Research uses artificial intelligence to identify priority area for combating deforestation in the Amazon

by

A method developed by Brazilian researchers based on satellite images and artificial intelligence shows that the priority area for actions to combat deforestation could be 27.8% smaller than that of the 11 municipalities currently monitored by the federal government in the Amazon Plan 2021/2022 . This monitoring, however, fails to consider new frontiers of forest clearing, which go beyond the limits of these target municipalities.

Research published in June in Conservation Letters, a magazine of the Society for Conservation Biology, points out that the regions with the highest deforestation rates in the Amazon, classified as “high priority”, cover 414,603 km² this year, compared to the total area included in the plan that , adding up all the municipalities, is 574,724 km². In other words, the area to be monitored would be 160 thousand km² smaller, an extension similar to that of Suriname.

But while the hotspots identified by the researchers accounted for 66% of the average annual rate of forest devastation, the plan’s 11 target municipalities accounted for 37% of the deforestation rate in the last three years (2019 to 2021).

In the article, scientists from the National Institute for Space Research (Inpe) and from American universities conclude that the proposed method, in addition to giving a more defined focus to inspection, reveals new frontiers of forest clearing, currently outside the monitoring plan for extrapolating. the limits of these target municipalities.

“With this new approach, we concluded that there is a gain in effectiveness by prioritizing areas with higher deforestation rates, not limiting by municipalities. This result is important, since more and more inspection bodies, such as Ibama and ICMBio, have suffered with cuts in resources, both in terms of servers and budgets. Some of these devastation hotspots are within the 11 target municipalities, but others are on the borders, opening new frontiers”, the corresponding author of the article Guilherme Augusto Verola Mataveli, told FAPESP, of the Earth Observation and Geoinformatics Division of Inpe.

The work received support from FAPESP through four projects: Influence of land use and land cover on the emission of fine particulate matter from fires in the Amazon and Cerrado biomes, Quantifying tree mortality with lasers, Use of Sentinel-2/MSI dense time series and machine learning algorithms to improve agricultural monitoring in the Cerrado biome and Interannual variation of the greenhouse gas balance in the Amazon Basin and its controls in a world under warming and climate change – Carbam: a long-term study of the carbon balance of the Amazon Basin Amazon.

Sought by Agência FAPESP, the National Council for the Legal Amazon (CNAL), responsible for the 2021/2022 Amazon Plan, informed through its advisory that “the objective [do plano] was to focus on where the occurrence of environmental offenses had the greatest impact on the results of Brazilian environmental management, without neglecting the performance in other areas of the Legal Amazon”. as a higher incidence of fires, including others that may be mapped by the Management and Operational Center of the Amazon Protection System [Censipam]”.

The CNAL also said that Inpe was one of the “protagonist institutions in the process of indicating the priorities established”, but that the scientists who worked on the research could “have contributed in a timely manner in an institutional way”. “The CNAL always works with official information managed, worked on and analyzed by official government agencies,” says the agency.

Advancement in data processing

In the study, the scientists point out that deforestation in the 11 municipalities targeted by the plan has been significant in recent years, motivating monitoring, but this would not be enough to prioritize only these locations. They are: São Félix do Xingu, Altamira, Novo Progresso, Pacajá, Portel, Itaituba and Rurópolis, in the State of Pará, Apuí and Lábrea, in Amazonas, Colniza, in Mato Grosso, and Porto Velho, in Rondônia.

The researchers mention that, even with inspection actions concentrated in the region, there was a 105% increase in the rate of devastation between February and April 2021 compared to the average in previous years (2017-2021). According to data from Deter, Brazil’s official deforestation alert program, there were 524.89 km² of new devastation points in these areas.

“This research validates the importance of Inpe, which for 60 years has been training leading researchers and producing science and technology, based on satellite data, for society and national development. The advances in data processing presented in this study with the use of artificial intelligence for planning the fight against deforestation are critical for the mitigation of national environmental problems and for the construction of a sustainable development plan for the country”, says the head of the Earth Observation and Geoinformatics Division at INPE, Luiz Aragão, and one of the authors of the article.

Priority areas

The group used, among other sources, data from Prodes – Inpe’s Satellite Monitoring Project for Deforestation in the Legal Amazon. Since 1988, Prodes has produced annual deforestation rates in the region used by the Brazilian government to define public policies. These rates are estimated based on the clear cut identified in each satellite image covering the Legal Amazon.

According to the latest Prodes report, the area deforested in the Amazon was 13,235 km² between August 2020 and July 2021, an increase of 22% compared to the previous period (August 2019 to July 2020) and the highest rate since 2006 .

“The idea for the article came up in February 2021, when the 2021/2022 Amazon Plan was launched. At the time, it was reported that deforestation in these 11 municipalities represented 70% of the total recorded in the Amazon, but Prodes was different. model, we realized that we could present a tool to help direct the inspection”, completes Mataveli.

To establish priority areas, the researchers first defined so-called grid cells regularly distributed over the Amazon – regions measuring 25 km² by 25 km². Based on the Random Forest algorithm, which automatically builds sets of multivariate regressions to predict deforestation hotspots in the following year, a priority class was determined for each of these cells. This method identifies a larger fraction of areas at risk of logging in relation to the total region and public lands.

Five predictors were considered: deforestation rates in previous years; the distance to grid cells with high cumulative deforestation in previous years; distance to infrastructure (such as highways and/or waterways); the total area protected in the grid cell and the number of hotspots.

Then there was the definition of three priority classes, based on estimates of predicted deforestation – low (values ​​below the 70th percentile); medium (between the 70th and 90th percentiles) and high (above 90th). The researchers then separated only the grid cells with percentiles above 90 (high) to map the priority areas for 2022, reaching 414,603 km².

The method also allows the definition of priority regions annually using the clear-cut rates of the previous period, not depending on geopolitical boundaries. According to the researchers, among the examples that are outside the 2021/2022 Amazon Plan, but appeared as a “high” priority, are Roraima and Acre.

“Prioritizing these 11 municipalities will be insufficient for Brazil to be able to fulfill international commitments, such as zero illegal deforestation by 2028 assumed at COP-26 [Conferência do Clima das Nações Unidas]. In addition, the 2021/2022 Amazon Plan has a target of combating deforestation of 8,719 km² per year. But a 2018 decree already established a limit of 3,925 km²/year after 2020. In other words, less ambitious”, completes Mataveli.

The researcher refers to decree 9,578, of 2018, which deals with the National Policy on Climate Change and establishes an 80% reduction in annual deforestation rates in the Legal Amazon in relation to the average between 1996 and 2005. This goal is among the actions to be adopted by Brazil to contain greenhouse gas emissions.

At COP-26, in addition to the commitment to eliminate illegal deforestation by 2028, Brazil committed to reducing greenhouse gas emissions by 50% by 2030 compared to 2005 levels and achieving climate neutrality by 2050. However, growing rates of deforestation in the Amazon contrast with these goals – around 11% of greenhouse gas emissions are caused by mismanagement of forests and land use, including deforestation and fires.

When the 2021/2022 Amazon Plan was launched, experts criticized the established targets, calling them insufficient. This is because the government has set itself the objective of reducing deforestation based on the average recorded in the 2016-2020 period, which was already about 35% higher than that of the previous ten years.

Complementary actions

Research suggests that, in addition to more direct methods of targeting public policy, a series of complementary actions are needed to combat the devastation. Among them, he points out environmental education and awareness; the identification and accountability of actors who violate environmental protection laws and profit from illegal deforestation; incentives for projects that invest in actions aimed at the green economy and to keep the forest standing, in addition to the regularization of public and indigenous lands.

“The code we use to generate the model and the priority areas is open. We are talking with the Terra Brasilis platform to try to include these areas in the information available to anyone who wants to access it. So, if any government, including state or municipal, is interested, it is possible to apply it in practice”, concludes Mataveli.

You May Also Like

Recommended for you