2011 GEOG 506 Projects

Emilie Roy-Dufresne: A Multi-Scale Analysis of the White-Footed Mouse’s Distribution Using MaxEnt Species Distribution Model

Abstract: Issues of modeling species geographic distributions are critical factors which need to be studied in conservation biology. One of these issues is the appropriate spatial scale at which studies need to be conducted. The identification of spatial patterns depends on the spatial scale at which patterns are measured. Ecologists face the issue that interpreting the data based on one scale, and to apply them to another scale may not accurately describe the existing pattern but instead reflect artefacts of the scale of measurement. There is therefore a critical need to understand the scale dependencies of a habitat model for the different inputs resolutions that is when the grain size of the input dataset changes.The purpose of this study was to present a litterature review of the importance of scale for the field study of species distribution modeling and an analysis of how scale issue affects three different habitat metrics (spatial heterogeneity, fragmentation, and edge characteristics indices) and a selected habitat model MaxEnt along with the distribution of the White-footed mouse (Peromyscus leucopus) in the Monteregie. The scale dependence of the metrics and the model was explored by varying the grain size and extent of the input data, which in this case were landcover maps. The datasets at the following spatial resolutions: 500, 750, 1000, 2500, 5000, 7500, and 10,000 m. A covariance analysis was performed to determine if the metrics selected are scale dependant, whereas a sensitivity analysis was performed to investigate the uncertainty of the model output data as a function of the different inputs resolutions implemented in the habitat model. The results show that both the metrics and the model performance and output are sensitive to the scale of the study, but more specifically are influenced by the landscape complexity. The results provide insights with regard to the scale issue of models that are tools commonly used in the scientific community.

Sean McBride: A Stochastic Data Model for Geospatial Simulation of Internal Migration

Abstract: This research proposes an agent-based data model for stochastically predicting population migration within a geographic area composed of discrete regions and represented in a Geographic Information System (GISystem). It overviews relevant context in the field of data modelling and explains in detail the functioning of the proposed model. Literature regarding migration studies is reviewed, and the model is applied in a case study of state to state internal migration in the United States. The results of this model are discussed in light of relevant literature and conclusions and recommendations are made regarding both how this model could be improved and its implications for the field of Geographic Information Science (GIScience).

Multiple Remote Methods of Spatial Data Mining: A case study on the food environment

Abstract: Advances in technology related to the virtual presentation of geographic space and the collection and storage of large amounts of data present geographers with new opportunities for obtaining data remotely. In the field of health geography, remote methods of data collection can allow for modeling the food environment of a community without extensive field work, which facilitates the study of the impact of access to healthy food on diet-related disease and obesity. This project combines multiple methods of remote data collection—virtual audit, web scraping, and geographic data mining—to develop models of the food environment in two economically disparate neighbourhoods of Cleveland, Ohio. By incorporating extensive background research and a localized, virtual street-level audit, context is established without contact. Data mining through web scraping and geographic data mining expand the area being surveyed through time-saving semi-automated methods of collection. Whereas conducting research remotely presents significant challenges, well-designed methodology can mitigate its inherent flaws and create viable datasets.

Johanna Bleecker: Spatial Autocorrelation in Practice: Kibale National Park, Western Uganda

Abstract: Spatial autocorrelation, the phenomenon of near things being more related to far things, has been an important topic in GIScience ever since Tobler’s First Law defined it as a fundamental concept in geographic thought. Though spatial autocorrelation is known in the statistical spatial analysis community, it is largely ignored in mainstream geographic research despite the confounding effects it can have on commonly used statistical tests and other aspects of spatial analysis. I wanted to investigate the degree to which spatial autocorrelation is affected by and affects common data processing procedures, using demographic attributes of Ugandan villages as sample data. First, the initial spatial autocorrelation was measured by use of the Moran’s I statistic, three common interpolations were applied to these data points, and the resulting coverages were resampled and tested again for Moran’s I. None of the interpolation methods appeared to be the best for consistently avoiding spatial autocorrelation while obtaining a relatively accurate surface—most of them inconsistently resulted in higher spatial autocorrelation than the original data. For the second part of the project, the spatial trend that confers spatial autocorrelation onto the data was removed while similar interpolations were carried out, theoretically resulting in a surface with lower spatial autocorrelation. However, though these surfaces were more accurate than the default interpolations performed earlier, their spatial autocorrelation remained high and unpredictable. One major limitation probably responsible for the inconclusive results was the imperfect method of testing all the interpolated coverages for their resulting Moran’s I. Though no conclusions were reached about which interpolation method is best used with or without spatial autocorrelation present, the unpredictability of the results highlights the importance of acknowledging and testing for spatial autocorrelation, and the need for it to be given much higher regard in geographic research than is currently done.

Amina Hasan: Spatial Data Uncertainty: Uncertainty in Sea Surface Temperature

Abstract: Quantifying spatial data uncertainty is often difficult as it is considered to be a complex concept with a number of interpretations across knowledge domains and application contexts. It is important to characterize and quantify the uncertainty of datasets as this will have major implications for its usability. This project focuses on the uncertainty related to climate data, particularly in the form of sea surface temperature and considers the varying characterizations of accuracy in the measurements of sea surface temperature.