Using Time Series Foundation Models for Atmospheric CO2 Concentration Forecasting

Recent advancements in foundation models for time series forecasting have outperformed traditional statistical and ML methods. In this study, we approach the problem of CO2 concentration forecasting using time series foundation models (TSFMs). We extensively evaluate the performance of TSFMs under zero-shot and fine-tuned settings against popular traditional forecasting baselines and discuss the spatial transfer learning capability of TSFMs across diverse geographic locations.