Authors: Saeed Harati*, Université de Montréal, Liliana Perez, Université de Montréal, Roberto Molowny-Horas, CREAF, Robert Gilmore Pontius Jr, Clark University
Topics: Geographic Information Science and Systems, Land Use and Land Cover Change
Keywords: Land Use and Cover Change, model validation, irreversible process, area partition, proximity, disagreement analysis
Session Type: Virtual Paper
Start / End Time: 3:05 PM / 4:20 PM
Room: Virtual 21
Presentation File: No File Uploaded
Models of Land Use and Cover Change are validated by comparing their simulated changes against reference changes. To that end, methods of map comparison help scientists identify levels of agreement and disagreement between simulated and reference changes. To avoid problems of subjective interpretation of such levels of agreement, scientists sometimes use another model as a baseline for comparison. In this study, we focus on validating models of irreversible land change with two land classes. These models are suitable for simulating a variety of phenomena that spread in space. We present a method to partition the study into near and far strata based on proximity to initial points of change. We define a proximity model based on this partition. The proximity model, which predicts all change in the near stratum and no change in the far stratum, is then used as a baseline for comparison with the land change simulation. We present a technique to simultaneously compare four maps of reference at initial time, reference at final time, simulation at final time, and baseline model output at final time. We demonstrate our methods through an example case of validating a forest insect infestation model, and we show that our methods lead to more detailed information about model performance. Specifically, our methods allow scientists to note where the model’s errors occur, to distinguish near and far errors with respect to initial points of change, and to identify relative strengths and weaknesses of the land change model in comparison with the proximity model.