An Adapted VAR-EM (AVAR-EM) Imputation Algorithm to Populate a Broken Historical Climate Record

• Activity Number: 190 - Contributed Poster Presentations: Section on Statistics and the Environment
• Type: Contributed
• Date/Time: Monday, July 29, 2019 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics and the Environment
• Author(s): Benjamin Washington* and Lynne Seymour
• Companies: The University of Georgia and University of Georgia
• Keywords: Expectation-Maximization; Vector Autoregressive; Multivariate Infilling; Climate Data; Puerto Rico; Spatiotemporal Data

Abstract

Climate data is readily available from a number of publicly available sources: NCEI, Daymet, WDCC, etc. However, climate data records are rarely complete and have many missing or inaccurate data points. Infilling climate networks, or collections of climate series, represents a unique challenge as these data are related not only in time, but also in space. Schneider (2001) found that iterative algorithms tended to outperform a number of more conventional noniterative algorithms studied by Smith et al. (1996) in a multivariate infilling setting.

In this study we propose a combination of the EM algorithm and an adapted vector autoregressive model (AVAR-EM) similar to the methods of Bashir and Wei (2018). The advantage of this iterative approach is that it will take into account both the spatial and temporal correlations amongst the stations. To quantify the performance of this method, we will run a pilot study using historical temperature data at seven stations in Florida and coastal Georgia whose NCEI temperature record is complete (no missing data). Then, we will examine the utility of this EM-VAR approach on a broken NCEI temperature record at 23 locations in Puerto Rico.