Authors: Rui Zhu*, University of California - Santa Barbara
Topics: Spatial Analysis & Modeling, Geographic Information Science and Systems
Keywords: Spatial Statistics, Land Cover/Land Use, Spatial Interactions, Uncertainty
Session Type: Paper
Start / End Time: 1:20 PM / 3:00 PM
Room: Proteus, Sheraton, 8th Floor
Presentation File: No File Uploaded
Modeling geospatial interactions is a core aspect of geographic information science. Traditional approaches concentrate on using two-point statistics, such as Moran’s I and correlation functions, to accomplish the goal. In these methods, the geospatial interactions are modeled as a combination of several pairwise relations between the target and its neighbors. These pairwise interactions have a wide application in geography. For instance, they are demonstrated to be effective on detecting some spatial patterns, like the randomness and clustering. Furthermore, the Kriging, which is the most commonly used stochastic model for quantifying spatial uncertainties, is built based on pairwise interactions as well. However, provided only pairwise interactions, it would become difficult to analyze, and subsequently assess the uncertainty of, rather complex spatial patterns (e.g., curvilinearity and polygonal geometries). Therefore, relations that are beyond pairs should be considered to quantify the geospatial interactions. In this work, we propose to use multiple-point geostatistics (MPS) to simultaneously consider interactions among multiple entities. Specifically, templates (i.e., a configuration of multiple locations) are designed in MPS to learn the interaction of a spatial pattern in a higher-ordered manner (i.e., the interaction is modeled as higher-ordered conditional probabilities), in contrast to the second-order semivariogram when applying Kriging. To evaluate our approach, we compare it with the Kriging on assessing the uncertainties of land use/land cover (LULC) change at Santa Barbara area.