| Growing environmental concerns require monitoring and modelling of greenhouse gases. These modelling efforts require processing of massive datasets in a timely fashion. This, in turn, can lead to feasibility problems when estimating values of missing data points. This thesis examines and compares multiple methods for estimating values of missing data points, including their spatiotemporal extensions. Resulting predictions are compared from the perspective of accuracy and computational efficiency. The results show that kriging based methods generally outperform the others in terms of accuracy, but took longer to process. Hierarchical methods prove to be a more suitable choice, providing slightly less accurate results at much shorter times, especially for dense datasets.;The second part of the thesis explores a scheme for updating emission inventories using socioeconomic data. Random forest and extreme machine learning techniques applied for this task show poor performance on real-world data. |
