Performance evaluation of multiple methods for landscape aesthetic suitability mapping: A comparative study between Multi-Criteria Evaluation, Logistic Regression and Multi-Layer Perceptron neural network

Abstract

Landscape aesthetics, as a cultural ecosystem service should be included in land-use planning. Therefore, appropriate mapping algorithms that allow quick and accurate visualization of the scenic beauty in a spatially-explicit manner are of significant importance. The present study implements and compares three mapping approaches including Multi-Criteria Evaluation (MCE), Logistic Regression (LR) and Multi-layer Perceptron (MLP) neural network in a GIS environment for landscape aesthetic suitability mapping in the Ziarat watershed basin of northeastern Iran. Ground truth data were collected during several field observations and landscape photographs were taken in winter and autumn. Mapping algorithms were compared for their spatial accuracy using the Receiving Operator Characteristic (ROC) method and the comparison was made for automatic identification of scenic beauty on routes applying landscape metrics. According to the results, the ROC statistic scored at 0.94, 0.93 and 0.88 for MLP, LR and MCE methods, respectively. In addition, landscape metrics-derived results depicted the MLP method as more successful for automated delineation of a connected network of scenic routes. Finally, due to acceptable spatial accuracy, this study suggests expert-based mapping methods such as MCE and statistical algorithms such as LR can be used as ground truth layers for a sampling of presence/absence data. The map containing sampled points can be used as a training layer for iterative artificial intelligence-based methods such as MLP for quick and accurate suitability mapping of landscape aesthetics in neighboring watersheds. This application demonstrates how landscape aesthetics as one of cultural ecosystem services can be integrated into land-use planning practices.