Integration of urban growth modelling products with image-based urban change analysis

Urban change detection using remotely sensed data has been extensively studied. One current application of detection products is the formulation of calibration data for urban change prediction models. As multi-temporal scenes become available and urban growth prediction models increase in popularity and accuracy, it is natural to envisage a bi-directional relationship where, in addition to detection products assisting prediction models, the prediction information acts as ancillary input to enhance spectral-based change detection products. This closed feedback loop has the potential to significantly increase the accuracy of both detection and prediction efforts. Consequently, our objective is to evaluate the integration of prediction information with spectral data for urban change monitoring. A case study was carried out in the Denver, Colorado metropolitan area. Probabilities of urban change generated from two existing urban prediction models (based on decision trees and logistic regression) are combined as additional information content with a Landsat Thematic Mapper (TM) scene. Detailed assessments at the pixel and block scales are implemented to evaluate urban change classification accuracy using different input data and training sample sizes. Results show that in pixel-based assessments, the fusion of decision tree change probabilities and TM spectral bands with sufficient training samples leads to improvements. In terms of overall accuracies, the improvement is 2.0-2.4%, from 87.3% (spectral-only model) and 87.7% (prediction probability model) to 89.7% for the fused model. Similarly, the corresponding kappa coefficients show increases of 0.07-0.08, from 0.60 for the spectral model and 0.61 for the urban prediction model to 0.68 for the fused model. Accuracies aggregated at the block scale present an approximate 2.1-4.3% increase when the fusion-based model is employed compared with the exclusive use of either spectral or prediction probability data, namely 87.6% (fused) vs 83.4% (spectral) and 85.7% (prediction). It is also important to state that the standard deviation of accuracies between blocks is significantly reduced by more than 3% (11.5% vs 14.9% and 14.7%), suggesting higher consistency in classification performance. This is a desirable attribute for subsequent use of these products, for example by the urban planning community. Statistical tests at the block scale also demonstrate that such improvements are significant. It is also observed that to receive the integration benefits, the remote sensing classifier needs a large but reasonable training data set size, while the prediction model should be based on advanced modelling methods. Further assessments on block accuracy with respect to urbanization conditions (i.e. urban presence and change sizes) indicate the ability of the fusion to address spectral limitations, especially in blocks with high relative change. These initial results encourage the expansion of spectral/prediction data fusion to other sites, modelling techniques, and input data.