Developing Bayesian Information Entropy-based Techniques for Spatially Explicit Model Assessment

TL;DR

This paper develops advanced Bayesian information entropy techniques for spatially explicit land use model assessment, enabling better understanding of model accuracy, uncertainty, and information dynamics across different spatial scales.

Contribution

It introduces novel spatial Bayesian assessment methods and metrics for analyzing land use models, emphasizing higher-order information and uncertainty quantification.

Findings

Enhanced understanding of model fit to landscape transitions

Estimated theoretical accuracy under incomplete information

Insights into spatial uncertainty patterns

Abstract

The aim of this paper is to explore and develop advanced spatial Bayesian assessment methods and techniques for land use modeling. The paper provides a comprehensive guide for assessing additional informational entropy value of model predictions at the spatially explicit domain of knowledge, and proposes a few alternative metrics and indicators for extracting higher-order information dynamics from simulation tournaments. A seven-county study area in South-Eastern Wisconsin (SEWI) has been used to simulate and assess the accuracy of historical land use changes (1963-1990) using artificial neural network simulations of the Land Transformation Model (LTM). The use of the analysis and the performance of the metrics helps: (a) understand and learn how well the model runs fits to different combinations of presence and absence of transitions in a landscape, not simply how well the model fits our given data; (b) derive (estimate) a theoretical accuracy that we would expect a model to assess under the presence of incomplete information and measurement; (c) understand the spatially explicit role and patterns of uncertainty in simulations and model estimations, by comparing results across simulation runs; (d) compare the significance or estimation contribution of transitional presence and absence (change versus no change) to model performance, and the contribution of the spatial drivers and variables to the explanatory value of our model; and (e) compare measurements of informational uncertainty at different scales of spatial resolution.