Student Research - Mission Earth

View Research Report

This analysis produces a robust, interpretable flagging system for identifying anomalies in surface temperature data collected as a part of the GLOBE mission, through metadata augmentation and the use of unsupervised clustering algorithms and statistical tests. Our research highlights the varied results of individual clustering models, justifying an ensemble scoring approach between multiple unsupervised clustering models. The partial correlations among various anomaly flags are also demonstrated, showing that the flagging system captures multiple dimensions of abnormality. The metadata augmentation and flagging approach added these columns to the GLOBE dataset: ➢ Metadata: Country, Country Code, Continent, Year, Month, Biome, Season ➢ Flags: latlon_range, high_elev_hot_temp, duplicated_coords, zscore_outlier, lof, mahalanobis, ensemble_score This flagging system enhances the trustworthiness and usability of the data, providing future researchers with a scalable strategy to detect anomalies in Earth science datasets.