Harmonized images from state-of-the-art satellites map soybean areas in the Brazilian Cerrado

A team of researchers from Embrapa, Unicamp and Inpe has been using the data set from the Harmonized Landsat Sentinel-2 program to identify land use and coverage and separate areas with soybean cultivation, Brazil's most important agricultural commodity, from other annual crops found in the region in 2022. This type of approach and results, using remote sensing, can contribute to public policies and the monitoring of agricultural crops on a large scale throughout the year/harvest, in a less costly and more accurate way.

In addition to the effective coverage of the harvest period provided by the images and analysis techniques applied, the ability to map sub-targets was observed, such as identifying soybean plantations under irrigation or rainfed systems more effectively, with 88% accuracy. .

According to Édson Bolfe, researcher at Embrapa Digital Agriculture and coordinator of the Agricultural Mapping in the Cerrado project via combination of multi-sensor images, financed by Fapesp – MultiCER –, the use of HLS data greatly increases the data coverage time (temporal resolution), with up to 7 scenes per month in the same locations, and a spatial resolution, which indicates the smallest unit of area represented, of 30 meters.

“In the study, using these satellites in an integrated manner, we had 13 images with less than 50% cloud coverage, between October and March, for mapping the harvest of summer agricultural crops, in the 2021/-2022 harvest year, in the study area. Areas such as western Bahia, specifically the work location, which partially covers the municipalities of Barreiras, Luís Eduardo Magalhães and Riachão das Neves, known as the 'Soy Ring' due to the intense production of this crop, are severely affected by cloud cover persistent, mainly between December and February, when around 70% of the information contained in the images is generally lost. This becomes an excellent example of the challenge of crop monitoring in Brazil: large areas, high productivity, fast cycle and low data availability”, comments Bolfe.

According to Taya Parreiras, a doctoral researcher in Geography at the Unicamp Institute of Geosciences and member of the project, “monitoring rainfed agricultural production, which accounts for approximately 75% of global food, can benefit from multi-sensor approaches, as they increase the periodicity of observations by taking advantage of complementary information from multiple satellite data sources, optimizing the ability to identify the type of crop”.

“During field visits, we also identified areas in the process of converting natural vegetation to possible annual cultivation or pasture, in the process of regenerating native vegetation, as well as forestry areas being transformed into pastures. Considering the dynamics of land use conversions in this modern agricultural frontier, where change decisions are driven mainly by agribusiness, situations like these are quite common” says Parreiras.

Challenges

Among the challenges for mapping and monitoring Brazilian agriculture, the country's continental dimensions, productive diversity and climatic conditions stand out, which are reflected in the high presence of clouds in the images. “Satellite images are traditionally used for agricultural monitoring, but with major restrictions on their use, one of which is the availability of sufficient images during a harvest period. Sensors such as the Landsat 8 OLI (Operational Land Imager), for example, have a satisfactory spatial resolution of 30 m, but provide data every 16 days, which, added to the cloud cover, which is very dense and constant in tropical regions, results in little availability of scenes for effective agricultural monitoring. Other large-scale sensors, such as MODIS (Moderate-Resolution Imaging Spectroradiometer), perform daily imaging, but have a spatial resolution of 250 m, in itself a major limitation for agricultural applications, which is one of the dilemmas of this type of remote sensing instrument. that HLS helps to resolve”, explains Embrapa Meio Ambiente researcher Luiz Eduardo Vicente, also a member of the project.

Another challenge is the high heterogeneity of crops which, with up to 3 annual harvests on central pivots, increases the spatial and temporal variation of targets, making their detection difficult. “In the Cerrado region, in addition to the increase in areas with crops such as soybeans, corn, cotton and sugar cane, the diversity of production systems is also increasing, such as crop-livestock integration, which has been increasingly adopted , mainly the association between corn and brachiaria”, highlights Bolfe.

Despite the new simplified access to this type of “harmonized” data, processing HLS images is not trivial, requiring specific algorithms and application tests, considering parameter adjustments such as: data acquired at different resolutions, spatial, spectral and radiometric , as well as different viewing angles and signal-to-noise ratios. These procedures have been carried out by researchers independently, but the North American space agency NASA makes the data practically ready for analysis available, which makes the use of HLS a relatively simple and accessible process”, comments Parreiras.

In general, applications of remote sensing tools in agriculture still require local information, and obtaining this information is one of the main challenges for accurately mapping and monitoring the main crops on a regional and national scale in Brazil. “Acquiring field data with the necessary volume and balance is not always possible. In this sense, we adopted a hybrid sampling strategy, using high-resolution images to aggregate information on targets whose identification is less complex, such as natural vegetation, for example, and data systematically collected in the field using the AgroTag application, which provided speed and robustness in the training and validation of classification processing algorithms, fundamental steps for generating agricultural maps from satellite images”, explains Vicente.

Therefore, different agricultural use mapping projects in Brazil can benefit from Harmonized Landsat Sentinel-2 data and the methodology proposed in the project, enhancing the monitoring of dynamics in rural areas and a better understanding of processes such as expansion, integration, diversification, agricultural conversion and intensification, emphasizes Bolfe.

NASA program to produce a virtual constellation of harmonized images ready for analysis, with a temporal frequency of two to four days, in surface reflectance values, acquired by the Landsat Operational Land Imager (OLI) and Sentinel-2 Multispectral (MSI) sensor systems , the first being the responsibility of NASA (National Aeronautics and Space Administration), and the second, of ESA (European Space Agency).