Imprimir Republish


The art of scanning clouds

Three satellites, Landsat, Cbers and Terra, monitor the Amazon Region's controversial deforestation

JEFFERSON RUDY/MMASelective exploitation of timber, without suitable management, leaves the soil exposed. For Prodes this is not vegetation clearance but for Deter this is an illegal actJEFFERSON RUDY/MMA

Since the second half of the 20th century, deforestation in the Amazon Region picked up speed, as the agricultural frontier expanded, roads were built and regional development centers set up. In the late 1970s, after the Transamazon highway was opened and at the request of the Superintendent’s Office for Development of the Amazon (Sudam), Inpe, the National Space Research Institute, took the first account of the damage done with the help of images taken by Landsat-1, a satellite belonging to Nasa, the North American space agency. It saw that 2.5% of the region had already been degraded.

Ten years later, when the region’s deliberate forest fires began drawing international attention ( smoke over the forest is visible from the porthole of the Columbia space shuttle) and the level of deforestation had risen to 10%, Brazil decided to embrace a deforestation control policy and in 1988 created Prodes, the Amazon Deforestation Monitoring Program.

Over 20 years, and at the cost of the devastation of 17% of the Amazon region’s biome, Brazil has developed a system for tracking the approximately 5 million sq. km of the legally-defined Brazilian Amazon region, a system which has been qualified as “enviable” by the journal Science, volume 316, April 27, 2007. In this period, monitoring technology evolved from an analogue system, with a resolution of 1:500,000 and images in black and white, to a digital system, with a scale of 1:250,000. “From the 1990s we started developing software and no longer monitoring  by using visual interpretation, but image processing,” recalls Dalton de Morisson Valeriano, coordinator of Inpe’s Amazon Program.

The raw material of this analysis are images sent by Landsat-5 and by the Chinese-Brazilian satellite, Cbers, which have a spatial resolution of 30 meters. The three bands of the two satellites capture the quantity of radiation reflected from the surface of the soil, discriminated by spectrum band – near visible red light and medium infrared light. This information, which when duly modeled allows green plants, soil, shade and water to be identified, is processed by logarithms that give the proportion of each of these components in each pixel of the image.

The algorithm for transforming the data from the satellites (developed in the 1980s by Yosio Shimabukuro, an Inpe researcher) follows the so-called linear mixture model; “‘linear’ because it starts from the assumption that the elements that comprise the image are independent and ‘mixture’ because down there they really are all mixed up,” explains Valeriano. In this image classification, smooth and homogenous vegetation corresponds to a light green shade; the forest appears as dark green because it has a stronger shadow component; soil is pinkish, because it mixes blue and red, and water is black. The image analysis method is based on segmentation, classification and editing, in such a way as to identify vegetation clearance covering an area as small as 6.25 hectares.

INPECbers SatelliteINPE

This information forms a map that has 229 digital images, each corresponding to a 160 x 160 km area. Of the total, 213 images are interpreted, because Prodes does not evaluate the status of the Cerrado [savannah]. “The Cerrado catches fire and becomes black, like deforested areas. The following year it becomes Cerrado again and this leads to a large number of mistakes,” explains Valeriano.

The data is processed using geo-referenced information software, called Spring, developed for Inpe by IBM and by the Brazilian Agriculture and Livestock Farming Research Company (Embrapa), and are interpreted and audited by specialists from Inpe. “Once this has been done, the information is placed under a mask of the past and the area that has been cleared of vegetation is identified,” sums up Valeriano. The conclusions of each batch of information collected are published every year, in December.

It was through Prodes that the planet learned of the record deforestation of 1995, when 29,100 sq. km of forest were destroyed, and saw that the situation was still out of control in 2004, when the Amazon lost 27,400 sq. km of its biome. The last report showed a 20% reduction in vegetation clearance in  2006-2007 vs. 2005-2006.

Since 2003, “in favor of transparency,” the Prodes information has been put on the Internet. “The measure had an interesting impact,” relates Valeriano. For example, it overthrew some of the “theses,” such as the inefficiency of the conservation areas for containing vegetation clearance or that indigenous areas were degraded. “We showed that this was not true,” he states, indicating the example of the Marabá region in Pará, where deforested areas surround Indian land. In 2005 the rate fell by 30% and in 2006 Inpe recorded a reduction of 14,000 sq. km of forest.

Despite being “enviable,” Prodes has its limitations. One of them is the amount of cloud-cover in the region. “The observation window for the Amazon runs from May to October. Outside this period it’s difficult to see because it’s all cloudy,” explains Valeriano. Another problem is the risk of dependence on Landsat- 5, a satellite launched in 1984 to operate for five years, but that is still active after almost a quarter century. “The answer is to resort to images from Cbers too,” comments Valeriano.

Landsat crosses the Amazon every 16 days and Cbers, every 26. Inpe also buys data from DMC (a consortium of micro-satellites led by England), from Spot and ResourceSat. These satellites function as a type of back-up for information collected by Prodes and they are also used to “sweep the clouds,” as he says. “When the amount of vegetation  deforestation has been calculated it’s also necessary to estimate how much of the forest it was impossible to observe because of clouds and to project the rate of deforestation in the area,” says Valeriano. The percentage of “non-observation,” as he highlights, varies from 10% to 15%.

The third problem is that Prodes data is a “posthumous image” of the forest and of the vegetation cleared; it does not provide a basis for government action to preempt the deforestation dynamics. It was to fill this gap that in 2004 Inpe started the Real Time Deforestation Detection project (Deter), which uses images from the Modis sensor of the Terra satellite launched in 1999 as part of the Earth Observing System (EOS) program, and the WFI sensor installed aboard Cbers- 2, to provide information every two weeks to Ibama, the Brazilian Institute of the Environment and Natural Renewable Resources “Deter was conceived to aid inspection work,” says Valeriano.

The spatial resolution of the Modis and Cbers sensors (250 meters) is not as good as Landsat’s 30 meters. “With this resolution we’re at the limit of technology for this type of application,” recognizes Valeriano. The main function of Deter, however, is not to calculate deforestation but to “indicate” risk situations to provide information for decisions on what should be inspected, notes the program coordinator. The information is transmitted every two weeks to Ibama and published every month on the Inpe website.

inpeIn the Cbers image, Manaus appears close to the Negro River (black) and the Solimões River (purple). The darker green is the forest and the lighter green and lilac are former forest areasinpe

Deter operated from January to October in 2004 and 2005 with this particular mission, thanks to Inpe resources. In 2006 it hired the Space Science, Applications and Technology Foundation to help to process data. The results so far are compatible with the Prodes information. “In 2006/2007, Prodes indicated a drop in deforestation, from 27,000 to 19,000 sq. km. Deter observed 12,000 of these 19,000 sq. km. It can see between 40% and 60% of what Prodes sees,” explains Valeriano.

In 2007, Deter gathered data from March to December. “We took various images from Modis and observed five out of every fifteen, one every three days, which is the frequency of the satellite cover. We selected and interpreted these images using the linear mixture model and we published the findings in January,” he says. The estimates indicated that from August to December, 2007, degraded areas in the Amazon totaled 3,235 sq. km. As Deter only “reads” some 40% to 60% of the vegetation clearance recorded by Prodes, deforestation may have reached 7,000 sq. km. If this figure is confirmed, vegetation clearance is averaging more than 1,000 sq. km a month.

The news elicited different reactions from various areas of the federal government and from states and the Inpe information was called into question. “Deter’s aim is not to estimate the total area cleared of vegetation in the Amazon,” insists Valeriano. Because of the resolution of the Modis and WFI/Cbers sensors, there is a risk of making a mistake. “To calculate the deforested area, Inpe will continue to use better resolution images from the Landsat and Cbers sensors,” he says. Deter, however, supplies the size of each polygon cleared of vegetation, enabling “inspection work to be prioritized”.

Valeriano explains that besides the spatial resolution of the satellites, the two projects have different “readings” of the vegetation clearance: Prodes identifies areas cut level with the ground – in other words quick clearance –, whereas Deter also captures the areas cleared by selective cutting of timber and recurrent burning, which are in a process of continuous degradation,. “We’re pointing out potential signs of vegetation clearance that are not accounted for as such. To the east of Pará and in Mato Grosso, timber is selectively exploited and the soil is exposed. For Prodes, depending on the amount of soil exposed, this is not necessarily vegetation clearance. But for Deter this is vegetation clearance. It’s a way of converting the forest without cutting everything down to the ground,” he highlights. “If Deter focused only on areas cut down to the ground it wouldn’t be monitoring the dynamics of vegetation clearance and the contribution of the Ministry of Science and Technology in the Interministerial Group for Fighting Vegetation Clearance, which is coordinated by the Civil Office, would be meaningless.”

Inpe is making plans to advance even further with monitoring technology. “We’re negotiating with India to have access to data from ResourceSat. Our plan is to be able to gather information on a multi-time basis, using multiple better resolution sensors of some 50 meters.”

Pixels without clouds
Imazon, the Institute of Man and the Amazon Region Environment operates a Vegetation Clearance Early Warning System(SAD) that uses IDA digital processing (interactive data language) to analyze images from the Modis sensor. The analysis model is the same one Inpe uses. SAD has been monitoring the states of Mato Grosso since 2000 and Pará since last year. Starting this year it will track the entire legally-defined Brazilian Amazon region.

However, Inpe detected a deterioration in vegetation clearance from August to December 2007 and Imazon observed a reduction of 21% in deforestation in Mato Grosso and an increase of 74% in Pará vs. 2006 data. “We work with the same image. Inpe chooses the best image of the period and we manage to filter it by substituting the cloud areas by pixels without clouds,” explains Carlos de Souza Júnior, Imazon researcher and executive secretary. Each pixel corresponds to an area 250 x 250 meters. In forested areas, 60% of the pixels are equivalent to vegetation and 40% are shade. “When there’s vegetation clearance it loses a percentage of the vegetation and signs of the soil appear, which should be zero,” he explains. “If the percentage of vegetation falls to 25% and the soil increases by more than 20%, then this pixel has changed.”

The Imazon measuring system, as he says, is conservative. “There are grey areas,” he admits, referring to burnt areas that are yet to be converted into farmland. “There’s controversy about how to deal with these degraded areas,” he ponders, offering a second explanation for the discrepancy between the data.