Incorporating Volunteered Geographic Information into Land Cover Monitoring

• Activity Number: 344
• Type: Contributed
• Date/Time: Tuesday, August 5, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics and the Environment
• Author(s): John Lombardi*+ and Stephen Stehman
• Companies: SUNY ESF and SUNY ESF
• Keywords: area estimation ; propensity scores ; model-assisted estimation ; design-based inference

Abstract

Volunteered geographic information (VGI) has grown in popularity as a source of land-cover data. A primary application of VGI in land-cover monitoring is estimating area of different land-cover classes. Most VGI is not collected via a probability sampling design, so questions of how to incorporate these data in a statistically rigorous manner must be resolved. We explore two options: 1) VGI locations are treated as the sample and estimators are based on propensity scores in lieu of estimation weights derived from inclusion probabilities, and 2) a probability sample has been selected separate from the VGI locations and a model-assisted estimators is constructed in which the VGI data are used to create an auxiliary variable. The estimators are evaluated for a case study focusing on estimating land-cover area in New York. Web sites (e.g., Panoramio) were used to identify VGI locations, and a population of land-cover data was obtained from the National Land Cover data. When volunteer data are plentiful (e.g. populated areas), propensity scores yielded area estimators with small bias. For model-assisted estimation, a kriging approach proved successful at reducing standard errors.