Reproducibility of a Climate Vulnerability Model for Africa
Topics: Geographic Information Science and Systems
, Human-Environment Geography
, Africa
Keywords: Reproducibility, Climate, Vulnerability, Africa, Malawi
Session Type: Virtual Poster
Day: Friday
Session Start / End Time: 4/9/2021 09:35 AM (Pacific Time (US & Canada)) - 4/9/2021 10:50 AM (Pacific Time (US & Canada))
Room: Virtual 52
Authors:
Joseph Holler, Middlebury College
Kufre Udoh,
,
,
,
,
,
,
,
,
Abstract
Successful reproduction studies enhance the credibility of the original study by generating identical outcomes using the same materials and methods. Reproducibility and credibility is particularly important if research about systems with high levels of complexity and uncertainty is to inform public policy. In this study, we attempt to reproduce a model of vulnerability to climate change in Malawi. Many studies of climate vulnerability explicitly aspire to broader impacts through informing policy and through the replication of methodologies and theories across geographic and temporal contexts. However, reproduction studies are rarely attempted in the human-environment and geographical sciences, and are increasingly difficult to conduct as research becomes more computational.
Our reproduction attempt was only partially successful, but has allowed us to articulate several types of challenges in reproducing computational research in human-environment and geosciences, and to suggest specific interventions for improving reproducibility of future studies. We found that some indicators used in the study were not publicly available, suggesting the need to provide data in supplementary materials or public repositories. We found inconsistencies between narrative data descriptions and metadata, suggesting the need for supplementary materials mapping conceptual terms precisely to data variables and values. We also found ambiguities in data processing decisions related to missing data, data errors, rescaling and reclassifying attributes, and spatial resolution, suggesting the need for complete documentation of the software environments and computational code for data processing from the first step of data collection through to published statistics, figures, and maps.
