Detecção de fósforo foliar na cultura da batata através de imagens multiespectrais

Abstract

Phosphorus is an essential nutrient for the development of potato crops (Solanum tuberosum L.); however, its natural availability in agricultural soils is low, requiring frequente applications. Given the scenario of depleting mineral reserves and the need for precision agriculture, a rapid and non-destructive diagnosis of leaf phosphorus content becomes necessary to reduce waste, avoid environmental impacts, and maximize production efficiency. Therefore, this study aimed to evaluate the feasibility of using multispectral images obtained through remote sensing with the MAPIR Survey 3 camera to estimate leaf phosphorus contente in potato plants. To generate data variability, an experiment was conducted in a greenhouse using 13-liter pots at the Federal Institute of Education, Science and Technology of the Triângulo Mineiro – Uberaba Campus (IFTM), in a randomized block design with eight treatments and 40 replications. The treatments consisted of increasing doses of P₂O₅ (0, 50, 100, 200, 400, 800, 1,600, and 3,200 kg ha⁻¹), totaling 320 pots, each considered an experimental unit. At the onset of tuberization, multispectral images of all plants were captured using the MAPIR Survey 3 camera. Immediately after imaging, leaf samples were collected to determine phosphorus content. Using the complete dataset, the following vegetation indices were calculated: normalized difference vegetation index (NDVI), green normalized difference vegetation index (GNDVI), and green–red vegetation index (GRVI). Linear regression models and machine learning methods (Random Forest, Multilayer Perceptron, and SMOreg) were employed to test the predictive ability of the indices and combined spectral bands. A significant inverse correlation was observed between NDVI and leaf phosphorus content (r = –0.71), physiologically explained by the nutritional dilution effect. The Random Forest algorithm showed the best performance (relative RMSE of 19.01%), standing out among the evaluated methods for its ability to integrate multiple spectral attributes. The combination of multispectral variables yielded significant gains in accuracy for phosphorus prediction when compared to univariate approaches, enhancing the robustness of the results for real-world scenarios. The findings reinforce NDVI as the primary indicator but also highlight the added value of multispectral integration coupled with machine learning for nutritional diagnosis of potato crops in precision agriculture systems.