Context
Investment portfolios linked to commodity prices represent US$400 billion. In commodities futures trading, knowing the future yield of a crop is a critical piece of information to assess price.
Issues
Remote sensing (e.g. satellite images) has been used extensively to aid in yield prediction. However, given the proprietary nature of most trading firms, there is little openly known about the robustness of this approach.
Objectives
The goal of our project was to assess the power of satellite images combined with weather data to predict yields and to do a comparative analysis of modeling strategies.
Research
Prior work by researchers in the academic, government, and private sectors is reviewed, highlighting key studies and techniques used. Market research was also conducted to verify the viability and relevance of our project.
Data Acquisition
The data sets for satellite imagery and weather information are large and varied. Our methods for acquisition, cleaning, and preparation are discussed.
Modelling
In addition to attempts to beat existing baseline models, we tested our ability to predict final yields at earlier stages in the soybean growth cycle.
Results
Our models and findings are documented for further examination and peer review in our quest for continued improvement of yield forecasting models.