Drones guarantee 66% accuracy in monitoring pastures

Embrapa research shows that the use of drones to monitor the coverage and height of pastures achieved 66% accuracy in the Cerrado of Bahia. The experiments, carried out between 2019 and 2021, reinforce the qualification of this remote sensing tool to increase the efficiency of agriculture, optimizing time, productivity of work in the field and expanding the capacity for observation and control of rural production.

The research was carried out at Fazenda Trijunção, in the municipality of Cocos, in the interior of Bahia, using a beef cattle farming system with rotational grazing and the use of brachiaria BRS Piatã. According to researcher at Embrapa Pecuária Sul (RS) Márcia Silveira, the work compared digital data on pasture height and soil cover, captured from images taken by drones, with values ​​observed in the field, using traditional methods of measurement, such as assessment by trained staff and measurement with a management ruler, as well as assessment of soil cover and forage sampling.

“Our objective was to verify whether a common drone, capable of being purchased by a producer, combined with machine training, can help in estimating vegetation cover and plant height. We wanted to evaluate whether the correct use of this tool can help them make decisions related to livestock management, through a comparison between the images generated by the drone and the information obtained in the field, considering height measurement, forage cutting and soil cover. by the forage plant”, explains Silveira.

The image bands taken by drones at different times, during two years, were compared with three classes of soil cover, which represent the management of a livestock property: pre-grazing, in grazing and post-grazing, in addition to a category called of exposed soil. The R-Studio data recovery software was used to validate the algorithm and analyze the images captured by the drone. The formula applied combined the different bands of the image to predict the cover class and pasture height. The accuracy assessment of the developed pattern was carried out based on the analysis of the confusion matrix (error) and the success matrix of the program.

Technology reinforces the potential of drones in Brazilian livestock farming

Flávia Santos, researcher at Embrapa Milho e Sorgo (MG) and leader of the Trijunção Project of which this study is part, highlights that the development of procedures for processing and analyzing drone images carried out by Embrapa during the research reinforce the benefits of using remote sensing as an auxiliary tool in pasture management. “In the future, the studies can serve as a basis for the creation of new products, such as applications for smartphones, further optimizing work in the field”, she adds.

“With the database covering just two years, it was already possible to visualize the potential of this type of information. We will continue monitoring to obtain more data and increase the robustness of the script for machine training. We hope, with more data, to extrapolate this type of information to different types of pastures”, points out the researcher.

The use of UAVs (unmanned aerial vehicles) is yet another strategy to increase the efficiency of livestock activity, assisting in the planning and management of pastures, based on the balance between supply and demand for food for the animals. The key point is the availability of forage in quantity and quality, in addition to maintaining conditions for plant persistence and regrowth quickly and vigorously.

“Height can be used as a practical criterion to define the ideal time for grazing, as well as allowing to identify the need or not to make animal load adjustments, aiming to establish optimal conditions for pasture use through the main processes involved in growth and use. of forage plants under grazing. In order for these height recommendations to be respected, it is necessary to monitor pasture areas more frequently, in order to make decisions to adjust the load and rotation of animals between areas more effectively. Therefore, the use of monitoring techniques, such as remote sensing, is promising in helping decision-making regarding pasture management”, concludes Silveira. 

Artificial intelligence helps correlate pasture management data

The methodology used used the machine learning technique in a digital environment. According to agronomist and doctoral student in Agronomy at the Federal University of Viçosa (UFV) Pedro Almeida, the script developed during the study is a compilation of several tools used for digital image classification. “We use the spectral response of drone images to correlate with pasture management classes. From the field data, we compared the variables with the readings carried out in the field. In the end, it was possible to establish relationships between soil cover and plant height, making it possible to automate the entire area, instead of just statistical sampling,” he says.

Considering the entire data set, the adjusted model achieved 66% accuracy and a Kappa index of 0,53 in predicting the four classes (pre-grazing, grazing, post-grazing and bare ground). The Kappa coefficient measures the agreement between two forms of assessment – ​​in this case, software analysis based on drone images and conventional measurement carried out in the field.

“In relation to the model training data set, the accuracy and Kappa index were 70% and 0,58 for the rainy season, and 68% and 0,56 for the dry season, respectively. Considering the entire training set, the accuracy and Kappa index were 66% and 0,53, within a two-year monitoring period, which included two dry and two rainy periods”, observes Silveira.

Integration of technologies in the field

For Manoel Filho, also a researcher at Embrapa Milho e Sorgo, the integration of technologies that increase the efficiency of rural property monitoring has been a constant search in modern agricultural production, aiming to assist the management and execution of production processes in a scenario of scarcity of labor. Western Bahia, where the study has been carried out, is characterized by production in large areas and a narrow period of production in rainfed systems. In this reality, the use of practical, wide-ranging and reliable monitoring methods, such as the use of drones, becomes essential.

“Our work has made the drone’s agility for collecting information very noticeable, as it only takes three hours to completely cover an experimental area of ​​more than 100 hectares”, highlights the geologist and doctoral student at the Federal University of Viçosa (UFV) Claudio Andrade.

Furthermore, the images produced automatically become a visual database that is easily accessible and available. The increase in accuracy, as a result of these results, indicates that this model has the potential to be a significant aid for large-scale pasture management. Additional monitoring is being carried out by researcher from the Secretariat of Agriculture, Livestock, Sustainable Production and Irrigation Rio Grande do Sul (SEAPDR-RS) Carolina Bremm, who works to improve the accuracy and validation of the model.