University College London
Deep Tuning of Urban Ecology.pdf (2.31 MB)

Deep Tuning of Urban Ecology

Download (2.31 MB)
posted on 2023-05-15, 11:27 authored by Sheng-Yang HuangSheng-Yang Huang, Enriqueta Llabres-VallsEnriqueta Llabres-Valls, Mochen Jiang, Yuankai Wang, Fei Chen

As revealed in Planetary Urbanisation, wilderness spaces consistently decay due to the unfettered  worldwide spread of the urban fabric. If the end of wildness is inevitable in the planetary urbanisation, how  can we recover the biodiversity and live with local wildlife in the perpetual urbanisation process? The  contemporary use of ecological corridor has raised serious concerns in environmental justice; its design  heavily relies on well-built, explicit knowledge and always confronts the ideology dilemmas with how to  position humans in the making of urban ecology. This paper proposes an integrated AI reasoning  framework for the anthropocentric design searching the latent spatial pattern of urban ecology in the  Anthropocene epoch. Our proposal combines the connectivity methodology and conditional generative  adversarial network (cGAN) to algorithmically enable the cross-domain reasoning between wildness and  humanity and to create design knowledge without determined rules and specific data of wildlife. The  proposed framework consists of three components: 1) connectivity modelling, 2) progressive reasoning  and 3) parametric adaptation. It is tested with a design exercise on a 1km*1km site in East London: The  human connectivity and wildlife connectivity of one hundred selected reference locations are modelled to  train a CycleGAN to suggest the wildlife connectivity based on the human connectivity of the site, and  another CycleGAN turns it into a materiality reference. With the reference, parametric prototypes are  adopted to adapt the extant urban landscape to the suggested condition in a parametric approach. The  study emphasises on how the framework leads to an Anthropocentric ‘sweet spot’ of urban ecology tuned  under the limited availability of local wildlife data instead of developing a design work fully. The result  suggests the creation of the discrete urban ecological field can be led by an eco-intensity-graded, mass?customisable framework.