Skip to main content
eScholarship
Open Access Publications from the University of California

Scholarly Works (2 results)

  • Thesis
  • Peer Reviewed
UC Berkeley Electronic Theses and Dissertations (2023)

This thesis explores statistical methods for characterizing the occurrences and impacts of climate. Data on climate and the environment more broadly show unique characteristics, posing methodological challenges.

Part I presents an overview of an emerging, accessible earth observation data source, satellite images, and explores its potential applications in statistical and causal inference. We introduce an approach for incorporating image data, represented by image features, into a regression framework as a potential proxy for confounding variables. The study examines the impact of image features on regression estimates, focusing on the characterization of bias with and without the inclusion of images, as well as the conditions under which the inclusion of image data reduces or amplifies bias.

Part II introduces an empirical and methodological problem related to estimating the economic values of the environment. This question is pertinent for understanding how various environmental qualities, such as flood risk and air pollution, are currently reflected in residential property values and how society manages climate and environmental risks. However, a methodological challenge in estimating capitalization lies in the difficulty of accounting for confounders, which invites the application of the method studied in Part I. The results indicate that while some risk-related factors are associated with lower housing values, others, such as PM 2.5 and flood risk score, are associated with higher housing prices. This provides a lens through which to discuss how risks could be reflected in property values.

Part III focuses on statistical methods for characterizing occurrences of extreme climate events. These data are spatio-temporal in nature, and effectively visualizing and summarizing such data is challenging, though crucial for monitoring and identifying hazardous events. A primary challenge stems from the context-dependent nature of extreme events, where definitions vary over time and space. While many adopted approaches involve pre-defining criteria for anomalous events, typically by setting an exceedance threshold, determining appropriate values for the exceedance threshold, time window, and spatial boundary is a non-trivial task. The study applies functional principal component analysis to characterize spatio-temporal trends of extreme precipitation. The method shows potential as a flexible way to identify both the temporal window and geographic location of anomalous events.

    Cover page: Statistical Methods for Characterizing Occurrences and Impacts of Climate
    • Article
    • Peer Reviewed
    UC Berkeley Previously Published Works (2021)
    Combining satellite imagery with machine learning (SIML) has the potential to address global challenges by remotely estimating socioeconomic and environmental conditions in data-poor regions, yet the resource requirements of SIML limit its accessibility and use. We show that a single encoding of satellite imagery can generalize across diverse prediction tasks (e.g., forest cover, house price, road length). Our method achieves accuracy competitive with deep neural networks at orders of magnitude lower computational cost, scales globally, delivers label super-resolution predictions, and facilitates characterizations of uncertainty. Since image encodings are shared across tasks, they can be centrally computed and distributed to unlimited researchers, who need only fit a linear regression to their own ground truth data in order to achieve state-of-the-art SIML performance.
      Cover page: A generalizable and accessible approach to machine learning with global satellite imagery
      Top