Combing deep learning and multi-source data to promote subjective perception of ecosystem services in urban landscape

Subjective perception of ecosystem services is an emerging topic to better understand nature-based solutions for human and natural sustainability. Survey-based methods for subjective perception has the difficulty to move their conclusions beyond site-specific applications. Potential data sources for subjective perception exist in many sources such as geo-tagged social media and street-view photos. In this paper, we develop a combined deep-learning, survey, and multi-source data big data approach to study and promote subjective ecosystem service perceptions beyond site-specific applications. Specifically, we use machine learning models trained to predict human perception from a large dataset of images to rate urban landscape photos from social media and street-view maps. The predictors include CNN-engineered photo features, geographic information, survey-based ratings as well as public ratings from social media and street-view maps. The method of this study can be applied to understand subjective perception of ecosystem services for a wide range of urban landscape site. The results contribute to a better understanding of connections between subjective perception and objective evaluation of ecosystem services value for urban landscape so that nature-based solutions can be better implemented for human well-being and sustainability.

Key Words: Deep learning; Multi-source big data; Subjective perception; Ecosystem services; Social media