Global Environmental Change 23 (2013) 1542–1555

Rivers as municipal infrastructure: Demand for environmentalservices in informal settlements along an Indonesian river

Derek Vollmer a,b,*, Adrienne Gret-Regamey b

a Future Cities Laboratory, Singapore-ETH Centre for Global Environmental Sustainability, 1 CREATE Way, #06-01 CREATE Tower,

Singapore 138602, Singaporeb Chair of Planning of Landscape and Urban Systems, ETH Zurich, HIL H 51.4, Wolfgang-Pauli-Str. 15, 8093 Zurich, Switzerland

A R T I C L E I N F O

Article history:

Received 8 April 2013

Received in revised form 23 September 2013

Accepted 2 October 2013

Keywords:

Environmental services

Rivers

Informal settlements

Vulnerability

Indonesia

A B S T R A C T

In many developing world cities, where municipal infrastructure lags urban growth, lower-income

communities may compensate by relying on local waterways to meet basic needs for water, sanitation,

and recreational space. Access to these environmental services is possible because residents settle in

floodplains, but thus entails elevated exposure to several water-related hazards, especially flooding. We

examine this complex relationship in the neighborhoods of Bukit Duri and Kampung Melayu on the

Ciliwung River in Jakarta, Indonesia. Based on a spatially referenced household survey, we analyze and

map the patterns of use of six environmental services provided by the river: direct sanitary use,

recreation, harvesting plants, groundwater use, solid waste disposal, and sewage disposal. Using spatial

interpolation and regression methods, we identify the most probable areas where services are being

used and analyze possible influences on this behavior. We find that proximity to the river significantly

influences households’ behavior toward the river, as do infrastructure-related variables and

neighborhing households’ behavior, while household demographic factors appear less significant.

These results indicate that many households rely on multiple environmental services, and that residents

most reliant on these services are also at greater risk of water-related hazards, service disruption (e.g., a

decline in water quality), and potentially, eviction. This pattern of floodplain development is prevalent in

many low-income countries, and a better understanding of how informal settlements rely on

environmental services can be used to assess their vulnerabilities and inform more sustainable courses

of development.

� 2013 Elsevier Ltd. All rights reserved.

Contents lists available at ScienceDirect

Global Environmental Change

jo ur n al h o mep ag e: www .e lsev ier . co m / loc ate /g lo envc h a

1. Introduction

Rivers and their floodplains provide fertile land for agriculture,water for consumption and irrigation, and a means of transportinggoods and discarding waste. We refer to these as environmentalservices, the benefits that humans derive, directly and indirectly,from hydrologic systems (Daily, 1997; Brauman et al., 2007). Mostcities have come to depend on hydrologic or watershed services insome form, and certain urban infrastructure can be thought of as ameans to replace or more efficiently harness these naturallyprovided services (Lee, 2006; Gomez-Baggethun and Barton,2012). Urban planners now incorporate green spaces, such asurban forests and riverside parks, as a form of ‘‘green and blue’’

* Corresponding author at: Future Cities Laboratory, Singapore-ETH Centre for

Global Environmental Sustainability, 1 CREATE Way, #06-01 CREATE Tower,

Singapore 138602, Singapore. Tel.: +65 8198 2124.

E-mail addresses: vollmer@arch.ethz.ch (D. Vollmer),

gret@nsl.ethz.ch (A. Gret-Regamey).

0959-3780/$ – see front matter � 2013 Elsevier Ltd. All rights reserved.

http://dx.doi.org/10.1016/j.gloenvcha.2013.10.001

infrastructure that supports public and ecosystem health (Tzoulaset al., 2007). Yet physical and psychological distance betweenurban residents and environmental services is predicted toincrease with time as suggested by Carpenter et al. (2006):‘‘[u]rban migrants will consume ecosystem services via longersupply chains, and they will experience greater separation fromnature (p. 29).’’ In other words, built infrastructure is expected toreplace many of the functions that ecosystems have long provided.

Yet most urban population growth will take place in so-called‘‘mega-deltas’’ in Africa and Asia, where cities are ill-equipped toaddress competing challenges of surging populations andenvironmental change (Seto, 2011). In these rapidly growingcities, where municipal infrastructure often lags urban growth(Crane and Daniere, 1996; Roberts and Kanaley, 2006; Schafflerand Swilling, 2013), we posit that lower-income communitiesmight be relying more directly on environmental services tomeet basic needs. This is particularly relevant in cities’ informalsettlements, colloquially referred to as ‘slums’, which areresponsible for housing roughly 40% and over 60% of urbanresidents in Asia and sub-Saharan Africa, respectively

D. Vollmer, A. Gret-Regamey / Global Environmental Change 23 (2013) 1542–1555 1543

(Un-Habitat, 2009). Urban informal settlements exist in a grayarea between legal and illegal status (Roy and Alsayyad, 2004)and, consequently, often lack municipal public goods such aswater supply, waste collection, and park space. Growth rates forinformal settlements have been comparable or only slightlybelow overall urban growth rates in most of Asia and sub-Saharan Africa, and so even if a greater proportion of urbanresidents are extended basic services over time, this ‘‘urbandivide’’ is likely to be persistent for some time (Un-Habitat,2010). A substantial share of informal settlements exist alongriverbanks and urban waterways (Un-Habitat, 2003) since theseareas are regarded as unsuitable for development and thuspresent a comparatively lower risk of eviction.

In the case of urban settlements in flood plains, exposure toflood risk and other water-related hazards is increased (throughdecreases in elevation, primarily) in exchange for cheaper land orimproved access to other services along waterways. We shouldemphasize that such tradeoffs are implicit, that is, a number ofsocio-economic determinants lead to residents settling in floodplains; their use of, or demand, for environmental services is ratheran adaptation to the particular environment. Importantly, thisrelationship between human settlements and demand for envi-ronmental services tends to be heterogenous at a fine scale(Cadenasso et al., 2007) and is often spatially dependent andinvolves tradeoffs between and among services (Nelson et al.,2009). Depicting these sorts of spatially dependent relationshipsrequires fine-scale data and a geographic information system (GIS).Spatially oriented analytical methods, from interpolation tospatial-weighted regressions, can help further investigate thisheterogeneity (Anselin, 1992) and refine our understanding ofthese complex social–ecological systems.

To our knowledge, little empirical analysis has been done on theinteractions between urban informal settlements and the envi-ronmental services they demand. Victor Savage (Savage, 2006)coined the term ‘‘ecosystem people,’’ typified by urban canaldwellers in Thailand meeting their daily needs through locallyavailable ecosystem services, but he did so to make a broader pointabout consumption disparities in Southeast Asia. Field studiesfrom the developing world on the link between environmentalservices and livelihoods typically focus on mangroves (Badola andHussain, 2005; Barbier, 2007; Das and Vincent, 2009; Concheddaet al., 2011), forests (Kosoy et al., 2007; Meyfroidt and Lambin,2008; Engel and Palmer, 2008; Chhatre and Agrawal, 2009; Sodhiet al., 2009; Jaung and Bae, 2012), or agro-ecosystems (Alpızar andMadrigal, 2008; Li et al., 2011; Arias et al., 2012). In each of thesestudies the primary environmental issue is whether the commu-nity can maintain a productive stock of locally harvestedmarketable goods without sacrificing the ecosystems’ regulatingfunctions. In their study of risk perception and vulnerability inmangrove ecosystems in East Africa, Crona et al. (2009) extend thisanalysis to a peri-urban context, examining the conflict betweenoutlying communities’ harvesting activities and central cities’ useof the mangroves for treatment of urban sewage.

A small minority of papers focus exclusively on environmentalservices and urban informal settlements, but generally in thecontext of informal land conversion’s impact on environmentalservice provision (Aguilar, 2008; Zeilhofer and Topanotti, 2008;Aguilar and Santos, 2011; Mendoza-Gonzalez et al., 2012;Harriden, 2012). Jayawickreme (2011) documents Sri Lanka’surban wetlands becoming unregulated solid waste dumps andnotes that the lack of sanitary infrastructure is threateningbiodiversity along the island’s western coast. And surveys suchas Resosudarmo et al. (1997) and Winayanti and Lang (2004) inJakarta, Noosorn and Niamkamnerd’s (2009) work in Thailand andSeftyono’s (2011, 2012) recent work in another Indonesian city,Yogyakarta, provide limited empirical evidence of how riverside

residents use nearby urban rivers, though that is not the primaryfocus of any of their research.

By studying informal settlements along the Ciliwung River inmetropolitan Jakarta, Indonesia, we aim to investigate therelationship among proximity to the river, household demographiccharacteristics, and use of environmental services, using spatiallyexplicit data. In the midst of this dense urban settlement, it isapparent that a substantial share of residents still rely directly onthe river to meet basic needs. In light of rapidly changing land usewithin the catchment, and the possibility that climate changecould exacerbate flooding and low-flow events, a better under-standing of this relationship has implications for planning andresource management. First, the spatial dimension of thisrelationship is underresearched, and spatially explicit informationcan help support context-appropriate, targeted planning or designinterventions. Second, such research provides insight into criticalissues regarding longer-term river rehabilitation programs in thedeveloping world. Jakarta, like many other cities, is facing growingdemands to improve conditions of its waterways, and the fate ofinformal settlements along these waterways is central to thedebate.

We begin by framing our fieldwork in a global context, with anexploratory spatial analysis of urban population density trendsalong waterways in different countries from 1990 to 2000. Thoughour population data in this case are restricted to density andproximity to waterways, we hope to demonstrate that thephenomena we observe in detail in our subsequent case studyof an urban river in Jakarta may be symptomatic of a broader globaltrend. We then elaborate on the spatially explicit methods we useto obtain and analyze our data, which allows for a more nuancedunderstanding of how residents engage with the river. Weconclude with a discussion of the multiple facets of vulnerabilitymany of these households are facing with respect to the river, andthe implications of these findings for similar settlements in rapidlyurbanizing Africa and Asia.

2. Setting the scene: from global to local

2.1. Global context

As many post-industrial cities have only recently begun torehabilitate degraded waterways to meet environmental oreconomic development goals (Findlay and Taylor, 2006), rapidlygrowing cities in the developing world are at a relatively earlystage of exploiting and managing their urban streams and waterbodies. Furthermore, the demands on their waterways arecomparatively more complex (Marcotullio, 2007), with pressuresto increase exploitation for social and economic developmentwhile, paradoxically, maintaining ecological functionality in thename of sustainability (Mcgranahan, 2007) or resilience (Palmer etal., 2008). This is further complicated by the fact that populationdensity in African and Asian settlements is typically 2–3 timesgreater than their Western counterparts (UNDESA, 2012). Ceterisparibus, more people means more environmental pressure, a largerpopulation at risk of perturbations, but also a larger potentialdemand for environmental services. Anecdotal evidence suggeststhat many urban migrants, particularly low-income residents, aresettling in floodplains and other environmentally sensitive areas,but there is a lack of data that would help us interpret the globalsignificance of this trend.

Thus, as a first step, we analyzed changes in population densityin proximity to urban waterways across the top 20 countries interms of urban population growth between 1985 and 2000(UNDESA, 2012). Countries were classified into three incomegroups based on Gross National Income (GNI) per capita, year 2000(current US$), using the World Bank’s cut points (World Bank,

D. Vollmer, A. Gret-Regamey / Global Environmental Change 23 (2013) 1542–15551544

2013). Additional data on countries is provided in the appendix.These countries together accounted for over 40% of the world’surban population growth over that time period and thus providea reasonable cross-section of global population change. Wethen combined global data on girded population density (CIESINet al., 2011a) recently released by the Global Rural-UrbanMapping Project (GRUMP v1), urban extents (CIESIN et al.,2011b) and the location of inland streams and water bodies(VMAP, 2000), to analyze country-level spatially explicit changesin densities over three time periods (1990, 1995, 2000). Weclassified urban waterways simply as the water courses presentwithin the urban extent data layer. We did not have data on thespatial extent of floodplains, and so we selected a 1 km bufferarea around urban waterways as a proxy for the zone most likelyto involve direct interactions such as direct sanitary use,recreation, groundwater use, solid waste disposal, or sewagedisposal, to name a few. A 1 km radius also allowed us to efficientlyuse the gridded population density data, which are at a resolutionof 3000, or slightly less than 1 km.

We calculated the means, medians and interquartile ranges forpopulation density within these buffers, and for the urban extentas a whole for all 20 countries in 1990, 1995, and 2000. We presentthese results in Fig. 1, aggregated into low, lower middle, andupper middle/high income countries while individual country dataand results are presented in Appendices A and B. Differencesbetween means for a particular year, and across years, are allstatistically significant. Mean population density appears toincrease by roughly 10% within 1 km of waterways for the lowincome countries, counter to trends observed in the middle andhigh income countries. This gives no information about qualitative

Fig. 1. Mean population densities in urban areas (gray) and within 1 km of urban waterwa

aggregated by income group. The median values (white circle) and interquartile range

differences between households living within the 1 km buffer oroutside of it, but it does suggest that a spatially different type ofsettlement pattern is taking place in the low-income countries ofAsia and Africa. The drivers of this pattern of development aremany; we are concerned with the implications of such a pattern.We postulate that under these conditions, a not insignificant shareof residents are relying directly on environmental services. Whilewe make no attempt to validate this on a global scale, we provideempirical evidence of such behavior in our case study area ofJakarta.

2.2. Description of Jakarta and study area

Indonesia is the fourth most populous country in the world,with an urban population of nearly 120 million and expanding atan average annual rate of greater than 2.5% as of 2010 (UNDESA,2012). Jakarta, the seat of Indonesia’s national government, is thelargest city in southeast Asia, the main economic growth pole forthe island of Java, and the center of an extended metropolitanregion which is home to over 28 million residents. Despite its sizeand prominence, the city is still frequently referred to as acollection of kampungs, or urban villages (Steinberg, 2007; Silver,2008). These kampungs are vestiges of villages originallyestablished in areas lying just outside the official Dutch colonialurban boundaries, but which in the second half of the 20th centurywere enveloped by urban growth. In terms of environmentalquality, there is still a dichotomy between these informalkampungs and the rest of the city, and as of the early 2000sroughly 80% of Jakarta’s slum dwellers (and over 20% of the totalurban population) resided there (Un-Habitat, 2003).

ys (blue) for top 20 countries in terms of urban population growth from 1985–2000,

are given as well. The list of countries can be found in Appendix B.

D. Vollmer, A. Gret-Regamey / Global Environmental Change 23 (2013) 1542–1555 1545

Thirteen rivers, the largest being the Ciliwung, flow throughthe city, with some river corridors hosting population densitiesin excess of 50,000 people per square kilometer (Fig. 2). Noneof these rivers meet water quality standards for raw drinkingwater supply, and some like the Ciliwung are also nottechnically fit for any type of use (Ali, 2010). Wet seasonaverage biological oxygen demand (20.6 mg/l), dissolved oxygen(6.6 mg/l), and total suspended solids (380 mg/l) far exceed theassimilative capacity of the Ciliwung, and high levels of fecalcoliform such as E. coli (200–1100 CFU/ml) are present as well,primarily a result of diffuse sewage and septic tank dischargethat begins upstream of Jakarta and continues throughout thecity (Djadjadilaga et al., 2008; Tjampakasari and Wahid, 2008;Phanuwan et al., 2006). These rivers also contribute to seasonalflooding, with the 2007 flood being Jakarta’s deadliest anddisplacing more than 400,000 residents (Texier, 2008). Floodingin January 2013 displaced fewer residents (�40,000) but stillclaimed over 40 lives and caused damages tentatively estimatedat US $3.3 billion (Sudoyo, 2013).

Our study site comprises the Bukit Duri and Kampung Melayuneighborhoods on either side of the Ciliwung River, near thegeographic center of Jakarta (Fig. 2). Bukit Duri and KampungMelayu are known administratively as kelurahans (literally,‘‘villages’’) that are subdivided into Rukun Warga (RW, communityunit) and, a level lower, into Rukun Tetangga (RT, neighborhoodunit). The 13 RW in our study site are home to over 48,000 people,within a territory of just over one square kilometer. Theseneighborhoods have been mostly informally settled, and arecharacterized by densely located one- or two-story dwellings,narrow roads and footpaths, low coverage for municipal water and

Fig. 2. Left image: population density (year 2000) along major rivers in and around Jakar

outline). Right image: Study site and location of households surveyed. The Ciliwung

municipalities and the Bukit Duri and Kampung Melayu neighborhoods.

waste handling services, and high vulnerability to flooding. SomeRTs within the site were under 4 m of water during the January2013 flood. The river is swift moving, with flows averaging 34 m3/sand 54 m3/s in the dry and wet seasons, respectivelyand with adepth ranging from 1 to 3 m and width of 5–10 m in our study areaduring non-flooding conditions. Despite the dense settlement andchronic flooding problem in the rainy season, this stretch of river,though encroached by settlements, has not been channelized. Onlysmall stretches have reinforced the banks with permanentmaterials such as concrete, and significant portions of the bankshave riparian vegetation. As a result, the river can be accessed atmost points in the study area, and residents have built dozens ofbamboo platforms and rafts along the river to facilitate a range ofuses of the Ciliwung’s resources.

2.3. Materials and methods

Based on initial field observations, six distinct environmentalservices along the river were of primary interest for our analysis:

� direct sanitary use: bathing, washing goods, and other sanitarypurposes excluding toilet use� recreation: passive (sitting or walking along river) and/or active

(fishing or swimming),� harvesting plants: collecting materials (fruits, fiber, medicinal

herbs) from plants growing directly in the soil along the riverbanks, i.e., not from potted plants or private gardens,� groundwater use: obtaining at least some of household’s water

supply from a groundwater source, including private wells andpublic wells,

ta (dashed line), and study site location within the Ciliwung River catchment (black

River forms a natural and administrative border between South and East Jakarta

Table 1Summary of variables used in analysis.

Code Description Type Unit/range

HH Number of persons in household Integer Persons

YRS_RES Tenure in present household Integer Years

OWN Ownership of dwelling Dummy 1 if self-owned, 0 otherwise

CONS_EXP Monthly household consumption expenditures Ordinal <500,000 Rpa

500,000–1.5 million Rp

1.6–2.5 million Rp

2.6–4.0 million Rp

>4.0 million Rp

H2O_SOURCE Source(s) of water for household use, excluding river Categorical Public

Public and vendor

Jet pump and vendor

Jet pump

Municipal

Municipal and other sources

TOILET Household with at least one private toilet Dummy 1 if yes, 0 otherwise

T_WO_S Toilet in household without septic tank Dummy 1 if yes, 0 otherwise

NONREC Washing goods,

bathing, and/or other direct sanitary uses

Dummy 1 if yes, 0 otherwise

REC Passive or active forms of recreation Dummy 1 if yes, 0 otherwise

DISP Dispose solid waste directly to river Dummy 1 if yes, 0 otherwise

PLANTS Harvest non-potted plants along river Dummy 1 if yes, 0 otherwise

PAM Household connection to municipal water supply (PAM Jaya) Dummy 1 if yes, 0 otherwise

GW Household relies on groundwater for some portion of water supply Dummy 1 if yes, 0 otherwise

ELEV Elevation relative to sea level Continuous Meters

DIST Linear distance from household to river Continuous Meters

WT_T_WO_S Weighted variable of neighboring households with toilet but not septic tank Continuous 0–1

WT_NONREC Weighted variable of neighboring households using the river for washing goods,

bathing, and/or other direct sanitary uses

Continuous 0–1

WT_REC Weighted variable of neighboring households using river for recreation Continuous 0–1

WT_DISP Weighted variable of neighboring households discarding solid waste directly to river Continuous 0–1

SERVICES Bundle of environmental services a household relies on (including groundwater use) Integer 0–6

a Exchange rates between the U.S. $ and Indonesian Rupiah (Rp) ranged from 9150 to 9400 Rp per US $1 over the course of the surveying period.

D. Vollmer, A. Gret-Regamey / Global Environmental Change 23 (2013) 1542–15551546

� solid waste disposal: domestic garbage that is discarded directlyto the river, and,� sewage disposal: private toilet waste that is discharged directly

to the river, or to open ditches that drain to the river.

Though solid waste disposal and sewage disposal wouldtypically be measured as a single ecosystem function (wasteassimilation) we subdivided them because the demand for eachservice is distinct. The sewage disposal we measured is likely anunderestimate of the actual demand for this service, since we werenot able to verify that all public toilets in the site (referred to asMCKs for mandi [bathing] cuci [washing] kakus [toilet]) wereconnected to septic tanks. Direct sanitary use refers to thehouseholds that use river water to bathe or wash items on eitherprivate platforms extending from their households to the river orthe privately owned bamboo rafts (getek). The getek sometimeshave two toilets installed on the downstream end, and so forhouseholds that reported using the toilets we included them in thesewage disposal category since the waste discharges directly to the

Table 2Cross-correlation and spatial correlation among environmental service variables.

Variable T_WO_S NONREC REC

T_WO_S – �.12 .15*

NONREC �.12 – .29**

REC .15* .29** –

DISP .13 .28** .37**

PLANTS �.05 .04 .05

GW .02 .12 .00

CONS_EXP �.00 �.17* �.08

SERVICES – – –

n = 213.* p < .05.** p < .01.

river. We initially divided recreation activities into passive andactive, because the latter would in theory depend more strongly onstream health. Groundwater use within the site is directly relatedto the river because the river recharges the shallow aquifer onwhich household wells depend. Microclimate regulation, erosionprevention, and flood wave attenuation are three additionalecosystem functions being provided by the streamside vegetationwithin the site, but we could not directly measure demand forthese services.

We developed a household survey to explore the share andspatial extent of residents using these services in the study area.The primary dataset is from a spatially referenced householdsurvey conducted March–July 2012. Surveys were conducted in-person at the household to reduce the non-response rate andensure complete responses. This also allowed interviewers toobtain geographic coordinates for each household, saving coordi-nates to a mobile phone tracking application and then integratingcoordinates into a geographic information system (GIS). No currentbuilding plot maps of the neighborhoods were available to verify

DISP PLANTS GW Getis-Ord G Statistic

Z-score

.13 �.05 .02 1.17

.28** .04 .12 7.65**

.37** .05 .00 7.14**

– .06 .08 14.3**

.06 – .02 3.05**

.08 .02 – 2.65**

�.16* �.07 �.02 �.86

– – – 11.7**

Fig. 3. Share of households demanding environmental services in study area, by

individual service. Codes correspond to variables in Table 1. Error bars represent a

95% confidence interval.

D. Vollmer, A. Gret-Regamey / Global Environmental Change 23 (2013) 1542–1555 1547

the accuracy of the coordinates at the individual structure level,but coordinates were cross-checked with satellite images and withmaps of the RT-level units. In cases where coordinates did notconform to expectations, the point was assigned to the centroid ofits RT unit, giving it, on average, an error of no more than 20 m. Thislevel of precision should be adequate for its intended purpose,which is to provide geographically dispersed data points that canbe used to interpolate values across neighborhoods, rather thanpredicting values for individual households.

The population under study was segmented into primarysampling units (PSU) at the RW level (see Fig. 2), to ensuresufficient geographic coverage on either side of the river for thespatial analysis. There was no specific limit on responses from agiven PSU, which would have affected the probability of beingselected once a cap in a particular PSU was reached. Surveyadministrators were also instructed to obtain at least onerespondent from the riverside in each PSU, since these householdsmay represent an important subset of the population and non-coverage of such populations is a significant source of bias inhousehold surveys (UNDESA, 2005). In total we interviewed 219households from which 213 were considered valid for our analysis.We collected standard demographic data from each household inorder to control for variables such as household size or income,using monthly consumption expenditures as a proxy. We added ariver shapefile into the GIS and calculated Euclidean distance to theriver for each point as an additional variable. We also added 1 melevation contours into the GIS and then assigned an elevationvariable to each point. Most of the data were recorded as eitherbinary (Yes/No) or categorical data (see Table 1).

We collected information on the frequency of consumption ofcertain environmental services (REC and PLANTS) but for thepurposes of totalling household consumption of services, we usedonly the binary responses. These binary responses referred tocurrent use, or at least one occurrence in the past year. Many morehouseholds reported having recreated and/or harvested plantsseveral years ago, but those responses were coded as ‘‘0’’ for ourpurposes. Though the six environmental services are not strictlycommensurable, we added them together as a bundle of distinctservices that might otherwise be provided by a municipalgovernment or procured in the marketplace. With the exceptionof groundwater use, households are actively (and knowingly)obtaining these services from the river. Reliable estimates of waterconsumption (aside from drinking water if the primary source wasprivate vendors) and expenditures were not feasible, though meanhousehold consumption appeared to be low (<100 L/capita/day),consistent with surveys documented in Crane and Daniere (1996)and Bakker et al. (2008). Monthly drinking water expenditureswere calculated for the subset of the population relying onvendors, but could not be reliably estimated for the overallpopulation.

2.3.1. Descriptive statistics for the study area

Households have a relatively long tenure (mean of 32 years),with many homes being intergenerational, and about 77% ofhomes were owned, though this refers only to the structure, notnecessarily to a formal land title. Using our consumptionexpenditure data as a rough proxy for poverty levels, theinformation suggests that approximately 40% of households wouldlikely fall under Jakarta’s poverty line as of 2011, which could bealternately defined in per capita terms (355,480 Rp/US$40.75 at2011 exchange rate) or as a monthly minimum wage of 1.529million Rp/US$175.26 (BPS, 2012). In other words, the study area ispoorer than the city overall, but is still economically mixed. Waterand sanitation service coverage is low – less than 30% ofhouseholds had some access to the municipal water provider(Perusahaan Daerah Air Minum Jakarta Raya [PAM Jaya]), which is

lower than the citywide figure of 45%. No households surveyedwere connected to municipal sewerage, meaning that liquid wastewas discharged to either private septic tanks or directly to theriver. Kampungs frequently have small, open drainage channelsalongside footpaths. We considered households discharging wasteinto these channels as discharging to the river, since these channelsall eventually reach the river.

Tallying the bundle of services (out of 6 total) consumed by eachhousehold, we estimated a mean of 2.5 � 1.5 services perhousehold, with groundwater use being the most common serviceconsumed. As Fig. 3 illustrates, approximately 1 of every 8 householdssurveyed is using the river for direct, sanitary purposes via thebamboo platforms known as getek. Thus the most visible consumersof the river’s services constitute a minority of overall consumers.Recreational users constituted a surprisingly high share of house-holds (�40%), given the bank conditions and water quality. Severaladditional households reported having recreated along the river indecades past, but have ceased due to worsening water quality orenvironmental conditions.

We calculated correlations (Spearman’s r) among our sixservices, as well as a correlation with (1) distance and (2) elevationwith respect to the river, since we hypothesized that householdscloser to the river will rely more strongly on these services. We alsocalculated correlations of our socioeconomic variables with servicevariables, and with respect to elevation and distance from the river,to test the conventional wisdom that poorer, shorter-tenuredhouseholds are located nearer the river. To test for global spatialcorrelation, and in particular, whether environmental serviceconsumers cluster, we calculated the Getis-Ord General G statisticusing ArcGIS 10 (ESRI). The General G statistic is an inferentialstatistic that can test for overall (global) spatial association, againsta null hypothesis that our binary values for service consumptionare completely spatially random (Getis and Ord, 1992). Wehypothesized that service consumption was likely influenced byneighboring households’ behavior (which is of course alsoinfluenced in part by proximity to the river). We specified aninverse distance weighting so that neighboring observations’influence decayed with distance, and we selected a thresholddistance of 152 m which was the smallest ‘‘sphere of influence’’ wecould conceptualize while still allowing each observation to haveat least one neighbor influencing it. Large positive z-scores(associated with statistically significant p values) would indicatethat we can reject the null hypothesis, and that the spatial

Fig. 4. Correlations between variables and elevation within the study site. A *p < .05

and **p < .01.

D. Vollmer, A. Gret-Regamey / Global Environmental Change 23 (2013) 1542–15551548

distribution of high values (i.e., ‘‘1’’ indicating a householdconsumes a particular service) is more clustered than we couldexpect under a completely random process.

2.3.2. Examining spatial dependence

Because we sampled primarily binary variables from arelatively small set of points, it was difficult to extrapolate andvisualize the results from conventional indicators of local spatialautocorrelation such as Anselin Local Moran’s I or Getis-Ord Gi*(Anselin, 1995). Thus in order to visualize the observed spatialclustering of our point data, we generated probability maps withthe indicator kriging function in ESRI’s ArcGIS 10 GeostatisticalWizard. Indicator kriging requires setting all values to either 0 or 1by establishing a threshold, so in our case an arbitrary threshold of0.5 was selected since the binary variables were already coded as 0or 1. We used the same threshold distance, 152 m, that we used totest for spatial autocorrelation. The resulting raster dataset was acontinuous surface of probabilities that a household consumes aparticular service. While probability maps are no substitute for amore detailed census, they do provide insight into hot (or cold)spots of activity.

To further investigate the relationship among service consump-tion and our other variables, we developed a linear regressionmodel estimated through ordinary least squares (OLS) and a spatialautoregressive (SAR) model to better account for any spatialdependence in the data (Anselin, 2002) using the bundled servicesvariable (SERVICES) as our dependent variable. We analyzed thenumber of services, rather than specify the mix of services thatwould constitute a bundle, because correlations among serviceswere typically low (almost all less than 0.3) and we had no priorexpectations of how these services might be bundled. Thus, by notspecifying the particular services in advance, we remained open toany combination of services. SAR models make use of a spatialweights matrix, W, which we generated using OpenGeoDa (Anselinet al., 2006) again using the same threshold distance of 152 m. Thevalue at each observation point was compared with the weightedaverage value of neighboring points within this threshold.Since elements of the matrix are nonstochastic and exogenousto the model, the matrix can be used directly in a spatialautoregressive model, or to construct spatially lagged variablesthat can be input as instrumental variables into a model and thensolved using OLS (Anselin, 2002).

We regressed our dependent variable SERVICES against sevenexplanatory variables: four demographic variables (household size[HH], years residing [YRS], home ownership [OWN], and consump-tion expenditures [CONS_EXP]), two water & sanitation-relatedvariables (having municipal water connection [PAM] and a toilet in

Table 3Linear regression model parameter estimates (SERVICES as dependent variable).

Variable Model 1 (OLS) Model 2 (OLS)

HH .035 .014

YRS_RES �.002 �.005

OWN .094 .165

CONS_EXP �.038 �.005

PAM �.920** �1.07**

DIST �.008**

ELEV �.273**

TOILET �1.13** �1.00**

Lambda

WT_T_WO_S

WT_DISP

WT_REC

WT_NONREC

R-squared .533 .601

n = 213.* p < .05.** p < .01.

the home [TOILET]), and one of two spatial variables, distance to theriver [DIST] or elevation [ELEV], which were highly correlated withone another and thus not used in tandem in the regression models.Formally, the linear model is:

SERVICES ¼ xb þ e (1)

where xb are the explanatory variables and e is the error term.We also specified a spatial error model in the case that there is

unobserved spatial dependence in the error term e, where:

SERVICES ¼ xb þ e; e ¼ lWe þ u (2)

The error term is divided into l, the coefficient for spatiallycorrelated errors (essentially unobserved location-specific effects),and the non-correlated component, u. A statistically significantvalue on the l coefficient would indicate that some spatial processnot captured directly by our covariates is influencing thedependent variable.

To try and decompose the influence of neighboring households’behavior, we developed spatially lagged variables for the fourservice variables that we hypothesized would have significantspatial dependence, i.e., households would be likely to follow whattheir neighbors were doing: discharging sewer waste(WT_T_WO_S), direct sanitary use (WT_NONREC), discarding solidwaste (WT_DISP), and recreating (WT_REC). We developed aconditional model (Anselin, 2002; Goetzke, 2008) where thevalues for neighboring observations are assumed to be exogenousto the model. We hypothesized that certain services would besensitive to a neighborhood effect, where environmentally

Model 3 (spatial error) Model 4 (conditional)

.043 .020

�.007* �.004

.176 .131

�.087 .000

�.912** �1.04**

�.268** �.201**

�1.17** �1.02**

.507**

�1.15

.978*

.416

.098

.576 .628

D. Vollmer, A. Gret-Regamey / Global Environmental Change 23 (2013) 1542–1555 1549

degrading behaviors would be more ‘‘accepted’’ and thus likelywhen clusters of neighbors engaged in the behavior, and recreationwould similarly be likely in areas where a high share of neighborsrecreate. The spatial lag terms we computed using the weightsmatrix W can be inserted into the equation as observablecharacteristics and estimated using OLS:

SERVICES ¼ Xb þ WTTwoS þ WTNONREC þ WTREC þ WTDISP þ e (3)

Some efficiency is lost when the neighboring dependent valuesare assumed to be independent (Griffith, 2004) but our resultssuggest that the overall models improved.

Finally, to investigate interactions among our individual servicevariables, we estimated a logistic model using recreation (REC) asour dependent variable. The REC variable was a suitable dependentvariable because it exhibited some of the highest correlations withother service variables (see Table 2 in next section) and thecorrelations were positive, which was counterintuitive. It alsoallowed us to test the hypothesis that households segregate theiruses of the river, e.g., those who use it for recreation would not alsodump their waste in it. We decided to keep REC as a binary variablerather than parse it by frequency or type (passive versus active), asthis allowed for a more parsimonious model. Our analysis here wasmore exploratory than explanatory, as we had not initiallydesigned the survey instrument to focus on recreational activitiesin detail.

3. Results

3.1. Correlations between services and proximity to the river

One of our primary objectives was to detect heterogeneity andspatial dependence within our study area, which could counter thenotion of kampung behavior as being uniform. In particular, wesought to challenge assumptions about households located nearerthe river, since rehabilitation plans appear to be focused on a fixeddistance from the river. Fig. 4 illustrates the correlations observedbetween several variables and elevation. Correlations withdistance are not depicted in Fig. 4 but were quite similar. Allcorrelations among services and proximity to the river (DIST and

ELEV) were significant and with signs as expected. Table 2 providesthe correlations among service variables and our proxy variable formonthly income, CONS_EXP. This table also provides the Getis-OrdG Statistic Z-score as evidence of the strength (or absence) ofspatial clustering of service users.

On the one hand, it should not be surprising that certainservices, such as discharging sewage (T_WO_S), would bedependent on being near the river. On the other hand, since allof these services, with the exception of recreation, havealternatives that can be procured through the marketplace moregenerically, it is nonetheless surprising to have found significantand, in some cases, strong correlations over such a small studyarea. While we initially assumed that household income wouldstrongly influence a household’s decision to rely directly onenvironmental services, CONS_EXP exhibited very low and in mostcases insignificant correlations with the services. Table 2 showsthis and additional results from service cross-correlations, as wellas the Z-scores for the General G statistic on spatial clustering. Wedid find CONS_EXP to be positively correlated (.246) with TOILET

but not significantly correlated with PAM. As we will discuss inSection 4, the marketed alternatives, such as municipal water andhome septic tanks, are often constrained by locational andinstitutional factors and not necessarily monetary costs.

We did not find statistically significant correlation betweenlength of tenure (YRS_RES) and proximity to the river, nor betweenhome ownership (OWN) and proximity, thus we found no supportthat relative newcomers to the area are renting properties along

the river’s edge. Typically, riverside communities will developcloser to the river (i.e., developing on riskier land) over time, butwithin our sample we did not detect such a signal. We did notgather information on when homes were constructed and,crucially, whether the property has been expanded over time. Itcould be the case that our variables failed to capture thespatiotemporal dynamics of land conversion within the site. Itcould also be that renters occupying a room within a house ownedby another family were underrepresented in our sample. We didfind a positive correlation between CONS_EXP and proximity,suggesting that household incomes increased with distance fromthe river. This conforms to expectations that poorer householdsmight be located in riskier, and thus lower-value, areas. Sinceconsumption expenditures is an imperfect measure of householdincome and wealth, and since we were attempting to subdivide asample population where consumption expenditures are gener-ally low, it may be too difficult to detect a strong signal related torelative wealth within the study area.

3.2. Mapping the consumption of environmental services

Another goal of our work was to represent, in a spatially explicitmanner, the use of these environmental services. The probabilitymaps that we generated for four services are presented in Fig. 5. Ingeneral, the stronger the clustering effect, the clearer and morepronounced are the hot (yellow) and cold (blue) spots of probableactivity. Solid waste assimilation is strongly concentrated inspecific neighborhoods, mostly on the Kampung Melayu side of theriver, and with active areas stretching more than 200 m from theriver’s edge. This is consistent with information gathered throughinterviews, where the RTs on the Bukit Duri side of the riverappeared to be better organized with regards to solid wastecollection. Those who do have organized solid waste collection payrelatively modest monthly fees (up to Rp 15,000 per month,collected at the RT level), suggesting that discarding waste directlyto the river is not a great monetary savings. Interestingly,households on the Bukit Duri side paid more on average thanhouseholds in Kampung Melayu, and some respondents suggestedthat fees being collected in Kampung Melayu were insufficient toensure reliable and adequate service from the private trashcollectors.

Groundwater use, as a service, was dispersed across the studyarea. In areas where municipal water service was readily available,some households still relied at least partly on private wells, or evenpublic wells, owing to poor quality or service from the municipalprovider. It should be noted that the municipal water supplies arenot potable and must be boiled before consuming, making itequivalent, quality-wise, to the water extracted from local wells,where leaky septic tanks are contaminating groundwater.Throughout the study area, the only directly potable water wasfrom the vendors, though even some of the (less costly) vendedsupplies needed to be boiled first, as they originated frommunicipal taps or well water from within the site.

Direct sanitary use is tightly concentrated in small clustersalong the river, mostly within approximately 50 m of the river’sedge, which is incidentally the distance Resosudarmo et al. (1997)used to define their ‘‘riverside communities’’ when surveyingperceptions of the Prokasih clean river program of the early 1990s.These clusters correspond neatly to the locations of the getek alongthe river, which are privately constructed and owned, withnominal fees paid to operators and use typically restricted toresidents of a particular RT. Recreational use is much moredispersed throughout the study area, though the highest con-centrations tend to be within roughly 100 m of the river’s edge, andmay also reflect stretches of the river bank that are comparativelyaccessible or open (i.e., clear of structures). This map for

Fig. 5. Probability maps of households consuming specific services. The lighter the color, the greater the probability that a household is using a particular service. Services are,

clockwise from top left: solid waste disposal (DISP), groundwater use (GW), recreation (REC), and direct sanitary use (NONREC). Maps were generated by indicator kriging

using the Geostatistical Wizard in ArcGIS 10 (ESRI, Inc.).

D. Vollmer, A. Gret-Regamey / Global Environmental Change 23 (2013) 1542–15551550

recreational use also does not convey information on frequency ofuse.

3.3. Regression results and interpretation

The main findings of the regression analysis affirm theimportant influence that proximity to the river has on theconsumption of environmental services (Table 3). Moreover,the results suggest that neighborhood characteristics influencehousehold behavior, and that households with access to

municipal infrastructure are significantly less dependent onthe river’s services.

In our initial linear model (Model 1) the coefficient on distancefrom the river suggests that, for every 125 m a household movesback, it consumes one less service. When we substituted elevationin for distance in the second linear model, we observed a higher R-squared value (.601 versus .533), thus elevation appears to be abetter measure of proximity than linear distance, at least in ourstudy area. In the subsequent spatial regression model we usedelevation as the proximity measure, and although it (Model 3)

Table 4Logistic regression model parameter estimates (REC as dependent variable).

Variable Model 5 Model 6 Model 7

DISP 1.286 (.323)*** .834 (.369)** .708 (.405)*

NONREC 1.600 (.516)*** 1.391 (.522)*** 1.268 (.541)**

T_WO_S .762 (.348)** .622 (.354)* .631 (.381)*

ELEV �.197 (.086)*** �.233 (.098)**

YRS_RES .000 (.009)

CONS_EXP .074 (.180)

Nagelkerke R-square .248 .277 .282

n = 213; standard errors are in parentheses.* p < .1** p < .05.*** p < .01.

D. Vollmer, A. Gret-Regamey / Global Environmental Change 23 (2013) 1542–1555 1551

returned significant coefficients on the spatial term, it had a lowerR-squared value than the second linear model. Thus, we retainedthe linear model and supplemented it with our four spatiallylagged service variables in Model 4, which delivered the highest R-squared of any of the models we tested.

At a household level, our two water and sanitationinfrastructure variables returned significant (p < .01) and largecoefficients in all six models. The coefficient on having amunicipal water connection ranged from �.833 to �1.06, whichwas as expected since many households with a municipalconnection (PAM) do not rely on groundwater, and thus wouldconsume one less service. The coefficient on having a toilet in thehousehold was always less than one, as low as �1.27. Althoughhouseholds with a toilet would be less likely to use the river fordirect sanitary purposes (the getek also have toilets on one end),roughly a third of households with toilets were discharging theirsewage into the river, and so we would not have expected acoefficient less than one.

Our spatial variables performed as expected, suggesting thatboth proximity to the river and neighboring activity had largeinfluences on the size of the bundle of services a householdconsumes. We ran the linear model without a proximity variableand it reduced the R-squared value to .383, indicating that theproximity variable has strong explanatory power. To test forrobustness, we varied the spatial lag distance that we used forconstructing the weights matrix and computing spatially laggedinstrumental variables. We were unable to examine neighbor-hood effects inside of our 152 m threshold distance, but whenwe estimated the spatial error model with progressively largerthreshold distances, we did find that the model fit (as measuredby the Akaike Information Criterion) deteriorated. Fig. 6represents this graphically, and it can be seen that there may

Fig. 6. Relationship between distance threshold and Akaike Information Criterion

(AIC) in spatial error model. Five models were estimated, beginning with the

minimum threshold of 152 m and increasing the threshold distance by

approximately 25 m for each subsequent run.

have been diminishing returns to an even smaller distancethreshold. In other words, 150 m appears to be a reasonableproxy for the neighborhood effect as it influences the consump-tion of environmental services in the study area.

For the logistic regressions, using REC as our dependentvariable, we found that other service variables seemed tosignificantly influence the likelihood of a household recreatingalong the river, although the explanatory power of our models wasconsistently low (Table 4).

In stepwise fashion, we incorporated our proximity variable(ELEV) and our two demographic variables that we hypothesizedcould influence the likelihood of recreating. The Nagelkerke R-square increases at each step, but the demographic variables arenot significant, though they do reduce the statistical significance ofthe three service variables.

Our results with this logistic model were not surprising in lightof the observed positive correlations between REC and the otherservice variables, but nonetheless counterintuitive, that house-holds using the river as a waste sink or kitchen sink would be thesame households using it as a recreation source. This suggests to usthat households which rely on the river for services rely on itintensively, and that residents are able to compartmentalize thedifferent functions of the river and adapt their behavioraccordingly. We did not collect individual-specific information,such as education levels, knowledge of the river’s water quality, orrisk attitudes, but of course these might have given us a morecomplete picture of behavior toward the river.

Our overall results suggest that standard demographic vari-ables play a limited role in determining the services a typicalhousehold consumes. To confirm this we estimated a logisticmodel with each service as a dependent variable, using the sameexplanatory variables from Model 2. While many of these logisticmodels had a good fit overall, only three of the six models returneda statistically significant parameter on a demographic variable. Wehad hypothesized that larger households, less tenured residents,renters, and poorer households would all rely on more services, butout of the four linear models reported in Table 3 and the six logisticmodels we ran on individual services, only two coefficients wereever significant. The YRS_RES was positive for groundwater use, butnegative (as expected) for solid waste disposal, and in both caseshad a small overall effect. Ownership had a more pronouncedeffect, and was negative for discharging sewage (T_WO_S) butpositive for disposing solid waste.

4. Discussion

Extant research on the environment in urban informalsettlements has tended to reduce this complex social–ecologicalsystem into a simplified stressor–response relationship. Ratherthan callous polluters, we might view residents as rationallyusing the river to compensate for a lack of infrastructure. Our

D. Vollmer, A. Gret-Regamey / Global Environmental Change 23 (2013) 1542–15551552

results demonstrate that households are in fact depending on arange of environmental services, and that this dependenceseems to decrease with elevation from the river. Alternatives tothese services are available, but are typically constrained byeconomic, institutional, and location issues. If block-level (RT)solid waste collection is not linked to a formalized landfill ortransfer station, then the service quickly breaks down—in someRTs households were paying a fee to have their waste collectedin a pushcart, only to have the pushcarts emptied into the river!Household septic tanks range in cost from US$150–500 andrequire maintenance fees of roughly US$25 at least twice a year.More critically, septic tanks need a buffer area so as not topollute groundwater wells, and given the density of wells in ourstudy area this was a near impossibility for some RTs.Connecting to the municipal water supply reduces this relianceon groundwater, but households must pay the costs of extendingthe pipe network to their homes (Bakker, 2007). This helpsexplain the clustering of municipal water users we observed inour study area, likely corresponding to nodes of the existingsupply network where extension costs would be lowest.

Bunch’s (2003) research on the Cooum River and its environs inChennai, India is instructive. He observes that, by encouragingstakeholders to consider the broader social–ecological system(and focus on human activity rather than just environmentalquality indicators), slum dwellers were no longer caricatured as aprimary source of pollution and were better understood as actorsin a more complex system. More often, policies are bluntlyformulated (Texier, 2008), such as Jakarta’s plan to ‘‘normalize’’ itsrivers by uniformly widening them, involving relocation ofpotentially hundreds of thousands of households. Remedialpolicies to improve environmental conditions ought to strive toreflect the heterogeneity we observed, and to seize on the positiveengagements with the river that already exist. Within our studyarea, we found evidence of community-based management(loosely referred to as gotong royong) in the forms compostingprograms for organic waste, organized solid waste collection andrecycling, drainage system maintenance, and restoration ofriparian vegetation. There are some challenges to scaling theseefforts up, but they suggest that some residents are activelyengaged in maintaining and rehabilitating their local environ-ment. Similar self-organizing was observed in Winayanti andLang’s (2004) study of an informal settlement along the CipinangRiver in East Jakarta.

Proximity to and reliance on the river makes these settlementsespecially vulnerable to floods as well as declines in water qualityand quantity. Given the close interaction between residents andthe natural environment in these spaces, health and well-beingshould be included as goals for plans to rehabilitate the rivercorridor (Parkes et al., 2010). Households relying on wells for theirwater supply are at risk during the dry season when river levels arelow – some households reported that their wells are going dry forup to a month. This so-called riparian ‘‘hydrologic drought’’(Groffman et al., 2003), where cities experience lower water tablesand a decline in certain hydrologic services due to processes ofurbanization, is arguably more critical in informal settlementswhere a low water table directly affects water supply for asubstantial portion of residents. Phanuwan et al.’s (2006) study offloods in the Kampung Melayu neighborhood found high levels ofE. coli and human enteric viruses present in the flood waters. Theauthors proposed that flooding was a more likely source ofgroundwater contamination than leaky septic tanks, and thatresidents faced a high risk of infection through contaminatedgroundwater or contact with the urban floodwaters. Residentsrarely possess such detailed information about the potential risksthey face. Crona et al. (2009) find that peri-urban residents in EastAfrica largely did not perceive the sewage being dumped into

nearby mangrove swamps as posing a health hazard. We did notlook specifically at individual risk perception but acknowledge thatit plays an important factor in households’ vulnerability within ourstudy area.

Vulnerability in these settlements is also induced byinstitutional or other social characteristics (Adger, 2006). Myllylaand Kuvaja (2005) note in their review of urban environmentaldevelopment in Lagos, New Delhi, Cairo, and Manila, that thechallenge of managing environmental services is primarily one ofsocietal constraints that perpetuate inequitable access. Bakker(2007) refers to this as ‘‘urban dualism’’ in the case of Jakarta’swater supply, where low-income informal communities paymore for lower-quality services, and we see evidence of thatdualism extending to solid waste and sanitation services in ourstudy area as well. Vulnerability frequently varies among socialunits (Turner et al., 2003) and our empirical results confirm this.Spatially explicit analysis of these settlements, which differ-entiates levels of exposure to environmental hazards and thesocial units (RTs, in our case), is thus crucial to understandingvulnerability in this context.

Institutions such as a property rights regime tend to createrigidity traps (Robards et al., 2011) – informal settlements inJakarta are tolerated but rarely incorporated into the city’s formalinfrastructural systems. Consequently, as settlements along therivers have grown, waste loads have increased, leading to, as someresidents noted during interviews, a noticeable tradeoff inenvironmental services and blame being centered on thesecommunities. This appears to be a fundamental challenge inJakarta, where water pollution and flooding issues are attributed(at least partially) to informal settlements, which are a highlyvisible and proximate cause even if the magnitude of theircontributions is not clear. Steinberg (2007) describes the discon-nect between Jakarta’s aggressive eviction policies to clearenvironmentally sensitive land, and the government’s complicitityin permitting large-scale developments along riverbanks and othergreen space. Residents are vulnerable to forced evictions by virtueof living in informal settlements, regardless of their efforts to avertenvironmental degradation, or their relative contribution toflooding and water quality problems, vis-a-vis upstream landuse changes taking place outside of Jakarta’s provincial boundaries.

It would indeed be useful to analyze the larger social–ecological system, at the catchment-level or greater, characteriz-ing spatial and sociodemographic differences and illuminatingthe complex, nested scales of vulnerability (Adger, 2006; Turner,2010). We have analyzed a very small portion of a 384 km2

catchment area because this particular site typifies the informalsettlements along a larger stretch of not only the Ciliwung butmany rivers in Indonesia and other rapidly urbanizing regions.Still, such a close-range analysis does not allow us to speculate asto how the river basin’s resources might be managed moresustainably. Most of the households we surveyed were no morethan 200 m from the river’s edge, and so presumably as we movefurther away we would see a different level of householdengagement with the river. The Ciliwung River catchment, likeriver systems in much of Asia, is host to rapid land conversion,including loss of irrigated rice paddies in the middle catchmentand forested land in the upper catchment, that will haveconsequences for services available to urban residents down-stream. A clearer picture of these dynamics within the catchmentwould also shed light onto the underlying social and economicpressures on the river system, the tradeoffs being made, and thepotential gains from various rehabilitation scenarios.

People continue to settle along urban waterways for a varietyof reasons, and major waterfront rehabilitation efforts arebecoming more common in developing world cities (Vollmer,2009). Still, the persistence and growth of informal settlements

D. Vollmer, A. Gret-Regamey / Global Environmental Change 23 (2013) 1542–1555 1553

along waterways seems to be an issue that many cities will faceinto the future. The environmental conditions of these informalsettlements vary across countries and even within cities. Weconsidered informal settlements along waterways unique pre-cisely because of the availability of environmental services. Themere presence of a river may make these sites attractive to notonly low-income residents but also to a growing middle classdemanding waterfront amenities. Assuming that a substantialportion of future urban migrants will move, at least initially, toinformal settlements, these urban waterways may become highlycontested areas, as recent large-scale evictions (Du Plessis, 2005;Padawangi, 2012) suggest. A more nuanced understanding of thesocial–ecological interactions there should help policymakersdevelop less intrusive, more targeted plans, whether the end goalsare environmental remediation, socio-economic development, orboth.

Acknowledgments

We would like to acknowledge the time that all respondentsdevoted to participating in the household survey and interviews.

We are grateful to the team of faculty and students from theUniversity of Indonesia’s Department of Architecture for their helpin translating and administering the survey: Herlily, AhmadGamal, Joko Adianto, Nur Hadianto, Lutfi Prayogi, Adlina Baridwan,Adriana Andhini, Ajeng Dwi Astuti, Ajeng Nadia Ilmiani, AyuMurhardiningtyas, Bagus, Fera Farwah, Libra Tono, M. ArlexSalahudin, and Yolanda Sembiring. We benefitted from theinsightful comments of two peer reviewers. Finally, we wouldlike to thank the Singapore National Research Foundation andSingapore-ETH Centre Future Cities Laboratory for providing thefinancial support to carry out the research and associatedfieldwork.

Appendix A

Mean population densities (year 2000) in urban areas (gray)and within 1 km of urban waterways (blue) for select countries.Median values and the interquartile range are also given for eachdata point. These countries witnessed the largest aggregate growthin urban population over the time period 1985–2000. Populationdata are taken from UNDESA (2012).

Appendix B Top 20 countries in terms of urban population growth, 1985–2000.

Country Change in urban population,

1985–2000 (thousands)

GNI per capita, year 2000

(current US$)

Income category,

year 2000

Percent of urban population

in slums, year 2001

Algeria 7960 1600 Lower middle 12

Argentina 7471 7460 Upper middle 33

Bangladesh 14,426 380 Low 85

Brazil 46,434 3860 Lower middle 37

China 213,646 930 Low 38

Colombia 8992 2350 Lower middle 22

India 100,575 450 Low 55

Indonesia 45,771 560 Low 23

Iran 17,019 1630 Lower middle 44

Malaysia 7278 3420 Lower middle 2

Mexico 22,170 5010 Upper middle 20

Nigeria 25,057 270 Low 79

Pakistan 19,877 470 Low 74

Philippines 13,834 1050 Low 44

Republic of Korea 10,335 9910 Upper middle 37

South Africa 9180 3050 Lower middle 33

Turkey 15,284 4170 Upper middle 18

United States 43,802 34,890 High No data

Venezuela 7739 4100 Upper middle 41

Viet Nam 7399 390 Low 47

Data are from the United Nations’ World Urbanization Prospects 2011 (UNDESA, 2012), the World Bank’s World Development Indicators (World Bank, 2013), and UN-Habitat’s

Global Urban Indicators database (www.unhabitat.org/stats/Default.aspx), last accessed 18 February 2013.

D. Vollmer, A. Gret-Regamey / Global Environmental Change 23 (2013) 1542–15551554

References

Adger, W.N., 2006. Vulnerability. Glob. Environ. Change 16, 268–281.Aguilar, A.G., 2008. Peri-urbanization, illegal settlements and environmental im-

pact in Mexico City. Cities 25, 133–145.Aguilar, A.G., Santos, C., 2011. Informal settlements’ needs and Environ Conserv in

Mexico City: an unsolved challenge for land-use policy. Land Use Policy 28,649–662.

Ali, F., 2010. Development of water stress index as a tool for the assessment of waterstress areas in the metropolitan Jakarta. In: 16th Annual International Sustain-able Development Research Conference 2010. May 30–June 1. The KadoorieInstitute, The University of Hong Kong.

Alpızar, F., Madrigal, R., 2008. Constructing payment schemes for ecosystemservices at the local level: methodological approach and some lessons learned.Economics and Conservation in the Tropics: A Strategic Dialogue. ConservationStrategy Fund, Gordon and Betty Moore Foundation, Resources for the Future.

Anselin, L., 1992. Spatial Data Analysis with GIS: An Introduction to Application inthe Social Sciences. NCGIA, Santa Barbara, USA, pp. 10–92.

Anselin, L., 1995. Local indicators of spatial association—LISA. Geogr. Anal. 27,93–115.

Anselin, L., 2002. Under the hood Issues in the specification and interpretation ofspatial regression models. Agric. Econ. 27, 247–267.

Anselin, L., Syabri, I., Kho, Y., 2006. GeoDa: an introduction to spatial data analysis.Geogr. Anal. 38, 5–22.

Arias, M.E., Cochrane, T.A., Piman, T., Kummu, M., Caruso, B.S., Killeen, T.J., 2012.Quantifying changes in flooding and habitats in the Tonle Sap Lake (Cambodia)caused by water infrastructure development and climate change in the MekongBasin. J. Environ. Manage. 112C, 53–66.

Badola, R., Hussain, S., 2005. Valuing ecosystem functions: an empirical study on thestorm protection function of Bhitarkanika mangrove ecosystem, India. Environ.Conserv. 32, 85–92.

Bakker, K., 2007. Trickle Down? Private sector participation and the pro-poor watersupply debate in Jakarta, Indonesia. Geoforum 38, 855–868.

Bakker, K., Kooy, M., Shofiani, N.E., Martijn, E.-J., 2008. Governance failure: rethink-ing the institutional dimensions of urban water supply to poor households.World Dev. 36, 1891–1915.

Barbier, E.B., 2007. Natural capital and labor allocation. J. Environ. Dev. 16, 398–431.BPS (Badan Pusat Statistik). Trends of selected socio-economic indicators of

Indonesia, February 2012.Brauman, K.A., Daily, G.C., Duarte, T.K.E., Mooney, H.A., 2007. The nature and value

of ecosystem services: an overview highlighting hydrologic services. Annu. Rev.Environ. Resour. 32, 67–98.

Bunch, M.J., 2003. Soft systems methodology and the ecosystem approach: a systemstudy of the Cooum River and environs in Chennai, India. Environ. Manage. 31,182–197.

Cadenasso, M.L., Pickett, S.T.A., Schwarz, K., 2007. Spatial heterogeneity in urbanecosystems: reconceptualizing land cover and a framework for classification.Front. Ecol. Environ. 5, 80–88.

Carpenter, S.R., Bennett, E.M., Peterson, G.D., 2006. Scenarios for ecosystem ser-vices: an overview. Ecol. Soc. 11, 29–43.

Chhatre, A., Agrawal, A., 2009. Trade-offs and synergies between carbon storage andlivelihood benefits from forest commons. Proc. Natl. Acad. Sci. U.S.A. 106,17667–17670.

CIESIN (Center for International Earth Science Information Network), IFPRI (Inter-national Food Policy Research Institute), The World Bank & CIAT (InternationalCenter for Tropical Agriculture), 2011. Global Rural-Urban Mapping Project,Version 1 (GRUMPv1): Population Density Grid. NASA Socioeconomic Data andApplications Center (SEDAC), Palisades, NY, USA.

CIESIN, IFPRI, The World Bank & CIAT, 2011. Global Rural-Urban Mapping Project,Version 1 (GRUMPv1): Urban Extents Grid. NASA Socioeconomic Data andApplications Center (SEDAC), Palisades, NY, USA.

Conchedda, G., Lambin, E.F., Mayaux, P., 2011. Between land and sea: livelihoodsand environmental changes in mangrove ecosystems of senegal. Ann. Assoc.Am. Geogr. 101, 1259–1284.

Crane, R., Daniere, A., 1996. Measuring access to basic services in global cities:descriptive and behavioral approaches. J. Am. Plann. Assoc. 62, 203–221.

Crona, B.I., Ronnback, P., Jiddawi, N., Ochiewo, J., Maghimbi, S., Bandeira, S., 2009.Murky water: analyzing risk perception and stakeholder vulnerability relatedto sewage impacts in mangroves of East Africa. Glob. Environ. Change 19,227–239.

Daily, G.C. (Ed.), 1997. Nature’s Services: Societal Dependence on Natural Ecosys-tems. Island Press, Washington, DC, USA.

Das, S., Vincent, J.R., 2009. Mangroves protected villages and reduced death tollduring Indian super cyclone. Proc. Natl. Acad. Sci. U.S.A. 106, 7357.

Djadjadilaga, M., Sigit, H., Tejalaksana, A., 2008. From Data to Policy—Ciliwung RiverWater Quality Management. The 3rd WEPA International Forum, Kuala Lumpur,Malaysia.

Du Plessis, J., 2005. The growing problem of forced evictions and the crucialimportance of community-based, locally appropriate alternatives. Environ.Urban 17, 123–134.

Engel, S., Palmer, C., 2008. Payments for environmental services as an alternativeto logging under weak property rights: the case of Indonesia. Ecol. Econ. 65,799–809.

Findlay, S.J., Taylor, M.P., 2006. Why rehabilitate urban river systems? Area 38,312–325.

Getis, A., Ord, J.K., 1992. The analysis of spatial association by use of distancestatistics. Geogr. Anal. 24, 189–206.

Goetzke, F., 2008. Network effects in public transit use: evidence from a spatiallyautoregressive mode choice model for New York. Urban Stud. 45, 407–417.

Gomez-Baggethun, E., Barton, D.N., 2012. Classifying and valuing ecosystem ser-vices for urban planning. Ecol. Econ..

Griffith, D.A., 2004. A spatial filtering specification for the autologistic model.Environ. Plann. A 36, 1791–1811.

Groffman, P.M., Bain, D.J., Band, L.E., Belt, K.T., Brush, G.S., Grove, J.M., Pouyat, R.V.,Yesilonis, I.C., Zipperer, W.C., 2003. Down by the riverside: urban riparianecology. Front. Ecol. Environ. 1, 315–321.

Harriden, K., 2012. Potential impacts of slum urbanisation on channel bank storagein the Bang Pakong River, Thailand. Water Pract. Technol. 7 .

Jaung, W., Bae, J.S., 2012. Evaluating socio-economic equity of REDD+ in arights-based approach: Rapid equity appraisal matrix. J. Environ. Sci. Pol. 22,1–12.

Jayawickreme, D.H., 2011. Threats to Sri Lanka’s urban wetlands. Front. Ecol.Environ. 9, 320–321.

Kosoy, N., Martinez-Tuna, M., Muradian, R., Martinez-Alier, J., 2007. Payments forenvironmental services in watersheds: Insights from a comparative study ofthree cases in Central America. Ecol. Econ. 61, 446–455.

D. Vollmer, A. Gret-Regamey / Global Environmental Change 23 (2013) 1542–1555 1555

Lee, K.N., 2006. Urban sustainability and the limits of classical environmentalism.Environ. Urban 18, 9–22.

Li, J., Feldman, M.W., Li, S., Daily, G.C., 2011. Rural household income and inequalityunder the Sloping Land Conversion Program in western China. Proc. Natl. Acad.Sci. U.S.A. 108, 7721–7726.

Marcotullio, P.J., 2007. Urban water-related environmental transitions in SoutheastAsia. Sust. Sci. 2, 27–54.

Mcgranahan, G., 2007. Urban transitions and the spatial displacement of environ-mental burdens. In: Marcotullio, P., Mcgranahan, G. (Eds.), Scaling UrbanEnvironmental Challenges: From Local to Global and Back. Earthscan, London,UK.

Mendoza-Gonzalez, G., Martınez, M.L., Lithgow, D., Perez-Maqueo, O., Simonin, P.,2012. Land use change and its effects on the value of ecosystem services alongthe coast of the Gulf of Mexico. Ecol. Econ. 82, 23–32.

Meyfroidt, P., Lambin, E.F., 2008. The causes of the reforestation in Vietnam. LandUse Policy 25, 182–197.

Myllyla, S., Kuvaja, K., 2005. Societal premises for sustainable development in largesouthern cities. Glob. Environ. Change 15, 224–237.

Nelson, E., Mendoza, G., Regetz, J., Polasky, S., Tallis, H., Cameron, D., Chan, K.M.A.,Daily, G.C., Goldstein, J., Kareiva, P.M., Lonsdorf, E., Naidoo, R., Ricketts, T.H.,Shaw, M., 2009. Modeling multiple ecosystem services, biodiversity conserva-tion, commodity production, and tradeoffs at landscape scales. Front. Ecol.Environ. 7, 4–11.

Noosorn, N., Niamkamnerd, P., 2009. Public participation to minimize waterbornediseases in Yom riverside communities. J. Public Health Epidemiol. 1, 37–40.

Padawangi, R., 2012. Climate change and the North coast of Jakarta: environmentaljustice and the social construction of space in urban poor communities. In: Holt,III, W.G. (Ed.), Research in Urban Sociology. Emerald, UK.

Palmer, M.A., Reidy Liermann, C.A., Nilsson, C., Florke, M., Alcamo, J., Lake, P.S., Bond,N., 2008. Climate change and the world’s river basins: anticipating managementoptions. Front. Ecol. Environ. 6, 81–89.

Parkes, M.W., Morrison, K.E., Bunch, M.J., Hallstrom, L.K., Neudoerffer, R.C., Venema,H.D., Waltner-Toews, D., 2010. Towards integrated governance for water, healthand social–ecological systems: the watershed governance prism. Glob. Environ.Change 20, 693–704.

Phanuwan, C., Takizawa, S., Oguma, K., Katayama, H., Yunika, A., Ohgaki, S., 2006.Monitoring of human enteric viruses and coliform bacteria in waters after urbanflood in Jakarta, Indonesia. Water Sci. Technol. 54, 203.

Resosudarmo, I., Resosudarmo, B.P., Isham, B., 1997. The Indonesian clean riverprogram (Prokasih) as perceived by the people residing along the rivers inJakarta. Indones J. Geogr. 29, 47–64.

Robards, M.D., Schoon, M.L., Meek, C.L., Engle, N.L., 2011. The importance of socialdrivers in the resilient provision of ecosystem services. Glob. Environ. Change21, 522–529.

Roberts, B., Kanaley, T., 2006. Urbanization and Sustainability in Asia: Case Studiesof Good Practice. Asian Development Bank, Manila, Philippines.

Roy, A., Alsayyad, N. (Eds.), 2004. Urban Informality: Transnational Perspectivesfrom the Middle East, Latin America, and South Asia. Lexington Books, Lanham,MD, USA.

Savage, V.R., 2006. Ecology matters: sustainable development in Southeast Asia.Sust. Sci. 1, 37–63.

Schaffler, A., Swilling, M., 2013. Valuing green infrastructure in an urban environ-ment under pressure—the Johannesburg case. Ecol. Econ. 86, 246–257.

Seftyono, C., 2011. Local Community in Valuing Ecosystem Services: Warga Kam-pung Code’s Perspective on Kali Code Existence.In: First Annual Indonesian

Scholars Conference in Taiwan, 2010. . Available at SSRN: http://ssrn.com/abstract=1763326 (accessed 2.04.13).

Seftyono, C., 2012. Kali Code: From Merapi to Sustainable Society.In: InternationalConference on Sustainable Innovation. . Available at SSRN: http://ssrn.com/abstract=2020132 (accessed 2.04.13).

Seto, K.C., 2011. Exploring the dynamics of migration to mega-delta cities in Asiaand Africa: contemporary drivers and future scenarios. Glob. Environ. Change21, S94–S107.

Silver, C., 2008. Planning the Megacity: Jakarta in the Twentieth Century. Routledge,London and New York.

Sodhi, N.S., Lee, T.M., Sekercioglu, C.H., Webb, E.L., Prawiradilaga, D.M., Lohman, D.J.,Pierce, N.E., Diesmos, A.C., Rao, M., Ehrlich, P.R., 2009. Local people valueenvironmental services provided by forested parks. Biodiv. Conserv. 19,1175–1188.

Steinberg, F., 2007. Jakarta: environmental problems and sustainability. Habitat. Int.31, 354–365.

Sudoyo, W., . Total Losses for Jakarta Flooding Hit Rp 32 Trillion. In: The JakartaGlobe.

Texier, P., 2008. Floods in Jakarta: when the extreme reveals daily structuralconstraints and mismanagement. Disaster Prev. Manage. 17, 358–372.

Tjampakasari, C.R., Wahid, M.H., 2008. Water quality of Angke River: microbiolog-ical point of view. Med. J. Indones 17, 82–87.

Turner 2nd, B.L., Kasperson, R.E., Matson, P.A., Mccarthy, J.J., Corell, R.W., Chris-tensen, L., Eckley, N., Kasperson, J.X., Luers, A., Martello, M.L., Polsky, C.,Pulsipher, A., Schiller, A., 2003. A framework for vulnerability analysis insustainability science. Proc. Natl. Acad. Sci. U.S.A. 100, 8074–8079.

Turner, B.L.I., 2010. Vulnerability and resilience: coalescing or parallelingapproaches for sustainability science? Glob. Environ. Change 20, 570–576.

Tzoulas, K., Korpela, K., Venn, S., Yli-Pelkonen, V., Kazmierczak, A., Niemela, J., James,P., 2007. Promoting ecosystem and human health in urban areas using GreenInfrastructure: a literature review. Landscape Urban Plann. 81, 167–178.

Un-Habitat, 2003. The Challenge of Slums: Global Report on Human Settlements2003. United Nations Human Settlements Programme, Nairobi, Kenya.

Un-Habitat, 2009. Planning Sustainable Cities: Policy Directions: Global Report onHuman Settlements 2009. Earthscan, London, UK and Sterling, VA, USA.

Un-Habitat, 2010. State of the World’s Cities 2010/2011: bridging the urban divide.In: Un-Habitat (Eds.), State of the World’s Cities. United Nations HumanSettlements Programme, Nairobi, Kenya.

UNDESA (United Nations Department of Social and Economic Affairs), 2005. De-signing Household Survey Samples: Practical Guidelines. Series F. UnitedNations, New York, USA.

UNDESA, 2012. World Urbanization Prospects: The 2011 Revision, CD-ROM Edition.UNDESA, Population Division.

VMAP (Vector Map Level 0). Castle Rock, CO, USA, Land Info Worldwide MappingLLC, 2000, available through DIVA-GIS, http://www.diva-gis.org.

Vollmer, D., 2009. Urban waterfront rehabilitation: can it contribute to environ-mental improvements in the developing world? Environ. Res. Lett. 4, 024003.

Winayanti, L., Lang, H.C., 2004. Provision of urban services in an informal settle-ment: a case study of Kampung Penas Tanggul, Jakarta. Habitat. Int. 28, 41–65.

World Bank. World Development Indicators. Washington, DC, The WorldBank Group. http://data.worldbank.org/data-catalog/world-development-indicators, Last accessed February 10, 2013.

Zeilhofer, P., Topanotti, V.P., 2008. GIS and ordination techniques for evaluation ofenvironmental impacts in informal settlements: a case study from Cuiaba,central Brazil. Appl. Geogr. 28, 1–15.