Temporal patterns of road network development in the Brazilian Amazon
- First Online:
- Cite this article as:
- Ahmed, S.E., Souza, C.M., Riberio, J. et al. Reg Environ Change (2013) 13: 927. doi:10.1007/s10113-012-0397-z
- 846 Views
The Brazilian Amazon is a globally important ecosystem that is undergoing rapid development and land-use change. Roads are a key spatial determinant of land-use conversion and strongly influence the rates and patterns of habitat loss and represent a key component of models that attempt to predict the spatio-temporal patterns of Amazonian land-use change and the consequences of such changes. However, the spatio-temporal patterns of road network development are poorly understood and seldom quantified. Here, we used manually digitised satellite imagery at multiple temporal and spatial scales across the Brazilian Amazon to quantify and model the rate at which road networks are proliferating. We found that the road network grew by almost 17,000 km per year between 2004 and 2007. There was large spatial variation in road network density, with some municipalities having road densities as high as 0.5 km/km2, and road network growth rates were highest in municipalities with an intermediate road network density. Simulations indicated that road network development within municipalities follows a logistic growth pattern through time, with most of the development occurring within a 39-year time period. This time period is similar to those of other boom and bust development dynamics observed in the Brazilian Amazon. Understanding the temporal patterns of road development will aid the development of better predictive land-use change models for the Amazon, given the key importance of roads as a predictor of deforestation in many existing models.
KeywordsDeforestationEconomic developmentLand-use changeLandsatLogging road
Global road networks have been expanding at a rapid rate since the 1900s (Forman et al. 2003). Roads are a distinctive feature in any landscape, with many countries giving 1–2 % of their land surface over to roads and roadsides (Forman 1998). However, the ecological effects of roads spread beyond the physical footprint of the network and may impact 15–20 % of the land or more (Forman and Alexander 1998). In the context of tropical deforestation, roads cause a relatively small amount of direct habitat loss, but exert a huge indirect influence on spatial patterns of deforestation by allowing easier access to new frontiers (Fearnside 2008; Geist and Lambin 2002; Perz et al. 2007, 2008). Roads also encourage extractive industries and further deforestation by settlers, thereby indirectly influencing deforestation rates. As roads are developed to access resources, roads may be seen as a cause of development and it is this causal relationship that led to the Brazilian development policies of the 1970s based around road construction (Alves 2002; Kirby et al. 2006). However, roads may also be a consequence of development, where, as economies grow, they require better infrastructure support and so road networks are developed.
The influence of roads on spatial patterns of deforestation ensures they also exert a strong influence of spatial patterns of biodiversity loss (Fearnside 2005; Finer et al. 2008; Forman 1998; Maki et al. 2001). Roads influence biodiversity directly through road kill events, but again the largest impact of roads is through indirect processes and ‘extended effects’. For example, roads can alter abiotic processes such as surface run-off (Forman and Alexander 1998) and microclimatic conditions including light levels, air and soil temperature, air and soil moisture, soil pH and nutrient levels (Delgado et al. 2007; Gehlhausen et al. 2000; Honu and Gibson 2006). Roads fragment forest habitats, creating new habitat edges and acting as barriers to the movement of animals (Arima et al. 2005; Dyer et al. 2002; Goosem 2007; Keller and Largiader 2003; McGregor et al. 2008; Richardson et al. 1997; Rico et al. 2007). These extended effects alter the abundance, distribution and behaviour of species over large areas (Bee and Swanson 2007; Blom et al. 2005; Eigenbrod et al. 2008; Potvin et al. 2005; Vos and Chardon 1998). Some species do appear to prosper with the presence of roads, benefitting from additional resources such as road-killed carrion (Laurian et al. 2008; Rydell 1992). However, negative effects are five times more prevalent than positive effects (Fahrig and Rytwinski 2009).
The Brazilian Amazon contains approximately one-third of the world’s remaining rainforest, covering an area of 4.1 million km2. The region is highly biodiverse with 10–20 per cent of the planet’s known species, is one of the three most bioculturally diverse areas in the world (Loh and Harmon 2005) and provides many valuable ecosystem services such as water regulation (Bradshaw et al. 2007; Fearnside 2005; Foley et al. 2007), carbon sequestration (Asner et al. 2004; Foley et al. 2007) and local and global climate regulation (Bonan 2008; Foley et al. 2007; Malhi et al. 2008). However, the Brazilian Amazon is also rapidly undergoing extensive development with widespread land-use conversion. Roads are a key spatial determinant of land-use conversion in this region, dictating the spatial pattern of deforestation by regulating access to standing forests which are logged for timber and later clear-cut to make way for agriculture or pasture, which secures land ownership via ‘productive use’ (Fearnside 2005; Kirby et al. 2006; Perz et al. 2008). This process of deforestation and agriculture following roads has been well documented (Fearnside 2008; Geist and Lambin 2002; Perz et al. 2007; Walker et al. 2004). Given that roads are a key spatial determinant of land-use conversion and that they have extensive impacts on rates and patterns of habitat loss, it is important that we know how many roads are being built, how fast road networks are developing and where they are developing in this globally important ecosystem.
Many studies have investigated temporal patterns of land-use change in the Amazon (e.g. Araujo et al. 2009; Carpentier et al. 2000; Dale et al. 1994; de Barros Ferraz et al. 2005; deKoning et al. 1999; Moreira et al. 2009; Muller et al. 2010; Soares-Filho et al. 2002; Walker et al. 2004; Walsh et al. 2008; Wassenaar et al. 2007), as well as the temporal patterns of change in the forces that drive land-use change such as economic and agricultural trends (Araujo et al. 2009; Ewers et al. 2008; Morton et al. 2006). However, knowledge of the temporal dynamics of road network development lags far behind, with just one study having examined this in the Brazilian Amazonian state of Para; Brandao and Souza (2006) mapped a total of 25,196 km of roads, of which 15,727 km were constructed in a 10-year time period (1991–2001). This stands in stark contrast to the acknowledged importance of temporal changes in road networks for understanding and predicting deforestation rates and patterns. For example, at least half of the land-use change models based in the Amazon that we are aware of use road networks as either a spatial or temporal predictor of deforestation (Araujo et al. 2009; Etter et al. 2006; Laurance et al. 2001; Mena et al. 2011; Messina and Walsh 2001; Michalski et al. 2008; Soares-Filho et al. 2006; Walker et al. 2004). In Central Africa, Laporte et al. (2007) showed that rates of road building in Central Africa increased over 31 years as a function of industrial logging. Yet no similar study has been carried out for the Amazon or any of its nine constituent countries despite the Amazon being at the forefront of global tropical deforestation.
Much of the existing literature on road network development in the Brazilian Amazon and elsewhere is qualitative rather than quantitative. For example, Taaffe et al. (1963) described an idealised progression of transport development using Ghana and Nigeria as examples. Perz et al. (2007) investigated the differing development histories of four road networks through interviews with local residents and qualitatively compared the social and spatial processes behind different patterns of unofficial road development. More rigorously, Maki et al. (2001) documented the development of a road connecting two urban centres in Peru (Iquitos and Nauta), and Brandao and Souza (2006) documented the growth rate of roads for an area of 546,000 km2 in the state of Para (Brazil), finding that the road network nearly doubled over a period of 10 years. There have also been some recent attempts to model the paths that individual roads take as they are constructed (Arima et al. 2008). Yet there have been no attempts made to understand how rapidly road networks are developed over large spatial scales, despite that knowledge being fundamental to understanding land-use change trajectories in the region (Barlow et al. 2011).
Here, we used pre-existing road maps, satellite imagery and simulations to document the historical, and predict the future, temporal dynamics of road network development in the Brazilian Amazon. We conducted our analysis at two spatial and temporal scales. First, we generated annual road maps for a 9-year period (2000–2008) from areas of low, medium and high road density to investigate annual patterns of network growth at relatively small spatial and temporal scales. Second, we used pre-existing road maps for the entire Brazilian Amazon from 2004 to 2007 to investigate rates of road network growth within municipalities. Based on our analyses of observed spatial and temporal variation in road network development, we constructed simulations to determine the temporal trajectory of road network growth in this region.
To examine how road networks change over time and if changes varied in different density classes, we calculated the cumulative amount of road (km) between 2000 and 2008 for each Landsat image. Changes in road length were analysed as a function of year and location using an ANCOVA in the statistics program R 2.10.1 (R Development Core Team 2009).
To investigate the spatio-temporal patterns of road network development at the scale of the entire Brazilian Amazon, we used two complete road maps for the region that were generated using the same manual digitisation methods on Landsat imagery taken in 2004 and 2007 (Brandao and Souza 2006). We used 443 municipalities of the Brazilian Amazon as spatial units for analysis, determining the initial road density in 2004 (km/km2) and the change in road density between 2004 and 2007 (km/km2) for each municipality. We chose to carry out analyses at the municipality level because this is the level at which decisions pertaining to development are made.
Using multiple regression, we tested the effect of the official road density and various socio-economic factors had on road network growth rates. Census data by municipality were obtained for the year 2004 from IPEA (2012) for permanent agricultural area, temporary agricultural area, total agriculture area, cattle head count, credit available for agriculture and credit available for cattle. The percentage of land area protected was also included. Rural and urban population counts for 2000 were used, because data for 2004 were unavailable. Data were log transformed where appropriate and agricultural area measures were converted to density to control for differences in municipality area. Starting with a complete model including all variables, we used model simplification to identify factors relevant to road network growth rates. We also tested for an effect of initial road density on road network growth rates using linear regression, including a quadratic term to allow for a non-linear relationship. Preliminary analyses showed that these models explained little of the variance in the data (R2 = 0.06), but visual inspection of figures suggested that maximum rates of road network development varied with initial road density and therefore that initial road density might act as a limiting factor on road network growth rates. The effect of a limiting factor can be quantified using quantile regression (Cade et al. 1999), so we estimated how initial density limits road network growth rates using quantile regression as implemented by the R package ‘Quantreg’ (Koenker 2010).
To investigate the likely temporal trajectory of road network development in the Brazilian Amazon, we constructed a simulation (essentially a Markov chain) based on the relationship we observed between initial road density and road network growth rate. The simulation was designed to estimate the length of time it would take for a municipality to develop a road network to such a density that new roads are no longer being created. The maximum road density observed in any municipality in 2004 was 0.5 km/km2, this municipality showed no development in 2007, and so we assumed that 0.5 km/km2 is the density at which road development stops. We assumed that municipalities would have an initial road density of 0.0 km/km2. In each annual time step of the simulation, we determined the amount of new road that would be developed in that municipality by creating a window (data bin) around the observed initial density of ±0.05 km/km2 and randomly selecting one observed growth rate that occurred within that window. For example, given an initial density of 0 km/km2, the window will subset out all growth rates corresponding to municipalities with densities of between 0 and 0.005 km/km2. A random growth rate is selected from the subset rates (e.g. 0.03 km/km2) and is added to 0 km/km2, giving the initial density for the next time step (i.e. 0.03 km/km2). All growth rates within a subset have an equal probability of being selected. This method takes into account the non-uniform distribution of growth rates observed among municipalities with similar initial road densities and allows for ‘kick-start’ development, as observed in some municipalities that had very low initial road network densities but very high rates of road network growth. The road network density in a simulation progressively increased through time, and we stopped all simulations when a road density of 0.5 km/km2 was reached. For each of 1,000 simulations, we recorded the cumulative increase in road density over time and fitted a logistic model to the simulated data using the R package ‘grofit’ (Kahm et al. 2010). We used the logistic model to quantify the length of the lag and boom phases of road network development. The lag phase tells us the number of years a municipality might take from initial colonisation until it experiences an exponential increase in road network growth. The boom phase, in turn, begins immediately at the end of the lag phase and represents the number of years during which road networks are rapidly growing. We used the second derivative of the logistic model to determine the start and end point of the boom phase. The local maxima and minima points on the second derivative of a logistic model indicate the two time points at which the rate of change in the road network is at its greatest (Ewers and Didham 2006), either accelerating at the end of the lag phase or decelerating at the end of the boom phase.
ANCOVA comparing differences in road development, in terms of total length (km) in three Landsat scene locations over a 9-year period (2000–2008)
The effect of socio-economic factors and official road density on the change in road density between 2004 and 2007, across the Brazilian Amazon
Standard error slope
Log official road density
Log permanent agriculture density
Log credit available for agriculture
Log protected area per cent
Log official road density interaction with log permanent agriculture
We found that, on average, 16,889 km of new roads are added every year to the Brazilian Amazon road network and that road network development within municipalities is a process that likely spans around four decades. Road network growth rates varied considerably among municipalities, likely reflecting the different economic and developmental histories of those regions, and the simulated patterns of road network development indicated it follows a boom and bust dynamic in which the majority of roads are built in a short period of time.
There is an evident spatial pattern of road density observed, with high road densities observed in the eastern and southern Amazon (Fig. 1), particularly in the states of Rondonia, Mato Grosso and Para. This pattern is likely associated with settlement and economic development of the municipalities within these states. In the 1970s, the Brazilian government instigated several policies to encourage the development and colonisation of the Amazon. Government incentives to encourage colonisation and cattle ranching projects (Kirby et al. 2006), as well as private colonisation projects via companies and co-operatives, were utilised to encourage settlement particularly in the frontier states of Mato Grosso, Rondonia and Para (Jepson 2006). At the centre of the colonisation initiatives was the building of road networks, notably the Cuiaba-Porto Velho (BR-364), Trans-Amazonian (BR-230) and Cuiaba-Santarem (BR-174) highways (Alves 2002; Kirby et al. 2006). Between 1970 and 1990, almost 700,000 families were relocated to the Amazon (Ludewigs et al. 2009), and it is unsurprising that road densities were highest in these states.
Within a given Landsat scene, we found that the road network growth rate was relatively constant through time over the 9-year period during which we tracked annual changes to the road network. There are two main causes of road network development: government initiatives to link populations and economies, and private sector development to aid the extraction of natural resources (Brandao and Souza 2006; Perz et al. 2007). Both causes have considerable momentum behind them, in that constructing roads require expensive machinery and personnel. If these resources are both limiting and used with the same level of efficiency among years, it follows that the rate of road construction in any given year will be similar to the rate the year before. This potentially explains why networks grew at remarkably constant rates within Landsat scenes, and the same momentum suggests that the rates of road network development we observed are likely to continue into the immediate future.
While road networks appear to grow at relatively stable rates within a region, we detected significant differences in road network growth rates among regions. Average growth rates were low in regions with low initial road density, increased in regions with medium road density and then reduced again in regions with high initial road density. This general pattern was defined largely by an apparent upper limit to road network growth rates at any given initial road density, with the most rapid rates of network growth occurring at intermediate road densities. By contrast, at all initial road densities, there were municipalities that experienced no road network development. Across the Brazilian Amazon, road networks were rapidly expanding in the set of municipalities that collectively form the ‘Arc of Deforestation’ along the south and eastern edges of the Amazon (Fearnside 2005). This is where most deforestation is concentrated in the Brazilian Amazon, and the level of congruence between road network growth rates and the Arc of Deforestation further reinforces the link between roads and deforestation (Fearnside 2005; Kirby et al. 2006).
A very small number of municipalities with very low initial road densities had road network growth rates that were much higher than the general pattern observed in the remaining municipalities (Fig. 4). Similarly, we detected a rapid jump in road construction between 2000 and 2001 in the high-density Landsat scene (Fig. 3). These apparently sporadic events of rapid road construction at particular places and times suggest that the general pattern of road network development can be accelerated under exceptional circumstances, perhaps through significant government investment in roads as part of regional development initiatives devised to kick-start economic development (Carvalho et al. 2002), or possibly through the establishment of new industries. The number of these exceptions was, however, very low, indicating that there is a general pattern by which road networks develop.
The observed pattern of road network development in the Brazilian Amazon likely reflects the economic trajectory of municipalities. The first new roads created in a municipality increase access to the region for extractive industries and colonisers, who further expand the road network by constructing unofficial roads to move their products to markets (Fearnside 2008; Geist and Lambin 2002; Perz et al. 2007, 2008). These unofficial roads are financed by the profits arising from exploiting resources and land in a previously unexploited region. This process of increasing economic returns within a municipality can stimulate further development in the region, generating a positive feedback loop that leads initially to exponential growth and maximum network growth rates as resource extraction and land development reaches the highest levels the region can support. Once resources begin to be depleted and the majority of the land had either had its resources removed or has been converted to other uses such as agriculture, the rate of road network growth might slow for one of two reasons. First, the economic impetus to expand and develop might be reduced, leading to slower road network growth rates. Alternatively, the road network might have approached a density that is high enough to provide access to all parts of the municipality. In this case, a sufficient road infrastructure already exists and would render further expansion unnecessary. This likely explains why the majority of observed road development is concentrated along the Arc of Deforestation (Fig. 2); new roads are developed in front of the arc to aid timber extraction, and once the ‘arc’ passes the rate of new road, construction will slow and eventually stop as resources are exhausted and road networks are developed to such an extent that the entire municipality is well connected.
The relationship between initial road density and change in road density (Fig. 4) provided a good relationship upon which to model the temporal trajectory of road development. The official road network has been used in the past to predict future road networks for use in land-use land cover change modelling (Messina and Walsh 2001; Soares-Filho et al. 2006). Here, we found that the density of official roads to be inversely related to change in road density, likely because the official road network has not changed much over the years and much of the development associated with official roads has already taken place. Thus, while areas with high official road densities are likely to be areas of high road density in general, they are unlikely to be areas of high changes in road density. Arable agriculture was also inversely related to change in road density, such that areas of high road development have less agricultural area, and this seems feasible if we consider roads to be a cause of development. In locations where roads have already developed (i.e. low changes in road density), we see high levels of agriculture (behind the Arc of Deforestation) but where they are developing, agriculture has yet to fully expand. Credit for agriculture is positively related to road density development, and it is possible that this positive relationship is linked to the amount of capital available generally for development; thus, in areas where there is high credit for agriculture, there are also capital resources available for road development. The negative relationship between the percentage of land area protected and road development highlights the role that protected areas play in determining road network development, with highly protected areas experiencing less network growth. While we found that official roads, in conjunction with agriculture, have a significant relationship with road development, they explain little variance in observed road growth over the study period. Further, the majority of socio-economic factors tested showed no significant relationship with road development. This, in addition to the fact that socio-economic factors are extremely difficult to predict, they, unlike initial density, do not form a suitable basis upon which to model future road development trajectories.
The temporal dynamics of road network development we describe above and simulated (Fig. 5) are similar to those of economic boom and bust development trajectories that have previously been observed in the Brazilian Amazon. The development of the Amazon has been economically dependent on boom and bust cycles of extractive industries (Godfrey 1990), with that dependence mirrored in many development indices in the Amazon, including life expectancy, literacy and standard of living (Rodrigues et al. 2009). It is thought that the boom in development and improvement in living standards occur as a result of people taking advantage of resources that become available in frontier areas where new roads have been laid (Fearnside 2008; Perz et al. 2008; Rodrigues et al. 2009). The temporal scale of economic boom and bust cycles appears to match the temporal scale of road network development that emerged from our simulations. For example, the rubber boom of the late eighteenth century lasted approximately 50 years (Godfrey 1990), cacao booms lasted between 20 and 40 years (Clough et al. 2009) and a boom in animal skins from Rio Preto lasted 20 years (Macedo and Anderson 1993). By comparison, our simulations suggested that the boom period of road development lasts for an average period of 39 years. Our model, then, appears to have generated realistic temporal trajectories of road network development for the Brazilian Amazon.
On average, it was found that it takes 75 years for an area to reach ‘maximum’ road density. Maximum is a relative term, as some areas may, in reality, ultimately reach marginally higher densities than those predicted by the simulation. Also, the simulation assumes that all areas will eventually reach the maximum density, whereas many places will not for various reasons including areas that may be protected or where resources are depleted quickly with little investment and thus little road development. Our simulation, then, reflects a scenario in which development continues unhindered until the maximum road density is attained. Further, the rate at which the road network is developed is likely to be highly variable because of differences in geographic environments and economic investment. Our simulation is based on empirical growth rates and as such implicitly takes into account such variable conditions. This is reflected in the range of length of time for road networks to develop (58–103 years).
The Amazon is an ecosystem that provides a multitude of ecosystem services that are globally important (Foley et al. 2007), but is undergoing rapid transformation as forests are progressively cleared for agriculture. As such, it is unsurprising that many models have been developed to predict future land uses of the Amazon. However, a key aspect of these models that has been largely ignored is the rates and patterns of road network development (Barlow et al. 2011). We have used a combination of spatial and temporal data to explore this issue, revealing the remarkable rate at which road networks are expanding and modelling the temporal trajectory of road networks. We anticipate these analyses will form a base for generating integrated land-use models that incorporate the economic development of the region, with a view to gaining a more comprehensive understanding of how this globally important ecosystem is changing.
We would like to thank the Grantham Institute for Climate Change and Microsoft Research for funding this work, and IMAZON (Instituto do Homem e Ambiente da Amazônia) for providing the Amazon wide road maps.