风景园林活力认知及其当代转型

Prospect of Landscape Architecture in the Era of Artificial Intelligence: A Profound Portrait of Intelligent Landscape

关键词:  风景园林  智慧景观  人工智能  深度计算  生态感知  数字孪生

Key words:  landscape architecture  intelligent landscape  artificial intelligence  DEEP-computing  ecological approach to perception  digital twin

摘要:

面对人工智能技术带来的变革，风景园林学科因其对象的复杂性、动态性与人文性，亟须构建区别于建筑及城市规划的独特智能化范式。旨在提出并系 统构建“智慧景观”的理论框架，将智慧景观定义为一个能动响应、自我优化的虚实融合“活系统”；并为其构建了“数字空间基底→数字语义主线→数字孪生模 拟→数字情景输出”的四模块闭环运作架构。该架构由风景园林空间艺术思维与人工智能计算思维深度融合的“人机协同”机制所驱动，实现了从数据整合、规则 生成、动态评估到沉浸体验的完整技术路径。本框架旨在为风景园林的智能化转型提供兼具理论深度与实践指导意义的系统性方案，推动学科向数据驱动、智能决 策与人本关怀深度融合的新阶段迈进

Abstract:

As artificial intelligence (AI) matures, landscape architecture faces a dual challenge: the biophysical and cultural complexity of landscapes, and the fragmentation of current "digital landscape" and "smart city" practices when scaled to landscape systems. This paper advances a distinct paradigm, "Intelligent Landscape", defined as a virtual-physical, proactive, and selfoptimizing living system. We operationalize it through a closed-loop architecture of four interlocking modules: 1) Digital Spatial Substrate, which consolidates geospatial, ecological, sensory, and behavioral data into a modular, machinereadable lexicon; 2) Digital Semantic Thread, which encodes ecological rules and human perceptual values as computable, generative constraints; 3) Digital Twin Simulation, enabling multi-scenario, dynamic performance assessment with feedback; and 4) Digital Scenario Output, providing immersive, interactive experiences that communicate options and capture stakeholder responses. Together these modules form an end-to-end stack from data to decision and back, supporting continuous learning and adaptation. Methodologically, the framework is driven by human - AI collaboration and a DEEP-Computing logic (Digital, Evaluation, Ensemble, Parameter + Computing). Designers' spatial and artistic reasoning is coupled with AI's capacities for pattern discovery, rule induction, and multi-objective optimization. Across concept, synthesis, testing, and interaction phases, site intelligence is built from data-rich diagnostics; perceptual and ecological semantics are jointly parameterized; digital twins quantify micro- to meso-scale effects on climate moderation, hydrology, biodiversity, accessibility, and experiential quality; and immersive outputs elicit feedback that updates models and rules for the next iteration. The Intelligent Landscape paradigm yields five capability dimensions that move beyond static digitization: 1) intelligent sensing creates multi-layered situational awareness across space and time; 2) perceptual quantification treats human experience as a first-class design variable by translating multi-sensory responses and cognitive processes into measurable targets and constraints; 3) process computation shifts evaluation from heuristic post-rationalization to ex-ante, model-based prediction of coupled nature-society dynamics; 4) virtualized space broadens design expression and review via VR/AR, enabling time-of-day, seasonal, and extreme-event stress tests prior to construction; and 5) contextualized decisionmaking supports adaptive operations after delivery, maintaining system vitality under shifting climate and use conditions. Applications illustrate practical value while remaining generalizable. In park operations, fusing real-time flows, microclimate, and facility status allows digital twins to forecast comfort and carrying capacity and to adjust irrigation, shading, and wayfinding dynamically. In regional corridor restoration, multi-scale data reveal ecological breaks and quantify biodiversity gains across planting or connectivity scenarios. In heritage landscapes, high-fidelity twins combined with semantic narratives support preventive conservation and immersive public interpretation while easing onsite pressure. Across these use cases, performance indicators bridge ecological function, user perception, and managerial feasibility, translating trade-offs into transparent choices. The paper's contributions are threefold: conceptually, it reframes Intelligent Landscape as a living, learning system at the landscape scale, distinct from infrastructure-centric smart-city logics and building-centric controls; methodologically, it systematizes a four-module loop that unifies data structures, generative rules, predictive simulation, and participatory visualization into a coherent pipeline; operationally, it shows how human-AI collaboration converts dispersed digital tools into an auditable, iterative, and transferable practice. Looking forward, two priorities merit attention: establishing standardized protocols and datasets for ecological perception, reducing subjectivity while preserving local meaning; and developing integrated, multiobjective evaluation to jointly optimize ecological, experiential, and technical performance under uncertainty. Advancing these fronts will accelerate the discipline's transition toward evidence-rich, intelligence-assisted, and carecentered landscape practice