Soil carbon stock prediction based on machine learning

Abstract

Understand the mechanisms and interactions that govern soil carbon dynamics is essential for mitigating climate change and increasing world food production. In this context, in this project we aim to assess the performance of two classifier ensembles, bagging and boosting, to recognize patterns and predict soil carbon stock and its variations at the time and space, in sugarcane areas under different cover crops and soil tillage in Brazil. This research proposal will be started after a six-month internship at the University of Bologna - Italy, whose aim will be to calibrate and assess soil carbon dynamics models using the software library UNIMI.CRONO. Then, for a period of six months, under the supervision of Professor Dr. Asim Biswas of the University of Guelph - Canada, It is intended to use the same dataset to assess the performance of the algorithms bagging and boosting to combine individual classifiers (base classifiers), such as Artificial Neural Networks, Random Forest, Logistic Regression and Support Vector Machines (SVM) to predict soil carbon stock. After a period of 12 months, we hope to find the best modeling solutions to understand and predict carbon dynamics in sugarcane areas under tropical conditions. In addition, we will be sharing those results by means of lectures, talks, articles and conferences. (AU)