Statistical Methods: Landscape Pseudo-History Analysis (LPHA)

From time to time scientists do not have a clear picture or a strong theory in relation to the processes or mechanisms under investigation. This problem becomes more prevalent at the current era of big data—data are explosively available, while techniques or metrics are still limited. As a result, there are no guidelines to guide scientists' choice of statistical models or sampling strategies.

The CHES group has been developing a new methodology called landscape pseudo-history analysis. This methodology employs agent-based models (ABMs) as data (e.g., space time data or longitudinal/time series spatial data) generators under a set of predetermined rules. Doing so eliminates darkness in the processes or mechanisms under investigation as the modeler has full understanding of the relevant ABMs and thus (nearly) complete control over the system of study. Then, modelers can apply a set of candidate statistical models or sampling strategies to analyze such data, and find out the one(s) that best uncover the preset rules.

The approach is able to offer new insights into the application domain of statistical methods and sampling strategies. One application is to study the similarities and differences in the development trajectories of San Diego and Tijuana over the past 60 years (Ninghua Wang dissertation). Below are a couple of seminal articles in this regard.

Readings and References:

An, L., D. G. Brown, S. E. Page, and W. Rand (2005). What statistical models can better detect land-change mechanisms? The 2005 GeoComputation Conference. August 1-3. Ann Arbor, Michigan, USA.

Wang, N., D.G. Brown, L. An, S. Yang, and A. Ligmann-Zielinska (2013). Comparative performance of logistic regression and survival analysis for detecting spatial predictors of land-use change. International Journal of Geographic Information Science 27(10):1960-1982 (10.1080:1-23).