Vous êtes sur la page 1sur 11

Habitat International 35 (2011) 316e326

Contents lists available at ScienceDirect

Habitat International
journal homepage: www.elsevier.com/locate/habitatint

Improved accessibility modeling and its relation to poverty e A case study in Southern Sri Lanka
Anders Ahlstrm a, *, Petter Pilesj b,1, Jonas Lindberg c, 2
a

Department of Earth and Ecosystem Sciences, Lund University, Slvegatan 12, SE-223 62 Lund, Sweden GIS Centre, Lund University, Slvegatan 12, SE-223 62 Lund, Sweden c Department of Human and Economic Geography, School of Business, Economics and Law, University of Gothenburg, Box 630, SE-40530 Gothenburg, Sweden
b

a b s t r a c t
Keywords: GIS Accessibility Sri Lanka Poverty

Many studies have found close relationships between accessibility and various socio-economic indicators. Yet, since accessibility tends to have differentiated effects, both socially and spatially, there is a need for a model which allows for a disaggregated analysis of accessibility. The model should be possible to use in areas where road network data is incomplete. In this paper such an accessibility model is developed, using a raster-based approach in a Geographical Information System (GIS). One important factor in accessibility modeling is to estimate the traveling speed on different landscape entities. This paper develops a method where local knowledge and physical geographical data are integrated in the GIS model. From the interview data the best door-to-door traveling speeds of three road classes were estimated. The results from these calculations have been used as frictions for a cost surface. The analysis shows strong relationships between poverty indicators and estimated spatial accessibility, stronger than the commonly used accessibility measure of Euclidian distance. 2010 Elsevier Ltd. All rights reserved.

Introduction An important trend in development thinking and planning is the refocus on spatial dimensions of poverty and development (see e.g. Sachs, 2005; World Bank, 2008). One such spatial dimension regards accessibility, which at a general level is about the ease with which one location can be reached from another location. Different measures of accessibility have been found to closely relate to various physical and social indicators. The relationship between accessibility and land use is widely recognized both for urban and rural environments (Castella, Manh, Kam, Villano, & Tronche, 2005; Etter, McAlpine, Wilson, Phinn, & Possingham, 2006; Hanson, 1986; Laurance et al., 2002; Nagendra, Southworth, & Tucker, 2003; Verburg, Overmars, & Witte, 2004). Accessibility is also related to socio-economic factors, such as poverty, health and production patterns (Guagliardo, 2004; Kam, Hossain, Bose, & Villano, 2005; Nanayakkara, 2006; Olsson, 2006; World Bank, 2007). Jalan and Ravallion (2002) have found that rural roads are important in explaining why some rural households are able to increase their

* Corresponding author. Tel.: 46 46 222 4609; fax: 46 46 222 03 21. E-mail addresses: anders.ahlstrom@nateko.lu.se (A. Ahlstrm), petter.pilesjo@ giscentrum.lu.se (P. Pilesj), jonas.lindberg@geography.gu.se (J. Lindberg). 1 Tel.: 46 46 2229654. 2 Tel.: 46 31 786 4458. 0197-3975/$ e see front matter 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.habitatint.2010.11.002

consumption levels more than others. Bryceson, Bradbury, and Bradbury (2008) discuss the complex and conditional linkages between rural road networks and poverty reduction in terms of mobility enhancement (e.g. enhanced possibilities to travel for work or services) and accessibility enhancement (e.g. increased attractiveness of rural areas for service infrastructure and stafng). Recent research however also stresses the importance of disaggregating the effects of accessibility, since transport infrastructure tends to have very differentiated impacts on regions and groups (Bryceson et al., 2008; Olsson, 2009). For this to be possible there is a need to develop models for more detailed analyses of accessibility; models which also are possible to use in situations when road network data is incomplete. This paper develops such a model for analyzing spatial accessibility. Hanson (1986) has argued that accessibility constitutes of two main parts: proximity and mobility. In this paper, proximity is represented by the distance from departure to destination, road quality and terrain, while mobility is represented by the availability of motor vehicles and the supply of, and reliability of, public transports. Accessibility is estimated as travel time, which depends on both proximity and mobility. In some studies, accessibility is modeled by simple measures, such as distances from roads, towns and other destinations (Etter et al., 2006; Nagendra et al., 2003). In other studies, accessibility is studied by comparing villages in a more interview based

A. Ahlstrm et al. / Habitat International 35 (2011) 316e326

317

approach (Olsson, 2006). In this paper, an accessibility model is developed based on raster (see e.g. Burrough & McDonnell, 1998) calculations. One of the reasons to use a raster-based approach is the incompleteness of data, a common feature of many areas in the Global South. Substantial parts of the study area in Southern Sri Lanka are served by small dirt tracks, footpaths, which not always could be found in the available road network data. The raster approach allows for simultaneous on-road and off-road modeling, which can compensate for the incompleteness of data. This is not possible with available network analysis tools. Off-road traveling is then calculated by the straight-line distance to the closest road. This approach is not suitable when data quality/completeness is low or the traveling habits in the study area to a considerable part consists of off-road or footpath traveling. One simple and quick measure of accessibility is Euclidian (as the crow ies) distance, often measured to roads (Etter et al., 2006; Nagendra et al., 2003). However, when measuring Euclidian distances to roads, the road becomes the destination. Destinations in the model presented here are chosen to represent where people are actually traveling to, or would want to travel to, like markets, work places, etc. Other studies have also recognized the need for more site specic and tailor made models (Verburg et al., 2004). Euclidian distance studies do not take road quality, road length, absence of roads, physical obstacles, or topography into account, all of which are incorporated in this model. The model presented here is also a tool which could be used to integrate methods and knowledge traditionally conned to either the social or the natural sciences. This research proposes an approach where local knowledge is integrated in a physically based GIS model, by using interview data as the base for assignment of frictions. Frictions are what dene how fast one can ctionally travel over a ground condition; a result of the estimated speed of traveling. The higher the friction of a ground condition, the slower the traveling speed is assumed to be. Generally there are two basic approaches to acquire frictions of roads for an accessibility model. The most common method is to use speed limits. However, to use speed limits is only feasible where the trafc ow has the same speed as the speed limits. A second approach is to use interviews of key informants or eld estimations to estimate frictions. In this paper, a method is developed to estimate frictions on three road classes using interview data on traveling times, departure locations and destinations. This approach quanties interview data and produces frictions based on local knowledge.

Study area The island of Sri Lanka, located just southeast of the Indian subcontinent, is classied as a lower middle-income country by the World Bank. The country has suffered from civil war for several decades, and although the war seems to be over since early 2009, data for the North and East of the country, which were most severely affected by the war, is often excluded in national calculations. The total population of Sri Lanka was estimated at 18.8 million in 2001 (DCS, 2006a), and the growth rate of the population was 1.2% the same year. The island has a proximate size of 65,610 km2. The study area of Hambantota district is located in the southern province of Sri Lanka (Fig. 1). Sri Lanka is divided in 4 administrative levels: provinces (largest), districts (second largest), divisions (third largest), and Grama Niladhari divisions (smallest). Grama Niladhari divisions will be referred to as GN-divisions throughout this paper. Hambantota district constitutes of 12 divisions and 583 GNdivisions. It is a largely agricultural district and it is usually referred to as one of the poorer districts of Sri Lanka (see e.g. Peiris, 2006). In 2002, Hambantota district had the fth highest rate of people below the national poverty line (DCS, 2006b), at least compared with other districts not directly affected by the civil war. Very recent large-scale infrastructure developments in terms of an international harbor and an international airport, as well as planned investments in railways and an express highway to Colombo, are likely to alter the marginality of the district to some extent. Accessibility e the situation in the study area The trafc of the roads in Sri Lanka and in Hambantota district is distinguished by the variations in means of transport. Many different vehicles, with different maximum speeds, trafc the same roads. There are no rules for which kind of vehicles that are allowed to trafc even the largest roads. This induces a dangerous and slow trafc situation. Fast cars and buses travel in the same lanes as slow landmasters, old trucks and bicycles. Hambantota district is very diverse in terms of accessibility. The eastern area is the least accessible, mostly due to the Yala National Park, which occupies a substantial part of the area (see Fig. 1). The area directly to the west of Yala NP is also less accessible than the central and western parts due to low quality of the roads and long distances. The western parts are the most populated and accessible area of the Hambantota district, the roads are relatively good and distances are shorter. The cost of traveling to markets and towns are relatively high in the whole district, but especially in the northcentral and eastern parts. Distances are longer, road conditions are worse and public transports run less frequent. According to literature accessibility is an important factor for poverty alleviation. Nanayakkara (2006) argues that geographical isolation is highly correlated to poverty both on a district level and on the sub-district level in Hambantota district. The World Bank (2007) shows a strong relationship at the division level between poverty and accessibility to Colombo for the whole of Sri Lanka. The Rural Economic Advancement Project (REAP) (http://www. spc.gov.lk/projects.html) has developed a total of 14 markets in Hambantota district. Some are more important than others and some are wholesale markets while others are not. According to a high ofcial at REAP (2007-09-19), 7 markets and 6 towns are clearly bigger and more important than the rest. The most important wholesale markets are (with the most important rst and major crops in parenthesis): Sooriyawewa (banana and vegetables); Pannagamuwa (banana and vegetables); Ranna (vegetables); Barawakumbura (banana); Ambalantota (banana and vegetables); Weeraketiya (banana, vegetables and fruits); and Katuwana

Objectives The main objective of this paper is to develop a model for the estimation of accessibility at a disaggregated level and in areas that suffer from incompleteness of data. A subsidiary objective of the paper is to use the model to investigate possible relationships between accessibility of agricultural markets and towns on one hand, and poverty indicators on the other hand, in Hambantota District, Sri Lanka. To fulll these objectives, a number of specic aims have been dened. Geographical data (roads, topography, etc) and data on destinations (markets/towns) have been collected. An interview based, method has been developed to estimate the traveling speed of roads from an interview study. Accessibility is modeled in a way that accounts for off-road traveling as well as on-road traveling. The research also investigated how reliable the attained road network data is in terms of both positional and thematic accuracy. Furthermore, appropriate traveling speeds of different ground conditions have been assessed.

318

A. Ahlstrm et al. / Habitat International 35 (2011) 316e326

Fig. 1. Map of Sri Lanka and the different administrative levels (data and names comes from Survey Department and DCS 2006).

(banana and vegetables). The most important towns, in terms of economic activities, are identied as Beliatta, Tangalle, Ambalantota, Tissamaharama, Walasmulla and Weeraketiya. These markets and the towns can be seen in Fig. 2 below. The REAPofcial interviewed furthermore argued that people in Hambantota District do not to any greater extent visit markets outside of Hambantota District except for Embilipitiya situated to the north of the study area.

DATA and pre-processing Road network data The existing road network data were of unknown thematic- and positional accuracy and needed to be evaluated and geometrically corrected. The road network data origins from digitalizations of 1:50,000 paper maps (Survey Department 2007-09-03). The paper

Fig. 2. The largest and most important markets and towns for the people in Hambantota District.

A. Ahlstrm et al. / Habitat International 35 (2011) 316e326

319

maps, in turn, originate from aerial photography from the 1960s. Testing the positional accuracy with a GPS receiver in eld reviled considerable errors. The low accuracy of the data is partly a consequence of generalizations in the map-making process. However, the map-making generalizations do not explain all the positional errors. The data is also of low thematic accuracy in the sense of road classication if compared to the real situation. The cause of the relatively low thematic accuracy is, according to a high ofcial at the Survey Department (2007-09-03), that the Road Development Authority classies roads according to intended road quality. However, in many cases, funds are insufcient either to maintain or to build the roads as to t the intended quality of the road classes. The collected data were classied in four main road classes, primary roads, secondary roads, tracks and footpaths, as well as railways, but the latter were not present within the district boundary. The extent of the road data layers collected was larger than Hambantota District. This was to assure that the modeled traveling routes represent the shortest route although it might not always be within the district boundary. In the re-classication used in this paper, the primary roads should be wide enough to carry trafc in three ctional lanes. A slower vehicle should be easily overtaken and they should be good enough to allow mean speeds of approximately 50 km/h. The denition applied for secondary roads is that they are paved roads with less than 3 lanes and that they allow for mean speeds of about 35 km/h. Secondary roads vary substantially in quality; some have a lot of potholes while others have close to perfect tarmac quality. In contrast to primary and secondary roads, the quality of tracks varies between seasons. In the denitions used, tracks are wider dirt roads passable by 4-wheeled vehicles in the dry season. Footpaths are common in the district, and these are narrow (maximum 3 m) dirt tracks. Although the name insinuates differently, they are in many cases fully passable with four wheeled motor vehicles although the speed is often much lower than on tracks. The road network was evaluated with regard to both positional and thematic accuracy by analyzing a set of distances between 232 GPS-measurements (see Fig. 3) of the roads and the corresponding locations on the map. In order to minimize the positional errors a 2nd order polynomial transformation was carried out, adjusting the position of the road network data to the independent GPS points. Furthermore, 21% of the roads originally classied as primary roads, 5% of the secondary roads and 0.4% of the tracks, were re-classied with the purpose of correcting the thematic errors. All corrections were checked in eld.

A measure of completeness was also applied, where completeness was dened as the percentage of the roads mapped by GPS found in the road network data. Primary and secondary roads both ended up at 100% completeness while track roads had a completeness of 95%. Footpaths had a completeness of 57%. Footpaths were on the base of their low completeness excluded from the analysis. However, since footpaths are present more or less everywhere one can say they were incorporated in the off-road classes (see below). Other data used Slope was estimated from a 90 m DEM acquired from USGS (2007). The slope was classied in 4 classes, 0e5% slope 5e10% slope, 10e20% slope and above 20% slope. Water bodies were acquired from Sandell (2008). These data origin from a satellite image (Landsat TM, 2006) classication, and it was stored as a raster image. The positions of towns and markets were digitized by hand and their positions were controlled in the eld by the use of hand held GPS receivers. Interview data Interview data have acted as the foundation of the proposed accessibility model. The results of the interviews indicated which means of transport that were important as well as traveling speeds on different roads and tracks. Interviews of ofcials also indicated the economic importance of different markets and towns. Three types of interviews were conducted. Four ofcials in Hambantota District were interviewed with the help of a semi-structured questionnaire. A survey of village households was carried out with the assistance of a semi-structured questionnaire and an interpreter. The general accessibility situation was addressed and 40 interviews were collected. Additionally, 139 (on top of these 15 training interviews were performed but not included in the analysis) structured interviews of village households were performed with the help of three eld assistants. The questionnaire used comprised of 40 quantitative, straightforward questions on traveling habits, traveling times, destinations and accessibility problems. All interviews were carried out in September and October 2007. The spatial distribution of the interviews is illustrated in Fig. 4. The semi-structured village interviews addressed the villagers general views of transportation and accessibility. They gave a better understanding of the transportation situation and accessibility in

Fig. 3. Map showing positions taken for data precision evaluation, re-classication and geometric correction (n 232).

320

A. Ahlstrm et al. / Habitat International 35 (2011) 316e326

Fig. 4. Distribution of interviews.

general. Public transportation was perceived as important by the respondents, even if some of them had to walk long distances to catch a bus. One person in a remote eastern village expressed that they had to walk for almost two hours one way to the main road to catch the bus, and when they reached the main road, they had to wait for the unreliable bus to come (2007-09-28). The village in question was located by a secondary road of medium to bad quality. However, the village was located at the very end of a road in a sparsely populated area. This implies that spatial separation, and not only road quality, is determining the transportation facilities available at a location. The structured interviews had the aim of quantifying travel speed for different means of transport on different roads and tracks as well as the perceptions of accessibility. Note that the total number of interviewees differ depending on the question as a result of different questionnaires and deletion of interviews with low quality (mainly due to training errors). Some of the main ndings were: 1. Out of 183 households questioned, 38% owned a motor vehicle, and the two most common vehicle types were motorcycles and landmasters. 2. On the question on what the biggest transportation problem was, 45% of the 168 interviewees mentioned lack of, or low supply of, public transportation, and lack of, and high expenses for, private motor vehicle. 39% answered that bad road conditions were the biggest problem while 4% stated that distances were too long. However, many respondents linked road conditions and the supply of public transports by suggesting

that the bad conditions of the roads to their village were the reason for the bad supply of public transport. Hence, the limited supply of public transportation was judged to be the biggest transportation problem according to the majority of the respondents. 3. Public transportation, i.e. bus, was the most frequent means of transport of the respondents for all asked purposes except for traveling to the market to sell goods such as agricultural produce (see Table 1). Another interesting nding is that ownership of motor vehicles does not seem to correlate well with the number of farmers taking their produce to market themselves and the amount of produce sold to middlemen in the village. Out of 88 farmers, 31% of those in possession of a motor vehicle take their produce to markets personally, while 34% of those not in possession of a motor vehicle take their produce to markets personally. A possible explanation is the relatively high petrol and diesel prices leading to high cost of transportation. The ndings of the interview survey also indicate that distance and the supply of public transportation are the most important factors explaining actual mobility. In other words, road quality, distance and the potential mobility (supply of public transportation, ownership of an own motor vehicle) are all inseparable for the measure of physical accessibility according to the results above. It seems as if mobility, one part of accessibility, is not a question about who owns a motor vehicle and who does not. Thus the method used to estimate accessibility has to account for all the main factors determining accessibility, not only examine e.g. the distance or the ownership of motor vehicles. Many factors are intertwined and have to be accounted for.

Table 1 The traveling habits of the respondents. MC motorcycle, LM landmaster 3W 3-wheeler. Type of journey Means of transport Bus 3-W MC LM Bicycle Lorrie Tractor Walking Total Traveling to market 12 0 to sell goods (no.) (%) 27 0 Traveling to market 109 11 to buy goods (no.) (%) 73 7 Traveling to 133 32 a hospital (no.) (%) 74 18 Traveling 107 6 to a town (no.) (%) 76 4 1 27 2 2 4 0 0 0 0 0 0 1 2 0 1 0 0 0 0 0 0 1 1 1 1 2 1 45 100 149 100 179 100 141 100

Table 2 Example of some of the combinations of departures and destinations. Departure Destination Length of roads Primary (m) Village1 Village2 Village3 Village3 Village4 Village4 Village5 Village5 Village6 Sooriyawewa Ridiyagama Sooriyawewa Hambantota Angunakolapelassa Mideniya Thanamalvila Lunugamwehera Tangalle 0 0 3039 18,196 0 2179 0 278 4803 Secondary (m) 26,655 10,001 8584 569 8299 6719 7772 4876 0 Track (m) 2037 0 3700 4951 0 2645 772 772 0

2 60 4 14 2 12 9 9 5 16 11 1 0 8 4

0 2 0 10 0 7

A. Ahlstrm et al. / Habitat International 35 (2011) 316e326 Table 3 Example of a travel matrix of bus traveling times used for the speed estimations. Given travel time (min) Estimated length of roads to the destinations Primary (m) 20 30 60 120 120 45 40 25 30 960 960 960 18,196 18,196 0 0 960 960 Secondary(m) 1882 1882 11,550 569 569 7772 7772 7898 7898 Track (m) 267 267 0 4951 4951 772 772 0 0

321

Method Data preparation for traveling speeds estimations As mentioned above, travel data were collected through structured interviews. Traveling times to different locations and with different means of transport were collected for a total of 139 households (15 interviews in 2 villages performed the rst day were regarded as training interviews). Each respondent gave the traveling time to 3 or 4 destinations with the means of transport most often used. For every journey, from departure (equal to the village centers) to the destination, the travel distances along the 3 road classes were measured. If there were, for example, 4 different destinations given by the respondents of one village, the distances needed to travel on primary, secondary and track roads to all these 4 destinations were measured in a GIS program. An example of the table of departures, destinations and traveling times produced can be seen in Table 2. The paths assumed to be taken from the departure to the destination were estimations of which roads the respondent travel on. However, the length of the chosen paths was cross-checked with the given distances by the respondents. In most cases the choice of path was considered to be obvious. A total of 13 villages spread out in the study area were used in the analysis. For the 13 villages 30 different combinations of departures and destinations were given. The next step was to create travel matrices for each means of transport, bus, landmaster (LM), 3-Wheeler (tuk-tuk, rikka), motorcycle and bicycle. The matrices hold information in four columns and a varying number of rows depending on the number of journeys given in the interviews. The rst column holds information on travel time given by the respondent from the departure to

a destination. The three following columns hold information on the length of the journey on primary roads, secondary roads and tracks from the village departure point to the destination that the travel time was given for. An example of a matrix is given below in Table 3. As can be seen in the rst two rows of Table 3, a given travel time from a departure to a destination was not always the same as another given travel time for the same departure to the same destination. Every row in the different matrixes (one for each means of transport) corresponds to one journey. Every journey has a departure, a destination, a traveling time and lengths of 3 road classes. However, the departure and the destination are irrelevant for the estimations of traveling speed. Only traveling time and length of roads in the 3 classes were kept in the matrices. As a result of the respondents traveling habits the matrices of different means of transport have different numbers of rows, or journeys, as a basis for the estimation of traveling speed. For bus transportation 326 journeys were collected, 111 journeys were collected for landmaster transportation, 39 for 3-wheeler, 25 for motorcycle and 20 journeys were collected for bicycle transportation. Speed estimation A program that uses the approach of a least square algorithm was developed to estimate the traveling speed of three road classes from the travel matrix. The program optimizes traveling speeds so that the mean absolute error (not the square of errors) between given and calculated traveling times are minimized. The sum of errors between given and estimated traveling times is the mean absolute error (in %) of that specic combination of speeds (Equation (1)).

P Meanabsoluteerror

jTe Tg j P 100 Tg

(1)

where Te is the estimated traveling time and Tg is the given traveling time. Two rules were set up in order to make sure that the program only gives one answer. The rules are as follows:

SpeedP ! SpeedS SpeedS ! SpeedT

(2) (3)

where SpeedP is the speed on primary roads, SpeedS is the speed on secondary roads and SpeedT is the speed on tracks.

Fig. 5. Different classes of the cost surface used in the accessibility modeling.

322 Table 4 Table of calculated traveling speeds. Means of transportation Road type Primary (km/h) Bus LM 3-Weehler MC Bicycle 33.0 7.0 22.0 50.0 11.0 Secondary (km/h) 15.5 6.5 22.0 27.0 11.0

A. Ahlstrm et al. / Habitat International 35 (2011) 316e326

No of journeys used Track (km/h) 3.5 6.5 7.0 27.0 10.5 325 74 38 22 20

Total no of journeys

Mean absolute error (%)

Length in estimations Primary (%) Secondary (%) 73 82 66 86 88 Track (%) 3 2 3 6 7

326 111 39 25 20

30 18 32 24 24

24 16 31 8 5

For every journey the mean speed for the whole trip is also calculated. This gives a possibility to lter out unrealistic given traveling time values. Two values can be set, maximum allowed mean speed and minimum allowed mean speed. Only journeys with a mean speed within these values are used in the optimization. Accessibility modeling The accessibility estimations were carried out in ArcGIS 9.2 (http:// www.esri.com). All analyses were based on raster calculations, meaning that data used in the accessibility calculations were represented by raster layers. These input rasters were combined to one main cost raster. Every cell in the cost raster has a value that represents its traveling speed. The least cost path is the path from a certain departure to a certain destination that induces the least cumulative traveling time. The least cost path tries to avoid the cells with the lowest traveling speed. However, it might cross a cell with low traveling speed if the passage over cells with higher traveling time brings an extra distance, or an extra number of cells, that induces a larger cumulative traveling time. The nal least cost path is the path with the lowest cumulative traveling time with mind on both the cost of the cells that are passed and the number of cells that are passed. The least cumulative traveling time according to the least cost path to the destination is assigned to every cell, giving a raster of traveling times to the destination along the paths that induces the least cumulative traveling time. This means that in the case of a homogeneous cost surface the cost path equals a straight line (Euclidean distance). Cost surface The cost surface constitutes of roads in 3 classes, slope in 4 classes and water bodies as can be seen in Fig. 5. The cost, or frictions, of the classes are dependent on which means of transport that is modeled.

For every cell in the raster, the value of the lowest friction of the ve different input raster layers was calculated. The layers were primary roads, secondary roads, tracks, slope and water bodies. The three road layers only hold information of the roads. Water bodies were excluded since the model does not take travel over water in to consideration, inducing that water bodies were impassable. The road cells of the cost surface were assigned with values corresponding to traveling speed originating from the speed estimation program. The off-road traveling speeds were set to represent transportation along winding footpaths which are present almost everywhere to a walking speed relating to topography/ steepness. Footpaths were assumed to be present everywhere except in forests. This simplication has its origin in the location of forests in the study area. Forests are mostly located on hills and mountains, were the slope or the soils are unsuitable for other use of the land. The second reason for not incorporating terrain in the cost surface was an insufcient basis for the decision of frictions for the different terrain classes. The off-road traveling speeds were dened according to eld estimations and they are in line with other studies (Verburg et al., 2004). The traveling speeds of the cost surface were not judged to decrease in the national parks Bundala and Yala, even though the access to the parks is partly restricted. There are villages in the national parks, and roads connecting other areas pass through the national parks. Those roads are not restricted. Accessibility and indicators of poverty As described in the introduction, previous studies have found different measures of accessibility closely related to a number of socio-economic indicators. However, the absence of high resolution data has called for the development of a method to generalize the accessibility measures to be able to compare it with the available statistical data.

Fig. 6. Estimated accessibility expressed as hours to selected markets traveling with landmaster.

A. Ahlstrm et al. / Habitat International 35 (2011) 316e326

323

Fig. 7. Estimated accessibility expressed as hours to selected towns traveling with bus.

The results of the accessibility modeling described above have a cell size of 90 m. The highest level of data on socio-economic indicators available is division level. The divisions of Hambantota District have a mean size of 219 km2. Socio-economic data was acquired from the department of census and statistics in Colombo. The acquired report holds a number of indicators and statistics for all 12 divisions of Hambantota District (DCS, 2006b). Two indicators and 3 estimated accessibility results were chosen for the comparison between the statistical data and the estimated accessibility. The statistical data chosen were the percentage of households without electrical lights and the percentage of households below poverty line. The poverty line is a measure of income, and all households having an income that is below the poverty line falls in the category of poor. The rst two chosen estimated accessibility measures consisted of estimated accessibility to market with landmaster and estimated accessibility to town with bus. These were considered to be the 2 most important accessibility measures according to the interview survey. Bus is according to the interview survey the main mean of transport overall and when traveling to town, and landmaster is the most common private mean of transport when traveling to a market to sell goods. The third accessibility measure was Euclidian distance to towns. To be able to compare the accessibility models results with the data of socio-economic indicators a method to generalize the accessibility data was developed. The aim of the generalization was to get a value for each accessibility measure at divisional level. To use a simple mean value within each division does not account for

population patterns since a mean value weights all cells in the accessibility rasters equally. Also a mean value does not compensate for extreme values, e.g. that people does not live on mountains. A high mountain in an otherwise relatively accessible division would greatly affect the mean value. The mean value would thus not represent the mean accessibility of the inhabitants of the region. Instead, the accessibility was rst generalized to GN-divisional level by using the minimum accessibility value per GNdivision and then weighted by the relative number of inhabitants in each GN-division (Fig. 1). The method developed gives a generalized accessibility value depending on an approximation of where people live. The population data on GN-level was attained from the district secretary in Hambantota town. Results The results of the calculations of best t speeds and information about number of journeys used, total number journeys, the maximum mean speed allowed, the minimum mean speed allowed and percentage error of the calculations can be seen in Table 4. The result of the accessibility modeling was a series of raster layers. The ones for landmaster and bus to selected markets and towns are presented in Figs. 6 and 7 below. The patterns of high accessibility along roads are clear, especially for bus travel (Fig. 7) where the speeds are higher than for LM (Fig. 6). In Fig. 9 a scatter plot visualizes the differences between accessibility and Euclidian distance (see Fig. 8). It is obvious that the accessibility estimations

Fig. 8. Euclidian, or straight line, distance from selected markets.

324

A. Ahlstrm et al. / Habitat International 35 (2011) 316e326

Fig. 9. Scatter plot of 1465 randomly sampled values of Euclidian distance and accessibility to towns when traveling with bus.

are not linearly correlated to distance, since the accessibility is dependant on presence/absence of roads as well as road quality while distance is not. It can be noted that accessibility decreases rapidly at longer distances, related to absence of roads or decreased road quality in the outskirts of the district. The generalization of bus travel time to towns from 90 m raster to GN-divisional level is illustrated in Fig. 10 below. Also here we clearly see the difference compared to distance. Fig. 11 illustrates the population-weighted generalization of the GN-division accessibility values of bus travel time to town on divisional level. In Table 5 the results of the correlation analyses are illustrated. The relationship between estimated accessibility, Euclidian distance and poverty indicators show the same pattern of relationships for the two selected poverty indicators. Correlation coefcients indicate an increase in poverty with high accessibility (shorter traveling times) to markets and the opposite to towns. Euclidian distance from towns shows a smaller degree of explanation in both instances as well as an inversed sign of the correlation coefcient compared to estimated accessibility. Discussion Speed estimation The speed estimation program incorporates all delays and traveling problems in the nal estimated traveling speeds. The respondents have given the traveling time from door to door.

The results of the program are by comparison to eld estimations realistic. The estimation of speeds for bus is the calculation with most journeys. Buses travel fast on primary roads, sometimes faster than the mean speed of the rest of the trafc. On secondary roads however, buses travel slower. Interviews also indicate that buses on secondary roads are often unreliable and late, and that the service can be very sparse or absent. The low speed on tracks is justied by the fact that the respondents often have to walk this distance to get to the bus. In the wet season buses either travel slowly or not at all on tracks. The walking time to the bus, the time it takes to change busses at a second bus stop, how often buses run or that tracks can be impassable in the wet season are all part of the nal door-to-door speeds. The calculations for motorcycle, landmaster, 3-wheeler and bicycle all have fewer journeys than for bus to base the calculations on. But the results of the program are strengthened by the fact that slower vehicles not affected much by road quality, e.g. landmaster and bicycle, get a similar travel speed for all road classes. Bicycles have a speed around 11 km/h on all road classes which is very realistic compared to eld estimations. The mean absolute error of calculated traveling time compared to given traveling time might seem high with values around 30% for some modes of transport. The 30% mean absolute error implies that the given average journey takes 30% longer or shorter time than the modeled journey. It is a measure of the deviations from the mean, and the mean (not absolute) error is 0. There are two main reasons for these errors. The rst is errors in the answers from the respondents in the interview survey. Some answers might be very inaccurate. Secondly there are considerable variations within every road class. Some respondents have to walk long distances along secondary roads, while others only need to wait for 10 min outside their door to catch a bus on another secondary road. The program calculates the best t for all these variations within every class for the whole study area, resulting in big errors depending on data errors and variations within the classes. One weakness in the estimations is the unbalanced relative distance of the classes. The distance used for secondary roads is the longest, and the distance used for primary roads is the second longest (see Table 4). The distance used for tracks is relative secondary and primary roads very short. Consequently the class with the longest relative distance used in the calculations gets the heaviest weight, affecting the results the most. This problem could be solved by using villages served with an even distribution of roads of the three classes.

Fig. 10. Generalized estimated accessibility to towns as traveling time when using bus for GN-divisions.

A. Ahlstrm et al. / Habitat International 35 (2011) 316e326

325

Fig. 11. Population-weighted GN-level estimated accessibility to towns as traveling time when using bus for divisions.

Accessibility modeling The choice of destinations is a critical part of the study. Even though the model can estimate the accessibility to different locations with high accuracy, the results are useless if the destinations are not relevant to the data the results are to be compared to. Therefore in a study of accessibility, a lot of effort should be put on selecting the important destinations. In this paper, all destinations are given the same importance, but in reality this is not the case. The choices of which markets and which towns to use as destinations in the modeling are based on interviews and on eld experience. All markets with wholesale trading were used, and none that does not have regular wholesale trading were used. For towns a certain size or economic activity was wanted. Hambantota town is an example of a medium sized town which in a rst look would be given as a destination as the districts administrative center. However, Hambantota town is not used as a destination due to low economic activity. Generalization of accessibility Due to socio-economic indicator data of low spatial resolution the continuous accessibility data needed to be generalized before comparison. Although being weighted with population, the results indicating that accessibility to markets increases poverty does not tell about the variation inside each division. The results should rather be interpreted on a division level. Divisions having only a market are poorer than divisions having a town. The intra divisional variation is hidden due to the coarser resolution. It is
Table 5 Results of the Pearsons linear correlation analyses of estimated accessibility and poverty indicators. A positive r value between estimated accessibility to a set of destinations (towns or markets) and the poverty indicators indicates that the % of population below poverty line or % without electric lights is decreasing with higher accessibility (shorter estimated traveling time). A negative sign as in accessibility to markets indicates that the % of population below poverty line and the % without electric lights is increasing with higher accessibility. Poverty indicator % of Population below poverty line % of Population without electric lights Estimated accessibility Estimated accessibility to town Estimated accessibility to market Euclidian distance from town Estimated accessibility to town Estimated accessibility to market Euclidian distance from town r 0.59* 0.71* 0.18 0.83** 0.63* 0.19

important to note that some of the towns and the wholesale markets used as destination are the same, they have the same localization. This means that the result of the relationship of the poverty indicators cannot be totally contradictory. To still see a difference in the relationship strengthens the results credibility. However, the results are uncertain due to the generalization of the accessibility data as well as standardized socio-economic data. Conclusions and nal discussion The study has found that it is possible to join the local knowledge of the transportation situation with geographical data to construct an accessibility model. The construction and presentation of the accessibility model is the main contribution of this article. However, when testing the model with available poverty data, it has also been found that estimated accessibility seems to be related to poverty, and that the relationships from using our model are stronger than the relationships presented in two recent studies on the same topic (World Bank, 2005, 2007). Hence, it is likely that, if more detailed poverty data were available, the model would greatly facilitate a deeper analysis of both direct and indirect effects of accessibility for different groups and areas, as has recently been called for in the literature (e.g. Olsson, 2009). With more disaggregated poverty data, the model could also help us in the search for explanations to spatial patterns, and to suggest whether the benets accrue mainly to mobility enhancement or accessibility enhancement (Bryceson et al., 2008). Another interesting nding, although uncertain, is that households with high estimated accessibility to markets seem to be poorer than households with high accessibility to towns. This distinction between markets and towns could also have important policy implications. To start with, the nding partly contradicts the results from the World Bank (2007), where there is no distinction made between the importance of markets and towns. In addition to this, key policy makers in the district suggest that it is market accessibility that is the most important form of accessibility in the area. One possible explanation for this difference e which is open for testing e is that high accessibility to towns is more helpful for households in their endeavors to diversify away from agriculture, for example by commuting to urban areas for employment or selfemployment. In relation to this, a report based on case studies in Sri Lanka, Indonesia and the Philippines (Hettige, 2006), shows a connection between rural road improvements and the propensity to shift away from agriculture for all socio-economic groups. Another possible explanation could be that high accessibility to

*Signicant level of p < 0.05, **Signicant level of p < 0.001.

326

A. Ahlstrm et al. / Habitat International 35 (2011) 316e326 Kam, S.-P., Hossain, M., Bose, M. L., & Villano, L. S. (2005). Spatial patterns of rural poverty and their relationship with welfare-inuencing factors in Bangladesh. Food Policy, 30, 551e567. Laurance, W. F., Albernaz, A. K. M., Schroth, G., Fearnside, P. M., Bergen, S., Venticinque, E. M., et al. (2002). Predictors of deforestation in the Brazilian Amazon. Journal of Biogeography, 29, 737e748. Lindberg, J. (2010). The changing accessibility of educational opportunities in southern Sri Lanka. Development and Change, 41, 131e151. Nagendra, H., Southworth, J., & Tucker, C. (2003). Accessibility as a determinant of landscape transformation in western Honduras: linking pattern and process. Landscape Ecology, 18, 141e158. Nanayakkara, A. G. W. (2006). Poverty in Sri Lanka e Issues and options. Department of Census and Statistics. Olsson, J. (2006). Responses to change in accessibility: Socio-economic impacts of road investment: The distributive outcomes in two rural peripheral Philippine municipalities. Department of Human and Economic Geography, School of Business, Economics and Law, Gteborg University. Olsson, J. (2009). Improved road accessibility and indirect development effects: evidence from rural Philippines. Journal of Transport Geography, 17, 476e483. Peiris, G. H. (2006). Sri Lanka: Challenges of the new millennium. Kandy Books. Sachs, J. D. (2005). The end of poverty: How we can make it happen in our lifetime. Penguin. Sandell, J. (2008). Detecting land cover changes in Hambantota District, Sri Lanka, using remote sensing and GIS. Masters thesis, Lund University. USGS. (2007). U.S. Geological Survey. Seamless Shuttle Radar Topography Mission. http://seamless.usgs.gov/ Collected 26.07.07. Verburg, P. H., Overmars, K. P., & Witte, N. (2004). Accessibility and land-use patterns at the forest fringe in the northeastern part of the Philippines. The Geographical Journal, 170, 238e255. World Bank. (2005). A poverty map for Sri Lanka e ndings and lessons. World Bank. (2007). Sri Lanka poverty assessment: engendering growth with equity: opportunities and challenges. World Bank. (2008). World development report. 2009, reshaping economic geography.

towns leads to accessibility enhancement through increased attractiveness of these places to government servants. A main problem in rural schools in Sri Lanka is for example the lack of teachers, which to a large extent is caused by the unwillingness of teachers to live and work in remote areas (Lindberg, 2010) References
Bryceson, D. F., Bradbury, A., & Bradbury, T. (2008). Roads to poverty reduction? Exploring rural roads impact on mobility in Africa and Asia. Development Policy Review, 26, 459e482. Burrough, P. A., & McDonnell, R. A. (1998). Principles of geographical information systems. Oxford University Press. Castella, J.-C., Manh, P. H., Kam, S. P., Villano, L., & Tronche, N. R. (2005). Analysis of village accessibility and its impact on land use dynamics in a mountainous province of northern Vietnam. Applied Geography, 25, 308e326. DCS. (2006a). Brief analysis of population and housing characteristics. http://www. statistics.gov.lk/PopHouSat/PDF/p7%20population%20and%20Housing%20Text11-12-06.pdf Collected 22.11.07. DCS. (2006b). Socio economic indicators for 119 selected divisional secretarys divisions. Department of Census and Statistics. Etter, A., McAlpine, C., Wilson, K., Phinn, S., & Possingham, H. (2006). Regional patterns of agricultural land use and deforestation in Colombia. Agriculture, Ecosystems & Environment, 114, 369e386. Guagliardo, M. (2004). Spatial accessibility of primary care: concepts, methods and challenges. International Journal of Health Geographics, 3(3). Hanson, S. (1986). Dimensions of the urban transportation problem. Geographic Perspectives on Urban Transportation, 3e23. Hettige, H. (2006). When do rural roads benet the poor and how? An in-depth analysis based on case studies. ADB. Jalan, J., & Ravallion, M. (2002). Geographic poverty traps? A micro model of consumption growth in rural China. Journal of Applied Econometrics, 17, 329e346.