How Reliable is a Map Generated from GIS-Based Modelling? An Analysis from a Model Validation Perspective

Abstract :
The development of a system or predictive model is in essence carried out based on scientific theory, which is aimed at mimicking the real system or process. Geographical Information Systems (GIS) have to date been popularly used for modeling purposes particularly on a spatial context. However, limited number of model-based studies gives a comprehensive validation exercise that could describes uncertainty. It is thus always necessary, particularly in complex GIS modeling, that the model built for explaining, understanding or performing performance of a system of reality be tested for its validity. The purpose of this study is to develop a methodology for validating and evaluating the outputs of land suitability evaluation (LSE) based on fuzzy decision approaches. Four different methods were employed: (i) sensitivity analysis; (ii) comparison against well-established systems of land suitability classification; (iii) comparison with expert judgments; and (iv) comparison against crop yields. Two sets of GIS-based modeling outputs were used. Different statistical methods were employed to deal with spatial comparison between two sets of information (i.e., outputs of the model and the performance standards representing a real system). The analyses undertaken here show how the relationships between two sets of data can be examined and reviewed to reveal the reliability of a map generated. Surprisingly, it was also found that land areas with high suitability classes commonly have less variation (or vice versa), so that that it illustrates the significance of the model for micro-mapping.