kwrS A (VI%
POLICY RESEARCH WORKING PAPER   2687
Geographic Patterns of Land                                          Nearly90percentof
agricultural land in the
Use and Land Intensity                                               Brazilian Amazon is used for
in the Brazilian Amazon                                              pasture, or has been cleared
and left unused. Pasture on
average is used with very low
Kenneth M. Chomitz                                                   productivity. Analysis based
Timothy S. Thomas                                                    on census tract data shows
that agricultural conversion of
forested areas in the wetter
Western Amazon would be
even less productive, using
current technologies.
The World Bank
Development Research Group
Infrastructure and Environment
October 2001



| POLICY RESEARCH WORKING PAPER 2687
Summary findings
Using census tract data from the Censo Agropecuario          infrastructure and market access, proximity to past
1995-96, Chomitz and Thomas map indicators of                conversion, and protection status. Chomitz and Thomas
current land use and agricultural productivity across        find precipitation to have a strong deterrent effect on
Brazil's Legal Amazon. These data permit geographical        agriculture. The probability that land is currently
resolution about 10 times finer than afforded by             claimed, or used for agriculture, or intensively stocked
municipio data used in previous studies. Chomitz and         with cattle, declines substantially with increasing
Thomas focus on the extent and productivity of pasture,      precipitation levels, holding other factors (such as road
the dominant land use in Amazonia today.                     access) constant. Proxies for land abandonment are also
Simple tabulations suggest that most agricultural land     higher in high rainfall areas. Together these findings
in Amazonia yields little private economic value. Nearly     suggest that the wetter Western Amazon is inhospitable
90 percent of agricultural land is either devoted to         to exploitation for pasture, using current technologies.
pasture or has been out of use for more than four years.       On the other hand, land conversion and stocking rates
About 40 percent of the currently used pastureland has a     are positively correlated with proximity to past clearing.
stocking ratio of less than 0.5 cattle per hectare.          This suggests that in the areas of active deforestation in
Tabulations also show a skewed distribution of land          eastern Amazonia, the frontier is not "hollow" and land
ownership: almost half of Amazonian farmland is located      use intensifies over time. But this area remains a mosaic
in the 1 percent of properties that contain more than        of lands with higher and lower potential agricultural
2,000 hectares.                                              value.
Multivariate analyses relate forest conversion and
pasture productivity to precipitation, soil quality,
This paper-a product of Infrastructure and Environment, Development Research Group-is part of a larger effort in the
group to understand the causes and consequences of land use change. Copies of the paper are available free from the World
Bank, 1818 H Street NW, Washington, DC 20433. Please contact Shannon Hendrickson, room MC3-640, telephone 202-
473-7118, fax 202-522-0932, email address shendrickson@worldbank.org. Policy Research Working Papers are also
posted on the Web at http://econ.worldbank.org. The authors may be contacted at kchomitz@worldbank.org or
tthomas2@worldbank.org. October 2001. (38 pages)
The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about
development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The
papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this
paper are entirely those of the authors. They do not necessarily represent the view of the World Bank, its Executive Directors, or the
countries they represent.
Produced by the Policy Research Dissemination Center



Geographic Patterns of Land Use and Land Intensity
in the Brazilian Amazon
Kenneth M. Chomitz and Timothy S. Thomas
Development Research Group, World Bank
with contributions by
IBGE, University of Washington CAMREX Project, and IMAZON






Geographical Patterns of Land Use and Land Intensi_y in the BrazjIian Amaton
Acknowledgements
We thank Robert Schneider for supports, useful discussions, and for formulating the 'rainfall
hypothesis' explored here. We are extremely grateful to our many colleagues who generously shared
their data, expertise, advice, and support. At IBGE, we thank Sergio Besserman Vianna for his
support and encouragement and Lidia Vales de Souza, Antonio Carlos Florido, and many others for
helping us to obtain and understand the Censo Agropecuario data that is the heart of this study.
Jeffrey Richey and Miles Logsdon of the University of Washington CAMREX project provided the
precipitation data. Hari Eswaran and Paul Reich at USDA World Soil Resources provided the soil
limitations data. Eugenio Arima, Paulo Barreto, and Carlos Souza of IMAZON contributed greatly
in helping to frame and interpret this work. We're grateful to David Kaimowitz, Benoit Mertens,
and Pedro Olinto for useful comments. This work was funded in part by the Global Overlay
Program, supported by the Danish government.
The findings, interpretations, and conclusions expressed in this paper are entirely those of the
authors. They do not necessarily represent the view of the World Bank, its Executive Directors, or
the countries they represent. The boundaries, colors, denominations and any other information
shown on maps herein do not imply, on the part of the World Bank Group, any judgement on the
legal status of any territory, or any endorsement or acceptance of such boundaries.
Comments on this working paper are welcome and may be sent to: kchomnitz(worldbank.org






Geographical Patterns of Land Use and Land Intensi_0 in the Brazilian Ama.Zon
Tables, Maps and Figures
Note: tables, maps andfiguresfollow the text.
Table of Tables
Table 1. Main sources of revenue for farm establishments by state
Table 2. Main sources of revenue for farm establishments by size of farm
Table 3. Main sources of agricultural revenue by rainfall category
Table 4. Overview of study area
Table 5. Regressions on proportion of census tract in agriculture land
Table 6. Regression on proportion of census tract in agricultural land, with dry months as categorical
variable
Table 7. Count of census tracts by rainfall and consecutive dry months
Table 8. Summary of variables used in proportion of census tract in agricultural land regressions
Table 9. Regressions on proportion of pasture in census tract
Table 10. Regressions on natural log of stocking density (cattle per total pasture area)
Table 11. Summary of variables used in stocking density regression
Table of Maps'
Map 1. Mean annual rainfall, 1970 to 1996
Map 2. Primary limiting factors of soils
Map 3. Underlying vegetation types
Map 4. Proportion of establishment area in total area of census tract
Map 5. Land with some type of protected status
Map 6. Stocking density (cattle per hectare of pasture)
Map 7. Land cleared by 1976, 1987, and 1991
Map 8. Proportion of productive but unutilized land in cleared area
Map 9. Proportion of natural pasture in cleared area
Map 10. Proportion of planted pasture in cleared area
Map 11. Proportion of total pasture in cleared area
Map 12. Mean farm size
Map 13. Proportion of cleared land in census tract
Map 14. Proportion cleared in 1995 (predicted by model)
I Maps in color are available on the Policy Research Working Papers Web site.



Geographical Patterns of Land Use and Land Intensiy in the Bran4lian Amapon
Table of Figures
Figure 1. Area of establishments since 1975, by State
Figure 2. Effect of rainfall on the proportion of a census tract in agriculture
Figure 3. Effect of rainfall on the proportion of a census tract in pasture
Figure 4. Lorenz curves for land distribution in the Amazon (excluding Maranhao)
Figure 5. Land use by precipitation, all census tracts
Figure 6. Land use by precipitation, census tracts with a portion < 25 km from a principal road
Figure 7. Land use by precipitation, census tracts with no portion < 25 km from a principal road



Geographical Patterns of Land Use and Land Intensity in the Brazlian Amanaon
Motivation and goals
Policies affecting development in Amaz6nia must balance a variety of competing options for land
use. These include pasture, crops, agroforestry, sustainable forest management, provision of
environmental services, and conservation to maintain future options. Because conversion to some
kinds of agriculture may preclude the option to devote the land to other uses in the future, it is
important to know:
* How public policies, especially with regard to infrastructure, affect the likelihood that land will
be converted to different kinds of agriculture.
* The potential economic benefits of conversion to agriculture.
The potential economic and noneconomic values of the land vary dramatically from place to place,
Mapping these variations can help us to understand which biophysical and socioeconomic
conditions favor productive agriculture, and to identify conditions under which land is at risk of
being converted to relatively unproductive agriculture.
As a modest first step, this paper uses census-tract-level data from the Censo Agopecuario 1995-96
(lBGE, 1998) to map indicators of current land use and agricultural productivity across the Legal
Amazon of Brazil. It relates these indicators to market proximity, infrastructure access, and
agroclimatic conditions. It focuses particularly on pasture for two reasons. First, pasture is by far
the dominant land use in Amaz6nia today, accounting for more than three-quarters of agricultural
land use. Second, this land use is characterized on average by low productivity and low employment
absorption, suggesting that in many cases it may not be a socially optimal use of the land.
The results of this analysis must be interpreted with caution. The historical data used here cannot, of
course, tell us what development patterns might be possible in the future using new or hypothetical
agricultural technologies. However, a record of the actual behavior of hundreds of thousands of
farmers across the wide and varied landscape of Amaz6nia does provide insight into the
geographical opportunities and constraints to agriculture as modulated by current technical and
institutional condtions.
The paper begins with a description of the biogeophysical and socioeconomic context. It then
describes broad patterns of land use in Amazonia, using simple descriptive statistics and maps. An
analytical section draws on these data to conduct two multivariate analyses: the determinants of
agricultural land use, and the determinants of stocking rates of pasture. A concluding section
summarizes finding and discusses their implications.
The biogeophysical and socioeconomic context
This section describes some of the basic features of the region - natural and human -- that shape
patterns of land use. The data described here constitute the main explanatory variables for the
multivariate analysis.
Climate, soils, and natural vegetation
Agriculture is constrained by biogeophysical factors: climate and soils. Sombroek (1999)
hypothesizes that high rainfall and lack of a dry season are important limiting factors to agriculture
in Amaz6nia. In high rainfall areas, he claims, humans and animals are more susceptible to disease;
Page 1



Geographical Patterns of Land Use and Land Intensi~y in the Brajlian AmaZon
forest burning is incomplete, complicating the establishment of crops or pasture; grains and many
other crops such as soybeans, are subject to rotting; mechanization is difficult; and rural access roads
are difficult to build and maintain. We use monthly precipitation data for 1970-96 kindly provided
by the CAMREX project (University of Washington). Each composite month is the mean of
individual months formed by interpolations of gauge records of the Agencia Nacional de Energia
Eletrica (ANEEL) to 0.05 degrees spatial resolution. Map 1 shows the mean annual precipitation
based on this data. There is a strong gradient from high precipitation in northwest towards lower
precipitation in the southeast, with an additional rainfall peak in the northeast. Number of dry
months (the statistic stressed by Sombroek as a key limiting factor) is highly correlated with mean
annual precipitation. Because the precipitation data extend only to 450 W, parts of the subsequent
analysis exclude the easternmost portion of the Legal Amazon (part of Maranhao comprising about
1.3 percent of the land area of the Legal Amazon).
Another important class of biogeophysical parameters are those related to soil types. There are
many different ways to classify soils based on their many underlying properties (such as texture,
slope, parent material, depth, and soil moisture and temperature regimes). Map 2 was kindly
provided by the Soil Survey Division of World Soil Resources of the U.S. Department of Agriculture
(Eswaran and Reich, nd). It sumnmarizes the soils by their primary limiting factor. In our study area,
the data distinguish thirteen soil categories, though worldwide their system notes about twice that
mnany.
Map 3 shows the Vegetation Map of Brazil (Ministerio da Agricultura, et al, 1988, and digitized by
USGS EROS Data Center). While natural vegetation may itself reflect soil and climatic
characteristics, it may provide additional biogeophysical and economic information related to the
ease and attractiveness of converting the land to agricultural use. To take an obvious example,
cerrado will have lower costs of clearing, but also lower revenues ftom sale of timber, than forest
areas.
The socioeconomic context
Land use in Amaz6nia - and in virtually all regions of agricultural expansion, worldwide - is strongly
shaped by past settlement patterns and by roads (Reis and Margulis, 1991; Chomitz and Gray, 1996;
Angelsen and Kaimowitz, 1999). Alves (1999), for instance, shows that deforestation in the
Amazon has tended to expand from areas already deforested by 1978. Map 7 shows the progressive
extent of clearing, based on remote sensing data. The relatively small amount of clearing in 1976 is a
strong predictor of current land use, as will be seen below. Principal roads are shown in Map 4;
their relation to agriculture is evident on inspection. Pfaff (1997) uses multivariate analysis to show
that proximity to roads is indeed a strong predictor of deforestation in Amaz6nia.
Protected areas can also shape land use, though their efficacy in deterring settlement has been
questioned. Map 5 shows the substantial area under protection as indigenous lands or for
conservation. The multivariate analysis later in this paper allows an estimate of the actual detertent
effect of protected status.
Agriculture in Amaz6nia
This section describes broad patterns of land use and agricultural output in Amaz6nia, using data
from the Censo Agropecudrio 1995-1996 (IBGE, 1998). We are extremely grateful to the Instittto
Brasikiri de Geografia e Estatistica (IBGE) for providing us with tabulations of land use, production,
Page 2



Geographical Patterns ofLand Use and Land Intensit in the Bra#iian Ama.Zon
labor, and cattle at the level of the census tract (seto), along with census tract boundary maps. We
merged very small sectors (less than 400 hectares) with adjacent sectors, yielding 6776 sectors or
agglometates as the units of analysis. This petrnits geographical tesolution about 10 times finer than
afforded by municipio data that has been the subject of previous study. We rely also on some
tabulations of municipio- and state-level data for variables and establishment-size breakdowns not
available in the census tract data.
Geographical pattems of land use and production
Land use
Table 4 presents basic statistics on land ownership and use in the Legal Amazon. The study area
includes 492.7 million hectares, of which just under one quarter is in agricultural establishments,
with a virtually identical extent in national parks, protected areas, conservation areas, and indigenous
areas. Of the area in establishments, 41.5 percent remains in native forest, 55.0 percent is in
agricultural land, and the remaining 3.5 percent is unutilizable (paved, rock-covered, etc.).
A total of 65.3 million hectares is agricultural land; that is, productive land in crops, pasture,
plantation forest, or previously used and now abandoned. As we shall in greater detail, the vast
majority of this territory devoted to very low-value uses. More than three quarters of this land is in
pasture, and another tenth is 'productive unutilized' - probably abandoned. About eight percent is
in annual crops; much of this is manioc, characterized by high per-hectare gross production value
but low net revenue per hectare given its high labor input requirements. Less than 2 percent of
agricultural land is in perennials or planted forest, often thought of as potentially sustainable and
higher-value land uses.
This paper uses the ratio of agricultural land (as defined above) to census tract area as a measure of
deforestation. Some caveats apply, since the Census categories were not designed for this purpose.
Based on our reading of the Census interviewers' guide and discussions with IBGE staff, we assume
that cerrado is classified as forest unless it is currently used for grazing or agriculture, or has been
abandoned recently. A cerrado area used for grazing is assumed to be classified as 'natural pasture',
an agricultural land use. We are not sure how Census interviewers classified natural grasslands that
are not used for grazing (if such areas exist). 'Unutilized' areas are defined as those that have not
been used for more than four years, and we presume them to be abandoned. However, it is possible
that some long-abandoned parts of current establishments may now be in advanced regeneration
and may be classified as natural forest. Also, it is possible that some establishments may have been
entirely abandoned and not included in the Census. Figure 1 shows that there were substantial
declines between Censuses in the area of establishments in Amazonas and in Acre.  Our
deforestation estimates will exclude degraded land in any such areas, and will also exclude areas
outside current establishments that have lost forest cover because of fires or logging.
As is well known, agricultural land use is largely concentrated along the Arc of Deforestation that
curves along the eastern and southern edges of the region. (See Map 13) This is true not only of
agricultural land, but of all land in agricultural establishments. Establishments in this region tend to
be quite large, ranging up to an average of several thousand hectares in northern Mato Grosso (map
12).
In contrast Map 4 shows that only a negligible proportion of areas in the Western Amazon is in
establishments; much is in protected areas. The establishments in this region tend to be quite small
Page 3



Geographical Patterns of Land Use and Land Intensiy in the Brazilian Amar(on
(often less than 20 hectares on average) and many are presumably subsistence-oriented. However,
wet areas tend to have a higher proportion of their agricultural land in perennial crops (table 4).
The dorninance of pasture is shown vividly in Maps 9 to 11, depicting the proportion of natural
pasture, planted pasture, and total pasture in agricultural area. Natural pasture areas, as might be
expected, correspond closely to areas where the natural vegetation is cerrado or 'pioneer.'
Map 8 shows the proportion of productive but unutilized land. This probably represents abandoned
land. It is prevalent along the Western Amazon and around Belem (both high-rainfall areas), but
also in Maranhao along the border with Tocantins.
Production
The census-tract level data provide fine geographic detail on land use, but, in our dataset, lack
information on specific production commodities or their values. For complementary information
on land use, we turn to municipio-level data on production value in Tables 1-3' and refer to
municipio-level maps of these data (not shown).  Gross value of production is dominated by a
handful of products: cattle, soybeans, manioc, milk, and logs. These tend to show strong geographic
patterns.
Sqybeans
Soybeans are concentrated in Mato Grosso and southern Maranhao (Table 1). Soybeans tend to be
important where rainfall is between 1,600 and 2,000 mm annually; where there are 3 or 4
consecutive dry months; where the primary soil limiting factor is "high phosphorus, nitrogen, and
organic matter retention"; and where the underlying vegetation is cerrado.
Milk
Milk is an important product in central Rond6nia, as well as the tri-state region of Pari, Tocantins,
and Maranhao. The location of dairy production is highly sensitive to road access and proxirnity of
processing plants, though the advent of ultraprocessed milk extends the range for establishing such
plants. Most dairy production is in areas with annual precipitation below 2,200 mm.
Manioc
Manioc is often the main crop in census tracts which do not have much agricultural activity, where
precipitation is high, and where average establishmnent size is small. This is true especially in
Amazonas, but also in central and western Pari, and parts of Acre and Arnapa. These are
presumably frontier regions where manioc is largely a subsistence crop. We note, however, areas
near the coast in northeast Pari and northern Maranhao, as well as other places such as south-
1Tables 1 and 2 are taken direcdy from state-level agricultural census data, and include data for the entire state, even
though a large part of Maranhao and a small part of Tocantins are not in the Legal Amazon. Table 3 indudes only the
census tracts in the Legal Amazon that are west of 45 degrees West, and because the data is extracted from municipio
data, did not include all of the agricultural products that were in Tables 1 and 2. One key commodity excluded was milk.
Many other products which may have been locally important but of low imnportance for the entire Amazon were also
excluded.
Page 4



Geo,graphical Patterns of Land Use and Land Intensity in the BraZilian AmaZon
central Mato Grosso, northern Roraima, and eastern Acre, where overall agricultural land use is
relatively high but where manioc constitutes a large share of production.
Cattle
Cattle are a major contributor to the value of production2 in the southern half of Para, all but the
western part of Mato Grosso, all of Tocantins, and especially the western part of Maranhao; and in
parts of Amapa, Roraima, and Acre. In areal terms, cattle are found across a variety of soils and
vegetation types, and across a range of precipitation levels. They seem to be concentrated, however,
in areas that have at least two consecutive dry months.
Extractive products
In northeast Para, northern Mato Grosso, and the north and southwest of Amazonas, extractive
activities represent a large portion of the agricultural production, though the areas in Amazonas have
a very low density of establishments. Logging constitutes most of the extractive activities, though
piacaba is important in northern Amazonas.
Land use, rainfall, and roads
Table 4 and Figures 5 through 7 present some simple cross tabulations of land use by precipitation
category and distance to the nearest principal road, for those census tracts for which we have
precipitation data. Approximately 40 percent of the Amazon on average receives between 1,300 and
2,000 mm of rain; another 40 percent receives between 2,000 and 2,400 mm or rain; and the
remaining 20 percent ranges up to around 3,500 mm. The driest category has 45 percent of the land
in establishments; the mniddle category, 13 percent; while only 8 percent of the wettest category is in
establishments.
A striking feature of the figure is the sharp drop-off in nonforest3 land as precipitation increases. In
part this is due to the increased proportion placed under ptotection in the wettest areas. But the
proportion of nonforest land outside protected areas also declines with higher precipitation. The
proportion of all land in agriculture generally declines with increasing rainfall, reaching near zero by
3,200 mm.
It is instructive to examine the apparently anomalous increase in nonforest land in the 2800-3000
mm precipitation range.  Does this provide a counterexample to the thesis that high rainfall areas
are unfriendly to agriculture? On closer examination, almost all of this high-rainfall agricultural land
is near the Bel6m, a city of more than a million inhabitants that has been settled for almost half a
millennium. About half of the high-rainfall agricultural land consists of natural grasslands on Marajo
Island currently being used for grazing. Of the remaining half, approximately half is unutilized and
presumed abandoned. This example might be viewed as an 'exception that proves the rule' - non-
perennial agricultural development is possible in high rainfall areas, but only in conditions of very
high local demand, centuries of effort, and unusual agroecological conditions - and even then with a
high failure rate.
2 The value of cattle was calculated by adding the value of cattle slaughtered to value of cattle sold, and subtracting value
of cattle purchased.
3 Almost all of this is agricultural land, with a small proportion of 'unutilizable.' Some of the latter category includes
settlements, areas paved over, etc.
Page 5



Geo,graphical Pauterns of Land Use and Land Intensi4y in the Brazilian Amazon
Figures 6 and 7 show that the association of land use with rainfall is not an artifact of the location of
roads; it holds even when census tracts are disaggregated by distance to the nearest principal road.
Comparison of the charts suggests that roads do affect the proportion in agriculture, especially in the
middle ranges of precipitation.
Production and land use by establishment size class
Land in the nine Amazonian states is overwhelming concentrated in large holdings (see Table 2,
which includes areas of Maranhao and Tocantins outside the Legal Amazon; and Figure 4). While
only about 1 percent of all establishments have more than 2,000 hectares, these establishments
control 52.7 percent of private land and account for 46.8 percent of all land converted from forest
or cerrado to agricultural use. In contrast, establishments with less than 20 hectares constitute 53.8
percent of the total number of establishments, but control only about 1.5 percent of the property or
agricultural land.
There is strong product differentiation by size class of establishment (Table 2). The smallest farms -
those under 10 hectares - appear to be strongly subsistence-oriented, with manioc and rice
constituting 30 to 40 percent of production. In the 20 to 100 hectare size range, manioc is still
important, but so are cash products such as milk and bananas. For large and very large
establishments in the 100 to 100,000 hectare range, cattle and soybeans predominate. Among the
few ultralarge establishments of more than 1,000 square kilometers, silviculture is dominant
Land Value
There is no question that land values, on average, are low in the Legal Amazon. Published data4
from Receita Federal show the mean declared unimproved land value in the Northern Region5 was
just R$46.84/ha in 1997, as compared to to a Brazil wide average of $339.88. Anecdotal reports
suggest typical values, for improved pasture, of R$200/hectare6.
We are interested in studying spatial variation in these land values. Unfortunately, direct valuation
data is limited. Declared property tax data are not available at a disaggregate level, and may be
subject to misrepresentation. We use therefore a variety of proxies for land value.
One basic proxy is land scarcity. In regions where only a small proportion of available land has been
claimed as private property, it is reasonable to assume that land is so abundant as to have essentially
no value. Another way of putting it is that the potential revenue from the land is less than the cost
of enforcing claims to it (Schneider, 1995). Land scarcity is shown in Map 4, which depicts the ratio
of land in establishments to total non-water area of each census tract. It indicates, as one would
expect, more scarcity near cities and roads. This reflects higher farnigate prices of products, lower
costs for agricultural inputs, and lower costs of enforcing claims.
This proxy shows tremendous land abundance in the Western Amazon. In the wettest regions of
Amaz6nia, as we have seen, the low ratio of land in establishments goes along with a high
proportion of land in protected areas. However, the designation of these areas as protected may
reflect, in part, a recognition that these areas are not suitable for agricultural development. In
4 http://www.receita.fazenda.gov.br/PessoaJunidica/itr/PerfilITR97/TerraNua.htmn#Valor da Terra Nua - UF
5 Acre, Amapa, Amazonas, Para, Rondonia, and Roraima.
6 Eugenio Arima, personal communication, based on a survey in progress.
Page 6



Geographical Patterns of Land Use and Land Intensiy in the Bra<ilian AmaZon
addition, the extent of land outside protected areas far exceeds the area incorporated in
establishments. A closely linked indicator is the proportion of each census tract in agricultural land
(that is, converted from natural habitat). On the simple model that land is converted if it has
positive value, higher conversion proportions track higher land values. Map 13 shows that this
proxy closely tracks the previous one.
For the three quarters of agricultural land in pasture, we would expect that per-hectare profits would
provide a guide to land value. Several attempts have been made to assess the profitability of
Amaz6nian pasture using farm models based on interviews with small samples of farmers. A recent
and thorough study is that of Arimna and Uhl (1997), based on three locations in Para. They find
annual profits ranging from US$23/ha (small dairy farmers) to $7/ha (self-reproducing herd,
medium to very large ranches in upland areas) to $20-$25/ha (range-fattening operations, medium to
very large ranches). Vosti, Witcover, and Carpentier (1998) report gross revenues of $96/ha for
small dairy operations in Acre and Rondonia, about 50 percent higher than the small dairy farmners
studied by Arima and Uhl. Annual profits in this range reported by Arima and Uhl, together with
assumed land price appreciation due to the growth of markets or improvement of transportation,
can provide modest real rates of return on capital (8 to 12 percent), assuming that land is costlessly
acquired and financed through sale of timber or through a subsidy of some kind. These rates of
return, together with assumed appreciation due to improvements over time in market access, might
justify land values in the $60 to $300/ha range.
Because of the difficulty in allocating Census-reported expenses between different activities, we do
not attempt to map net revenues from pasture. Instead, we use stocking density (cattle/hectare of
pasture) as a proxy for value. In general, one would expect better-endowed land, or land closer to
markets, to profitably support more cattle per hectare. This is admittedly an imperfect proxy for
several reasons. First, natural pasture with a low stocking rate may possibly be more profitable (and
thus command a higher price) than planted pasture with a higher stocking rate. Second, very high
stocking rates may indicate unsustainable overgrazing, or stall-feeding. Finally, small, subsistence-
oriented farms of a few hectares may not be comparable to larger establishments, and stocking rate
estimates are very sensitive to errors in measuring pasture area for these farms. Nonetheless, the
stocking rate provides a simple and intuitively appealing metric for assessing land use intensity across
much of Amaz6nia.
Overall, statistics on stocking rate show very low levels of pasture utilization. About 40 percent of
currently-utilized pasture in the Legal Amazon has a stocking rate of less than 0.5 (that is, two
hectares per animal); the mean for this area is 0.3. (The denominator does not include abandoned or
fallow areas; their inclusion would bring the rate down substantially). In the remaining 60 percent,
the mean stocking rate is about 0.95.
Map 6 shows the average stocking rate by census tract (total cattle divided by total area of pasture -
both natural and planted pasture - for the census tract). It excludes census tracts where there are
fewer than 5 hectares of pasture per farm with cattle, for the reasons mentioned above. The map
shows a trend toward lower stocking density in cerrado areas, with stocking densities below 0.4 cows
per hectare in southwest Maranhao, most of Tocantins, northern Roraima, the east half of Amapa,
and southern Mato Grosso. The stocking density appears to be relatively high along the main roads
in Acre and Rond6nia, along the Amazon river in western Para and eastern Amazonas, in northern
Mato Grosso, and in small pockets in other parts of Amazonas, Para, and Maranhao. As we shall
see, stocking rates are sensitive to farm size and to the use of unpaid labor, complicating
comparisons across regions where farm sizes vary.
Page 7



Geographical Patterns of Land Use and Land Intensi in the Brawilian Amazon
Summary
According to the Censo Agropecudrio, about one eighth of the Legal Amazon was converted to
agricultural use by 1996; this land is overwhelmingly concentrated in the hands of large landholders.
Several proxies suggest that utilization and productivity of this land is very low. About one-eighth
of the agricultural land is not in current use; of the remainder, ninety percent is in mosdy extensive
pasture, with low average stocking rates.
Visual inspection of maps, and simple cross-tabulations, suggest that agricultural values in high
rainfall areas are particularly low on average. In these areas, only a small fraction of available land has
been claimed by agricultural establishments, and a smaller fraction has been converted to agricultural
use. In these areas, and in cerrado areas, proxies for land use intensity or value are very low.
However, there is some cultivation of potentially high-value perennials in the higher rainfall areas.
The next section uses multivariate analysis to assess whether the apparently low potential of high-
rainfall and cerrado areas reflects only the lack of roads and settlements, or is a more fundamental
reflection of biogeophysical constraints.
Analysis
Land use
Model
What are the determinants of land use and deforestation in the Legal Amazon? Alves (1999) and
Pfaff (1997) used remote sensing data on land cover to address this question, with important
findings. Alves shows that clearing has tended to expand outwards from its 1978 location. Pfaff,
using data summarized at the municipio level, shows the importance of the road network in
determinling the location of deforestation; this multivariate analysis controls for urban proximity and
for nitrogen density of the soil.
The analysis reported here builds on and complements this earlier work. Taking advantage of the
Censo Agropecudrio, it uses reported land use (based on census-tract data) rather than remote-sensing
based land cover.  This provides a useful cross-check, since available remote sensing data
distinguishes only between forest and nonforest and is subject to classification errors. (For instance,
capoeira - i.e., regrowth - may be classified as forest.) The work reported here also uses precipitation
as an explanatory variable, a key factor not available to eatlier work.
The land use data have shortcomings of their own. They are potentially subject to reporting error
by the landholders. Our calculation of non-water area in the sectors is subject to registration error
in superimposing maps.7 We assume that areas outside agricultural establishments are in natural
vegetation; this will not be true in settled areas which must therefore be excluded from
consideration.
To explain spatial variation in land use, we apply the model of Chomitz and Gray (1996): propensity
to clear land depends on the potential profits nI (or land rent) per hectare from converting the land
to agricultural use. Potential profits depend on:
7 We estimate registration to be accurate within approximately plus or minus 2 kilometers.
Page 8



Geographical Patterns of Land Use and Land Intensiy in the Brahian Amazon
* farmgate prices, which are related to road, river and city proximity.
* costs of clearing, which we expect to be higher in forest areas than in cerrado areas. We also expect
that protected area status increases the cost of clearing (because of expected penalties).
*  revenuefmm clearing, which will be higher in forest-biome areas, closer to roads.
*  agroclimatic suitabii&y. Agricultural productivity depends on soil quality and climate. This relation
differs among agricultural products: conditions favoring perennials may not favor pasture, for
instance. In general, however, we expect that soils with the more setious physical and chemical
constraints will discourage pasture and annual crops. We also hypothesize that high levels of
precipitation will discourage these land uses.
*  subsidies and othergovernment policies promoting agriculture. Unfortunately we have no information on
policies that might have had identifiable regional effects, except for road-building. It is not clear
that subsidized credit and other incentives had disproportionate impacts on certain regions.
Given random variation of land quality within an observation unit, the proportion of landp that can
profitably be converted to a particular land use is an increasing function of mean benefits and a
decreasing function of mean costs. In the dynamic context of frontier expansion, where not every
profitable situation is yet exploited, the probability of land conversion is related both to the
potential profitability of conversion and to another variable, prox4miy to prior conversion. In this
simple model
Pt = P(HI,p-)
A standard functional form for this model is the tobit:
p*= XP+u
p = 0 if p* < 0; otherwise p = p*
where X is a vector of explanatory variables reptesenting 1X and p, u is a random disturbance term
andp* is a latent variable. Censoring at zero captures the intuition that there will be no conversion
in unprofitable areas, and the reality that many census tracts lack any agricultural land.
However, the tobit model, while standard, is consistent only if the disturbance term u is
homoscedastic. In other words, it assumes that all unobserved variables have the same variance
across census tracts. If this assumption fails, the tobit is not trustworthy. Given the great
heterogeneity of census tracts in size and other characteristics, the assumption of homoscedasticity
is questionable. A spatially autoregressive error structure would also preclude homoscedasticity
(Anselin, 1999). We therefore follow the suggestion of Deaton (1997, pp 89-90) that iterated
quantile regressions (Powell, 1984, 1986; and Buchinsky, 1994) can be used as a consistent
alternative to the tobit. In an iterated quantile regression, the median of p, conditioned on X, is
estimated as a linear function of X. Observations for which the predicted value ofp is less than zero
are then dropped, and the estimate is repeated. The process is iterated until it yields a stable subset
of observations. Standard errors are obtained via bootstrapping. The predicted value of p IX is
simply:
Xp, if X,3>O
0, if XP<0
Page 9



Geographical Patterns of Land Use and Land Intensity in the Braklian Amar<on
As Deaton points out, this is a simpler and more appealing representation than the nonlinear
function (derived from a probably incorrect assumption about disturbance variances) which arises
from a tobit prediction.
Data
The sample of census tracts analyzed consisted census tracts in the Legal Amazon west of 45
degrees West latitude (i.e., those for which precipitation data was available). Some urban census
tracts9 were excluded; however the sample probably includes census tracts with settlements.
The dependent variable was the ratio of agricultural area, as reported by the Census, to computed
non-water area of the census tract. In some cases the computed ratio was greater than one. This
may reflect establishments that straddle a census tract border, but whose total area is recorded
(according to standard Census procedure) in just one census tract. It may also reflect inaccurate
estimates of area, overlapping land claims, or registration error in computing areas. We assume that
reported proportions greater than 1 should be treated as equal to 1, censoring both the tobit and
iterated quantile regressions.
Results
The results are presented in Table 5. The preferred specification is the quantile regression in the
middle column. It is helpful to think of the predicted value of the regression as an index
corresponding to the expected proportion of the census tract used for agriculture, with the proviso
that the expected proportion is zero when the index is negative, and unity when the index is greater
than 1. Most of the variables are of the form: proportion of the census tract with a given
characteristic or location. Spatial patterns of model predictions, in Map 14, can be compared to
actual proportion of agriculture in Map 13.
State dummies were introduced to allow for policy differences between states, but these coefficients
also capture some biogeophysical effects, weakening the measured effects of natural vegetation. The
most striking divergences among states were a -0.13 differential for Amapa, a +0.06 effect for Mato
Grosso and a +0.12 differential for Tocantins, compared to a clustering of the other states. When
state dummies are excluded, location in the cerrado is found to have a strong positive effect on
agricultural use.
Probably the single most important influence on current agricultural use was proximity to pre-1976
clearing. Other things equal, location in an area that had been cleared by 1976 boosts the index of
current agricultural use by 0.43; location in the 50 km band outside the limits of 1976 clearing boosts
the index by 0.27; location in the 50 to 200 km band boosts the index by about 0.2. These effects
may to a large extent represent road impacts, especially the impacts of secondary roads not
otherwise included as explanatory variables. In addition, it is possible that the first spots to be
cleared were attractive for reasons not captured in our explanatory variables. To the extent that this
9 We considered a census tract polygon as urban if ten or more census tracts had to be combined into one polygon for
mapping purposes, due to the small size of the census tracts. We believed the reason for the small size was that the high
population density.
Page 10



Geographical Patterns of Land Use and Land Intensioy in the Brar-4ian AmaZon
is true, early clearing is a proxy for an unmeasured land characteristic. However, the strong effect of
early clearing very likely represents a dynamic process of settlement and natural 'sprawl' of
development.
Precipitation has a strong, highly significant negative effect, all else constant. For example, consider
a location in Para that has typical soils, is more than 25 km from a principal road, between 100 and
200 km from the nearest location that was already cleared in 1976, and is far from cities. With 1600
mm of rain, the predicted proportion of a census tract in agriculture is 22 percent. That proportion
drops to 8 percent at 2000 mm of rain and nearly 0 percent at 2300 mm.
Proximity to principal roads has a surprisingly mild measured effect. Location within 50 km of a
good quality principal road boosts the agricultural use index by about 0.06. Location within 25 km
of a poor quality road boosts the index also by 0.06. Location at 25 to 50 km from a poor quality
principal road appears to have a perverse effect, but this probably reflects the coarseness of road
buffer widths'0 and the correlation between the proportion of the census tract at 0 to 25 and 25 to
50 km.
We believe that these coefficients understate the impact of roads, for several reasons. First, we took
a very conservative approach to road inclusion, using only those principal roads which, arguably,
could be taken as exogenous causal drivers of land use change.  We excluded secondary roads
because some of them may represent responses to agricultural development, rather than causes.
However, this exclusion is almost certainly too severe - many of these roads did in fact stimulate
subsequent agricultural development, and for this reason the measured road effect is
underestimated. Second, we did not adjust for the length of time that the road has been in place.
Recently-constructed roads will have less measured impact. Third, most road effects probably occur
within 25 km. Because the census tracts are relatively large, and because registration of the census
tract boundaries is subject to some error, road impacts may be obscured. Impacts might be much
easier to measure using remote sensing data with 30 meter resolution. Finally, as noted above, the
impact of roads may be confounded with the impact of prior clearing.  When the variables
measuring earlier settlement are removed from the regression (right column of Table 5), the road
coefficients are boosted by about 50 percent.
Areas near rivers had somewhat lower agricultural use, other things equal - but riverine associations
with soil types may complicate the interpretation of this finding. Soils in general have detectable but
mild impacts on agricultural use, with more marked impacts in a few rather unusual soil types.
Protected area status does appear to substantially and significantly deter agricultural use; the effect is
not simply due to the location of protected areas in more remote or less agriculturally attractive
areas.11
Small cities - those of 25,000 to 100,000 population - have a very strong impact on agricultural use
of surrounding areas, boosting the use index by 0.23 for areas within 50 km. Larger cities actually
had smaller impacts, perhaps because land is converted to settlements and hence does not appear in
the agricultural use measure.
10 That is, the effect of poor roads may extend only 10 to 15 kilometers; the regression tries to take account of this
through a weighted average of the two available buffer categories (0-25 and 25-50 km), which are highly correlated.
11 It is possible, of course, that protected areas have been situated in agriculturally unattractive areas, and that this is not
detected by our available measures of agroclimatic suitability; see Cropper and Puri (1999)
Page 1 1



Geo,graphical Patterns of Land Use and Land Intensi_y in the Branibian Ama.Zon
We tested the sensitivity of the results to several alternative specifications. The first column of
Table 5 shows a traditional tobit analysis. The results are qualitatively very similar; the main
difference is a diminished impact of pre-1976 clearing. There is still a 0.40 drop in the index as
rainfall increases from 1,300 mm to 2,500 mm, with a leveling off thereafter.'2 Table 6 repeats the
regression with an alternative specification for precipitation. The alternative has a linear term in
mean precipitation, a dummy for a 'dry season' - at least two consecutive months of less than 50
mm rainfall - and an interaction between these two variables. It tells the same story as the previous
regression: at low levels of precipitation, where most areas have at least two dry months (see Table
7), the index drops rapidly with increasing precipitation. Above about 2,800 mm, where most areas
have no more than 1 dry month, the index continues to decline with increased precipitation, but
more slowly.
Table 9 shows the original regressions reproduced for a different dependent variable: proportion of
the census tract in pasture. The results are qualitatively very similar to those for all kinds of
agriculture. The effect of precipitation is now even steeper; the two quadratic terms are jointly very
significant, but receive low Z-statistics because the combined effect is nearly linear.
The deterrent effect of precipitation on agriculture is quite powerful, according to these estimates.
Using the coefficients of Table 5, little immediate clearing is expected on soils characterized by "low
nutrient holding capacity", where precipitation exceeds 1,800 mm even in the presence of a road -
unless the location is near a previous clearing or close to a small city. Clearing may occur over time
in the vicinity of previous agriculture or near small cities. Even in the presence of these stimuli,
cleating declines asymptotically to zero as rainfall increases.
Because the model crudely incorporates dynamics, it can be used to predict, in an indicative way,
patterns of future agricultural expansion. We do so by assuming that dynamics have not changed
appreciably since 1976. This is a very strong, perhaps untenable, assumption. Clearly, for instance,
various government programs, including fiscal incentives, subsidized credit, and colonization
programs, have been discontinued during the past decade. In addition, the drastic reduction in
inflation may have removed the incentive to claim land as an inflation hedge. Nonetheless, for
exploratory purposes we use the estimated coefficient for "cleating in 1976" to multiply the variable
"clearing in 1987."  The resultant predictions might be taken as representing predicted land use
around 2006. The precise date should not be taken seriously, given our relative ignorance of
dynamics; what is more important is the geographical pattern of areas predicted to be at risk. The
predicted patterns in Map 15 should be compared to the predicted pattern for 1995 shown in Map
14. Despite the caveats that attach to this prediction, it is instructive because it predicts relatively
little sprawl outward from the nuclei of clearing along the principal roads in western Amazonas: the
predicted agrocirmatic deterrent is too great. In contrast, the predictions show substantial forest loss
in drier areas such as southwest Pari, and northwest Mato Grosso. It should be stressed that these
predictions do not take into account the fire dynamics that have been so well described by the
IPAM/WHRC (Nepstad, et al, 1999) research team, but the predicted areas of clearance overlap
with areas at risk for the spread of fire. The predictions also do not allow for the effects of road
construction or improvement. Both these factors might lead to greater clearing than predicted in
the drier areas.
12 There is a slight upturn in the index after about 3000 mm, almnost surely an artifact of the functional form given the
few observations in this range.
Page 12



Geographical Patterns of Land Use and Land Intensiy in the Brazilian Ama-<on
Analysis of stocking rate
To examine the determinants of land value, we concentrate on the stocking ratio as an objective,
easily understood proxy. The analysis proceeds in two steps:
* What determines the location of commercially oriented pasture (proxied by mean pasture size
greater than five hectares)?
* Within these areas, what are the determinants of the stocking ratio?
We set this up as a sample-selection problem:
r =X,B+u
y*  Zy+e
y* >0 => y= 1; y <�0  'y= 0
where r is the natural logarithm of the stocking ratio,_y   1 is an indicator that pasture exists and
mean pasture size is greater than 5 hectares, u and e are unobserved, possibly correlated,
disturbances, and the stocking ratio equation is estimated only wheny = 1. Correlation of the
disturbances allows for the possibility that areas with pasture greater than five hectares may be
systematically different from other areas, controlling for observed variables. We specify that the
presence of protected areas affects the likelihood of finding large pastures (as opposed to finding
small pastures or none at all) but does not affect the stocking rate on converted land. A maximum
likelihood estimate of the sample-selection model did not reject the hypothesis of independence
between the two equations. That is, we can estimate the stocking ratio equation on areas with mean
pasture greater than 5 hectares without the necessity of a sample selection adjustment, and can
impute the predicted values outside the sample.
Table 10 shows alternative estimates of the stocking ratio equation. The first column includes farm
size and the ratio of unpaid labor to farmn area as explanatory variables. Holding agroclimatic
conditions constant, farm size is strongly negatively correlated with stocking rate'3. A 10 percent
increase in farm size reduces the stocking rate by about 1.6 percent; a 10 percent increase in the ratio
of family labor to agricultural land increases the stocking rate by about 0.4 percent. Assuming
unpaid family labor does not increase with farm size, a 50 hectare farm is predicted to have a
stocking rate 65 percent higher than a 500 hectare farm.
Other things equal, location in the cerrado decreases stocking rates by 38 percent; location in
Tocantins decreases stocking rates by a similar factor. Proximity to good-quality principal roads
(within 25 kin) boosts the stocking rate by about 10 percent; there is no statistically significant
impact at greater distances, or from poor-quality principal roads. However, proximity to clearing at
1976 has a large, statistically significant effect. Areas that had been cleared by that date have stocking
rates about 47 percent higher than otherwise comparable areas; the effect persists, at slightly lower
magnitudes, out to 200 km from the boundary of the 1976 clearing. This is an encouraging sign that
pasture use intensifies over time. But the coefficient on past clearing may also capture road and
matket access impacts. Location near a medium-sized city has a negligible measured effect on the
stocking density, as does location near roads. Location near a large city tends to substantially reduce
the stocking rate. This is surprising, given the presumed effect of urban demand on dairy farming,
13 Of course, farn size and labor use may tiemselves respond to agroclimatic and market variables. For this reason an
alternative specification excludes these variables as endogenous.
Page 13



Geographical Patterns of Land Use and Land Intensiy in the Bra!ti1ian Amazon
and requires further investigation, but it may simply reflect the poor agroclimatic conditions
surrounding Manaus and Belem.
Holding these and other factors constant, a 1,000 mm increase in precipitation decreases stocking
rates by about 38 percent. Consider a 500 hectare farm in Pari with the characteristics noted in the
previous example. At 1,600 mm, the predicted stocking rate is 0.38; at 2,000 mm, the stocking rate
is 0.31; and at 2,300, 0.27.
The remaining columns show the results of alternative specifications. Dropping the state dummies
(column 2) intensifies the impact of cerrado location, of higher precipitation, and of prior clearing.
Dropping the farm size and labor variables (arguably endogenous) reduces the effect of
precipitation, but a 1,000 mm increase still reduces stocking by a factor of 28 percent   Cerrado
location maintains its depressing impact in this specification.
Conclusions
Findings
Principal findings are as follows:
Most agricultural land in Ama.Zoniayields little private economic value.
Nearly 90% of agricultural land in the Amazon is either devoted to pasture or has been out of use
for more than four years. About 40%/0 of the currently-utilized pasture has a stocking ratio of less
than 0.5 cattle/hectare, with a mean of about 0.3; the remainder has a mean stocking ratio of 0.95.
Farm level studies suggest that most of this extensive pasture yield very low private returns to the
landholder. This is consistent with data from the agricultural census showing negligible net revenue
for most of the Legal Amazon. However, the costs of forest conversion to society are potentially
large. Clearing is associated with large-scale runaway fires that impose substantial costs to Brazil in
respiratory disease, disruption of economic activity, and damage to timber, pastures, crops, and
fencing. Clearing and associated fires may trigger local climate changes; it has been suggested that in
dry years, smoke from fires could inhibit rainfall, triggering prolonged regional droughts (IPAM,
2000). And clearing imposes national and global costs through loss of biodiversity and emissions of
greenhouse gases.
Land in Amaro6nia is overwhelming concentrated in large properties.
Almost half of Amazonian farmland is located in the one percent of properties than have more than
2,000 hectares. This snapshot of current landholding patterns is consistent with remote sensing
measures of deforestation, which show that 52 percent of total clearing occurs in individual patches
of more than 100 hectares (INPE, 1999). This suggests that the modest private gains associated
with agriculture in the Amazon accrue mostly to large landholders. There is also evidence that,
other things equal, larger landholders utilize pasture at a substantially lower stocking rate than
smaller ones.
Land in the vey moist regions of the western Ama.Zon has been extremely unattractive for agricultural development as
currenty and historically practiced.
Multivariate analysis shows that the probability that land is currently claimed, or used for agriculture,
or intensively stocked with cattle, declines substantially with increasing precipitation levels, holding
other factors (such as road access) constant. Proxies for land abandonment are higher in high
rainfall areas. This suggests that the returns to agriculture in these regions have been lower than in
Page 14



Geographical Patterns of Land Use and Land Intensity in the Brazilian Amanon
Amaz6nia as a whole. At the very least, we can say that it has been more attractive to develop other
areas first, even controlling for road access. Although it is possible that spatial patterns of
development reflect geographically targeted development plans, the climatic story is consistent with
agronomic hypotheses about the effect of high rainfall levels and short dry seasons on production.
There are indications, however, that some of the high-precipitation areas might be suitable for
agroforestry or some kinds of perennial crops.
Much cerrado land is also used with very low intensiy.
Other things equal, stocking rates are very low in cerrado areas.
Land use intensifies over time.
An encouraging finding is that the frontier is not 'hollow.' In general, areas near medium sized cities
and older settlements have both higher rates of overall agricultural use, higher proportions of area in
active pasture, and higher stocking rates on pasture.
Discussion
These findings suggest a number of issues for discussion. Deforestation in the eastern Amazon has
led overwhelmingly to the creation of low-productivity extensive pasture, and available evidence
suggests that replication of this strategy in the West would be even less successful. It is possible, of
course, that new technologies and institutions could provide favorable models for agricultural
development in the Western Amazon, and there are indications that perennial cultivation could be
suitable. However, the analysis of past experience sounds a strong cautionary note: we have no
evidence to suggest that large-scale pasture or grain cultivation will be successful in the wetter
Western regions. It implies that the agricultural opportunity cost of maintaining these areas under
forest cover is very low and may easily be outweighed by extractive values or option values of
preservation.
Provision of new roads in these very moist areas might have limited initial impact on clearing,
because of their inherent unattractiveness for agriculture. Over the long run, as communities form
along these roads, clearing would increase except in the most humid areas, fragmenting the forest
and disrupting biological processes.
The analysis draws particular attention to the potential impact of road-building in drier areas. In
these areas, roads will have a larger immediate effect on forest conversion and cerrado use, and are
more likely to trigger a dynamic process of settlement and clearance. As shown in Nepstad et al
(1999), clearance in these drier areas is more likely to result in runaway fires.  In addition, the
cerrado in these areas may be more biologically unique, and more threatened, than more moist
forest areas. While some of these areas offer relatively high agricultural returns - especially around
medium sized cities and in places suitable for soybeans - others are destined for pasture with very
low stocking rates.
The proximity of higher and lower value land uses in the drier areas of the Amazon raises interesting
policy issues. From the viewpoints both of fire prevention and biodiversity conservation, mosaic
patterns of agriculture and forest reserves may be undesirable. This pattern could increase fire
susceptibility and result in fragmented habitat. Policies that restrict agriculture to suitable areas
might therefore be socially preferable.
In this context, the current discussion of 'relocation' of legal reserve obligations is of interest. There
has been a long-standing obligation of landowners in Brazil to maintain 20 percent of each property
Page 15



Geographical Patterns of Land Use and Land Intensiy in the Braz!lian AmaZon
(higher in the Legal Amazon) in natural vegetation, as a legal forest reserve (Chomitz, 1999;
Bernardes, 1999). Recent discussions, as well as policy innovations in the states of Minas Gerais and
Parana, focus on allowing landowners to achieve compliance by mamitaining a forest reserve on a
remote property with similar biological features. In principle, this kind of trade can greatly reduce
the costs of achieving the desired aggregate forest reserve (Chomitz, 1999). However, the success of
this policy depends on:
* the attractiveness to agriculture of both the buying and selling properties. If properties that are not attractive
to conversion are allowed to sell legal reserve to properties under severe pressure for conversion,
total deforestation will decrease relative to enforcement of the property-by-property rule.
*  the substitutabiliy, for environmental purposes, of natmral vegetation on the two properties. Under what
conditions would we be willing to accept conservation of one forest as compensation for loss of
another? There is no easy answer. Tighter restrictions on substitutability (e.g., restricting
compensation to within a microwatershed) provides a surer guarantee of representivity of
biological features - but restricts the possible gains from specializing agricultural production in
the most suitable areas.
The analyses presented in this paper provide a starting point for examining the implications of
alternative ways to implement compensation mechanisms for the legal forest reserve.
References
Alves, Di6genes. 1999. "An Analysis of the Geographical Patterns of Deforestation in Brazilian
Amaz6nia in the 1991-1996 period." Prepared for the Conference on Patterns and Processes of
Land Use and Forest Change in the Amazon, Center for Latin American Studies, University of
Florida. March 23-26, 1999.
Angelsen, Arild and David Kaimowitz. 1998. "Rethinking the Causes of Deforestation: Lessons
from Economic Models," World Bank Research Observer 14(1):73-98, February.
Anselin, Luc. 1999. "Spatial Economettics," working paper, Bruton Center, University of Texas at
Dallas.
Arimna, Eugenio Yatsuda and Christopher Uhl. 1997. "Ranching in the Brazilian Amazon in a
National Context: Economics, Policy, and Practice," Sociey and Natural Resources 10:433-451.
Bernardes, Aline. 1999. "Some mechanisms for biodiversity protection in Brazil, with emphasis on
their application in the State of Minas Gerais." Prepared for Brazil Global Overlay Project,
DECRG, World Bank, Washington, D.C.
Buchinsky, Moshe. 1994. "Changes in the U.S. Wage Structure 1963-1987: Application of Quantile
Regression," Econometfica 62:405-458.
Chomitz, Kenneth M. 1999. "Transferable Development Rights And Forest Protection: An
Exploratory Analysis". Paper prepared for Workshop on Market-Based Instruments, July 18-20
1999, Kennedy School of Government. World Bank: Development Research Gtoup.
Page 16



Geographical Patterns of Land Use and Land Intensi_y in the Brazilian Amazon
Chomitz, Kenneth M. and David Gray. 1996. "Roads, Land Use, and Deforestation: A Spatial
Model Applied to Belize," World Bank Economic Review 10(3), September.
Deaton, Angus. 1997. The Analysis of Household Sarveys. Washington, D.C.: Johns Hopkins
University Press for the World Bank.
Eswaran, H. and P. Reich. GlobalMajor Land Resource Stresses Map (Unpublished). USDA-NRCS, Soil
Surv-ey Division, World Soil Resources. Washington, D.C.
IBGE (Instituto Brasileiro de Geografia e Estatistica). 1998. Censo Agropecu4rio 1995-1996: nAmero 2
(Rondonia), nu.mero 3 (Am, Roraima e Amapa), numem 4 (Ama.Zonas), nAmnero 5 (Pard), numero 6
(Iocantins), nAzmero 7 (Maranhdo), and n4mero 24 (Mato Grosso). Rio de Janeiro: IBGE.
INPE (Instituto Nacional de Pesquisas Espacias). 1999. Monitoring of the Brazilian Amazonian Forest
by Satellite (http://www.inpe.br/Informacoes_Eventos/amz/index.htm).
IPAM (Instituto de Pesquisa Ambiental da Amag.8nia). 2000. "Avanfa Brasil: Os C(usos Ambientais Para a
Amaoarnia" (http//www.ipam.org.br).
Ministerio da Agricultura (Brasil), Secretaria de Plangamento e Coordenacao da Presidencia da Republica (Brasil),
and IBGE. 1988. Mapa de Vegetacao do Brasil.
Nepstad, Daniel C., Adalberto Verissimo, Ane Alencar, Carlos NObre, Eirivelthon Lima, Paul
Lefebvre, Peter Schlesinger, Christopher Potter, Paulo Moutinho, Elsa Mendoza, Mark
Cochrane, and Vanessa Brooks. 1999. "Large-Scale Impoverishment of Amazonian Forests by
Logging and Fire," Nature 398, April 8.
Pfaff, Alexander S. P. 1997. "What Drives Deforestation in the Brazilian Amazon? Evidence from
Satellite and Socioeconomic Data." World Bank Policy Research Working Paper no. 1772,
Washington, D.C.
Powell, James L. 1984. "Least Absolute Deviations Estimation for the Censored Regression
Model," Journal of Econometrics 25:303-325.
Powell, James L. 1986. "Symmetrically Trimmed Least Squares Estimation for Tobit Models,"
Econometrica 54:1435-1460.
Reis, Eustaquio J. and Sergio Margulis. 1991. "Options for Slowing Amazon Jungle Clearing," in
Global Warming: Economic Policy Responses, ed. by Rudiger Dombusch and James M. Poterba.
Cambridge, MA: MIT.
Schneider. 1995. "Government and the Economy on the Amazon Frontier," World Bank
Environment Paper, no. 11, Washington, D.C.
Sombroek, Wim G. 1999. "Annual Rainfall And Dry-Season Strength In The Amazon Region And
Their Environmental Consequences," presented at the Seminar on Ecologic-Economic Zoning
and Environmental Management, held December 1999 in Manaus, Brazil.
Page 17



Geographical Patterns of Land Use and Land Intensi_y in the Bra#ilian AmaZon
Vosti, Stephen A., Julie Witcover, and Chantal Line Carpentier. 1998. "Arresting Deforestation and
Resource Degradation in the Forest Margins of the Hutnid Tropics: Policy, Technology, and
Institutional Options for Western Brazil". Draft of IFPRI Working Paper, IFPRI, Washington,
D.C.
Page 18



Geographical Patterns of Land Use and Land Intensi_y in the Brazjlian AmaZon
Table 1. Main sources of revenue for farm establishments by state
23,788     3,183      2.1%      107,201   Manioc        30.6%   Cattle         12.4%   Cow's milk     8.9%   Maize           5.2%
3,349       700      4.9%       68,872   Silviculture   51.2%   Cattle        11.7%   Manioc         9.2%   Oil palm        5.8%
83,289     3,323      2.1%      366,496   Manioc        48.8%   Banana         10.5%
368,191    12,561     37.6%      698,163   Rice          15.8%   Cattle         14.4%   Manioc         8.1%   Cow's milk    7.9%
78,762    49,840     55.0%    1,990,221   Soybeans    36.8%   Cattle           17.5%   Sugarcane    10.2%   Maize            5.7%
.; i;    206,404    22,520     18.0%    1,026,711   Manioc        15.4%   Cattle        14.5%   Logs           9.4%   Cow's milk    7.4%
76,956     8,890     37.2%      334,210   Cow's milk   18.3%   Coffee          16.0%   Cattle        13.4%   Beans           6.6%
7,476     2,977     13.2%       62,084   Rice          16.8%   Cow's milk   10.6%   Manioc           6.8%    Maize          6.1%
hsml  44,913    16,766      60.2%      356,366   Cattle        37.9%    Rice          13.3%   Cow's milk    10.2%
M1~mw    893,128   120,759         18.6%    5,010,324   Cattle       16.2%   Soybeans       15.3%   Manioc         9.2%   Cow's milk    6.9%
Notes;
1) "Cattle" and "Chicken" are values of sales plus value of slaughtered, minus purchases.
2) Silviculture is used to mean all type of tree plantations (other than fruit, coffee, and cocoa).
3) Percent is based on total value of agricultural production as given in Table 23 of the Agricultural Census.
4) The values in this table are for the entire state, not just the portion of the state within the Legal Amazon.
Page 19



Geographical Patterns of Land Use and Land Intensiy in the BrasiIian A'na-n
Table 2. Main sources of revenue for farm establishments by size of farm
16,873         0      NA        16,041   Babagu       53.7%   Charcoal       15.2%   Logs           9.2%
II@.   154,502       89   96.9%       145,500   Manioc        17.8%   Rice          14.2%   Babagu          8.1%   Charcoal       5.4%
106,671       147   94.6%       127,784   Manioc        23.0%   Rice          17.7%   Babacu          5.9%
.5   101,017       310   85.4%       226,266   Manioc        29.9%   Rice           9.9%   Banana          6.7%
0    54,514        376   71.4%       167,014   Manioc       29.3%   Banana          7.7%   Chickens       5.1%
64,028       867   66.2%       216,836   Manioc       27.3%   Cow's milk    6.3%   Banana           5.2%
139,442     4,521   62.9%       447,840   Manioc        22.6%   Cow's milk   11.7%   Rice             7.5%   Cattle         5.5%
110,063     7,305   55.8%       445,456   Cow's milk   15.7%   Manioc         14.7%   Cattle         9.9%   Rice            7.5%
74,001     9,130   55.1%       411,641   Cow's milk    17.2%   Cattle        14.6%   Manioc        11.5%   Rice            6.3%
39,434    11,892   63.4%       445,454   CaKtle        20.7%   Cow's milk   13.5%   Soybeans       12.1%   Rice            5.2%
14,869    10,276   64.9%       399,941   Soybeans    27.3%   Cattle          21.6%   Cow's milk    6.5%   Sugarcane        6.5%
8,793    12,202   65.8%       496,439   Soybeans    34.2%   Cattle          25.4%   Maize          5.4%   Sugarcane       5.1%
5,829    17,492   62.3%       597,268   Soybeans    34.2%   Cattle          21.9%   Sugarcane      8.3%
1,843    12,701   56.2%       350,049   Soybeans    33.7%   Cattle          26.4%   Rice           6.3%
1,218    26,094   45.6%      455,256   Cattle        28.1%   Soybeans    22.8%   Sugarcane    19.9%   Rice                7.6%
31     7,358   17.6%        61,539   Silviculture   64.2%   Sugarcane    11.3%   Cattle         10.9%
*   893,128   120,759   55.3%    5,010,324   Cattle         16.2%   Soybeans    15.3%   Manioc            9.2%   Cow's milk      6.9%
Notes:
1) "Cattle" and "Chicken" are values of sales plus value of slaughtered, minus purchases.
2) Silviculture is used to mean all type of tree plantations (other than fruit, coffee, and cocoa).
3) Percent is based on total value of agricultural production as given in Table 23 of the Agricultural Census.
4) The values in this table are for the entire state, not just the portion of the state within the Legal Amazon.
Page 20



Geographical Patterns of Land Use and Land Intensi_y in the Brajlian Amaaqon
Table 3. Main sources of agricultural revenue by rainfall category
Area of Mean size Percent of        Gross                Ranking of gross value of production (top 3;
Number of        farm    of farm     area in    value of
Number        farm  establish- establish-       farm  agricultural          First                 Second                   Third
of Muni- establish-    ments       ments  establih- production                   Percent                 Percent                 Percent
ciplos     ments  (O0Os ha)        (ha)    ments (OOOs reais)   Product         of total  Product       of total  Product      of total
561    663,638    116,054        175      24.0%    3,534,820   Cattle         21.9%    Soybeans    21.7%    Manioc            12.1%
f6 1   wil 14   11,731      6,935        591      64.0%      114,909   Cattle          65.6%    Sugarcane      6.8%    Maize           6.6%
>     .>      168    112,890      39,744        352      55.0%    1,127,283   Soybeans       32.9%    Cattle         27.3%    Sugarcane      9.9%
r;g-^t   129    130,414     35,243        270      35.0%    1,072,935   Soybeans        36.9%    Cattle         19.7%    Sugarcane    10.6%
9 3    ; 8   63    121,339  17,459       144     21.0%       357,613   Cattle         28.0%    Manioc         15.0%    Rice          12.3%
:4    >70    101,641         8,046         79       7.0%      266,346   Manioc          38.6%    Cattle        12.9%    Bananas         8.3%
30     54,702      3,030         55       7.0%      163,219   Manioc         45.2%    Cattle         12.1%    Logs           10.5%
41     88,583      4,008         45      11.0%      262,524   Manioc          36.7%    Logs          20.4%    Cattle          7.0%
42     38,524      1,381         36      15.0%      149,905   Chickens        33.6%    Manioc        27.0%    Bananas         6.8%
2      1,061        197        185       3.0%        4,158   Manioc         46.5%    Bananas        25.5%    Cattle         16.1%
2      2,753         11          4       0.0%       15,927   Manioc         71.3%    Bananas        10.8%    Pineapples      9.1%
Notes:
1) "Cattle" and "Chicken" are values of sales plus value of slaughtered, minus purchases.
2) The values in this table are for municipios in the Legal Amazon of Brazil with centroids west of 45W.
3) Percent of area in farms is based on sector data, and is therefore on approximate for municipios.
4) PJercent is oaseo on rotai vaiue oT agncultural production as avaiiaDie at ne municipio level. i nis exciuaeo milK ana various commodities proouceo in small percentages. i ne value oT
agricultural production east of 45W is approximately 21 9m reais. The value of milk production is approximately 347m reais. The small percentage commodities and other exclusions total
approximately lb reais.
Page 21



Geographical Patterns of Land Use and Land Intensioy in the Brazilian A          Zagon
Table 4. Overview of study area
* §W ~~~~~~~~~,20   8, '53.58  7374  101]244  81297N 11368   46-,744  29,460  10,998    1 8336   739    3,314 486 ,4i9   6,25  4927491
m W ~~~~~~~~~~97  11,116 106,688 140,321 122,414 105,694  53,629  81,307  37,869   4,410   1,758                          378 665,681 127,069 792,750
41.4%   63.8%   55.0%   35.5%   20.8%    7.1%    7.1%   10.8%   15.1%    3.5%    0.1%    0.0%   23.9%   41.2%   24.1%
% |225%   25.2%   26.4%   46.7%   63.5%   62.6%   56.1%   48.2%   29.1%   22.4%   16.5%    2.0%   42.0%   18.8%   41.5%
~\    646%   67.8%   69.0%   50.9%   34.6%   35.3%   40.9%   47.6%   67.2%   69.8%   75.1%   84.7%   54.5%   77.7%   55.0%
m  l |    N0A2%    1.5%    4.0%    5.2%    1.9%    2.5%    3.3%    4.5%    3.5%    1.7%   50.1%   54.3%    3.8%    7.5%    3.9%
090%    0.9%    0.3%    0.6%    1.2%    1.8%    1.1%    3.2%    2.6%    1.6%   19.9%   29.1%    0.8%    0.8%    0.8%
g64%   60.7%   56.6%   39.5%   27.2%   22.0%   25.4%   18.1%   41.0%   63.2%    1.9%    0.2%   42.7%   41.4%   42.6%
00%    0.0%    0.1%    0.2%    0.2%    1.8%    0.7%    0.2%    0.1%    0.0%    0.0%    0.0%    0.3%    0.1%    0.3%
0.% 1.2%    1.9%    1.5%    1.4%    1.9%    2.3%    5.7%    3.4%    0.2%    1.6%    0.0%    1.8%   11.5%    2.0%
NA    7.6%   1.7%    3.2%   236%    64.%    817%    2.18%   1845%   6.4%   81.%   9.9%    19.%   10.%   1.3%
&s,6@.k. s .,2^.g. .XtY  y,jez is g , .Census tracts with ano poron closer thn 25 km from a principal road
^. 2 ~~~~~~~~~~0                      10      859      707........ ...    . 48  45415  8292    58 316741  16 45     3 2   10 3,09200       42 3,429
NA    7.6%   13.7%   23.2/°2631%   16.8%   17.4%    3.1    12.8% 8 6 294%   81772 648  948.9    19.7 93193
97~~~~~~~~~~~~ 1,030,120 761,488 3,74852,32 839,2404 532,187 14,225193,1209 21,050  28                            9      13083( 357,3591  9070448,001
)S_   41.4%NA   64.2%   58.9%   43.4%   18.7%   10.4%   10.0%   17.0%   32.7%    1.5%    0.0%    0.1%   31.5%   44.8%   31.8%
NA   24.9%   26.3%   46.5%   53.2%   62.5%   43.3%   28.1%   34.1%   43.3%    0.4%    0.5%   38.6%   18.9%   38.0%
NA   68.3%   69.2%   51.1%   44.9%   35.3%   53.8%   67.8%   61.5%   55.2%   92.7%   85.0%   57.9%   77.7%   58.5%
NA    2.1%    4.7%    5.9%    3.0%    2.0%    3.1%    5.5%    3.7%    2.8%   67.9%   50.1%    4.5%    7.0%    4.6%
NA    1.2%    0.3%    0.7%    1.7%    1.3%    0.8%    2.2%    3.4%    1.7%   22.6%   34.4%    0.8%    0.6%    0.8%
NA   59.2%   56.0%   39.0%   33.6%   23.2%   38.5%   29.7%   31.3%   48.4%    0.0%    0.5%   45.0%   42.6%   44.9%
PUd j       l_.       ......NA    0.1%    0.2%    0.2%    0.4%    2.7%    0.7%    0.3%    0.2%    0.0%    0.0%    0.0%    0.4%    0.1%    0.4%
NA    1.5%    1.9%    1.6%    2.3%    1.4%    2.1%    7.3%    4.3%    0.0%    0.0%    0.0%    1.9%   11.6%    2.2%
NA    4.3%    6.1%    3.7%    3.9%    4.7%    8.6%   22.8%   18.5%    2.3%    2.2%    0.0%    5.3%   15.7%    5.7%
Census tracts with no portion closer than 25 km from a Principal road
P4 25         415     623      392     582     316      481     169       45      32       10   3,094     200   3,294
to~~~I#r      ) ~~~~1,620   2,720  25,700  50,868  41,620  63,167  32,383  24,264   8,389   4,772   6,482   2,948 264,932   1,488 266,420
of totl ~y~J 8.5%    7.4%   15.8%   35.9%   33.1%   23.8%   23.6%   25.8%    6.5%   33.3%   81.2%   82.2%   28.4%    0.0%   28.2%
~ ~~j~~~g ~~97   1,004  30,200  61,658  39,174  52,326  39,404  62,187  16,819   3,609   1,664                 248 308,390  36,359 344,749
41.4%   62.9%   47.7%   27.6%   22.8%    4.6%    5.9%    9.4%    9.7%    5.0%    0.1%    0.0%   17.5%   29.8%   17.5%
~      ~~~~ ~~~~22.5%   25.9%   26.7%   46.9%   71.6%   62.7%   65.6%   55.9%   23.7%   17.7%   17.3%    3.0%   47.1%   18.4%   46.8%
6456%   66.8%   68~5%   50.5%   26.4%   35.3%   31.3%   39.8%   73.3%   73.1%   74.2%   84.4%   49.4%   77.7%   49.7%
Anni~~~a1~r~~p~         2%    O23%    2.31%    4.2%    11. 1%    3.3%    3,5%    4.1%    3.3%    1.5%   49.2%   57.0%    2.8%   10.1%    2 8%
Por~~~n~~I~~W          0 %       I1%    .2%!o     0.6   0.8%     2.6%    1.30/   3.6%    1.7%    1.6%   19,8%   25,7%. u .8%         .7%    0.5%
&34%   43. 8%5 37,9,%  41.1% 2/ 72.2%   20.0%A  15.5%   13.7%   51.2%   66.5%    2.0%    0. 0%P  '32. D    20.6%   3 9.1 %
2Total~~~~~iro~0 %    (0.0%    Oi%0  01.2%    01%    0.1%    0.8%    0.1%    0.0%.   0.%    0.0%    0~0%    0I1    0.%                       0.1
0. 3%    0.5%                   CD%.~   17 %    2 7%    2.4%    5.1%    2.6%    0.3%  .7%0/   0.0%    i 65%   111.4%~~ 1.7%c~
P~~fuJv~~~zf ~ ~ifl   5%C 29            02%1     .. .!c 16%    6.7%    7.7%  i13.1 %   14.5%    3.3%    1. 6       1.7%    5.0Cp%   18.9%    5.1%
Note-.establishment area=agricultural area+native forest+ unutilizable area(not shown)
Agricultural land= annuals+perennials+total pasture+tree plantations+fallow+productive nonutilized
Page 22



Geographical Patterns of Land Use and Land Intensiy in the Brazjian AmaZon
Table 5. Regressions on proportion of census tract in agriculture land
LyMnt^;                             FlaPtram    t-stat    Param   t-stat        Param      WIN
State (omitted Rondonia)
Acre                              -0.1323   -5.73       -0.1804   -7.57        -0.1870   -4.68
Amazonas                          -0.0190   -1.13       -0.0927   -6.19        -0.1016   -4.35
Roraima                            0.0179    0.57         0.190    3.48        -0.0764   -3.17
Para                               0.0031    0.21        0.0242    1.47        0.0188    0.95
Amapa                             -0.0633   -1.87        -0.0915   -3.26       -0.1849   -5.32
Tocantins                          0.1213    6.86        0.1251    5.25        0.1543    3.98
Maranhao                          -0.0074   -0.44       -0.0365   -2.19        -0.0252   -0.68
Mato Grosso                        0.0589    3.89        0.0614    2.86        0.0950    3.16
Distance to land cleared bv 1976. proportion of sector in (omitted > 200 km)
Cleared bY 1976                    0.2378    9.96        0.4386   11.52            NA      NA
1 -50 km buffer                    0.0973    4.50        0.2759    8.82            NA      NA
50 - 100 km buffer                 0.0224    1.01        0.2000    5.78            NA      NA
100 - 200 km buffer                0.0085    0.39        0.1869    5.63            NA      NA
Annual rainfall at sector centroid (mm)
Annual                          -1.34E-03  -11.17      -7.87E-04   -3.07     -1.76E-03   -8.26
Annual, squared                  2.64E-07   10.41      1.23E-07    2.19      3.52E-07    7.79
Buffers around principal roads (omitted > 50 km)
Poor qualitv. 0 - 25 km            0.0668    4.41        0.0602    3.54        0.0920    3.38
Poor qualitv, 25 - 50 km           0.0122    0.62       -0.0773   -4.48        -0.0722   -2.21
Good quality. 0 - 25 km            0.0727    6.91        0.0766    5.92        0.0962    4.61
Good qualitv. 25 - 50 km           0.0668    5.58        0.0679    5.01         0.0974    4.31
Buffers around orincinal rivers (omitted is >50 km)
0 - 25 km                         -0.0755   -5.13       -0.1200   -8.71        -0.1245   -4.71
25 - 50 km                        -0.0740   -4.07        -0.0894   -5.69       -0.0804   -2.68
Protected areas of any type,   -0.2219  -15.51          -0.2877  -10.67        -0.3174   -6.59
Primarv limiting factors of soils (omitted "low organic matter")
Seasonal excess water              0.2519    2.69        0.2361    1.31        0.3344    1.67
Minor root restrictinq layer      -0.0693   -2.66       -0.0942   -3.46        -0.0617   -1.70
Impeded drainaqe                  -0.0801   -3.21       -0.0914   -3.39        -0.1103   -4.09
Seasonal moisture stress          -0.0634   -2.75       -0.0694   -2.61        -0.0680   -2.77
Hiqh aluminum                     -0.0101   -0.28       -0.0357   -0.64        -0.0558   -0.93
Excessive nutrient leaching        0.0173    0.59        0.0073    0.21        0.0156    0.38
Low nutrient holdinq capacitv     -0.0931   -4.01       -0.0949   -3.94        -0.1252   -5.55
High P. N. & orqanic retention    -0.0940   -2.32       -0.1285   -2.09        -0.0718   -1.09
Low water holdinq caDacity        -0.0617   -2.34       -0.0490   -1.43        -0.0564   -1.34
Salinity or alkalinitv            -0.1050   -3.23       -0.1341   -3.70        -0.1960   -5.10
Shallow soils                     -0.1631   -3.99       -0.1670   -2.22        -0.1947   -3.44
Buffers around cities with nopulations of 100,000 or more (omitted > 250 km)
0 - 50 km                         -0.1610   -8.34       -0.1492   -6.60        -0.1210   -4.73
50 - 100 km                       -0.0381   -2.65       -0.0435   -2.22        -0.0065   -0.35
100 - 250 km                      -0.0027   -0.28       -0.0166   -1.10        -0.0054   -0.30
Buffers around cities with Populations of 25.000 or more (omitted > 250 km)
0 - 50 km                          0.1628    7.73        0.2279    6.66        0.2970    8.04
50 - 100 km                        0.0850    4.11        0.1347    4.01        0.1751    4.93
100 - 250 km                      -0.0006   -0.03        0.0346    1.05        0.0491    1.17
Vegetation classes (omitted 'forest)
Pioneer                            0.0177    0.77        0.0100    0.52        0.0279    0.85
Cerrado                            0.0206    1.37        0.0396    1.57        -0.0381   -1.09
Cerrado-forest                    -0.0321   -1.84       -0.0550   -2.42        -0.0720   -1.94
Constant                           1.8025   12.75        1.0470    3.59        2.2635    9.38
Notes:
1) Bootstrap t-statistcs are based on 50 repettions.
2) The first iterated quantle converged to a pattem that gave the same parameter estmates every iteration, starting at iteration 15.
3) The second iterated quantile converged to a pattem that repeated identical parameter estimates every 5 iterations, startng at iteration 14
(i.e, 14 and 19 had the same results, 15 and 20, etc.). Following Deaton (p. 90), out of the 5 possible choices, we chose the iteraton with
the highest criteron, which was iteration 18.
4) We dropped low structural stability, campinarana, and forest-campinarana due to the number of non-zero values, which was causing
some iterations of the quantile regression to not converge.
5) Regressions were on those sectors located west of 45 degrees west. Excluded were those with computed areas less than 400 hectares,
and those with ten or more sectors merged together (an indicator of being an urban area).
Page 23



Geographical Patterns of Land Use and Land Intensi_y in the BraAlian Amaazon
Table 6. Regression on proportion of census tract in agricultural land, with dry months
as categorical variable
State (omitted is Rondonia)
Acre                                                                         -0.1957       -7.62
Amazonas                                                                     -0.0880       -4.03
Roraima                                                                      0.1820        3.03
Para                                                                         0.0314        1.52
Amapa                                                                        -0.2375       -0.85
Tocantins                                                                    0.1302        5.83
Maranhao                                                                    -0.0392        -1.70
Mato Grosso                                                                  0.0575        2.72
Distance to land cleared by 1976, proportion of sector in (omitted is > 200 km)
Cleared by 1976                                                              0.4463       14.88
1 - 50 km buffer                                                             0.2688        9.22
50 - 100 km buffer                                                           0.1827        5.91
100 - 200 km buffer                                                          0.1738        5.04
Rainfall at centroid of census tract
At least 2 consecutive dry months (1 if yes, 0 if no)                        0.4007        3.85
Annual (mm)                                                                -1.31E-04       -2.81
Annual * At least 2 consecutive dry months                                 -1.62E-04       -3.58
Buffers around principal roads, proportion of sector in (omitted is > 50 km)
Poor quality, 0 - 25 km                                                      0.0759        3.44
Poor quality, 25 - 50 km                                                    -0.0444        -1.43
Good quality, 0 - 25 km                                                      0.0621        3.86
Good quality, 25 - 50 km                                                     0.0602        3.63
Buffers around principal rivers, proportion of sector in (omifted is > 50 km)
0 - 25 km                                                                   -0.1333        -7.70
25 - 50 km                                                                  -0.0749        -3.08
Protected areas of any type, proportion of sector in                        -0.3062        -3.11
Primary limiting factors of soils, proportion of sector in (omitted is "low organic matter")
Seasonal excess water                                                        0.2671        1.16
Minor root restricting layer                                                -0.0968        -2.49
Impeded drainage                                                            -0.0932        -2.92
Seasonal moisture stress                                                     -0.0744       -2.80
High aluminum                                                                -0.0311       -0.56
Excessive nutrient leaching                                                 -0.0059        -0.19
Low nutrient holding capacity                                               -0.0971        -3.37
High P, N, & organic retention                                              -0.1260        -2.30
Low water holding capacity                                                   -0.0570       -1.80
Salinity or alkalinity                                                       -0.1241       -3.12
Shallow soils                                                               -0.1720        -2.74
Buffers around cities with populations of 100,000 or more, proportion of sector in (omitted is > 250 km)
0 - 50 km                                                                    -0.1550       -6.52
50 - 100 km                                                                 -0.0488        -2.24
100 - 250 km                                                                -0.0074        -0.58
Buffers around cities with populations of 25,000 or more, proportion of sector in (omitted is > 250 km)
0 - 50 km                                                                    0.2425        3.09
50 - 100 km                                                                  0.1451        1.72
100 - 250 km                                                                 0.0459        0.50
Vegetation classes, proportion of sector in (omitted is 'forest)
Pioneer                                                                      0.0354        1.55
Cerrado                                                                      0.0484        1.95
Cerrado-forest                                                              -0.0501        -1.86
Constant                                                                     0.1631        2.09
Notes:
1) Bootstrap t-statistics are based on 50 repettions.
2) The iterated quantile converged at iteraton 21.
3) We dropped low structural stability, campinarana, and forest-campinarana due to the number of non-zero values, which was
causing some iterations of the quantile regression to not converge.
4) Regressions were on those sectors located west of 45 degrees west. Excluded were those with computed areas less than 400
hectares, and those with ten or more sectors merged together (an indicator of being an urban area).
Page 24



Geographical Patterns of Land Use and Land Intensity in the Bra Ifan Ama'aon
Table 7. Count of census tracts by rainfall and consecutive dry months
A_
1 or less    2 to 5
- 14A ;9      0        4
t.4e4.8       1      124
1.861 8      15    1,259
- 1J.G      143    1,187
Z--Z2       371      510
-5 - 2-A    760      273
2.4 -Z6     341       64
2. .2.8     437      159
s8!.4.0     303       37
3.04.62      52        8
3.2 34       35        0
-.--36       11        0
Total   2,469    3,625
A. census tuncs     6,094
Page 25



Geographical Patterns of Land Use and Land Intensit in the Bra#lian Amazon
Table 8. Summary of variables used in census tracV agricultural land regressions
-  i*     i Ci             ilb1
Proportion of sector converted for agriculture       5933       0.3097      0.5662           0      21.0805
Proportion of sector in agricultural establishments  5933       0.2243      0.4829           0      20.9654
State
Rondonia                                             5933       0.1360      0.3428           0           1
Acre                                                 5933       0.0329      0.1783           0            1
Amazonas                                             5933       0.1643      0.3706           0            1
Roraima                                              5933       0.0300      0.1706           0            1
Para                                                 5933       0.2660      0.4419           0            1
Amapa                                                5933       0.0138      0.1168           0            1
Tocantins                                            5933       0.0971      0.2961           0            1
Maranhao                                             5933       0.1038      0.3051           0            1
Mato Grosso                                          5933       0.1561      0.3630           0            1
Distance to land cleared by 1976, proportion of sector in (omitted is > 200 km)
Cleared by 1976                                      5933       0.1435      0.3075           0            1
i - 0 km buffer                                      5933       0.4086      0.4369           0            1
50 - 100 km buffer                                   5933       0.1914      0.3461           0            1
100 - 200 km buffer                                  5933       0.1419      0.3214           0            1
Rainfall
Annual (mm)                                          5933        2,140         395        1,331       3,513
Annual, squared                                      5933    4,735,192    1,798,265    1,771,561   12,300,000
* of consecutive months < 50mm                       5933       1.8716       1.4974          0            5
# of months squared                                  5933       5.7445      5.3462           0           25
1, If # of consecutive months >= 2; 0 otherwise      5933       0.5951      0.4909           0            1
Buffers around principal roads, proporton of sector In
Poor quality, 0 - 25 km                              5933       0.0843      0.2520           0            1
Poor quality, 25 - 50 km                             5933       0.0593      0.1823           0            1
Good quality, 0 - 25 km                              5933       0.2698      0.4086           0            1
Good quality, 25 - 50 km                             5933       0.1938      0.3263           0            1
Buffers around principal rivers, proportion of sector in
0 - 25 km                                            5933       0.1958      0.3777           0            1
25 -0 km                                             5933       0.0738      0.2151           0            1
Protected areas of any type, proportion of sector In  5933      0.1236      0.2828           0            1
Primary limiting factors of soils, proportion of sector In
Low organic matter                                   5933       0.0381      0.1693           0            1
Seasonal excess water                                5933       0.0025      0.0378           0            1
Minor root restricting layer                         5933       0.0747      0.2291           0            1
Low structural stability                             5933       0.0001      0.0072           0       0.5560
Impeded drainage                                     5933       0.0931      0.2629           0            1
Seasonal moisture stress                             5933       0.3124      0.4175           0            1
High aluminum                                        5933       0.0190      0.1202           0            1
Excessive nutrient leaching                          5933       0.0468      0.1875           0            1
Low nutrient holding capacity                        5933       0.2631      0.4014           0            1
High P, N, & organic retention                       5933       0.0161      0.1061           0            1
Low water holding capacity                           5933       0.0869      0.2551           0            1
Salinity or alkalinity                               5933       0.0275      0.1544           0            1
Shallow soils                                        5933       0.0199      0.1074           0            1
Buffers around cities with populatons of 100,000 or morm, proportion of sector in
0 - 50 km                                            5933       0.0553      0.2165           0            1
50-100km                                             5933       0.1137      0.2948           0            1
100 - 250 km                                         5933       0.3868      0.4691           0            1
Buffers around cities with populations of 25,000 or more, proportion of sector in
0 - 50 km                                            5933       0.2631      0.4163           0            1
50 - 100 km                                          5933       0.3202      0.4199           0            1
100 - 250 km                                         5933       0.3128      0.4320           0            1
Vegetation classes, proportion of sector In
Campinarana                                          5933       0.0033      0.0408           0            1
Forest                                               5933       0.6724      0.4374           0       1.0004
Forest-campinarana                                   5933       0.0139      0.1031           0            1
Pioneer                                              5933       0.0481      0.1906           0            1
Cerrado                                              5933       0.1834      0.3598           0       1.0001
Cerrado-forest                                       5933       0.0780      0.2198           0            1
Page 26



Geo,graphical Patterns of Land Use and Land Intensity in the Bra4lian Ama.Zon
Table 9. Regressions on proportion of pasture in census tract
Tobit            Iterated QLJantile    Iterated Quantile
-Variable                           Pararn   t-stat        Param    t-stat        Param    t-stat
S i t   o i d R n o i ).   . ...... .. . .. .    ... ....... ........... ...................... ............ .... .... . . ""   '  
State (omitted RondonJa)
Acre                               -0.1237   -5.84        -0.1544   -6.37        -0.1330   -2.76
Amazonas                           -0.0381    -2.42       -0.2444   -8.00        -0.0609   -3.26
Roraima                             0.0709    2.39         0.2072    5.31        -0.0606   -2.19
Para                               -0.0120   -0.89         0.0131     0.81        0.0158    0.53
Amapa                               0.0000    0.00        -0.2045   -1.99        -0.1160   -1.88
Tocantins                           0.1227    7.58         0.1374    6.33         0.1613    4.54
Maranhao                           -0.0278   -1.80        -0.0733   -3.41        -0.0697   -2.86
Mato Grosso                         0.0807    5.82         0.0769    4.19         0.1182    7.42
Land cleared by 1976. orooortfon of sector in (omitted > 200 km)
0-1 km                              0.1781    7.87         0.4199   13.23            NA       NA
1 - 50 km                           0.1260    6.10         0.3127   13.27            NA       NA
50 - 100 km                         0.0722    3.42         0.2379    8.57            NA       NA
100-200 km                          0.0511    2.40         0.2215    8.43            NA       NA
Annual rainfall at sector centroid (mm)
Annual                           -1.22E-03  -10.26      -6.27E-04   -2.81      -1.04E-03   -1.87
Annual. squared                   2.20E-07    8.59       6.51 E-08    1.32      1.62E-07    1.40
Buffers around principal roads (omitted > 50 km)
Poor qualitv. 0 - 25 km             0.0743    5.29         0.0477    2.64         0.0465    1.72
Poor qualitv. 25 - 50 km            0.0524    2.87        -0.0188   -0.76        -0.0274   -1.07
Good quality, 0 - 25 km             0.0674    6.99         0.0671    5.05         0.0891    7.35
Good cualIty, 25 - 50 km            0.0439    3.99         0.0355    3.01         0.0694    2.90
Buffers around Principal rivers (omitted is 50 km)
0 - 25 km                          -0.0564   -4.08        -0.1253   -6.43        -0.0928   -4.19
25 - 50 km                         -0.0708   -4.13        -0.1057   -7.48        -0.0647   -1.40
Protected  areas of any type.   -0.2281  -16.35           -0.2214   -8.30        -0.2684   -7.55
Primarv limitina factors of soils (omitted 'low organic matter")
Minor root restrictinq laver       -0.0654   -2.79        -0.0948   -3.42        -0.0629   -1.31
Impeded drainaae                   -0.1292   -5.63        -0.1273   -4.91        -0.1470   -3.46
Seasonal moisture stress           -0.0964   -4.71        -0.0822   -4.08        -0.0932   -1.90
Hiah aluminum                       0.0142    0.44         0.0133    0.23        -0.0192   -0.45
Excessive nutrient leachinq         0.0032    0.12         0.0139    0.47        -0.0161    -0.24
Low nutrient holdinq capacitv      -0.1083   -5.18        -0.0519   -2.55        -0.1299   -4.98
High P. N, & oraanic retention     -0.1784   -4.89        -0.2232   -5.26        -0.1824   -2.12
Low water holdinq caDacitv         -0.0794   -3.36        -0.0466   -1.74        -0.0530   -1.04
Salinity or alkalinitv             -0.0872   -2.91        -0.2002   -5.92        -0.2284   -8.12
Shallow soils                      -0.1759   -4.57        -0.1293   -3.02        -0.1513   -2.19
Buffers around cities with populations of 100.000 or more (omitted > 250 km)
0 - 50 km                          -0.1559   -8.79        -0.1843   -5.63        -0.1683   -3.74
50 - 100 km                        -0.0327   -2.46        -0.0462   -2.61        -0.0301    -0.55
100 - 250 km                        0.0049    0.55        -0.0128   -0.91        -0.0037   -0.13
Buffers around cities with populations of 25,000 or more (omitted > 250 km)
0 - 50 km                           0.1266    6.44         0.1710    5.80         0.2192    4.03
50 - 100 km                         0.0610    3.16         0.1006    3.71         0.1131    2.51
100 - 250 km                       -0.0210   -1.15        -0.0066   -0.25        -0.0143   -0.34
Vecetation classes (omitted "forest")
Pioneer                             0.0258    1.20         0.0510    1.70         0.0429    1.81
Cerrado                            -0.0169   -1.23        -0.0008   -0.03        -0.0679   -1.24
Cerrado-forest                     -0.0392   -2.45        -0.0748   -3.31        -0.0942   -1.83
Constant                            1.6929   12.37         0.9003    3.66         1.5994    2.67
Notes:
1) Bootstrap t-statistics are based on 50 repetions.
2) The first iterated quantile did not completely converge in 40 iterations.
3) The second iterated quantile converged after 26 iterations.
4) We dropped low structural stability and seasonal excess water as soil limiting factors, and campinarana and forest-campinarana as
veaetation tvDes. due to the number of non-zero values. which was causina some iterations of the ouantile rearession to not converce.
5) Regressions were on those sectors located west of 45 degrees west Excluded were those with computed areas less than 400 hectares,
and those with ten or more sectors meraed tooether fan indicator of beina an urban area).
Page 27



Geographical Patterns of Land Use and Land Intensft in the Bra#lian AmaZon
Table 10. Regressions on natural log of stocking density (cattle per total pasture area)
State (omitted is Rondonia)
Acre                                     0.0401     0.55                                0.1493     1.94
Amazonas                                 0.0413     0.69                                0.1969     3.10
Roraima                                  -0.6718    -5.73                              -0.7628    -6.14
Para                                    -0.3319    -7.12                               -0.3242    -6.56
Amapa                                   -1.2828   -11.77                               -1.3467   -11.65
Tocantins                                -0.4717    -8.72                              -0.5085    -9.01
Maranhao                                 -0.4465    -8.21                              -0.2446    -4.32
Mato Grosso                              -0.0067    -0.14                              -0.2058    -4.22
Distance to land cleared by 1976, proportion of sector in (omitted is > 200 km)
Cleared by 1976                          0.3866     4.30        0.5901     8.44         0.5500     5.82
1-50km buffer                            0.2994     3.63        0.5821    10.00         0.3597     4.11
50 - 100 km buffer                       0.2530     3.02        0.4752     7.66         0.3174     3.57
100 - 200 km buffer                      0.3274     3.79        0.5054     7.64         0.3901     4.25
Ln(labor/ ha of cleared land)            0.0437     2.48        0.0827     4.69
Ln(mean farm establishment size)        -0.1685    -9.36        -0.1235    -6.81
Annual rainfall (mm)                   -4.93E-04    -8.77     -6.51E-04   -11.89      -3.29E-04    -5.55
Buffers around principal roads, proportion of sector in (omitted is > 50 km)
Poor quality, 0-25 km                    0.0394     0.81        -0.0253    -0.54        0.0131     0.26
Poor quality, 25 - 50 km                 0.0510     0.80        -0.0412    -0.64        0.0446     0.66
Good quality, 0 - 25 km                  0.0970     2.97        0.0734     2.22         0.0984     2.84
Good quality, 25 - 50 km                 0.0191     0.50        -0.0175    -0.45        0.0337     0.84
Buffers around principal rivers, proportion of sector in (omitted is > 50 km)
0 - 25 km                                0.0541     1.06        0.0544     1.10         0.1475     2.74
25 -50 km                                0.1442     2.32        0.1579     2.61         0.1476     2.24
Primary limiting factors of soils, proportion of sector in (omifted "low organic matter")
Seasonal excess water                    -0.4579    -1.76       -0.2063    -0.77       -0.5568    -2.02
Minor root restricting layer            -0.2113    -2.49        -0.2516    -2.93       -0.3059    -3.40
Impeded drainage                         -0.2666    -3.11       -0.1406    -1.60       -0.1751    -1.93
Seasonal moisture stress                 0.0417     0.58        0.0078     0.11         0.0210     0.27
High aluminum                            0.2965     2.71        -0.0315    -0.28        0.1319     1.14
Excessive nutrient leaching              0.0989     1.09        0.2316     2.59         0.0875     0.91
Low nutrient holding capacity            0.0911     1.22        0.1009     1.35         0.1017     1.29
High P, N, & organic retention           0.2092     1.66        0.4444     3.51        -0.0475    -0.36
Low water holding capacity              -0.0963    -1.18        -0.1363    -1.63       -0.1162    -1.34
Salinity or alkalinity                   -0.0696    -0.65       -0.0570    -0.52       -0.1164    -1.02
Shallow soils                            0.0060     0.05        -0.0247    -0.18       -0.0454    -0.32
Buffers around cities with populations of 100,000 or more, proportion of sector in (omitted is > 250 km)
0 - 50 km                                -0.3494    -5.77       -0.4202    -6.90       -0.3299    -5.13
60 - 100 km                              -0.1235    -2.65       -0.2525    -5.60       -0.1371    -2.77
100 - 250 km                             -0.0493    -1.60       -0.1240    -4.24       -0.0266    -0.81
Buffers around cities with populations of 25,000 or more, proportion of sector in (omitted is > 250 km)
0 - 50 km                                0.0681     0.92         0.0101    0.14         0.0916     1.17
50 - 100 km                              -0.0267    -0.37       -0.1265    -1.79       -0.0444    -0.58
100 - 250 km                            -0.1565    -2.21        -0.2647    -3.82       -0.2454    -3.26
Vegetation classes, proportion of sector in (omiffed is "forest")
Pioneer                                  0.0256     0.32        -0.1380    -1.66        0.0585     0.68
Cerrado                                 -0.4810   -10.90        -0.6708   -16.12       -0.5943   -12.85
Cerrado-forest                          -0.1436    -2.70        -0.1846    -3.40       -0.1860    -3.30
Constant                                 1.8409    10.61         1.8456    11.61        0.4879     2.84
Notes:
1) We dropped low structural stability, campinarana, and forest-campinarana due to the number of non-zero values.
2) Regressions were on those census tracts west of 45 degrees W. We exduded census tracts with less than 400 hectares total, and those with 10
or more sectors merged together (indicator of being an urban area).
Page 28



Geo,graphical Patterns of Land Use and Land Intensity in the Bra#ihIan Amayon
Table 11. Summary of variables used in stocking density regression
Number of
CtZl;Fi:lx;                                   observations       Mean    Std. Dev.         Min         Max
Natlai log of cattle per hectare of pasture)         4407      -0.2476      0.8938     -7.3620       2.2765
Cattle per hectare of pasture                        4407       1.0582      0.8561      0.0006       9.7426
Natural log of adult unpaid labor on farm            4407      -3.4501      1.8019    -11.2243       2.6589
Adult unpaid labor on farm                           4407       0.1168      0.3741      0.0000      14.2806
Natural log of mean farm size                        4407       5.0037      1.6552      0.1764      12.7827
Mean farm size                                       4407         956        7,157      1.1929      356,000
State
Rondonia                                             4407       0.1536      0.3606           0           1
Acre                                                 4407       0.0372      0.1893           0           1
Amazonas                                             4407       0.0917      0.2886           0           1
Roraima                                              4407       0.0254      0.1574           0           1
Para                                                 4407       0.2532      0.4349           0           1
Amapa                                                4407       0.0134      0.1149           0           1
Tocantins                                            4407       0.1246      0.3303           0           1
Maranhao                                             4407       0.1135      0.3172           0           1
Mato Grosso                                          4407       0.1874      0.3903           0           1
Distance to land cleared by 1976, proportion of sector in (omitted is > 200 km)
Cleared by 1976                                      4407       0.1538      0.3130           0           1
1 - 50 km buffer                                     4407       0.4633      0.4391           0           1
50-100 km buffer                                     4407       0.1987      0.3522           0           1
100 - 200 km buffer                                  4407       0.1259      0.3077           0           1
Rainfall
Annual (mm)                                          4407        2,059         366        1,372       3,372
Annual, squared                                      4407    4,375,412    1,628,559    1,882,384   11,400,000
Buffers around principal roads, proportion of sector in
Poor quality, 0 - 25 km                              4407       0.0925      0.2634           0           1
Poorquality, 25-50 km                                4407       0.0616      0.1864           0           1
Good quality, 0 - 25 km                              4407       0.3327      0.4303           0           1
Good quality, 25 - 50 km                             4407       0.2217      0.3371           0           1
Buffers around principal rivers, proportion of sector in
0 - 25 km                                            4407       0.1562      0.3455           0       1.0001
255-50km                                             4407       0.0692      0.2128           0       1.0001
Primary limiting factors of soils, proportion of sector in
Low organic matter                                   4407       0.0444      0.1811           0           1
Seasonal excess water                                4407       0.0034      0.0439           0           1
Minor root restricting layer                         4407       0.0703      0.2231           0           1
Low structural stability                             4407       0.0001      0.0084           0       0.5560
Impeded drainage                                     4407       0.0574      0.2070           0           1
Seasonal moisture stress                             4407       0.3719      0.4332           0           1
High aluminum                                        4407       0.0234      0.1326           0           1
Excessive nutrient leaching                          4407       0.0557      0.2058           0           1
Low nutrient holding capacity                        4407       0.2083      0.3702           0           1
High P, N, & organic retention                       4407       0.0181      0.1105           0           1
Low water holding capacity                           4407       0.1057      0.2785           0           1
Salinity or alkalinity                               4407       0.0243      0.1430           0           1
Shallow soils                                        4407       0.0170      0.0975           0           1
Buffers around cities with populations of 100,000 or more, proportion of sector in
0 - 50 km                                            4407       0.0617      0.2262           0           1
50 -100 km                                           4407       0.1243      0.3050           0           1
100 - 250 km                                         4407       0.3965      0.4720           0           1
Buffers around cities with populations of 25,000 or more, proportion of sector in
0 - 50 km                                            4407       0.3010      0.4312           0           1
50 -100 km                                           4407       0.3526      0.4257           0           1
100 - 250 km                                         4407       0.3010      0.4279           0           1
Vegetation classes, proportion of sector in
Campinarana                                          4407       0.0008      0.0164           0           1
Forest                                               4407       0.6305      0.4506           0           1
Forest-campinarana                                   4407       0.0053      0.0621           0           1
Pioneer                                              4407       0.0428      0.1783           0           1
Cerrado                                              4407       0.2288      0.3901           0           1
Cerrado-forest                                       4407       0.0912      0.2331           0           1
Page 29



Map 1. Mean annual rainfall, 1970 to 19961
=     State boidaries
Annual rai (mm)
400-1200
1200 - 100
1600-1800
2000 -2200
2200 -240D
2600 -2800
2800 -3200
2800 -320D
3600 -6729
No Data
Map 2. Primary limiting factors of soils
= State boundaries
Primary lim'iiin fafcoo ofo sil
6 Low o,Uanc rnatter
7 Seasonal excess water
8 Minor root rstricting layer
-10 Low structural sabbflty
12 Impeded draiunage
13 Seasonal moisttuae stress
14 High alwni um
16 Excessive nutrient leaching
1T Low nutrient holding capacity
-10 High P, N, & organic ratentioni
21 Low water holing capacit
23 Salkltylsikainity
~  24 Shallow sofa
500             0             500           1000 Kilometers
'Maps in color are available on the Policy Research Working Papers Web site.



Map 3. Underlying vegetation types
v  State boundaries
Major vegtation types
Campinarana
Forest
Forestcarnpinara na 
Pioree m,, -we.'''  
Corrado 
_Cerrado-forest_. 
No Data 
S00           0           Soo           1000 Kilomneters 
Map 4. Proportion of establishment area in total area of census tract
Cities I Cldades
y  > 600,ODO
100,000 -5 00,000
25,000 -1 00,000
* w25,000
State boundaries
Principal roads
Principal rivers
Fann areal sector area
, 0.02
E  0.02-0.1                                                                                  R
0. 1.  0.2
0.2- 0.                                _
0.5-.
500           0            500          1000 Klomneters
INIO*N'



Map 5. Land with some type of protected status
=  State boundaries
Areas with protected status
Conservation arma
ENational parks
Indigenous areas
*  Protected areas
Combination of two or more
600         0          600         1000 Kllonutems



Map 6. Stocking density (cattle per hectare of pasture)
z State boundaries
LJS than 5 ha pasture per cattle fafm
Cattle per ha of pasture
< OA
OA - 0.8
OA8- 1.2
1.2-.2
2. 10
~>10
jMssing data
'                  ^'X' -& # ...  W
500          0           500          1000 KIlometrstX
Map 7. Land cleared by 1976,1987, and 1991
L     State boundaries
Cleared land, by year
1976
_   1987
,1991
500          0            500         1000 KIlometers
.  - ' N [~~~~~iii



Map 8. Proportion of productive but unutilized land in cleared area
C:: $tae boundaries
No converted lend
Productive but vWilzed
=c 0.06
&d0.05.0.2                                ,
0.2 -0.4
> OA
Missing data
S00            0             S00           1000 Kilometers



Map 9. Proportion of natural pasture in cleared area
EJState boundarles
No converted iand
Natural pasture In convertedland          *
0.3-.
m  0.6 -Ci
E>~ 0.8
Missing data
500            0             oo500        1000 Klomters
Map 10. Proportion of planted pasture in cleared area
No onvrtd land
Planted pasture In converted land
(03
am 0.3- 0.6
0.6 -0.8
0.8
Missing data
500            0            50D           1000 KIlometers



Map 11. Proportion of total pasture in cleared area
m  State boundaries
No converted land
Total pasture in converted land
< 0.6
0.6 . 0.35
0.85 - 0.95
>. 0.95
EMissing datai
So0           0            500           1000 Kilometers
Map 12. Mean farm size
[    State boundaries
No land in farms
Mean farm size (hectares)                         i6
0 -20
U20 - 100
-100 -400
400 - 2000
2000 -750650
MAssing data
500           0            500           1000 Kilometers



Map 13. Proportion of cleared land in census tract
State boundaries                                                            X     fi    %
Proportion cleared, 1995                                                          i..........   
0.01   0.05                                                g t 
0.2 -0.5 
0.5-08                                                                           .-          U
'0.8;-1.1                                                                       s        f'>
Missing data                                                                       r  
300          0           500         1000 Kilometers



Map 14. Proportion cleared in 1995 (predicted by model)
:   State boundaries
Proportnio  cleared, 1995 prediction
-0.0.01
-0.01 . 0.05
0.05- 0.2
0.2 - 0.5
0.5 - 0.8
N Missing data
500           0            500           1000 Kilometers
Map 15. Proportion cleared in 2006 (predicted by model)
1 State boundaries
Proportionl cleared, 2006 prediction
~0- 0.01
0.01 - 0.05
0.05 .0.2
0.2 .0.5
0.5 - 0.8
0.8.-1
_M sing data
500           0            500           1000 Kilometers



Policy Research Working Paper Series
Contact
Title                            Author                   Date              for paper
WPS2667 Trade Reform and Household Welfare: Elena lanchovichina    August 2001        L. Tabada
The Case of Mexico               Alessandro Nicita                          36896
Isidro Soloaga
WPS2668 Comparative Life Expectancy in Africa F. Desmond McCarthy  August 2001        H. Sladovich
Holger Wolf                                37698
WPS2669 The Impact of NAFTA and Options for Jorge Martinez-Vazquez   September 2001   S. Everhart
Tax Reform in Mexico             Duanjie Chen                               30128
WPS2670 Stock Markets, Banks, and Growth:  Thorsten Beck           September 2001     A. Yaptenco
Correlation or Causality?        Ross Levine                                31823
WPS2671 Who Participates? The Supply of    Norbert R. Schady       September 2001     T. Gomez
Volunteer Labor and the Distribution                                        32127
of Government Programs in Rural Peru
WPS2672 Do Workfare Participants Recover    Martin Ravallion       September 2001     C. Cunanan
Quickly from Retrenchment?       Emanuela Galasso                           32301
Teodoro Lazo
Ernesto Philipp
WPS2673 Pollution Havens and Foreign Direct  Beata K. Smarzynska   September 2001     L. Tabada
Investment: Dirty Secret or Popular  Shang-Jin Wei                          36896
Myth?
WPS2674 Measuring Economic Downside        Yan Wang                September 2001     A. Rivas
Risk and Severity: Growth at Risk    Yudong Yao                             36270
WPS2675 Road Infrastructure Concession     Franck Bousquet         September 2001     G. Chenet-Smith
Practice in Europe               Alain Fayard                               36370
WPS2676 An Alternative Unifying Measure of   Philippe Auffret      September 2001     K. Tomlinson
Welfare Gains from Risk-Sharing                                             39763
WPS2677Can Local Institutions Reduce Poverty? Paula Donnelly-Roark  September 2001    E. Hornsby
Rural Decentralization in Burkina Faso Karim Ouedraogo                      33375
Xiao Ye
WPS2678 Emerging Markets Instability: Do   Graciela Kaminsky       September 2001     E. Khine
Sovereign Ratings Affect Country    Sergio Schmukler                        37471
Risk and Stock Returns?
WPS2679 "Deposit Insurance Around the Globe: Asl1 Demirgu,c-Kunt   September 2001     K. Labrie
Where Does It Work?              Edward J. Kane                             31001
WPS2680 International Cartel Enforcement:    Simon J. Evenett      September 2001     L. Tabada
Lessons from the 1990s           Margaret C. Levenstein                     36896
Valerie Y. Suslow



Policy Research Working Paper Series
Contact
Title                            Author                   Date              for paper
WPS2681 On the Duration of Civil War       Paul Collier            September 2001     P. Collier
Anke Hoeffler                              88208
Mans S6derbom
WPS2682 Deposit Insurance and Financial    Robert Cull             September 2001     K. Labrie
Development                      Lemma W. Senbet                            31001
Marco Sorge
WPS2683 Financial Policies and the Prevention  Frederic S. Mishkin  October 2001      R. Vo
of Financial Crises in Emerging                                             33722
Market Economies
WPS2684 From Monetary Targeting to Inflation  Frederic S. Mishkin  October 2001       R. Vo
Targeting: Lessons from Industrialized                                      33722
Countries
WPS2685 Monetary Policy Strategies for     Frederic S. Mishkin     October 2001       R. Vo
Latin America                    Miguel A. Savastano                        33722
WPS2686 Education, Earnings, and Inequality  Andreas Blom          October 2001       S. Benbouzid
in Brazil, 1982-98: Implications for   Lauritz Holm-Nielsen                 88469
Education Policy                 Dorte Verner