Modeling Amazon deforestation for policy purposes

An alternative approach to analysing deforestation, to take account of the conservationist and development priorities

By Clive W J Granger

Introduction

Despite a significant increase in environmental consciousness in both developed and developing countries, tropical deforestation continues at a worrying pace. According to estimates based on satellite photos, deforestation rates in the Brazilian Legal Amazonia fluctuate at around 20,000 km²/year depending on economic conditions (Figure 1). By 2005, almost 1 million km², or close to 20 per cent of Legal Amazonia, had been deforested.

Figure 1: Annual deforestation rates in Legal Amazonia, Brazil.

Source: Official National Institute of Space Research (INPE) figures.

While part of the deforested area has turned into highly productive agricultural land, a lot has been wasted with very little social benefit. It is in everybody’s interest to ensure that wasteful deforestation is minimised and that future deforestation is limited to areas where it brings substantial social benefits.

It is now widely agreed that a lot of wasteful deforestation in the past has been induced by special government incentives, such as land concessions based on deforested area, highly subsidised credit and tax breaks. These perverse policies brought large private gains irrespective of land productivity, resulting in socially wasteful deforestation. Most of these incentives have been dropped, both because they were expensive for the government, and because the occupation of the Amazon has gained so much momentum that artificial incentives are no longer necessary.

However, one highly controversial policy remains: road building. Many writers are very critical of the construction of roads through the Amazon, as roads invariably attract farmers who deforest along the road. Others suggest that not all kinds of road building are necessarily bad. If existing unpaved roads are paved, for example, this may encourage farmers to settle down close to the road and intensify their agricultural activities on a relatively small plot, instead of deforesting and practicing extensive agriculture further into the forest. Thus, while roads may cause an increase in deforestation close to the road, they may avoid deforestation further away from the road.

Whether this substitution effect actually occurs, and to what extent, is an empirical question that we have hitherto not been very good at analysing. While we now have excellent data from constant satellite surveillance of the Amazon, the econometric methods applied so far have been inadequate. The purpose of this paper is to explain why the methods usually applied give misleading results, and to point out some alternatives.

Geographic Information System (GIS) analysis of roads and deforestation

The raw material used to analyse deforestation consists of satellite pictures of the forest. Analysis of these pictures has shown that almost all deforestation has taken place within 50 kilometres of a main road. Based on this, it is possible to calculate the percentage of vegetation that has been cleared within a 100 kilometre-wide buffer zone around all main roads that are at least, say, 20 years old. Let us call this percentage X. It would then seem logical to expect that if you construct one kilometre of new road, this would cause X% x 1 km x 100 km = X square kilometres of additional deforestation within 20 years.

While simple and intuitive, this kind of GIS analysis has one major problem: it does not take into account correlation between deforestation in different areas (pixels). Spatial correlation can be either positive or negative. It is probably positive for plots close together, as the clearing of one plot makes it much easer to clear an adjacent plot, both because of easier access and because fragmented forest burns more easily. But correlation may be negative for plots farther away from each other. This would, for example, be the case if labour is a scarce factor. If farmers spend all their time clearing and cultivating one plot, they cannot at the same time clear another plot. Thus, if the number of people dedicated to agriculture is constant, then a rise in deforestation in one place would be matched by a decline in another.
In fact, the rural population in Legal Amazonia is growing over time but several farm-level studies suggest that labour is very scarce in the Amazon. This means that there is likely a negative correlation between the rate of deforestation in different areas, but it is difficult to say how strong it is.

In order to test the impact of a road on overall deforestation, it would be very important to take into account this possible negative correlation whereby, if deforestation goes up in one place, it would have to go down in another. The simple GIS model cannot capture this effect because, while it captures all the additional deforestation along the road well, it does not and cannot capture the avoided deforestation further into the forest.

Municipal level regressions

Another widespread technique used to analyse deforestation in the Amazon is municipal level cross-section regressions with some measure of deforestation as the dependent variable and policy variables, including roads, among the explanatory variables.

Since municipal level regressions use much larger spatial units than GIS analysis, the problems which can arise if we neglect spatial correlation are not as great as in the GIS analysis. This is because at least part of the avoided deforestation will be in the same municipality, but farther away from the road. Thus, part of the avoided deforestation is included in the municipal level data.

Some municipal level regression analyses do attempt to include spatial correlation explicitly, but due to the rudimentary nature of the spatial information, it is necessarily done in a rather crude manner.

The main problem with municipal level regression analyses is that the findings are not at all robust. Deforestation processes have changed dramatically over the last few decades, and a regression made on data from the 1970s is likely to yield completely different results from one made on data from the 1990s. These differences are very likely real, which implies that data from different time periods cannot be pooled to create one large sample. Regressions have to be made individually for each time period, which means that the number of observations is limited by the number of municipalities - 257 to 628 in Legal Amazonia, depending on the time periods included. Since there are hundreds of possible explanatory variables available at the municipal level for Legal Amazonia, and since the dependent variable itself can be expressed in many different ways (for example, levels, logs, shares, changes, changes in shares, based on either satellite information or on agricultural census information), it is possible to get virtually any result you might be looking for, if you try hard enough. This means that the reader should be highly sceptical when presented with a particular regression result. It may easily be the result of conscious or unconscious data mining.

Given that both GIS analyses and municipality level regression analyses have severe limitations, the following section outlines some alternative approaches.

Alternative analytical approaches

Deforestation is simultaneously a spatial, dynamic and economic process, and it is important to recognise that limiting deforestation is not the only objective we have to take into account. The impact on the living standards of the local people is an additional objective that even environmentalists must now consider. This means that we have a highly complex and dynamic spatial system with many economic constraints and interactions, and two objective functions. If we ignore any part of this, our analysis may be seriously flawed.

Given enough time, a well-funded team of experts in Computable General Equilibrium modeling, GIS analysis and deforestation might be able to develop a model that includes all these dimensions satisfactorily using available data. Whether the results of the model will be trusted, depends a lot on the composition of the research team. If it is dominated by environmentalists and funded by conservation institutions, people worried about local development may be concerned that the analysis is biased towards conservation, and vice versa.

Here is an attractive alternative, which is less demanding in terms of modeling and estimation, and which requires less cooperation and trust between environmentalists and developmentalists. Let environmentalists develop a map of conservation priorities (on a scale of 1 to 10, say) and let developmentalists create a map of development priorities, and then overlay the two maps to create a mosaic of land uses for development and conservation. There will probably be some areas of conflict between the two objectives, but in land-abundant countries – all Amazonian countries – it should be possible to accommodate both development and conservation interests in a rational way.

The map of conservation priorities could be constructed based on sub-maps of species diversity, endemism, carbon sequestration capacity, erosion risk, watershed protection services, and other variables judged important by the conservation community. The map of development priorities is likely to take into account soil quality, existing infrastructure, current and expected future population concentrations, and the existence of other natural resources, such as oil and minerals. Combining the information in the two maps should result in a map that indicates areas that should be conserved in their natural state, areas where development can be encouraged without too much environmental damage, and areas of conflict where care has to be taken as to which kind of development is encouraged.

Advantages with the priority approach

The big advantage with the priority approach is that the two underlying maps can be done independently. The specialists in conservation sciences can concentrate on counting birds and beetles, estimating biodiversity, measuring carbon density and assessing environmental services. This is by no means an easy task, but at least they do not have to worry about land prices, globalisation and poverty as well. They do have to acknowledge, however, that there are geographical variations in conservation priorities. The approach will not work if they assign top priority to all remaining natural areas.

The specialists in development will clearly have to take into account population growth and the corresponding growth in demand for agricultural products. Their main tasks will be to make spatially explicit population projections and estimate aggregate demand for agricultural land. They have to analyse soil quality and other agricultural conditions as well as market conditions, in order to pinpoint where agricultural activities should be ideally located.
Once the environmentalists and the developmentalists have made their priorities spatially explicit, an independent third party can overlay the maps and create a mosaic of ‘optimal’ land uses that gives sufficient room for expected agricultural expansion and at the same time protects the areas that are most important for conservation purposes. This third party will undoubtedly encounter certain areas that both conservationists and developmentalists claim to be top priority. These are conflict areas, where specific interventions will be necessary in order to ensure that the pressing development forces do not cause too much environmental damage. Such interventions could include, for example, encouraging eco-tourism, fish farming, high value perennial crops, or other relatively benign activities as alternatives to extensive agriculture. Further micro-level zoning can also be made within the areas of conflict, in order to reduce the environmental impact of the inevitable human presence in the area.

While it is not always politically feasible to implement the ‘optimal’ land use mosaic, at least policy-makers receive important guidance, and thus have concrete arguments to back up their proposals and decisions.

Conclusions

Road building in the Amazon remains a highly controversial subject. Brazil’s determination to develop the region is met with strong opposition from environmentalists, as is the less organised encroachment occurring in other Amazonian countries. While data is now abundantly available, the methods typically used to analyse the impact of roads on natural vegetation cover are methodologically weak and not very helpful to guide public policy.

This paper discussed the respective weaknesses of typical GIS analysis and typical municipality level regression analysis, and showed what would be needed to construct an ideal model of deforestation processes. It also presented an alternative approach that is much less demanding in terms of modeling and estimation, and which requires less cooperation and trust between environmentalists and developmentalists. This approach involves developing maps of conservation priorities and development priorities, and superimposing the two to create a land use mosaic, which takes into account both priorities at the same time. By doing so, the process acknowledges that both conservation and local development are important objectives.

Top ^

A condensed version of a public lecture given at the University of Melbourne in September 2006.

Professor Sir Clive Granger is Emeritus Professor of Economics at the University of California, San Diego and is a Visiting Eminent Scholar to the Department of Economics at the University of Melbourne. In 2003, Professor Granger was awarded the Nobel Prize in Economics (with Robert Engle) for his discoveries in the field of time series data analysis.

Volume 2 OCT 2007