ABSTRACT

Tropical deforestation is highest in Latin America compared to Africa and Asia. Settlers were interviewed in two Amazon colonies about land use and rates and causes of deforestation. Farmers in Pedro Peixoto, Acre, cleared about two ha per year per family and settlers in Theobroma, Rondonia, cleared some three ha per year to produce first rice followed by beans, maize, and cassava. Settlers then converted lands to pasture not only to raise cattle, but also as a way to add substantial value to lands for "improvements"--i.e., for more clearing, pasture, fencing, corrals and ponds. GIS analysis of satellite images and cadastral maps of the two colonies provided invaluable data on the dynamics of deforestation--i.e., on deforestation patterns and rates as influenced by distance to roads, wet-season access, land tenure, parcel size, and rates over time.

1. INTRODUCTION

Tropical deforestation, due in part to slash-and-burn agriculture, contributes to global warming via burning and release of CO2 into the atmosphere; and Brazil is now the fourth atmospheric carbon contributor--after the US, ex-Soviet Union, and China (Moran 1993). Deforestation is also leading to losses of genetic (Phillips et al 1994) and cultural diversity. Decreasing transpiration and precipitation within and outside of areas cleared may also be a consequence of deforestation (Fearnside 1985, Salati 1989).

Farmer settlers in the government colonization projects of Pedro Peixoto in the Brazilian Amazonian state of Acre and in Theobroma, Rondonia (Figure 1), were interviewed and GIS analysis was conducted as a part of activities to characterize local land use systems and the dynamics of deforestation. Settlement in such projects has been facilitated by government policies to populate frontier areas in the Amazon, road construction, and direct and indirect subsidies. This paper addresses dynamics of deforestation at the farm level.

The Pedro Peixoto colonisation site, established in 1972, covers 370,000 ha divided into 3700 lots distributed among 3200 families. Lots are located between 50 and 100 km from the state capital of Rio Branco, with the major highway BR364 passing from Rio Branco through the settlement and on to Porto Vehlo in Rondonia.

Theobroma, established in 1979 (albeit spontaneously settled much earlier), officially covers 300,000 ha divided into 3000 lots (reportedly) distributed among 3000 families. The project area is located some 350 km from the state capital of Porto Vehlo, also via BR364 which connects Porto Vehlo and Cuiaba in the south (D'Oliveira, unpublished)

2. METHODS

Farmer-settler interviews. A draft questionnaire was prepared by researchers from CIAT, the Empresa Brasileira de Pesquisa Agropecuaria (EMBRAPA), the International Centre for Research in Agroforestry (ICRAF), and the International Food Policy Research Institute as a first step in a series of characterization research activities, all, in turn, a part of the ICRAF-coordinated project "Alternatives to Slash-and-Burn" (ASB). The questionnaire was field tested and further modified by CIAT, IFPRI, and EMBRAPA researchers to ensure appropriateness to local conditions (similar work being conducted in Cameroon and Indonesia) and to facilitate ease of use in the field and data coding. Eighty-one randomly sampled farmers in Pedro Peixoto and 74 in Theobroma were interviewed in late August and early September 1994. Either or both male and female heads of household were included. We reviewed and cross-checked data, revised the coding system as needed, set up data archives, and entered and tabulated data.

Image selection. Landsat Thematic Mapper (TM) imagery was used in this study. Selection of imagery considered project goals, availabilility, and cost. Unlike SPOT (Multi Spectral, Panchromatic) data, Thematic Mapper, Multi Spectral Scanner (MSS), and ERS1 radar could accomodate either project area within a single scene. Aerial photographs were eliminated from consideration due to the substantial labor required to georeference and mosaic them. Use of MSS data--although available as part of a large historical archive--was discounted due to its spatial resolution of approximately 70 m2 because farmer's parcels ranged between 200 and 450 m in width; and data would not have afforded the detail of landcover desired. Radar imagery was not considered due to the lack of an historical archive.

Scenes were acquired from two sources. Three Pedro Peixoto images from 1984, 1987 and 1992 were obtained from the Pathfinder project at the Universities of Maryland and New Hampshire. The single TM image (1994) of Theobroma was purchased from the Brazilian Instituto Nacional De Pesquisas Espaciais (INPE). As with the Pedro Peixoto images, this was recorded in the dry season.

Cadastral maps. Digital cadastral maps of both areas were obtained from the Instituto Nacional de Colonizacao e Reforma Agraria (INCRA). Lot boundaries, partial road networks, population centres and drainage systems were present in both cadastral maps. The Pedro Peixoto coverage also indicated lots whose owners had obtained land titles prior to 1992. The cadastral maps were georeferenced to a UTM projection.

Ground truthing. Fieldwork was conducted in Pedro Peixoto and Theobroma. We interviewed individual farmers regarding their land management practices. Landcover samples were taken by interviewing farmers and walking lots with a handheld GPS (+/- 100m) while making field sketches.

Image preprocessing. An atmospheric correction (dark pixel subtraction method) was applied to all images to provide comparable classifications. Occasional line dropout was removed using a filter targeted at spurious lines. All images were georeferenced to the cadastral overlay for two reasons. First, topographic map coverage at the 1:100,000 scale was available but had not been updated since 1977. Second, as we wanted analysis at the parcel level it was more important that the imagery was accurately referenced to the cadastral lots than to the national projection.

The number of ground control points (GCPs) used in the correction varied according to relief. Pedro Peixoto, an area of low relief, was georeferenced using 75 GCPs, giving a Root Mean Square (RMS) error of less than one. Due to its slightly more undulating terrain, Theobroma was corrected using 90 GCPs, again to an RMS error of less than one.

Classifications. To quantify rates of deforestation the project required a forest/non-forest classification. Non-forest included grazed and ungrazed pasture, agricultural plots and secondary regrowth. Forest included only forest (a category which included "disturbed" forest). All images were classified to the 95% threshold of assignment using isodata clustering. Twenty five cover classes were created and visually interpreted using the ground truth data into either forest/non-forest. This allowed the study of deforestation over time in Pedro Peixoto. As the Theobroma study required measures of secondary forest regrowth, the non-forest category was split into secondary forest regrowth and land currently worked by the colonist.

A majority filter was run on all classified images to improve accuracy.

3. RESULTS

Land use from farmer interviews. A main research objective was to determine land use and patterns of land use change. The national government granted settlers large parcels of forested land which they cleared for agricultural use, starting with slash-and-burn agriculture. Parcels were a mean 88 ha in Pedro Peixoto and 76 ha in Theobroma. By 1993-94, settlers had cleared a mean 27 ha (31%) of these lands in Pedro Peixoto and 35 ha (46%) in Theobroma (difference not significant at 5% using the t-test). The 31% of cleared land in Pedro Peixoto were divided into a mean 20% (of the total parcel) in pasture, 6% in fallow, and 4% in annual crops. The 46% of cleared lands in Theobroma were divided into 26% pasture, 8% fallow, 7% annual crops, and 5% perennial crops (Table 1).

Changes in land use from 1993-94 to 1994-95 could be calculated because interviews were conducted in late 1994 after field clearing and burning as farmers prepared for the 1994-95 cropping season: deforested portions of the settlers' parcels increased a mean 2.0 ha and from 31% to 34% in Pedro Peixoto and a mean 2.7 ha and from 46% to 50% in Theobroma. Overall, some 40% of the settlers' land in the two colonies has been deforested, with more than half of the cleared area converted to pasture. Only 7% of settlers' lands were in fallow (Table 1).

Table 1.

Land use (mean areas), Pedro Peixoto, Acre (n=81) & Theobroma, Rondonia (n=74), 1993/94 & 1994/95

The interviewed farmers had converted a mean 19 ha in Pedro Peixoto and 30 ha in Theobroma of primary forest at the time of the interviews. A mean 3.3 ha in Pedro Peixoto and 8.5 ha in Theobroma had been cleared at the time of arrival and parcel occupation. As indicated above, farmers had occupied parcels in Pedro Peixoto a mean 9 years and Theobroma a mean 8 years. The rate of primary forest clearing was thus calculated as 1.8 ha per year in Pedro Peixoto and 2.8 ha per year in Theobroma--figures very close to those calculated for 1994-95.

In both colonies, individual farmers' areas cleared and area in pasture were correlated to overall parcel size; and area in pasture was highly correlated to area cleared. For Theobroma, farmers who did not clear forest land in late 1994 had significantly (at the 5% level) more fallow land (8.6 ha) than those clearing (4.0 ha), suggesting fallow use by some Theobroma farmers. Although Pedro Peixoto farmers clearing forest in late 1994 also had less fallow land (5.6 ha) than those not clearing (4.5 ha), the difference was not significant.

Rice was the major crop for both consumption and sales of surpluses at both sites. Farmers (92% in Pedro Peixoto and 70% in Theobroma) planted rice in the first year of cultivation of what was primary forest; and cultivated maize, cassava (in Pedro Peixoto), and pasture in the second year. Rice was not grown in the second or subsequent years of plot use.

Farmers cultivated lands cleared from primary forest for a mean 2.1 (Pedro Peixoto) to 2.5 years (Theobroma). Sixty percent in Pedro Peixoto cultivated such plots for two years; while Theobroma farmers used their newly cleared lands for from one to more than three years in somewhat equal proportions. Farmers reported that discontinuation of annual cropping on lands cleared from forest was due to the not mutually exclusive reasons of lower productivity, weeds--especially Imperata sp (locally sape), and insects and diseases.

After food crops, two-thirds of Pedro Peixoto and nearly half of Theobroma farmers converted their lands to pasture. About a third in both areas left some land in fallow (although much of the "fallow" could also serve as unimproved pasture). Theobroma (20%) but not Pedro Peixoto farmers also converted some land from annual to perennial crop use. Farmers at both sites "normally" left any fields which they fallowed for a mean 2.5 years, although they thought that 3.0-3.5 years of fallow would be ideal. Rice followed by maize and beans were the main crops planted in re-opened fallows.

Cattle and pasture formation are perhaps the major driving force behind deforestation in the settlements. Most settlers (91% in Pedro Peixoto and 81% in Theobroma) had cattle. Herd size was a mean 18 head (with 6 giving milk) in Pedro Peixoto and 26 (4 giving milk) in Theobroma.

Settlers' main source of wealth appears to be the appreciation of the value of their lands due to conversion to pasture: 93% of Pedro Peixoto and 97% of Theobroma settlers perceived their land values as having risen (values were discussed in terms of equivalent numbers of cattle), at annual rates of 74% in the former and 157% in the latter site. Farmers reported total increases since occupying their parcels in value of about 800% in Pedro Peixoto and 950% in Theobroma, with main reasons for increases attributed to addition of pasture or cleared areas, fencing, ponds, and corrals (Table 2).

Table 2.

Respondents' evaluation of and reasons for (% of respondents) changing land values, Pedro Peixoto (n=69) and Theobroma (n=70)

* calculated in terms of numbers of cattle at time of parcel acquisition and at time of interview

GIS: distance to roads and deforestation, Pedro Peixoto. Accessibility of areas has been shown to play an important role in the spatial distribution of deforestation. "Accessibility" in this case was defined in terms of distances to roads and farmers' evaluations of wet-season inaccessibility. No modern road network map existed for Pedro Peixoto. A partial road network was extracted from the cadastral map. The "missing links" were digitised to an accuracy of +/- 30 metres using an edge-enhanced georeferenced TM image, expert knowledge, and the cadastral map. Roads were classified into two categories, the major highway--the trans Amazonian BR364--and project roads, which eventually connect to highway. Distance was measured by buffering the roads. A preliminary study was conducted to assess the optimal buffer distance, a compromise between a high level of detail and a manageable amount of data to process. Two hundred and fifty metres was selected as optimal. Buffering was conducted across the whole study area with the exception of road junctions. Confusion created by proximity to both road classes exists at these locations. Deforestation rates were calculated within the GIS which held both the vector buffer zones and the classified raster images.

From 1984 to 1992, deforestaion increased: a) from 58% to 84% for areas up to 250m away from the main roads; b) from 25% to 48% at distances between two and three kilometres from the main road; an c) at intermediate levels for the intermediate distances. For secondary roads, deforestation similarly increased over the 1984-92 period from 10% to 51% for the area up to 250 m from the road; and from 4% to 13% over the same period and at two to three kilometres from the secondary roads. Intermediate distances (and the intermediate 1987 image) from the secondary roads again provided intermediate values (Figure 2).

GIS: wet-season access and deforestation, Pedro Peixoto. Two hundred and fifty lots were randomly selected for accessibility analysis. With the aid of the Empresa Brasilera de Pesquisa Agropecuario (EMBRAPA) the lots were categorised into those accessible and those inaccessible in the wet season. Again, rates of deforestation by lot type were calculated using the GIS. Deforestation was initially lower in less accessible compared to more accessible lots. In 1984, 5% of the inaccessible compared to 9% of the accessible lots were deforested. By 1992, however, the gap had narrowed to an insignificant difference of 25% of inaccessible vs 27% of accessible lots (Table 3).

Table 3.

Deforestation & wet-season access, Pedro Peixoto

GIS: land tenure and deforestation, Pedro Peixoto. Data regarding parcel ownership was available for 1992. We hypothesized that land ownership has an effect--positive or negative--on deforestation rates. Three hundred and twenty four lots were randomly selected, 160 with titles and 164 without. The 1992 forest/non-forest classification was vectorised and unioned with the cadastral map to give the forest/non-forest boundaries within sampled lots. This coverage was interrogated to give forest/non-forest areas in hectares for each sample lot. Data was assimilated for lots with and without title to give actual hectarages and landcover as a percentage of total lot area.

Deforestation amounts were converted to percentages as a function of lot area. Lots were sorted into percentiles of deforestion according to ownership (Table 4). Differences in levels of deforestation were significant at the 99.9% confidence level for classes of less than 30%, between 30 and 49.9% and over 50% deforestation. Colonists with title by 1992 were clearing a greater percentage of forest than those without. This finding is somewhat contrary to the idea that secure land tenure results in more careful resource management.

Table 4.

Deforestation x land tenure, Pedro Peixoto

GIS: parcel sizes and deforestation, Theobroma. The colonisation scheme is compromised of 100 ha and 50 ha lots. A hypothesis was that the area and extent of primary forest clearing per lot, and its replacement with both secondary regrowth and managed land, would not be affected by lot size. As each lot was occupied by a single family, it was assumed that the same workforce was available for agriculture and forest clearing. Two hundred and fifty nine lots were randomly selected: one hundred and twenty 100 ha lots and one hundred and thirty-nine 50 ha lots. The three-class--primary forest, secondary regrowth and managed lands--classification was converted into a vector coverage and unioned with the cadastral overlay of land parcel boundaries. This coverage was queried and values of areal extent and extent as a percentage of the lot area obtained.

Large lot holders had cleared significantly greater areas than small lot holders; absolute amounts of secondary regrowth were greater for large lots; and amounts of cultivated and pasture land in the large lots was significantly greater than that in small lots. Percents of lot areas under secondary regrowth for large and small lots, however, were not significantly different. Although large lot holders may have cleared their lands at a faster rate, we conclude that they have been clearing at the same rate as small holders, but started sooner. The situation is confounded by access: large lots are also located closer--i.e., are more accessible--to the main highway (Table 5).

Table 5.

Deforestation x parcel size, Theobroma

* All differences except indicated are significant at 0.01 using the student's-t

GIS: deforestation over time, Pedro Peixoto. The time series of classified Pedro Peixoto images was used to create a visual representation of deforestation. By using image subtraction a map indicating land cleared up to 1984, land cleared between 1984 and 1987, and land cleared between 1987 and 1992 was produced (Figure 3).

Analysis tends to confirm rates calculated from farmer-reported data. The images cover an area of 357,000 ha, of which there are 276,000 ha of colonists' parcels, 22,000 ha of haciendas, and 56,000 ha of "other" (the officially reported 370,000 ha of the Pedro Peixoto project thus appears to include haciendas and land uses besides settlers' parcels).

Overall, the percent area cleared was 8.8% in 1984, 12.1% in 1987, and 25.0% in 1992. Colonists' cleared areas increased from 8.1% in 1984, to 13.1% in 1987 and 26.4% cleared in 1992 (Table 6, Figure 4). This latter figure and the cleared area calculated from farmer interviews of 31% in 1993-94 are mutually supporting: both analysis provide annual rates of about 3% and total deforestation in Pedro Peixoto colonists' lots at about 30% in 1993.

The large cattle ranches (fazendas) included in the images increased cleared area from 16.8% to 32.4% from 1984 to 1992. The area cleared on the ranches approximiately doubled over the eight year period (1984-92); while clearing on the colonists' parcels increased by more than 300% for the same period (Table 6).

Table 6.

Forest clearing (satellite image analysis), Pedro Peixoto, 1994, 1987, and 1992.

For per parcel deforestation in Pedro Peixoto, images show that there was an increase from 22,388 ha cleared in 1984 to 72,970 ha cleared in 1992 for 3,141 farmer's parcels. That is, a total of 6,323 ha were cleared per year over the eight year period, or a rate of 2.0 ha cleared per parcel per year. This rate again corresponds to the calculation based on farmer-reported data.

Satellite image analysis was also used to determine the range and distribution of deforestation by deciles (i.e., frequency of lots showing 0-9.9%, 10-19.9%, 70% of the settlers' parcels were less than 10% deforested; and only 3% of the parcels were 40% or more cleared. By 1992 only 22% of the parcels remained less than 10% deforested; while another 22% were 40% or more deforested (Figure 5). The obvious and expected trend is that settlers' lots will steadily "move" towards the higher deforestation deciles and away from the lower. Modelling a date when, for example, no lots would remain in the less than 10% deforested and 10% of the parels would be deforested at the 90-100% level is complicated, however, by differential rates of forest clearing as a function of distances from roads and of access and other factors.

4. DISCUSSION

Data acquisition. The greatest difficulty was acquisition of satellite images. Very few cloudfree, or near cloudfree, images were available. Technical problems experienced by the South American receiving stations also dramatically reduced availability of recent imagery. As a result, fieldwork was conducted before the most recent images were acquired; which in turn meant that optimal strategies could not be developed for landcover sampling and farmer interviews (e.g., in areas of high and low deforestation). The problems encountered in image acquisition suggest that for future work, where a long historical archive is not required, radar or the more flexible airborne scanner (ATM) may be appropriate.

5. CONCLUSIONS

Deforestation at the farm level averaged two ha per year per parcel in Pedro Peixoto and three ha per year in Theobroma. Reasons for the significantly higher rate of deforestation in Theobroma were not determined, although the contrasting possibilities that clearing accelerates as the settlements age (Theobroma having been settled earlier) or that colonists decrease rates over time (the interviewed settlers in Theobroma had lived in that settlement for less time than had the Pedro Peixoto settlers) will be investigated in the future.

Settlers still had more than half of their lands in forest and will most likely continue to slash, burn, cultivate, and convert more primary forest as their currently most (economically) viable option. Two main factors driving land clearing at the farm level were the need to produce food and incentives (i.e., in the sense of increased land values) to convert land into pasture. In terms of food production, farmers consumed and sold rice, and to a lesser extent, beans, maize, and cassava (or cassava meal). Rice cultivation may "drive" some deforestation in that although farmers planted cleared fields for up to three years, they could not--for technical reasons--sow rice other than in the first year after clearing.

Farmers were clearly motivated to convert lands cleared from forest into pasture because of real or at least perceived resulting increases in land values. Farmers not only maintained cattle as standing "bank accounts" and obtained cash from sales of animals and milk, but built savings by investing time and resources in pasture, fencing, corrals, and ponds. As observed at the two sites, local ranchers and urban-based speculators have purchased continuous blocks of colonists' parcels to form new ranches or to expand the size of adjacent ranches; and payments were much higher for cleared vs forested portions of parcels.

Finally, we conclude that the combination of GIS analysis with intensive fieldwork (i.e., farmer interviews) provided the best optimal mix of methods to understand the dynamics of deforesation and land use in the Brasilian Amazon colonies studied.

REFERENCES

Fearnside, P.M. 1985. Environmental change and deforestation in the Amazon. In: J. Hemming (Editor), Man's Impacts on Forests and Rivers. Manchester: Manchester University Press.

Moran, E.F., 1993. Deforestation and land use in the Brazilian Amazon. Human Ecology 21:1:1-21.

Phillips, O.L., Hall, P., Gentry, A.H., Sawyer, S.A. and Vasquez, R., 1994. Dynamics and species richness of tropical rain forests. Proceedings of the National Academy of Sciences, USA. 91:2805-2809.

Salati, E., 1989. Deforestation and climatic changes in the Amazon Basin. In: Climate and Food Security. Los Banos, Philippines: International Rice Research Institute and the American Association for the Advancement of Science.