Human-machine collaboration in landscape architecture:Multi-stage integration of ai into landscape site analysisworkflow

Abstract

The 21st century's multifaceted problems such as rapidly increasing population, climate crisis, and uncontrolled urbanization have necessitated the acceleration of the production of functional solutions in landscape architecture, going beyond mere aesthetics. This situation underscores the need to work with more precise data sets in the planning process and the importance of interdisciplinary approaches. Within today's dynamic urbanization practices, landscape architecture plays a critical role in maintaining ecosystem services and increasing urban resilience. However, the speed required to produce solutions to these complex problems often strains the manual capacity of traditional design and analysis methods, forcing designers to choose between the need for up-to-date data and limited time. At this point, data-driven design approaches are no longer just technical support, but a fundamental paradigm shift for producing rational and ecologically consistent solutions. Especially in the last decade, remote sensing in obtaining geographic data, artificial intelligence (AI) tools in architectural visualization and modeling, cloud-based systems in automation, and satellite data in disaster prevention strategies have been used more frequently. The integration of these techniques with AI has led to revolutionary transformations in site analysis processes, which are the most critical stage of the design process. The time lost with traditional methods is compensated for by machine learning (ML), facilitating the decision-making process. This contributes to increased productivity and improved built environment quality in today's frequently encountered emergency and disaster scenarios. However, a review of existing academic literature and professional practice reveals a significant technical barrier to the use of these technologies. Maximizing the effectiveness of AI tools requires expertise and programming knowledge. This thesis aims to overcome these technical barriers by proposing a new workflow that allows designers without professional coding training to benefit from the power of artificial intelligence in a transparent and auditable way. The fundamental philosophy of the research is to position artificial intelligence not as an automation mechanism replacing humans, but as a 'collaborator' that supports human creative intelligence and intuitive capacity. At the same time, the thesis examines both the advantages and limitations of using AI-assisted tools in the context of landscape architecture, while also addressing ethical issues. This alternative Human-in-the-Loop (HITL) approach is practical and repeatable, especially for landscape architects, and encourages the effective use of programming and AI tools in education and professional practice. The HITL model is based on the continuous optimization of AI-generated data through user feedback. The AI-assisted coding (vibe-coding) concept, discussed within the theoretical framework of this study, is a method that enables the generation of functional code blocks via AI according to the user's needs. Thanks to this method, integration between complex tools such as Python libraries and ArcGIS can be achieved without requiring technical software expertise from the user. The use of natural language processing capabilities in this way reduces the cognitive burden created by technical details in the analysis process, allowing for a focus on interpreting spatial data and making strategic decisions. The fundamental element defining the original value of this study and its departure from existing approaches in the literature is its conceptualization of AI not merely as a producer of end products in landscape architecture practice, but as a proposal for an environment that radically transforms the designer's relationship with spatial data. While most existing research defines processes requiring advanced data analysis and coding as a field of technical expertise, this study creates a space for technical democratization in professional practice by offering a methodology that overcomes complex software barriers. Unlike autonomous systems where the process cannot be controlled, in the proposed model, the landscape architect's intuition and knowledge are preserved as a controlling filter at every step of the analysis process. Furthermore, testing the proposed model on three sites with diverse geographical and urban characteristics proves that the method has universal validity, independent of scale and location, and thus distinguishes it from studies based on individual samples. The research has a multi-layered structure consisting of literature review, tool comparison, and application phases. This study, conducted in 2024-2025, clarified the breaking points at the intersection of AI and landscape architecture and related technical terms; tested the visual generation capabilities of commonly used generative AI (GenAI) tools such as Midjourney, ArchiVinci, and ChatGPT; and compared the production performance and data analysis consistency of various tools (Python, ArcGIS, and web-based GIS platforms). Multi-layered analyses were performed using vibe-coding, and the results were automatically visualized (e.g. plant health, building function, land use analyses). To test the validity of the proposed workflow, three significant urban open spaces with different characteristics, climates, and urban textures were selected as examples from around the world. The research steps were carried out for Izmir Kültürpark, New York Central Park, and Singapore Gardens by the Bay. These sites were chosen as the focus of the research due to their biodiversity and multi-purpose uses, as well as their data density and the different ecological-urban problems they represent. In all three sites, it was observed that AI-assisted analyses reduced data processing times, which could take weeks with traditional methods, to minutes and minimized the margin of error. The aim is to enable landscape architects to focus on interpreting the obtained spatial data in a practical way, rather than getting lost in complex details. Data obtained during the implementation process revealed that incorporating AI into the landscape architecture workflow dramatically increased production efficiency. However, the discussion also highlighted the potential for misleading results when combined with data that has not been critically examined by the landscape architect. Algorithmic biases, data privacy, and the ethical implications of AI-generated results were among the critical questions addressed in the research. It was emphasized that the obtained outputs should always be validated against the local context and ecological realities. This situation demonstrates that technology is not replacing the landscape architect, but rather taking their data processing capabilities to the next level, enabling them to conduct more in-depth analyses. It is predicted that landscape architecture practice will become more democratic and data-driven by overcoming technical barriers and adopting approaches such as 'vibe-coding'. The HITL-based workflow proposed in this research not only offers an academic model but also serves as a practical roadmap that can be used in professional practice and educational curricula. Future projections emphasize that integrating AI into the decision-making mechanisms of local governments and using real-time ecological analyses in planning processes will make it possible to create more socially inclusive cities that are more resilient to physical change. This study demonstrates, within a scientific framework, that technology can be used not merely as a technical necessity, but as a strategic lever to create a more livable and sustainable built environment.