AN ASSESSM ENT OF ROAD NETWORK OPTIMALITY IN TIGRAY, ETHIOPIA

(1)

AN ASSESSMENT OF ROAD

NETWORK OPTIMALITY IN

TIGRAY, ETHIOPIA

Geographic remoteness and poverty connected

INTERDISCIPLINARY PROJECT

By Axel Hirschel (10656146), Wai Kee Man (10580514) & Claudia Schwennen (10655808)

Date: 8 May 2016 Words: 6585

(2)

Abstract

In recent years major public investments have been made in infrastructure development in Ethiopia. The Ethiopian Roads Authority (ERA) has expressed its goal for this road development, which is mainly poverty alleviation. This study is focused on evaluating the optimality of the current road network regarding this goal by taking geographic remoteness and poverty per woreda, which is a part of a province, into consideration. There are just a few towns in Tigray that are not geographically remote. Poverty is also not evenly distributed throughout the province. The poverty rate is generally higher in areas with higher population density. With the use of Geographic Information System (GIS) data files and a self-‐made algorithm, it is concluded that the current network is not optimal yet.

(3)

1. Introduction

Ethiopia is one of the least developed and poorest countries in the world. It has gone through various types of regimes and disasters such as drought, famine and civil war. As it is an agrarian economy, 81% of the population in 2014 lives in rural areas (World Bank, n.d.) where the poverty rate is exceptionally high. Therefore, strategies focused on developing agricultural growth are essential in reducing the country’s poverty (Diao & Pratt 2007; Easterly, 2002).

Road development is one of the status quo strategies to tackle these problems. According to World Bank (2009), they are the fundament of a country’s infrastructure, support growth in agriculture and industry and they provide access to internal markets and social infrastructure. Various studies support this theory as they have showed that better road quality and more access to rural areas have a positive impact on poverty reduction (Dercon, Gilligan, Hoddinott & Woldehanna, 2009; Khandker, Bakht & Koolwal, 2009).

Ethiopia’s government tackles problems with road development, because most research suggests that roads have a significant influence on poverty reduction. Still, parts of Ethiopia are unconnected and/or reliant on economic policy of the government. As a lot of money is invested in Ethiopia’s road development plans (World Bank, 2009) and it holds great potential to reduce poverty, it is essential that these roads are allocated optimally within Ethiopia. This means that the plans should have an impact on what it is meant to be for, in this case poverty reduction. Especially impoverished regions should have some priority, while also keeping in mind the investments that need to be put into the project. A factor that greatly determines the investments is the geographic remoteness of regions. This could significantly increase the investments, which might not be feasible (Bird et al., 2010). If the network is not optimal, it means millions of dollars are gone to waste and an unnecessary amount of people still suffer in poverty. This research will assess both this factor and the poverty rate in order to conduct an assessment of the optimality of the current road network in Ethiopia. Moreover, only asphalt road will be considered in this research, due to the available resources of this research and because these form a stronger factor in the accessibility to large urban centres (Hearn, 2011).

Optimality within infrastructure design is often defined as network reliability. The general idea within the multiple forms of network reliability is that the network can perform its proposed service level adequately for the period of time intended under the operating conditions encountered (Billington & Allan, 1992). This implies a pragmatic approach in which the results of a network are most important and should fulfill the wishes of the users. Within most research, the proposed service level is a directly measurable unit. However, the intended service of a network is subjective and is therefore reliant on interpretations of its function.

Within the context of Ethiopian road development, the government had specified the services that roads were to perform. The two large-‐scale road sector development programs (RSDP) during the period between 1997 and 2009 had selection criteria of the Ethiopian Roads Authority (ERA), shown in table 1. These criteria and their weighting indicate the

(5)

goals of RSDPs: economic development and the provision of access to remote regions to alleviate poverty (Shiferaw, Soederbom, Siba & Alemu, 2012).

Table 1 ERA’s weight for Road Placements (Shiferaw, Soederbom, Siba & Alemu, 2012)

So within this paper the optimality of Ethiopian province Tigray’s current road network is used as a case study. It should be noted that this assessment will prove that poverty can be alleviated. The current placement of roads is analyzed on basis of poverty and remoteness. Tigray was suitable since a lot of data was available for this province, and the province contains a variety of remote locations. It is therefore possible to see which areas have been prioritized, and evaluate whether this was optimal.

The research question is: “To what extent can the road network in Tigray, Ethiopia, be optimized to lessen poverty based on poverty rates and geographic remoteness?” The sub questions are listed below.

● How is the degree of geographic remoteness distributed in Tigray? ● How is the degree of poverty distributed in Tigray?

● How optimal is the current road network in Tigray?

Since Ethiopia has limited budgets for road network development, not every town can be connected. This means that certain towns have been prioritized within the government. Using a self-‐made algorithm and Geographic Information System (GIS) data, it is possible to assess this prioritization on basis of the services roads should provide according to the ERA. Firstly, the theoretical framework of the research is explained further in this report. The integration of the various disciplines is also mentioned within this chapter. Secondly, the methodological approach is explained, and the choices within the algorithm are clarified. Thirdly, the results of the research is presented. Fourthly, the results and the limitations of the research are discussed. Lastly, a few clear recommendations for network improvement and the overall consequences of the research are given.

(6)

2. Theoretical framework

2.1 Impact of roads on poverty

Numerous researches have been conducted, but overall several factors can be identified of the impact of roads on poverty in the scientific literature. First, roads facilitate the provision of basic needs, such as health and education. Second, roads give access to markets and with a greater input, prices will be reduced due to lower transport costs. High transport costs can be partially explained by geographic disadvantages, such as being landlocked and the remoteness of a region. Local economies do not have access to and from global market centers. Thirdly, transport infrastructure is able to reduce poverty by creating employment and new job opportunities (Calderón & Servén, 2008; Beuran, Gachassin & Raballand, 2015; Porter, 2002).

Dercon et al. (2009) used longitudinal household data for their quantitative research and showed that public investment in road quality and increased access to agricultural extension services led to faster consumption growth rates and lower poverty rates. Roads indirectly provide benefits for the economy by supplying access to opportunities.

Ali and Pernia (2003) compiled several results from studies that point to a significant impact of roads on poverty reduction through economic growth. They conclude that roads cause this impact through agricultural productivity, nonfarm employment, and an increase in consumption expenditure and time savings. The two authors have summarized the links from infrastructure investments to eventually poverty reduction in figure 1. For example, when investments are made into the building of roads, employees are needed and new job opportunities that are non-‐agricultural will arise. This influences the poor’s wages and, hence, stimulates economic growth. Higher productivity and expanded employment then affect the supply and prices of goods and thus, the poor’s well-‐being (Ali & Pernia, 2003).

(7)

Figure 1. Simple analytical framework with links between Infrastructure and Poverty Reduction (Ali & Pernia, 2003)

Poverty is a complex concept that can be defined and measured in many ways. One of the reasons why this concept is complicated is because it is socially constructed, thus a range of variables are correlated (Bevan & Joireman, 1997). When looking at what to measure, measures can be objective or subjective. Bevan and Joireman (1997) define ‘objective’ as a measure that is relatively simple to administer and one that the observer has decided on to measure. Subjective measures are thus locally defined or by the person who is being researched and who usually combines a number of indicator that are locally regarded as important. For instance, a subjective measure of poverty would be how poor the local observer ranks himself. For this research, we aim to find data that is suitable within the scope of this research. Objective measurements also tend to be quantitative which complements our analysis as these data can be easily converted into the programs that will be used.

2.2 Network optimality

Within the rural provinces in Ethiopia the poverty rate is exceptionally high. Since road development is a major strategy to combat poverty within Ethiopia, it is necessary to assess whether the limited means are spent optimally. However, optimality needs to be defined in order to assess it.

(8)

As stated in the introduction, optimality is often defined as reliability. The reliability of a network is the probability of the network performing its proposed service level adequately for the period of time intended under the operating conditions encountered (Billington & Allan, 1992). Wakabayashi and Lida (1992) propose a similar definition for network reliability. Since the road infrastructure of Tigray is a network, it is possible to use this definition. Wakabayashi and Iida (1992) have linked the definition of reliability to road networks; road network reliability is the probability that two nodes are connected for traffic.

To assure the positive impacts of roads on poverty alleviation, the roads need to be accessible for the population within the woredas, which are the second administrative units above the smallest unit of local government in rural communities in Ethiopia. The effects are namely bound to the location of the road. Since these positive effects of roads are vital for economic development for impoverished regions and since the ERA has placed these roads with the intention to combat poverty, the required service of the road is to tackle poverty. Chen, Yang, Lo and Tangs (2002) explicate three categories in reliability namely connectivity reliability, capacity reliability, and travel time reliability. Connectivity reliability is concerned with the nodes being connected. A path between two nodes always needs to be present (Wakabayashi & Iida, 1989). The network can work in only two states, either there is a connection between the nodes, or there is none. The binary state approach can be suitable for extreme situations, such as natural disasters (Chen et al., 2002).

Travel time reliability is applied when evaluating whether all nodes can be reached within certain time intervals (Asakura & Kashiwadani, 1991). The travel time between nodes is influenced by differences in traffic flows, such as the occurrence of daily traffic jams. Asakura (1999) later included that degradation of roads can also influence the travel time. He defined a ratio in which the more degraded a road is, the less reliable it is. This is in line with connectivity reliability, as severely degraded and therefore inaccessible roads are under both definitions unreliable.

Capacity reliability is concerned whether the capacity of a road is great enough for a specific demand at all times (Chen et al., 2002). The maximum road capacity therefore always need to be equal or greater to the specific demand at that time. In capacity reliability the everyday disturbances are also accounted for.

Most network reliability research had been done to developed infrastructure networks in Western countries. These papers focused on optimizing transportation flows and preventing traffic jams using both the definitions of travel time reliability and capacity reliability. These interpretations of reliability are unsuited for this research though. First of all, while traffic jams may exist within Ethiopia, the RSDPs were not intended to combat these. Moreover, most variables used within both fields are inaccessible.

However, the concept of reliability can still be used within the current road network in Ethiopia. Within connectivity reliability, one of the major challenges is to connect as much nodes as possible. The ERA has also specifically aimed to connect yet unconnected regions and therefore one road is sufficient. The critique on the binary state of connectivity is also less impactful in this case study; although a town can benefit from extra links when already

(9)

connected, the effects of one connection already provide the benefits mentioned within the previous paragraph.

In conclusion, the optimality of an infrastructure networks can be evaluated on basis of its reliability of its connectivity. Tigray’s network should combat poverty and therefore it needs to connect towns to the current network in order for the roads to have a positive impact. The network is therefore optimal if it connects as many impoverished people as possible.

2.3 Geographic remoteness

The concept geographic remoteness is very much related to the connectivity between certain communities and large towns, as it describes how large the access is to large towns. The Australian Institute of Health and Welfare (AIHW, n.d.) also describes this concept as “the level of access to certain services and goods”. It is an important concept that needs to be considered, as it also describes what the socioeconomic impacts could have on unconnected communities, as suggested by the AIHW. This is for example supported by Sunderlin et al. (2008), who specifically found a correlation between poverty rate and forest cover, although the poverty density is quite low in densely vegetated areas. Epprecht et al. (2011) conducted a study in Vietnam in order to observe whether poverty is related to physical accessibility or to the fact that some groups are ethnic minorities. They found that the physical accessibility is a strong predictor of poverty. However, they also found that the accessibility to small urban centres are of greater importance than large urban centres. Bird et al. (2010) also found that higher poverty rates exist in remote areas and that poor infrastructure is especially a problem, as this increases the travel costs as well. As it might be difficult to put large investments into road development in remote areas and especially in mountainous areas, it might be interesting to determine whether the location of political and administrative centres, so that small urban centres can develop in these areas (ibid.). The geographic remoteness greatly depends on the physical distance between certain communities and larger towns (AIHW, n.d.). Physical features in the landscape are crucial for the determination of the geographic remoteness of certain locations, as the physical distance is greatly dependent on these physical features (Pugh & Cheers, 2010). The physical features that will be focused on in this research are mountainous areas, rivers and forests. The physical distance is also significantly related to socioeconomic developments. An example of this is the privatization of public services, and in combination with large distances, this will increase the transport costs for the citizens of the communities of interest, which indirectly increases the physical distance and also the distance to markets (Huskey, 2005; Pugh & Cheers, 2010). This demonstrates how the geographic remoteness could have an impact on socioeconomic developments and therefore also on poverty reduction.

2.4 Integration of the different concepts and theories

In order to find common ground between the different concepts and theories from the three disciplines, one concept was used as an overarching concept, which is optimality. The concepts and theories have been reorganized into one system where optimality is the overarching concept. This type of integration was introduced by Repko (2012), which is called the technique of (re)organization. Essentially, commonalities between different

(10)

concepts and theories are determined, after which they are linked together and (re)organized into one system.

As described earlier, optimality is defined as that something can perform its proposed service level adequately for the period of time intended under the operating conditions encountered. For this study, this can be described through the geographic remoteness and poverty reduction of the communities of interest. As described in the previous section, the geographic remoteness gives an indication of the connectivity between the communities and towns, which in turn also describes the access to certain goods and services of the inhabitants. It affects the travel time, the travel distance and the travel costs, but it also determines the road construction costs when road development plans are considered. This in turn could determine whether investments should be put into certain projects and, because these might not be feasible in geographically remote areas. In cases like this, it might be difficult to combat poverty and the proposed service level of road development, namely poverty reduction, might not be possible to be reached. In short, it indirectly affects the possibilities for poverty reduction at certain locations. However, in non-‐geographically remote areas, road development could have a great impact on poverty reduction as explained before. Roads cause this impact through agricultural productivity, nonfarm employment, and an increase in consumption expenditure and time savings. These will all increase the economic welfare of the poor (Ali & Pernia, 2003).

The geographic remoteness will be assessed by mapping the physical features in the landscape. As reducing poverty is the ultimate goal of road development, poverty will be assessed by mapping poverty rates. This is of importance, because the placement of asphalt roads will also be mapped in order to do the optimization analysis and see whether there are any general connections that can be made.

Figure 2. Integrated concepts and theories

(11)

3. Methodology

As stated in the introduction, for the assessment of the geographic remoteness and poverty rates in Tigray, ArcGIS will be used to map these variables. Four different situations can be defined, which will determine whether a road should be placed in the optimal network. Figure 3 gives a quick overview of these four situations along an x-‐axis and y-‐axis together with the two variables. The first situation is when the poverty rate is high, but the degree of geographic remoteness is low. As stated before, impoverished regions should be connected to the current network in order to possibly alleviate poverty. As the degree of geographic remoteness is low, a road will be placed in this situation for the optimal network. The second situation is when both the poverty rate and the degree of geographic remoteness are high. In this case, it might not be feasible to invest into road development, especially since the region is very poor, which might not bring about the expected economic returns. As explained earlier, alternatives could be considered in future research. The third situation is when both the poverty rate and the degree of geographic remoteness are low. For this situation a road can be placed, as it might be likely that road development in regions like these will bring about the economic returns. Lastly, the fourth situation is when the poverty rate is low, but the degree of geographic remoteness is high. Again, in this case it might not be feasible to invest into road development.

Figure 3. Different situations regarding the economic potential and poverty

3.1 Design of the optimality algorithm

Within the research it is most important that the produced roads are optimal. As described within the theoretical framework, the network is optimal if it connects as many poor people as possible. It is essential to base the optimality of the network on the currently existing roads, since Ethiopia cannot spend more money on roads than it did. So firstly, all current connections between 2 towns have been entered in a database. The length of each road has

(12)

been calculated by the Euclidean distance between the coordinates of different towns. The real length of a road will be larger than the calculated length. However, since both existing and suggested links’ lengths will be based on this distance, and many of these links will be calculated, it can be assumed that the deviation of the Euclidean distance is not problematic. The total length of all roads combined is the maximum length of the network.

The optimality of a network can be calculated by designing a suited algorithm for the problem. For the optimality implementation, an algorithm in Python 3 was written, which is included in appendix C. The basis for the algorithm was the database with the information of certain woredas and towns. The information in the database includes the coordinates, names, poverty-‐indicators and remoteness of all the used towns. Since not all specific data on every town is present, we chose to use the information of the town’s woreda. This is a fair assumption, since woredas are pretty small and because asphalt roads to the specific towns are also important for the surrounding region. How the information of this database is gathered will be explained in section 3.3 and 3.4. It should be noted though that this means there is not actual information about the regions in between the towns and that this is not included in the algorithm, but rather the information about the landscape is included in the score of the town.

The central idea of optimal road placement is connecting large groups of people within Tigray, and prioritizing impoverished regions. The score of every town is based on its poverty and inhabitants and portrayed as a number, as well as for the degree of geographic remoteness. Towns that are geographically remote are not considered further. The number is calculated by multiplying the poverty rate of the town by the inhabitants of it. This way we get the amount of poor people within the town. The network optimality is then the sum of scores of every connected town, so the total of poor people connected. Within the calculation multiple roads to one town do not increase the score, but it might be beneficial when from that specific point roads can be created to unconnected towns nearby.

The optimal networks can be created using a recursive function to create possible links between nodes. The input for this function is a list of nodes, a list of roads and the maximum length of the total road. Every node within a possible optimal network is evaluated whether they are connected. If the node is unconnected, it will go on to the next node. If the node is connected, it will check all nodes again whether they are connected and if a node is not connected, it will make a connection between the connected node and the unconnected node. For every connection the distance between both nodes is subtracted from the maximum road. Every time a connection has been made, it will enter the updated list of nodes and whether these are connected, the updated list of roads and the new available road length in this function. If every node is connected or if the available road is fully used, that network will be given as output of the function. It will now continue with the for-‐loops. This way every possible optimal network can be created.

A possibility for the creation of the most optimal network is to calculate the optimality-‐score of every possible network with the length of the current network and then choose the one with the highest score. This is not possible though, since the information that needs to be remembered to both create and compare the various networks will exceed the amount of

(13)

atoms on earth. It is therefore necessary to eliminate bad suggestions beforehand to adhere to reasonable computer power and memory.

The algorithm will step-‐by-‐step calculate possible networks. Where normally the recursive function will call upon itself until every network is fully created, this algorithm will stop after connecting 3 roads. All networks are now within a long list. The optimality of all the created 3-‐road-‐networks and the length of these networks are compared on the basis of score divided by unit road placed. The list is sorted on this criteria and the first network in the list is now the most optimal network. From the most optimal network, the first of the three roads is saved. Now the program again calls upon the algorithm, but with the input including the roads previously formed. The program continues creating new roads until all of the available road is placed, by constantly creating 3-‐road-‐networks and evaluating these networks. Besides the maximum depth, the maximum amount of roads placed can also be entered as input. With this function possible extensions to the current network can be created.

3.3 Poverty assessment

As stated earlier, the poorest people in Ethiopia live in rural areas. For this reason we will focus on a woreda level in our research. A woreda is the second administrative unit above the smallest unit of local government in rural communities in Ethiopia. Furthermore, data on woreda-‐level is broadly more available. Consequently, when data on woreda level is being used, data about towns will be very contrasting and hard to visualize in a map. For these reasons and the limited time we have for this research, urban poverty will not be taken into our research and only rural poverty by woreda will be mapped, unless stated otherwise. However, towns will be displayed on the maps, this allows us to see if there may be any remarkable connections between the rural poverty rate of woredas and placement of roads and towns.

As economic data was not available, we opted for the food poverty line as our measure of poverty. Households which are below the poverty line are households who could not consume enough to get the minimum calorie requirement of 2200 Kcal per day (Nega et al., 2011). The head count ratio of rural poverty per woreda will eventually be mapped in ArcGIS. The food security will also be visualized in ArcGIS in order to see if there is a connection between food security and poverty. If this is present, it suggests that the theory that agriculture has a big influence on poverty is positive.

Firstly, the current asphalt road network needs to be visualized. Next, the population density will be mapped in order to see which areas are most rural and urban. Hereafter, a map with the poverty rate per woreda will be made. Once again, these maps will be made using ArcGIS. This data and information about which towns are present in which woreda will be clearly ordered as it is needed for the poverty assessment and the making of the algorithm.

(14)

3.3 Geographic remoteness assessment

A dataset provided by Dr. Crelis Rammelt contains most of the data that was used in this research. Other missing data was found either through ArcGIS online or through literature. Appendix A contains the complete metadata of all the data that was used in ArcGIS.

For the geographic remoteness mountainous areas, rivers and forests were mapped. More divisions of physical features could be used, but the scale of this research is too large to do this, which is why the focus was on three physical features only. Firstly, a Digital Elevation Model (DEM) was used to construct the slopes of Tigray. The DEM is visualized in figure 4. Since no exact definition could be found for lowlands regarding the corresponding slopes, lowlands were estimated at slopes between 0° and 10°. Data that included the rivers and forests were found through ArcGIS online. Ideally, rivers could be constructed using satellite imagery and by drawing polygons or lines in ArcGIS. Also, forests could be mapped through ERDAS by using Remote Sensing (RS). However, due to limited time, the data from ArcGIS online was used in this research.

Figure 4. Digital Elevation Model (DEM) of Tigray

Larger weight was given to mountain ranges than to rivers. Hearn (2011) studied the construction and maintenance of roads in lowlands and mountainous areas and concluded that the construction costs in mountainous areas can be two to three times higher than the construction costs in lowlands. An example of a project that he studied was a project in Ethiopia where a road was built across the Blue Nile gorge in 2010. The costs of the construction on the plateau were approximately $300 000. In contrast, the construction costs in the gorge were approximately $900 000. In comparison, the United States Department of Agriculture (2011) estimated the construction costs of bridges between $2100 and $3350 per feet and on top of that $30 000 -‐ $40 000 for deep spread footings.

(15)

This is why larger weight was given to mountainous areas. In addition, Hearn (2011) also found that the lowest construction costs possible in mountainous areas are often part of unsustainable construction strategies. He stresses that there is a significant need in efficient planning of the costs and that there is a need of the right technical resources.

The geographic remoteness will be clarified through two scores, namely that a region is not or slightly geographically remote or that a region is geographically remote. The regions that are geographically remote will automatically not be connected to the current network. Furthermore, coordinates of specific areas will also be retrieved for the making of the algorithm. These scores and coordinates are put into a table, which are needed for the algorithm.

4. Results

Figure 5. The road network of Tigray in 2007

Figure 5 shows the road network of Tigray as it was in 2007.

(16)

4.1 Results of the poverty assessment

Figure 6 shows the map of Tigray with the different rural poverty rates per woreda. Tigray counts 32 woredas. The mean poverty level is 44%. The highest head count ratio with 75% is in Irob and Mereb Hele. The woreda with the lowest rural poverty is Kafta Humera with just 10% (Nega et al., 2011). It is remarkable that high and low poverty woredas are bordering each other but overall, it seems like poverty rates are higher in the Eastern part of Tigray and lower in the West. The exact poverty rate per woreda can be found in Appendix B with additional information.

Based off of figure 7, it can be assumed that the population density in Western Tigray is sparse: less than 50 inhabitants per square kilometer. Lower population density in an area with enough successful agricultural practices may explain why the West of this province has lower poverty headcount ratio.

Figure 6. Poverty percentage per woreda in Tigray, Irob and Mereb Hele selected in North-‐East and Kafta Humera selected on the West side

(17)

Figure 7. Population density per woreda in Tigray

4.2 Results of the geographic remoteness assessment

Figure 9 shows the results of the geographic remoteness assessment. It contains the slope of the hills and mountains, rivers and the woody biomass/forests. Furthermore, it visualizes the towns and woredas of interest and the asphalt roads as well. The towns and their woredas that are not or slightly geographically remote are selected with a bright blue color. The selected towns are Chercher in Raya Azebo in the south of Tigray, Haiqmeshal in Atsbi Wenberta in the East and Mai Gaba in Welkait in the West. Raya Azebo and Mai Gaba are the towns that are slightly remote, as bridge costs need to be taken into account. The other towns are geographically remote due to the mountain ranges, and construction costs will be substantially high in this case as explained earlier. Because the forests were mainly situated on the mountain ranges, they did not need to be considered further.

(18)

Figure 8. Map showing the geographic remoteness of towns in Tigray, through mountainous areas, rivers and forests

4.3 Results of the optimality algorithm

The current network has a total length of 1.121.423 units, and consists of 44 roads. It connects 42 of the 60 towns within Tigray. The total population connected to the current network is 2.008.277. It has connected 1.059.973 impoverished inhabitants. The optimal network created by the algorithm has a total length of 1.047.961, and consists of 37 roads. It connects 38 of the 60 towns in Tigray. The total population connected to the current network is 2.865.350. It has connected 1.530.680 poor inhabitants. Both networks are listed in Appendix C.

12 roads occur in both networks, which is relatively few. 25 towns have been connected in both networks. Within the optimal network, towns far in the west have not been connected, whereas they have been connected in current network. This is due to the relatively low poverty rates in western Tigray. The optimal network scores especially better in the connection of all places near other nodes, since it prefers these links over long connections. The current network connects 31% less impoverished people and 30% less people overall. A possible explanation for the differences between the two networks might be that some towns have been connected multiple times and that this does not reflect within the score. However, the optimal network uses 0.97 road per town and the current network 1.05 road per town. This does not explain the major difference in optimality though. Most probable is that the ERA prefers asphalt roads throughout Tigray and connecting adjacent villages with rural roads.

(19)

Current Network • 44 roads

• 42 out of 60 towns • 2.0 million people

• 1.1 million people beneath the poverty line

Optimal Network • 37 roads

• 38 out of 60 towns • 2.9 million people

• 1.5 million people beneath the poverty line

Overall • 12 roads exactly the same

• 25 towns connected in both networks • 31 % less people in current network

• 30 % less impoverished people in current network Table 2. Optimality assessment summary results

Since the current network cannot be revised, it is also interesting to see which connections can improve the current network the most. Firstly, Rama should be connected, since a lot of people live in Rama. This connection can best be established from Axum. Also, Semema can be linked rather easily, since it is nearby Shire Endalase. Lastly, Mahbre Degue has a relatively high amount of people as well, yet it is currently unconnected. This link can also be established from Axum.

5. Discussion

As stated in the introduction, the model presented here is still quite simplistic. Due to limited resources, the focus of this research was narrowed down to a few factors. Moreover, due to limited availability of data of Ethiopia, the focus needed to be narrowed down even further. Even though the model is quite simplistic, it can serve as a base or starting point for future research and it offers great possibilities for automatic road planning, as manual road planning can be quite intensive (Mohammadi Samani et al., 2010). This has also been recognized by others and this has resulted in the recent development of automatic road planning. Again, it should be noted that actual poverty reduction was not assessed in this research, but rather new roads are proposed that are part of the most optimal road network that resulted from this research.

Another similar research was conducted by Mohammadi Samani et al. (2010). They used GIS and AHP (Analytical Hierarchy Process) for the planning of forest roads and compared this to the traditional way of road planning. Although they did not look at socioeconomic factors, they took into account more geographical and ecological factors, such as hydrology, slope, geology, soil, tree volume, tree type. They classified areas of the research area into 5 classes of suitability for forest roads. They found that this way of planning showed better results than the traditional way of road planning. This also suggests that in future research a baseline is needed in order to compare the result and determine the performance of the model. A research that did include socioeconomic factors was conducted by Yang and Bell (2007). They presented new developments concerning the Network Design Problem (NDP), which means that the travel demand is growing at higher rates than that the network can

(20)

grow. At the same time, resources to expand this network are limited. They developed an economic based function for the optimization of the road network by including the elasticity of the travel demand, but they also included the preferential routes and capacity improvements. Both studies give some indication of how different factors can be included into optimization models and how the model in this research can be improved.

What also need to be taken into further consideration is the fact that only rural data and asphalt roads have been assessed in this research. Even though most people live in rural areas, the urban population also matters and may even be even taken into more consideration than rural poverty heads when road placement is being planned. Urban areas ultimately also produce and involve economic development. In our case study we focus on Ethiopia while the first two RSDPs were running from 1997 to 2009. In these two RSDPs trunk roads and link roads were constructed and upgraded. The RSDPs do not focus on unpaved rural roads while rural roads also have potential to lead to economic growth and poverty reduction (Escobal & Ponce, 2002). The inclusion of travel time reliability as definition for optimality could be useful for the assessment of the network including rural roads, since it allows differences between roads.

Another factor that could be considered is that not only the physical distance plays a role in poverty. As explained earlier, Epprecht et al. (2011) showed that the geographic remoteness is a strong predictor of poverty. However, they show that the socio-‐cultural distance is probably a stronger predictor. In this perspective, barriers include the language and the cultural differences. It could be interesting to focus on this aspect as well in future research. Moreover, as stated earlier, it might be difficult for governments to decide whether large investments should be put into the road development in remote areas (Bird et al., 2010). Epprecht et al. (2011) stated that small urban centres might be of greater importance than large urban centres in the poverty distribution in a country. Therefore, it might be interesting to look into the alternative to invest into the development of small urban centres, by locating small political and administrative centres into these areas. This should be done in future research.

The result presented in this research only represents one moment in time. However, it might be interesting to conduct a long term research of the past and see whether the optimality of the network was better or worse in the past, since the poverty distribution might have been different in the past.

Concerning the geographic remoteness assessment, there are some points that could be improved in future research. Ideally the rivers and forests are determined using recent satellite imagery when there are enough resources to do this. Furthermore, it is difficult to get a quick idea of at which point the road construction costs substantially increase for mountainous areas. Collaborations with road engineers is particularly important for this type of research to get a clearer idea of how the construction costs increase. Finally, depending on which scale is being used, it might be interesting to look at other subdivisions of physical features, which should be done at lower scales.

(21)

6. Conclusion

Since roads are capable of empowering and developing areas, it is needed to create connection to these towns. The socioeconomic benefits of roads, such as access to larger markets, education and job opportunities, need to be available for as many people as possible. Most importantly, these effects directly benefit the poor within the woredas that are dependent on help from the government.

One of the problems in connecting all regions across Tigray is their geographical remoteness. Mountain ranges and the forest across them make the design and engineering of roads to these towns that are geographically remote more expensive in construction. However, also some reasonably not-‐remote towns have not been connected. Furthermore, poverty is not evenly distributed through Tigray. The poverty rate is generally higher in areas with higher population density and of course, in areas with higher food stress. With the algorithm it can be concluded that the current network is not optimal, since with the total length of roads in the current network more people could have been linked to the current road network. This difference in optimality between the current and the optimal network can for some extent be explained by the inclusion of other data. However, the difference needs to be researched further as the suggested optimal network scores 30% better. For further road development, suggestions within this research need to be considered as they improve the optimality.

AN ASSESSM ENT OF ROAD NETWORK OPTIMALITY IN TIGRAY, ETHIOPIA