AI+ Landscapes: Deep Learning in the Generation and Analysis of Landscape Form

From the 1960s, the field of landscape architecture has transitioned from focusing on static forms to dynamic processes and from determinism to indeterminism, influenced by systems thinking and ecological principles. Traditionally, complex systems have been approached using computational design tools that favor simplicity and reductionism. However, recent advancements in Artificial Intelligence (AI) have introduced new possibilities for managing this complexity. AI—machines exhibiting intelligence through perception, synthesis, and inference—has begun to revolutionize various fields, including landscape architecture. While AI-driven methods remain in their early stages within landscape architecture, they offer potential for fostering creative design expressions, evaluating complex system performance more holistically, and developing alternative design strategies. Although landscape architects are not expected to have the same expertise in AI as engineers and computer scientists, it is essential for those aiming to produce ambitious designs responsibly to understand how different AI tools function.
This dissertation focuses on the concepts and applications of AI, specifically Deep Learning (DL), within landscape architecture. The primary question guiding this research is: How will AI affect the practice of landscape design? The study explores the untapped potential of AI technologies in design workflows, considering how they can be integrated seamlessly into the process while acknowledging their limitations and inherent biases.
On the conceptual level, this study examines how AI tools contribute to the formation and development of landscape forms. It draws a parallel between the nature of Deep Learning Algorithms—how they handle big data, images, time, simulation, and meanings—and landscape architecture’s core principles, which are data-driven, cognitive, evolutionary, dynamic, and deeply contextualized. On the application level, the dissertation explores how AI can be used as a tool for evidence-based design, making it more scientific and automated. AI's ability to handle large-scale, multi-modal data provides unprecedented accuracy, making it ideal for studying complex and unknown systems. On the methodological level, the dissertation considers AI as a new way of seeing and reimagining landscape design. Unlike rule-based methods, AI is inherently qualitative, indeterminate, and less predictable, which lends it a sense of creativity and innovation. The research considers how these qualities of AI could inspire new approaches to landscape design. By embracing AI's unpredictability, landscape architects may bring fresh insights to designing within complex, living systems.
The dissertation is organized into three parts:
Part I situates the theoretical framework and explores the history and theories of computational design in landscape architecture, examining methodological shifts and potentials introduced by DL across various phases of research and design.
Part II reviews various deep learning algorithms, including Artificial Neural Networks (ANN), Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), Generative Adversarial Networks (GAN), and Large Language Models (LLM). Each algorithm corresponds to a core disciplinary area in landscape architecture (data, imaging, time, simulation, form). The chapters in this section explore why these issues are important in landscape architecture, the specific potential of DL methods, how they function, and their applications in landscape analysis and design.
In Part III, the author presents three projects focused on streetscape, vegetal systems, and design ideation. These projects showcase the practical application of DL methods across different design tasks.
Overall, this dissertation demonstrates how DL can enrich landscape architecture by offering new methods for analysis, design, and implementation. It advocates for an integrated and innovative approach to landscape design. The findings suggest that DL enhances both the precision and creativity of landscape design while providing a foundation for future research and development in the field.