Simulating Uncertainty in Volunteered Geographic Information
MetadataShow full metadata
Facilitated Volunteered Geographic Information (VGI), crowdsourced data that includes a geographical reference that is solicited for a specific purpose, holds great promise for environmental monitoring, yet a major limitation of VGI is unknown information quality. Prior to initiating a VGI project, it is difficult to know if the collected data will be useful for the intended project purpose. This research explores the use of computer simulation to inform the design and implementation of a facilitated VGI project, specifically an urban neighborhood white-tailed deer survey. The project was conducted in two phases; first, a computer simulation phase, and second, a simulation validation phase including a VGI neighborhood deer count. During the simulation phase of the project five different data collection methods were tested, each subject to various types of uncertainty, including observation location uncertainty, distance estimation uncertainty, and deer detection and classification uncertainty. Methods were tested under permutations of four levels of volunteer participation and three levels of deer density. Additional simulation refined and optimized the most promising data collection methods. Simulation results suggested a neighborhood counting protocol based on predefined observation areas with focused counting times to increase observed area, and the inclusion of zero-deer observations, that is, reports of areas searched that did not contain any deer. During the simulation validation phase of the project, results from the simulation phase guided development of the facilitated VGI neighborhood deer count. The 28-day volunteer deer count was conducted in October, 2012 in two adjacent neighborhoods in San Marcos, Texas. Aggregate results from volunteer observations were used to estimate the neighborhood deer population. Concurrent with the volunteer deer count, an Infrared Triggered Camera (ITC) deer survey, a scientifically accepted survey method, was conducted in the same area. The VGI population estimate was 72% of the ITC population estimate. Although the volunteer population estimate fell outside of the targeted range of 75% - 125% of the ITC population estimate, simulation was nonetheless useful for testing alternative data collection procedures, optimizations to data collection procedures and the relative performance of those procedures under differing conditions of deer density and participation. Simulation results also informed interpretation of VGI results, but simulation was not useful for predicting volunteer behavior or participation level. This research introduces the use of computer simulation to inform and improve the design and implementation of facilitated VGI initiatives.