MULTI-SENSOR, DATA-FUSION AND PREDICTIVE COMPUTATIONAL MODELS TO ACCESS SOIL FERTILITY ATTRIBUTES

• Grant number: 19/05345-2
• Support Opportunities: Scholarships abroad - Research Internship - Doctorate
• Effective date (Start): August 01, 2019
• Effective date (End): April 30, 2020
• Field of knowledge: Agronomical Sciences - Agricultural Engineering - Agricultural Machinery and Implements
• Principal Investigator: José Paulo Molin
• Grantee: Tiago Rodrigues Tavares
• Supervisor: Abdul M. Mouazen
• Host Institution: Escola Superior de Agricultura Luiz de Queiroz (ESALQ). Universidade de São Paulo (USP). Piracicaba , SP, Brazil
• Research place: Ghent University (UGent), Belgium
• Associated to the scholarship: 17/21969-0 - Evaluation of alternative techniques to traditional laboratory analysis for prediction of attributes in agricultural soils: approaches using VisNIR, XRF and LIBS spectroscopy, BP.DR

Abstract

Since conventional sampling and laboratory soil analysis do not provide a cost-effective capability to obtain georeferenced measurements with adequate frequency to precision agriculture approaches, different sensing techniques have been attempted. Our original project is exploring "new" sensing tools (XRF, LIBS and Vis-NIR sensors) and the synergism between then to access fertility attributes in Brazilian tropical soils. The first results are promising and we would like to sophisticate our predictive modelling by using machine learning tools in order to enhance the performance of these sensors, as well as compare the predictive ability obtained by our methodology in other types of soils. Our group (Precision Ag Laboratory, ESALQ, USP) already had the opportunity to interact with Professor Abdul M. Mouazen - who is a full Professor at Ghent University and expert in proximal soil sensing and precision agriculture approaches - and could help us better explore the potential of our data. We propose in this project of research internship abroad (BEPE/FAPESP) conduct the research activities in two steps. In the first step, the data already obtained with the different PSS will be used to apply machine learning tools to predict the attributes of interest and, then, the performance of these predictions will be compared with the determinations made using linear models. In the second step, a Belgian agricultural field will be chosen to apply our methodology to obtain soil data in high spatial density in order to validate the satisfactory performance of the synergistic use of Vis-NIR and XRF techniques to access the soil fertility. In the case of positive results, we will construct and evaluate maps of soil fertility attributes using the PSS's information in order to elaborate VR prescriptions of lime and K and P fertilizers. In parallel to these research activities, I will also take the course of Predictive Modeling offered by the Faculty of Bioscience Engineering (Ghent University).