1T. R. Tavares, 2M. T. Eitelwein, 2R. G. Trevisan, 1L. F. Maldaner, 1L. D. Corrêdo, 1L. G. Mendes, 3J. P. Molin
1. Biosystems Engineering Department, Luiz de Queiroz College of Agriculture, University of São Paulo
2. Smart Agri Technological Solutions
3. Biosystems Engineering Department, University of São Paulo
Sample density for mapping the spatial variability of soil attributes is limited due to the costs of laboratory analysis and the operational feasibility of the method. Furthermore, researches using geostatistical analyzes usually demonstrate that the density employed is not sufficient to characterize the spatial distribution of most soil chemical attributes. About 15% of the grain production area in Brazil is managed with precision agriculture tools including grid soil sampling and variable rate application of lime and fertilizers. Around 70% of this area is sampled with less than 0.5 samples ha-1, which is not adequate in most situations. In this context, proximal soil sensing (PSS) emerges as an alternative technique for surveying soil information in high density, without releasing chemical waste and at a reduced cost. These work aims to present and discuss some results and challenges faced by the Precision Agriculture Laboratory (LAP - ESALQ / USP) in the research involving PSS techniques. The PSS researches under development for tropical soils include on laboratory and on-site evaluation of electrical conductivity sensors (EC), ion-selective electrodes (ISE) and visible to near-infrared spectroradiometers (vis-NIR). Using on-the-go data collection, good results were reported for the prediction of clay content - using EC (R²> 0.83) and vis-NIR (R²> 0.60); and for pH (R² > 0.75) and K (R²> 0.59), using ISE. In controlled environments, vis-NIR spectroradiometer has presented good predictions (R²> 0.8) for soil texture and soil organic matter. In some cases, these results could be extended for available nutrients (P, K, Ca, Mg) - when these are correlated with spectrally active attributes in the vis-NIR region. However, these results are not always consistent among different surveys, which makes it difficult to establish generalized data collection and analysis protocols for agronomic recommendations. Finally, we present a case study using data from a 10 ha experimental area. Temporal variability was addressed by collecting data twice with a 76-day interval, using EC and ISE sensors on the Veris MSP® platform (Veris Technologies, Salina, Kansas, USA). The relationships between conventional laboratory soil analyses and PSS (EC, ISE and vis-NIR) were also evaluated using grid soil sampling with 2.4 samples ha-1.