Objective Planting design is a non-linear process that must simultaneously consider the scientific configuration of plant communities and the artistic and practical aspects of spatial creation. The diverse requirements in this field make traditional planting design rely heavily on the experience accumulation of designers, resulting in a certain degree of subjectivity. Computational design organizes the elements of planting design through algorithms, potentially replacing human cognitive process with machine. This presents new opportunities for intelligent planting design. Therefore, this research aims to summarize the research progress of computational design in the context of planting design.

Methods This research utilizes Web of Science (WoS) and China National Knowledge Infrastructure (CNKI) as data sources, and employs both bibliometric analysis and narrative review methodologies to statistically analyze and synthesize the literature collected. 1) Bibliometric analysis: By virtue of COOC 14.9 and Citespace 6.3.R2, visual knowledge maps of 330 publications in the landscape architecture field are constructed with a focus on development stages and emergence timelines, based on which the trends and dynamics of computational design in landscape architecture are elucidated, with a specific emphasis on planting design. 2) Narrative review: Of the 330 publications, by reviewing their titles and abstracts, 24 ones relevant to planting and computational design are selected. These publications are then categorized and analyzed based on computational design techniques, research directions, and specific research objects within planting design.

Results Between 2009 and 2023, the number of research on computational design in planting design has shown an overall upward trend. In terms of research objects, urban parks serve as the most significant research object with the largest number of publications, with recent trends extending to urban forests. Avenue trees and vertical greening also show significant potential. Regarding research directions, plant community configuration is a hot topic, followed by plant spatial layout. The research on planting direction still needs further enhancement. Computational design in planting design has progressed through three stages, namely parametric design, algorithmic design, and generative design. 1) Parametric design emphasizes the regulation of quantifiable parameters, aiming to deal with the complex qualitative characteristics of planting design on a one-to-one basis. It is the earliest and most widely used and mature design method in planting design, commonly applied in plant community configurations of urban parks. 2) Algorithmic design focuses on rule formulation, commonly using cellular automata (CA) and multi-agent system (MAS). CA simulates the “dynamic system” of natural communities and is adept at predicting competitive relationships within plant communities, which is widely used in natural plant community configurations in urban forests. MAS, suitable for interactions involving different goals and proprietary information, has significant implications for collaborative design among designers, the public, and the government, and has been applied to the spatial layout of avenue trees. 3) Generative design possesses strong rule-learning and implicit rule-capturing capabilities, with the potential for automated design generation. Algorithms like deep neural networks and hybrid intelligent systems are frequently used in planting design. Generative adversarial networks (GANs), such as Pix2Pix, CycleGAN, and StyleGAN, excel in style transfer, can generate images similar to input data, and are mainly used in plantscape layout in urban parks. For the aforesaid design methods, although the current focus is on their suitability for planting design, rapid development suggests they may become the primary technical approach for planting design in the future. By integrating multiple intelligent algorithms, hybrid intelligent systems can overcome the limitations of individual algorithms and combine the advantages of various techniques, which are currently applied in the integration of cellular automata and GANs.

Conclusion Computational design can effectively address the issues of ambiguity and high dimensionality in planting design, but still requires improvements. Presently, planting design lacks a data sharing platform, with various plant database resources dispersed across multiple platforms, leading to inconsistent data quality. Given the high demands of computational design for data quantity and quality, there is a need to promote the construction of planting design data sharing platforms and provide data evaluation channels to filter diverse and high-quality data. Moreover, planting design is highly related to human subjective aesthetics. Existing research primarily uses quantifiable spatial factors, lacking qualitative factors such as aesthetics and culture. Although generative design can mimic the design capabilities of designers, current generative results are unstable and lack detail. Future research needs to distinguish between quantitative and qualitative factors, with repetitive and quantifiable tasks handled by computational design, while qualitative factors judged by designers’ experience, thus forming a collaborative workflow between designer and computer. Additionally, most planting design research uses single techniques, failing to address the complexity of planting design practice comprehensively. Hybrid intelligent systems capable of integrating multiple intelligent methods, offer new ideas for planting design. Finally, computational design in planting design remains primarily at the experimental and research stages, necessitating further integration with practical application in the future.