Artificial intelligence makes mapping agricultural intensification in the Cerrado more accurate

Pioneering methodology was developed with digital satellite image classification algorithms based on Artificial Intelligence (AI)

12.09.2023 | 13:59 (UTC -3)
Valéria Cristina Costa
HLS image crop, true color composition (Bands 4-3-2), from January 26, 2022 in the municipality of Sorriso (MT)
HLS image crop, true color composition (Bands 4-3-2), from January 26, 2022 in the municipality of Sorriso (MT)

Pioneering methodology, developed with the support of Artificial Intelligence (AI), allowed the achievement of an accuracy level of up to 97%, when applied to analysis of satellite images of the Cerrado in the municipality of Sorriso (MT), one of the main agricultural producers in the region. Country. Accuracy is a relevant aspect in surveys carried out using remote sensing.

The tool provides greater precision to studies, monitoring and planning related to land use and the practice of agricultural intensification, and contributes to decision-making, in the public and private spheres, based on qualified geospatial information.

The methodology was developed with AI-based digital satellite image classification algorithms. It results from the work of researchers from Embrapa, the State University of Campinas (Unicamp), the National Institute for Space Research (Inpe) and the Federal University of Uberlândia (UFU), published in the International Journal of Geo-Information (IJGI), in the July issue 2023, with free access for the general public.

“The results demonstrate the robustness of the methodology developed with a focus on identifying processes of land use dynamics, such as agricultural intensification”, assesses researcher Édson Bolfe, from Embrapa Digital Agriculture, and coordinator of the Agricultural Mapping in the Cerrado project via combination of multisensor images - MultiCER, funded by the São Paulo Research Support Foundation (Fapesp).

Bolfe explains that, among the main differences of the methodology, is the generation of an expanded geospatial database based on harmonized images from the Landsat satellites, from the United States Aerospace Agency (NASA), and Sentinel-2, from the European Space Agency ( ESA), called HLS, and the use of AI-based digital classification algorithms. The approach made it possible to map agricultural crops at three different hierarchical levels, indicating areas with one, two and even three harvests in the same agricultural year.

The succession of harvests of different agricultural crops in the same area and on the same agricultural calendar, aiming to increase production without involving the suppression of new native areas, is a growing practice in Brazil, and its mapping and monitoring can guide decision makers in analyzes focused on agri-environmental planning, in particular.

Methodology is available for consultation and use

Members of academia, public authorities and the productive sector can access details of the methodology, results and maps generated in the Embrapa Research Data Repository (Redape). The methodological structure is replicable in other regions of the Cerrado with similar characteristics. More information about the Field Data Collection Platform is available at AgroTag.

Synthesis of the methodological approach generated for mapping agricultural intensification
Synthesis of the methodological approach generated for mapping agricultural intensification

Agility and precision, the role of AgroTag

Remote sensing products and AI models for pixel-by-pixel image classification have demonstrated high reliability in agricultural mapping, explains Bolfe. With HLS it is possible to obtain up to two images per week in the same agricultural regions of interest.

One of the challenges for the research team is obtaining qualitative and quantitative information from the field, which is fundamental in remote sensing in agriculture. To do this, the researchers used the AgroTag application, developed by Embrapa Meio Ambiente to provide agility and precision to the mapping of the main agricultural crops on regional and national scales.

“AI-based algorithms rely heavily on a massive amount of input data to carry out so-called ‘training’. The latter are processes in which reference sample data, or field truths, are used to teach algorithms to identify targets under investigation in large areas, in this case using satellite images, that is, large-scale mapping”, comments Luiz Eduardo Vicente, researcher at Embrapa Meio Ambiente, specialist in remote sensing and one of the coordinators of the AgroTag project.

In this sense, according to Vicente, the use of AgroTag was fundamental, as it allowed the quick and accurate collection of field information, such as the type of land use and coverage at each sampling point, automatically transferring them to the online data cloud. line, enabling its use in the aforementioned algorithms.

In contrast to traditional collection methods, AgroTag represented, during the project, an increase of 25% more in sampled areas. “The project reaffirms one of the reasons why Agrotag was created”, highlights Vicente.

Examples of agricultural crops produced in the second and third harvests, in the municipality of Sorriso-MT, in June 2022; photo credits: Taya Parreiras, 2022
Examples of agricultural crops produced in the second and third harvests, in the municipality of Sorriso-MT, in June 2022; photo credits: Taya Parreiras, 2022

Dynamic Mappings

“The study mapped agricultural production in 2021-2022 in Sorriso/MT, a municipality chosen for its economic and agro-environmental relevance in the context of the Cerrado and the country”, highlights Edson Sano, researcher at Embrapa Cerrados and member of the MultiCER project.

Most existing mapping does not follow the evolution of “land-saving” agricultural intensification practices - such as the production of up to three crops in the same area - remaining at the level of the first harvest. “Some surveys evolved to identify the number of crops planted, without, however, detecting specific crops”, concludes Sano.

“To produce dynamic, detailed and accurate mappings, a large volume of 'ground truth' information is needed, which are labeled samples of the types of land use or cover, obtained during field activities”, observes Taya Parreiras, a doctoral student at the Institute of Geosciences at Unicamp and member of the MultiCER Project.

According to the researcher, regular time series of satellite images with high temporal resolution are also necessary and, in this sense, the harmonization of Landsat and Sentinel-2 data is a different approach. Taya Parreiras indicates that, to deal with the size of these databases and information, machine learning algorithms, such as Random Forest or Extreme Gradient Boost, are fundamental.

“As part of AI, these algorithms are capable of analyzing and learning complex spectral and textural patterns from extensive and varied agricultural datasets, enabling the precise identification of different crop types, soil conditions and environmental variables,” he argues.

Random Forest, by creating several independent decision trees and combining them, can produce more reliable estimates. Extreme Gradient Boost also creates several decision trees, but with the advantage of allowing those with low predictive power to be adjusted. “Both algorithms are highly scalable, which allows them to process large volumes of data quickly, contributing to the generation of detailed and updated agricultural maps”, he concludes.

Digital classification of land use and cover, with emphasis on second-crop crops in the Municipality of Sorriso (MT) in January 2022, generated from HLS images and algorithms based on artificial intelligence
Digital classification of land use and cover, with emphasis on second-crop crops in the Municipality of Sorriso (MT) in January 2022, generated from HLS images and algorithms based on artificial intelligence
LS Tractor February