Can empirically based model results be fed into mathematical models? MCE for neural network and logistic regression in tourism landscape planning

A wide variety of mathematical and empirical models have been implemented as practical tools for land-use planning, and multilayer perceptron (MLP), logistic regression or LR (mathematical model) and multi-criteria evaluation or MCE (empirical) are among widely applied models. One of the main drawbacks of the mathematical models is that they require dependent data and the process of data collection can be so costly and time-consuming for large areas. As such, we investigated the possibility of providing dependent data set through the MCE method for tourism planning in Golestan Province, Iran. The accuracy of MCE-based algorithms was investigated using ground truth data collected during field observations from early spring up to late summer 2016. The MCE-based and ground-based outputs were investigated and compared for spatial accuracy and connectivity and compactness of the results using receiving operator characteristic (ROC) and landscape configuration metrics. ROC statistics were scored at 0.886, 0.834, 0.82 and 0.814 for ground-based MLP, ground-based LR, MCE-based MLP and MCE-based LR, respectively, showing no meaningful differences between MCE-based and ground-based methods in terms of spatial accuracy. Landscape metrics also indicated that MCE-based methods have resulted in a more connected and manageable pattern for tourism planning. According to the results of this study, MCE can serve as a preliminary approach to define field sampling spots or even as an alternative to field observation efforts in case of limited time and financial resources.