Speaker
Description
Introduction and research problem
Digital planning increasingly relies on GIS, machine learning, and artificial intelligence to transform complex environmental and spatial information into classifications, indicators, weights, and spatial models. These techniques strengthen the operational capacity of planning by producing apparently precise maps, scores, and decision rules. Yet numerical precision does not necessarily imply epistemic certainty. This tension becomes particularly important when digital methods are applied to culturally embedded landscape knowledge, whose meanings are situated, interpretive, and often resistant to direct measurement. In such contexts, knowledge must pass through a sequence of choices—what is selected as evidence, how it is classified, which qualities are formalized, and how those qualities are weighted—before it can enter a computational planning workflow.
This paper asks: how is uncertainty produced, transformed, and obscured when cultural knowledge is translated into digital planning evidence? It builds on a previously developed shanshui-informed digital rural planning framework in which traditional Chinese landscape paintings were treated as visual proxies for cultural-spatial cognition and translated into GIS-recognizable parameters for rural planning (Luo & Li, 2026). The earlier study demonstrated that such a framework could support ecological connectivity, landscape-visual quality, and service accessibility. The present study shifts attention from whether the model can operate to how knowledge becomes model-ready in the first place. Rather than treating uncertainty only as missing data or model error, it examines uncertainty as a consequence of knowledge translation within model-mediated planning.
Conceptual or theoretical framework
Planning theory has long recognized uncertainty as a constitutive condition of planning rather than a temporary information deficit. Christensen (1985) showed that uncertainty may concern both means and ends, requiring learning, bargaining, and judgement rather than the simple application of established knowledge. This concern becomes more complex in data-intensive planning environments, where models do not merely process evidence but also participate in defining what can count as evidence. Davoudi (2006, 2015) similarly challenges a linear view in which objective knowledge is produced first and then straightforwardly applied to planning action, emphasizing instead the situated and practical character of knowing.
Building on these perspectives, this study conceptualizes cultural digitalization as a process of knowledge translation. Cultural knowledge is first represented through selected materials, then classified and interpreted, formalized into measurable proxies, quantified through parameters and weights, and finally mobilized as planning evidence. Each transition can preserve some meanings while suppressing or simplifying others. Bowker and Star (1999) demonstrate that classification systems are not neutral containers of information: categories organize visibility and stabilize particular ways of knowing. Porter (1995) likewise shows that quantification can confer an appearance of impersonality and objectivity on decisions whose underlying assumptions remain socially and institutionally produced. More recent work on smart and data-driven urbanism further warns that computational forms of knowing can narrow the range of knowledge recognized in planning unless they remain open to situated and non-digital forms of knowledge (Kitchin, 2016; Nochta et al., 2021; Cook & Karvonen, 2024).
On this basis, the paper advances two propositions. First, digital modelling relocates human judgement rather than eliminating it. Second, quantification can make cultural knowledge more operational while simultaneously making the assumptions behind its translation less visible.
Methods
The study conducts a retrospective methodological audit of the previously developed shanshui-informed digital planning framework. The earlier workflow used Wufeng Village in Shaanxi Province, China, as an empirical case and translated traditional shanshui knowledge through several interconnected stages. A corpus of 112 high-resolution rural-settlement image samples was manually compiled from 35 Song-dynasty landscape paintings. CLIP was used to derive high-dimensional visual-semantic feature vectors, followed by HDBSCAN clustering. Because automated clusters did not constitute culturally meaningful planning categories by themselves, the results were manually calibrated and semantically labelled against site landscape characteristics and development conditions, producing six site-oriented pictorial archetypes (Luo & Li, 2026).
The audit does not rerun the GIS model or test alternative weights. Instead, it treats the completed workflow as an object of methodological analysis and traces the knowledge transformations embedded in it. For each transition, the analysis asks what form of knowledge entered the step, what was changed through the translation, where machine inference or human judgement intervened, and what assumptions became embedded in the resulting planning evidence. Particular attention is given to the transition from cultural concepts such as the Six Distances, gazing–traveling–wandering–dwelling, liubai (void), and host–guest hierarchy into GIS-compatible indicators, including viewshed, terrain relief, path connectivity, service accessibility, ecological buffers, elevation contrast, thresholds, and weighted suitability criteria.
Results or expected results
The audit identifies three interrelated forms of uncertainty.
Representational uncertainty arises before computation begins. The model does not directly process “shanshui culture”; it processes selected landscape paintings and selected image fragments that are treated as proxies for cultural-spatial knowledge. The choice of Song-dynasty paintings, the prioritization of images containing dwellings, terraced fields, forests, and other site-relevant elements, and the extraction of particular image samples already establish a boundary around what is considered representative. The dataset is therefore not a transparent container of cultural knowledge but the outcome of historically situated media and research choices.
Interpretive uncertainty emerges in the relationship between algorithmic classification and human judgement. CLIP and HDBSCAN identify visual-semantic similarities and clusters, but the six final shanshui archetypes were produced only after human calibration and semantic labelling in relation to Wufeng’s existing landscape character and planned development potential. The resulting categories are therefore hybrid epistemic objects: they are co-produced through machine pattern recognition, cultural interpretation, and site-specific planning judgement. This reveals that digital modelling does not remove interpretation from planning; it redistributes interpretation across less visible stages such as sample selection, cluster validation, semantic naming, and model calibration.
Formalization uncertainty occurs when culturally rich concepts are converted into measurable proxies. In the original framework, the Six Distances were operationalized through viewshed analysis, terrain relief, view-corridor width, and height control; gazing–traveling–wandering–dwelling was linked to path connectivity, viewpoint density, ten-minute service accessibility, and residential clustering; liubai was translated through ecological buffers, non-development land, and distance to water; and host–guest relations were represented through elevation contrast and visual dominance (Luo & Li, 2026). These correspondences are analytically productive, but they are better understood as operational equivalences than as literal translations. There is no natural one-to-one identity between liubai and buffer width, or between “habitable” cultural ideals and ten-minute accessibility. Their value lies in making aspects of cultural meaning actionable within planning, while their limitation lies in the risk that interpretive choices become naturalized once expressed as indicators.
The same issue intensifies when indicators become numerical weights and thresholds. Values such as 0.20, 0.30, 0.45, or 0.50 appear exact and comparable, yet their precision rests on prior decisions about category construction, proxy selection, planning priorities, and site interpretation. Numerical precision can therefore conceal epistemic uncertainty rather than resolve it. The audit suggests that uncertainty is not removed by the progression from culture to data; instead, it changes form, moving from explicit interpretive judgement into increasingly technical components of the model.
Discussion and conclusion
The findings do not invalidate the use of cultural knowledge in digital planning. The Wufeng case showed that the original framework could organize landscape structure across macro, meso, and micro scales and improve measurable ecological, visual, and accessibility outcomes (Luo & Li, 2026). These results demonstrate operational and performance value. They do not, however, establish that the model has fully measured or reproduced shanshui cultural experience. The original study already acknowledged that GIS evaluates physical prerequisites and geometric armatures rather than directly quantifying yijing, emotional resonance, or the full experiential depth of shanshui.
The paper therefore argues for a shift from eliminating uncertainty to making uncertainty explicit and traceable. Cultural-data workflows should document why particular sources are selected, where human and machine judgements enter, how cultural concepts are mapped onto proxies, and how weights or thresholds are justified. Such transparency would allow assumptions to remain contestable and revisable rather than being hidden behind the apparent objectivity of model outputs. In model-mediated planning, the central question is not simply whether a model produces precise results, but whether planners and affected communities can understand how those results were constructed.
For culturally informed digital planning, this implies a more modest but potentially more robust role for computation. Models can serve as transparent decision-support systems that organize evidence, reveal spatial relationships, and test planning propositions while leaving room for interpretation, local knowledge, and revision. Uncertainty, in this sense, is not a failure to be engineered away; it is a condition that must remain visible if culturally complex knowledge is to enter digital planning without being reduced to a false certainty.
References
Bowker, G. C., & Star, S. L. (1999). Sorting things out: Classification and its consequences. The MIT Press. https://doi.org/10.7551/mitpress/6352.001.0001
Christensen, K. S. (1985). Coping with uncertainty in planning. Journal of the American Planning Association, 51(1), 63–73. https://doi.org/10.1080/01944368508976801
Cook, M., & Karvonen, A. (2024). Urban planning and the knowledge politics of the smart city. Urban Studies, 61(2), 370–382. https://doi.org/10.1177/00420980231177688
Davoudi, S. (2006). Evidence-based planning: Rhetoric and reality. disP – The Planning Review, 42(165), 14–24. https://doi.org/10.1080/02513625.2006.10556951
Davoudi, S. (2015). Planning as practice of knowing. Planning Theory, 14(3), 316–331. https://doi.org/10.1177/1473095215575919
Kitchin, R. (2016). The ethics of smart cities and urban science. Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences, 374(2083), 20160115. https://doi.org/10.1098/rsta.2016.0115
Luo, T., & Li, Y. (2026). Shanshui-informed digital rural planning: Modular aesthetic–spatial mapping with GIS ecological analytics. Journal of Digital Landscape Architecture, 11, 739–750. https://doi.org/10.14627/537770070
Nochta, T., Wahby, N., & Schooling, J. M. (2021). Knowledge politics in the smart city: A case study of strategic urban planning in Cambridge, UK. Data & Policy, 3, e31. https://doi.org/10.1017/dap.2021.28
Porter, T. M. (1995). Trust in numbers: The pursuit of objectivity in science and public life. Princeton University Press.