Agent-based model validation using Bayesian networks and vector spatial data

Abstract

Validation of agent-based models (ABMs) of land-use change is a significant challenge in current spatial-modelling research and application. During the validation process, model performance and accuracy assessment depend mostly on pixel-by-pixel comparisons. However, in urban land-use planning problems the use of vector spatial data to develop ABMs is becoming more necessary. Hence, improved and robust validation approaches are required for vector-based ABMs. This study presents a novel validation approach for an ABM using vector-based geographic information system and Bayesian networks. The approach creates a unique-polygons map and an object-oriented database. Three indicator variables are calculated to assess the probability of agreement. The indicator variables are nodes in a Bayesian network that is used to evaluate the final agreement of each unique polygon. Further, an index of overall agreement is calculated. The approach was applied to a simulation outcome map generated by an existing Bayesian network-based agent-system (BNAS) model. The BNAS model simulation of land-use change for the year 2001 was compared with the actual land-use change for the same year using the proposed validation approach. The results obtained indicate that significant agreement between the maps was achieved. The approach is well suited for validating vector-based ABMs and can be used as an aid in model designs for improved model performance.

An agent-integrated irregular automata model of urban land-use dynamics

“…Although the raster data model is computationally more efficient than the vector counterpart, it is less effective in capturing the geographic and geometric details of the urban objects (Benenson and Torrens 2004). The grid cells of equal size and regular shape are incongruous to irregular land features such as city blocks, subdivisions, and parcels (Kocabas andDragicevic 2009, Crooks 2010). In order to address the limitation, researchers developed irregular CA models by modifying the basic CA assumption of squared-grid space to irregular geometries (e.g.…”

“…Dahal and T. Edwin Chow accuracy surface map is an effective technique of error assessment. One way to visualize the errors is to create an agreement probability surface map as illustrated by Kocabas and Dragicevic (2009). Another way is to create a surface map with categorical data in terms of agreement classes, where each and every developed site in the predicted map is assigned to a class to indicate its degree of spatial agreement with the developed site in the reference map.…”

“Space, the Final Frontier”: How Good are Agent-Based Models at Simulating Individuals and Space in Cities?

“…Here results are validated by comparison at the smallest unit (normally pixel-by-pixel). However, this form of analysis is restricted to raster formats of data and as evidenced above, validation of models using vector data is less developed [80]. While much work has been carried out on validation with respect to the similarity of model outputs to macro structures, little attention has been paid to those models that exhibit not only quantitative agreement with empirical macro structures but also quantitative agreement with empirical micro-structures (i.e., the agents being modeled [48]).…”

Integrating spatial relations into case-based reasoning to solve geographic problems

“…The BNs have been widely used to study land use classification and land use change (e.g. [32,31,5,22,33,23]). The BNs essentially are a probabilistic graphical model that represents the integration of conditional probability functions corresponding to a series of nodes into a joint probability distribution function [3].…”