Master of Science
University of New Brunswick
Management of nitrogen (N) fertilizers is an important agricultural practice and field of research to increase productivity, minimize environmental impacts and the cost of production. To apply N fertilizer at the right rate, time, and place depends on the crop type, desired yield, and field conditions. The objective of this study is to use Unmanned Aerial Vehicle (UAV) multispectral imagery, PlanetScope satellite imagery, vegetation indices (VI), crop height, leaf area index (LAI), field topographic metrics, and soil properties to predict canopy nitrogen weight (g/m2) of corn and wheat fields in southwestern Ontario, Canada. Random Forests (RF) and Support Vector Regression (SVR) machine learning models were tested with combinations of variable datasets and evaluated for accuracy of canopy nitrogen weight prediction. The results demonstrate that UAV and satellite-based prediction models including spectral variables, crop biophysical parameters, and field conditions can provide accurate and useful information for fertilizer management.
Summary for Lay Audience
The practice of agriculture first began thousands of years ago, when humans began building permanent settlements, growing their own crops, and raising livestock for reliable survival resources. Today, agriculture continues to be one of the most important industries around the world providing food and materials for over 7.7 billion people. As the world population is projected to continue growing, for agriculture to meet increasing demands requires sustainable production, adaptability to changing climates, and better methods of farming. Advances in technology have opened the way for precision agriculture, a management technique that gathers many types of data about a crop to improve resource use (e.g., water, fertilizer, pesticides), quality of production, profitability, and sustainability.
Remote sensing is the process of obtaining information about an object or phenomenon at a distance. In precision agriculture, remote sensing is especially useful without the need to make physical contact with plants to gather valuable crop data for analysis. With our eyes, we see colours such as blue, green, and red, through the way light is reflected or emitted from a surface. With special cameras designed for certain wavelengths of light, we can see beyond the visible spectrum and understand more about a plant’s characteristics (i.e., healthy plants may reflect more infrared light than an unhealthy plant, but we cannot see that with our eyes). This thesis studied how we may estimate nitrogen levels in corn and wheat plants using both satellite and unmanned aerial vehicle (UAV; a.k.a. drone) remotely sensed data.
The growth of a plant depends on many different factors including light, water, nutrients, and more. Using machine learning in this thesis, we built computer models from gathered data of a crop’s response to light, biophysical characteristics, and environmental conditions to improve nitrogen level estimation of crop fields. Nitrogen prediction maps of a field can be created from the best models showing areas of a field that need certain amounts of fertilizers. Following the map, a farmer can practice sustainable precision agriculture by applying resources at the right rate, time, and place for a bountiful crop.
Yu, Jody Seymon, "Intra-field Nitrogen Estimation for Wheat and Corn using Unmanned Aerial Vehicle-based and Satellite Multispectral Imagery, Plant Biophysical Variables, Field Properties, and Machine Learning Methods" (2021). Electronic Thesis and Dissertation Repository. 8346.