Predicting Vegetation Carbon Uptake in the Midwestern United States Using Environmental Variables and Machine Learning
Session Number
2
Advisor(s)
Dr. Bayarbadrakh Baramsai, NASA
Location
A117
Discipline
Environmental Science
Start Date
15-4-2026 11:10 AM
End Date
15-4-2026 11:55 AM
Abstract
Understanding how carbon moves through Earth’s ecosystem is essential for predicting climate change. However, about 20-30% of the carbon emitted each year remains unaccounted for in global carbon cycle models, which is often referred to as the “missing carbon sink.” Although resolving the missing carbon sink requires large-scale observations and global modeling efforts, smaller regional investigations can provide insight into the processes that influence carbon uptake. This project investigates whether monthly vegetation carbon uptake in the Midwestern United States can be predicted using environmental variables, including rainfall, temperature, and vegetation greenness indices (NDVI/EVI). A predictive model was developed to quantify the relationships between climate conditions and Gross Primary Productivity (GPP). Initial analysis explores correlations between variables and multiple modeling approaches, which include sinusoidal regression and machine learning methods such as multiple linear regression and random forest. These were used to predict GPP trends. The model was then tested for generalizability by applying it to regions with similar vegetation types to determine if observed relationships hold across spatial scales. This study aims to improve understanding of the environmental controls on carbon uptake, contributing to better characterization of the “missing carbon sink” and enhancing predictive capabilities for ecosystem carbon dynamics.
Predicting Vegetation Carbon Uptake in the Midwestern United States Using Environmental Variables and Machine Learning
A117
Understanding how carbon moves through Earth’s ecosystem is essential for predicting climate change. However, about 20-30% of the carbon emitted each year remains unaccounted for in global carbon cycle models, which is often referred to as the “missing carbon sink.” Although resolving the missing carbon sink requires large-scale observations and global modeling efforts, smaller regional investigations can provide insight into the processes that influence carbon uptake. This project investigates whether monthly vegetation carbon uptake in the Midwestern United States can be predicted using environmental variables, including rainfall, temperature, and vegetation greenness indices (NDVI/EVI). A predictive model was developed to quantify the relationships between climate conditions and Gross Primary Productivity (GPP). Initial analysis explores correlations between variables and multiple modeling approaches, which include sinusoidal regression and machine learning methods such as multiple linear regression and random forest. These were used to predict GPP trends. The model was then tested for generalizability by applying it to regions with similar vegetation types to determine if observed relationships hold across spatial scales. This study aims to improve understanding of the environmental controls on carbon uptake, contributing to better characterization of the “missing carbon sink” and enhancing predictive capabilities for ecosystem carbon dynamics.