Climate Change and Environmental Research
Our research group aims to address critical challenges in analyzing functional data to understand climate change better and provide accurate environmental data predictions. Over the next three years, we plan to tackle three interconnected problems.
- Quality Issues in Functional Data:
Problem: Outliers and missing values in functional datasets can lead to erroneous statistical inference and inferior predictions, but defining functional outliers is challenging.
Approach: We will investigate computationally efficient measures to quantify the location, shape, and directional outlyingness of functional data, potentially generalizing band-depth methods. Visualization and computing are vital bottlenecks we aim to overcome.
- Classification and Clustering of Functional Data:
Problem: Outlying data must be mitigated before performing classification or clustering of functional data, similar to traditional datasets.
Approach: We will develop non-parametric and model-based classification and clustering approaches. We aim to propose effective methods while considering computational efficiency, potentially leveraging federated learning and deep neural networks.
- Prediction of Functional Data and Spatio-Temporal Kriging:
Problem: Predicting functional data, such as environmental variables, requires accounting for common factors and incorporating exogenous predictors.
Approach: We will develop sparse factor models and study functional models for common factors, considering factors like humidity and wind speed for accurate predictions. We will also explore spatiotemporal kriging methods.
In addition to addressing these problems, we will analyze functional data from the ARGO dataset, focusing on spatial, temporal, and spatiotemporal aspects. We aim to provide practical solutions for climate change and environmental challenges by applying our methods to real-world environmental data. Our ultimate goal is to publish advanced research results and apply them to benefit Taiwan and global environmental efforts.
This research plan outlines a comprehensive approach to functional data analysis, encompassing outlier detection, classification, clustering, prediction, and real-world applications in climate change research. We aim to contribute significantly to the field through collaboration and innovative methodologies.