Heritage–tourism coordination in Chinese cities is highly scale-sensitive

A study published in npj Heritage Science examines how historic buildings and tourism development align across China’s historical and cultural cities, and its central message is one of caution: the relationship between built heritage and tourism is highly sensitive to how it is measured, and it resists confident prediction. Jing Fan and Maodong Tian screened 143 candidate city-level units and retained 127 for analysis, drawing on 1,176 historic-building records alongside data on domestic and international tourism. Their toolkit combined spatial analysis, a coupling coordination degree (CCD) model, repeated nested cross-validation, province-grouped validation, and SHAP interpretation of held-out Random Forest predictions.

The authors found that the resulting classifications depend heavily on the spatial unit used to normalize the data. When CCD values were normalized by administrative area, the classes matched the baseline in 76.4 percent of cases for domestic tourism and 83.5 percent for international tourism. When population was used instead, agreement dropped sharply, to 35.4 percent for domestic and 48.8 percent for international cases. In other words, a city’s apparent heritage–tourism coordination can change substantially simply because of the denominator chosen, a scale sensitivity the authors describe as substantial.

Typology shifts reinforced that caution. Thirty-six cities, 28.3 percent of the sample, moved between high and low resource–tourism categories when the analysis switched between domestic and international markets. A city that appears well coordinated for one market may not be for the other, which limits how far single-market conclusions can be generalized.

Predictive generalization proved weak. Across 100 outer folds of nested cross-validation, the Random Forest model achieved R² = 0.028 ± 0.228 for domestic tourism and R² = 0.065 ± 0.223 for international tourism. Province-grouped validation produced negative R², indicating that models fitted in some provinces did not transfer to others. For a field where local case studies often report strong heritage–tourism relationships, this nationwide test suggests those patterns may not hold as general rules.

SHAP interpretation of held-out predictions offered the most stable, though still limited, signals. Domestically, rankings were most stable for CLCD and DEM, variables reflecting land cover and terrain; internationally, GDP and SHDI, a landscape diversity index, ranked most consistently. The authors stress that these associations support only cautious, non-causal interpretation, since the model’s overall predictive power was minimal.

The study’s contribution lies less in identifying drivers of heritage–tourism coordination than in documenting how fragile such findings are. By testing 1,176 building records across 127 cities with repeated validation, the authors show that coordination classes shift with normalization choices, typologies shift between markets, and machine-learning models fail to generalize across provinces. For researchers and planners, the implication is that heritage–tourism assessments in China’s historical and cultural cities should report scale choices explicitly and avoid causal claims, because the reported coordination between historic buildings and tourism is far from robust to measurement decisions.

Subject of Research: Spatial coupling between historic buildings and tourism development in China's historical and cultural cities

Article Title: Spatial coupling between historic buildings and tourism development in China’s historical and cultural cities

Article References: Fan, J., & Tian, M. (2026). Spatial coupling between historic buildings and tourism development in China’s historical and cultural cities. npj Heritage Science. https://doi.org/10.1038/s40494-026-03004-6

Image Credits: AI Generated

DOI: 10.1038/s40494-026-03004-6

Keywords: historic buildings, tourism development, China, spatial analysis, coupling coordination degree, Random Forest, SHAP, cross-validation, heritage science, scale sensitivity, built heritage, predictive modeling