Nitrogen Status Prediction on Pasture Fields Can Be Reached Using Visible Light UAV Data Combined with Sentinel-2 Imagery
1F. R. Pereira, 2J. P. Lima, 2R. G. Freitas, 3A. A. Dos Reis, 2L. R. Amaral, 2G. K. Figueiredo, 3R. A. Lamparelli, 1J. C. Pereira, 3P. S. Magalhães
1. Federal Institute of Education, Science and Technology of Alagoas, 57120-000, Satuba, Alagoas, Brazil
2. School of Agricultural Engineering, University of Campinas, 13083-875, Campinas, São Paulo, Brazil
3. Interdisciplinary Centre of Energy Planning, University of Campinas, 13083-896, Campinas, São Paulo, Brazil
Pasture fields under integrated crop-livestock system usually receive low or no nitrogen fertilization rates, since the expectation is that nitrogen demand will be provided by the soybean remaining straw cropped previously. However, keeping nitrogen at suitable levels in the entire field is the key to achieving sustainability in agricultural production systems. In this sense, remote sensing technologies play an essential role in nitrogen monitoring in pastures and crops. With the launch of the Sentinel-2 missions, new opportunities have arisen for nitrogen status monitoring. Additionally, an RGB sensor boarded on a UAV is usually an option for low-cost UAV devices. However, few studies investigate the combination of UAV and satellites information to assess nitrogen status variability. Thus, to estimate the nitrogen variability on pasture fields under an integrated crop-livestock system, we tested the performance of an exclusively visible light UAV data (i.e., RGB – red, green and blue), named UAV_RGB and Sentinel-2 data (S2) (both individually and combined) to monitor plant N content (PNC), aboveground biomass (AGB), and nutritional nitrogen index (NNI). The study area was composed of four fields of 50 hectares located in the western region of São Paulo State, Brazil. This study focused on assessing the nitrogen status in pasture fields with ruzi grass, the most used grasses in ICLS in Brazil. During the forage growing season, field data collection was carried out in three field campaigns. We used the original bands from UAV_RGB and S2 and various vegetation indices (VIs) to capture the vegetation conditions during the study period. To evaluate and compare the RF models performance, we employed the Root Mean Square Error (RMSE), Mean Absolute Error (MAE) in absolute and percentage terms, the coefficient of determination (R2) and the RMSE-observations standard deviation ratio (RSR), calculated based on the field measurements of PNC, AGB and NNI in the testing datasets. Our results showed that the UAV visible data combined with the information from the Sentinel-2 data were complementary and benefited each other in the estimation of PNC, AGB, and NNI compared with the performance of using individual data. Therefore, we concluded that using UAV_RGB data with multispectral Sentinel 2 data is also efficient for monitoring nitrogen variability in commercial pasture fields.
