthe figures in the cross-tabulation, however, has often beenan arbitrary process. To avoid such arbitrariness, the Ana-lytic Hierarchy Process (AHP) has conventionally been ap-plied (Kontos et al., 2005; Obayashi and Kojima, 1996),but the AHP is based on the perspectives of subject people.If a problem includes more than two dimensions, AHP is auseful method to integrate them, such as for determininglandfill suitability. However, the problem of this study in-

cludes only one dimension, the probability of illegal dump-ing, and it is possible to establish an objective model orequation without applying subjective-based methods suchas the AHP. We therefore dismissed the matrix approach,and examined the adequacy of the addition and multiplica-tion approaches as follows.

First of all, each phenomenon is written as: D is theoccurrence of illegal dumping, A j belongs to category i of the geographical attributes of A, and B j belongs tocategory j of the geographical attributes of B . Then,the probability of occurrence of illegal dumping withina parent population (e.g., within a nation) is written as

P (D), while the probability of occurrence of illegaldumping under category i of the geographicalattributes of A is written as a conditional probabilityP (DjA j ).

Therefore, the relative probability of Eq. (3) can berewritten as:

P ai ¼ I ð AiÞ ¼ P ð Dj AiÞ= P ð DÞ ðA3Þ

where I () is the probability of illegal dumping under certaincondition(s). Generally speaking, the conditional probabil-ity of events X and Y is defined as:

P ð X jY Þ ¼ P ðY \ X Þ= P ðY Þ ðA4Þ



12/12

Therefore,

I ð AiÞ ¼ P ð D \ AiÞ=f P ð DÞ P ð AiÞg ðA5Þ

When X and Y are independent of each other,

P ð X \ Y Þ ¼ P ð X Þ P ðY Þ ðA6Þ

Therefore, if D and Ai are independent of each other, I (Ai )of Eq. (A5) becomes 1. In other words, when Ai is irrele-vant to illegal dumping, the relative probability P ai be-comes 1.

Next, the relative probability of Eq. (A3) is extendedfrom the case of a single condition to that of twoconditions.

I ð Ai \ B jÞ ¼ P ð Dj Ai \ B jÞ= P ð DÞ

¼ P ð D \ Ai \ B jÞ=f P ð DÞ P ð Ai \ B jÞg ðA7Þ

Incidentally, the product of relative probability with a sin-gle condition is written as:

I ð AiÞ I ð B jÞ ¼ ½ P ð D \ AiÞ=f P ð DÞ P ð AiÞg

½ P ð D \ BiÞ=f P ð DÞ P ð B jÞg

¼ f P ð D \ AiÞ P ð D \ B jÞg=

f P ð DÞ2 P ð AiÞ P ð B jÞg ðA8Þ

If A and B are independent of each other, Eq. (A8) isrewritten as:

I ð AiÞ I ð B jÞ ¼ f P ð DÞ P ð Aij DÞ P ð DÞ P ð B jj DÞg=

f P ð DÞ2

P ð AiÞ P ð B jÞg

¼ f P ð DÞ2

P ð Ai \ B jj DÞg=f P ð DÞ2

P ð Ai \ B jÞg

¼ P ð Ai \ B j \ DÞ=f P ð DÞ P ð Ai \ B jÞg ðA9Þ

Eq. (A9) is the same form as Eq. (A7). Therefore, we canobtain the relative probability of two or more conditionsby multiplying each relative probability, provided theseprobabilities are independent of each other.

Consequently, we applied the multiplication approachin this study as the preferred option, although thisapproach is associated with a certain degree of error inthe case that the probabilities are inter-dependent.

References

Chang, Y.C., Chang, N.B., Ma, G.D., 2001. Internet Web-basedinformation system for scrap vehicle disposal in Taiwan. Environ-mental Modeling and Assessment 6, 237–248.

Choe, C., Fraser, I., 1999. An economic analysis of household wastemanagement. Journal of Environmental Economics and Management

38, 234–246.Fullerton, D., Kinnaman, T.C., 1995. Garbage, recycling, and illicitburning or dumping. Journal of Environmental Economics andManagement 29, 78–91.

Kontos, T.D., Komilis, D.P., Halvadakis, C.P., 2005. Siting MSWlandfills with a special multiple criteria analysis methodology. WasteManagement 25, 818–832.

Japan Industrial Newspaper, 2002. Special Issue: Manual of preventingillegal dumping of wastes. Environmental Local Government 5, 7–31(in Japanese).

Japan Industrial Waste Technology Center, 2002. Special Issue: TacklingEnvironmental Crime. JW 1, 4–16 (in Japanese).

Ministry of Environment of Japan, 2000–2002. Current State of IllegalDumping of Industrial Wastes, Tokyo (in Japanese).

National Institute for Environmental Studies, 2002. A Report of Devel-

opment of Illegal Dumping Surveillance System Using ArtificialSatellites, fiscal year of Heisei 13, Section 3. Available from: http://www-cycle.nies.go.jp/research/datalib/hokoku/RC-01-001.pdf (inJapanese).

Obayashi, S., Kojima, H., 1996. Construction of the land use capabilityclassification algorithm applying the analytic hierarchy processmethod. Journal of Construction Management and Engineering 546(VI-32), 169–179 (in Japanese).

Sigman, H., 1998. Midnight dumping: public policies and illegal disposalof used oil. Rand Journal of Economics 29, 157–178.

Tasaki, T., Matsui, Y., Kawahata, T., Osako, M., Takagishi, S., Morita,A., 2004. Analysis of geographic attributes and probabilities related toillegal dumping. Japan Society of Waste Management Experts 15, 1–10(in Japanese with English abstract).

US EPA, Region 5, 1997. Site visit reports (Greater Chicago, Mississippi

Gateway, Northwest Indiana, Southeast Michigan, Greater Cleve-land). Available from http://www.epa.gov/region5/illegaldumping/.

US EPA, Region 5, 1998. Illegal Dumping Prevention Guidebook.EPA905-B-97-001. Available from: http://www.epa.gov/reg5rcra/wpt-div/illegal_dumping/downloads/il-dmpng.pdf .

US EPA, Region 5, 2000. IDEA (Illegal dumping economic assessment)Cost Estimating Model User’s Guide. EPA 905-B-00-002. Availablefrom: http://www.epa.gov/reg5rcra/wptdiv/illegal_dumping/down-loads/ideafctsht.pdf .

Xepapadeas, A.P., 1991. Environmental policy under imperfect informa-tion – incentives and moral hazard. Journal of EnvironmentalEconomics and Management 20, 113–126.

http://-/?-http://www-cycle.nies.go.jp/research/datalib/hokoku/RC-01-001.pdfhttp://www-cycle.nies.go.jp/research/datalib/hokoku/RC-01-001.pdfhttp://www.epa.gov/region5/illegaldumping/http://www.epa.gov/reg5rcra/wptdiv/illegal_dumping/downloads/il-dmpng.pdfhttp://www.epa.gov/reg5rcra/wptdiv/illegal_dumping/downloads/il-dmpng.pdfhttp://www.epa.gov/reg5rcra/wptdiv/illegal_dumping/downloads/ideafctsht.pdfhttp://www.epa.gov/reg5rcra/wptdiv/illegal_dumping/downloads/ideafctsht.pdfhttp://www.epa.gov/reg5rcra/wptdiv/illegal_dumping/downloads/ideafctsht.pdfhttp://www.epa.gov/reg5rcra/wptdiv/illegal_dumping/downloads/ideafctsht.pdfhttp://www.epa.gov/reg5rcra/wptdiv/illegal_dumping/downloads/il-dmpng.pdfhttp://www.epa.gov/reg5rcra/wptdiv/illegal_dumping/downloads/il-dmpng.pdfhttp://www.epa.gov/region5/illegaldumping/http://www-cycle.nies.go.jp/research/datalib/hokoku/RC-01-001.pdfhttp://www-cycle.nies.go.jp/research/datalib/hokoku/RC-01-001.pdfhttp://-/?-

A GIS-based zoning of illegal dumping potential for efficient surveillance.pdf

Documents

Transcript of A GIS-based zoning of illegal dumping potential for efficient surveillance.pdf