A Machine Learning Evaluation Method For Sustainable Urban Landscape Designs
In the Anthropocene, the urban population continues to grow on account of natural habitats 1. Urban sprawl, densification, and development on open land contribute to the ongoing degradation of existing ecosystem services in cities, which due to climate changes, result in severe urban ailments such as increased vulnerability to hurricanes, flooding, wildfires, and heatwaves 2. Sustainable development of urban landscapes that enhance ecologies and strengthen communities’ resiliency is considered one of the significant ways to fight climate change and increase urban dwellers’ wellbeing 3. Recent technological advances make computational sustainability research, which promotes automation, analysis, and visualization of sustainability, an emerging field of research 4,5. Computational sustainability evaluation tools were developed to enhance the implementation of sustainable design in architectural models. Nevertheless, although significant work has been undertaken in this area, little work covers early design phase evaluation of urban landscapes and socio-ecological sustainability aspects of the design. The study aims to propose computational sustainability evaluations method for urban landscapes in early design phases using machine learning (artificial neural network [ANN] and transfer learning) methods 6. The method rates the sustainability performance based on image detection and analysis of the design. In our experiment, we used a 750 sustainability-professional evaluations dataset of neighborhood designs. The evaluations rated landscape sustainability performance in three categories: Ecology: vegetation- patch connectivity and patch diversity, Resilience: heat island mitigation and stormwater management, and Social: exposure to nature. Each project was represented with one rendered plan followed by the five parameters ranking them from one to seven. Using the data collected from the experts, we then explored whether the ANN algorithm can rate projects as they are being designed based on the same five parameters in the dataset. Early results demonstrate that the test sample achieved high prediction rates for the varied sustainability aspects of the design. These findings support our hypothesis that ANN machine learning tools can be instrumental for improving sustainability-conscious design.
Keywords: Sdg9, Sdg11, Sdg13, Sdg15, Urban Design, Landscape Architecture, Sustainability, Machine Learning, Artificial Neural Network, Early Design, Design Evaluation